Safety and Economic Evaluations of Electric Public Buses Based on Driving Behavior

Abstract

The widespread adoption of electric public buses has a positive effect on energy conservation and emission reduction, which is significant for sustainable development. This study aims to assess the safety and economy of electric buses based on drivers’ behavior. To this end, based on the remotely acquired travel data of buses, the driving operation behavior is analyzed, and four safety and four economic characteristic indicators are respectively extracted via safety analysis, correlation examination, and an R² test. Then, the extreme learning machine (ELM) is leveraged to establish the safety evaluation model, and Elman neural network is employed to construct the economic evaluation model. A comparative analysis and a comprehensive evaluation are conducted for the behaviors of ten drivers. The results highlight that the proposed evaluation model that us based on the ELM and Elman neural network algorithm can efficiently distinguish the safety and economy of driving behavior. Furthermore, a graph of driving operation behavior is constructed and the analysis results also manifest that the change of driving operation behavior of buses with higher safety and economy lead to relatively stable characteristics. When the fluctuation of vehicle speed is smooth, the driver can implement moderate driving operation in real-time. One critical conclusion that was revealed through the study is that there exists a certain correlation between driving safety and economy, and buses with higher safety tend to be more economical. This study can provide a theoretical basis for planning the maneuvering and operation of electric buses, including driving speed, braking, and acceleration operations.

1. Introduction

Nowadays, energy consumption and safety in transportation have been widely considered, and drivers play an important role in road traffic safety. The study of driver behavior mechanisms has attracted wide attention [1,2,3]. Previous investigations show that 30% of vehicle fuel economy promotion is attributed to driver behavior [4]. However, the research on the energy-saving driving strategy of pure electric vehicles in China started relatively late, and the research on the energy-saving driving of pure electric buses was less [5]. In addition, there is a significant correlation between driving behavior and road safety [6] and studying the relationship between driving behavior and the nature of accidents can differentiate safe driving from unsafe operations [7]. Driving behavior can be facilitated by providing drivers with guidance on driving behavior [8,9]. At present, the evaluation of driving speed behavior in China is generally carried out through a series of driving behavior subtraction units, such as sharp acceleration subtraction unit, sharp deceleration subtraction unit, sharp turning subtraction unit, etc. The score is based on rapid acceleration and deceleration, the number of sharp turns, etc., data to realize the calculation of a driving behavior score [10]. The evaluation process is highly subjective, and the potential relationship between driving behavior and road safety is rarely studied from the perspective of data mining.

To now, a variety of researchers have been concerned with the evaluation of safety and economy of driver driving behavior. Different methods and parameters have been developed to assess driving behavior and rate and classify drivers based on their driving behavior. The concept of a driver behavior profile (DBP) is introduced and widely referred to in research and practice. A DBP is a comprehensive and standardized measure/rating of driver behavior based on historical driver operation data, including the frequency and magnitude of speeding, rapid acceleration, and rapid braking. By combining DBP and temporal and spatial data, the effectiveness of driving behavior interventions can be properly examined [11]. In addition, driver behavior analysis is often considered as a reference for insurance investment, of which the premiums are determined based on safety and fuel consumption levels that are evaluated from historical driver data. Related studies have shown that insurance premiums or fuel taxes are strongly related with road safety [12,13,14,15,16]. With the development of vehicle networking and information technology, usage-based insurance, or user behavior insurance (UBI) has been widely proposed and is favored by insurers and drivers [17,18]. UBI is priced based on vehicle usage and drivers’ daily driving habits, which are collected through telematics systems [19]. Furthermore, the variables that are related to driving behavior impose stronger correlations with traffic accidents than traditional risk factors and can be referred to promote the rationality of insurance pricing [20].

Optimizing the speed profile of a vehicle during driving is an effective way to reduce energy consumption, and this is also called eco-driving, which has been verified as efficient in reducing fuel consumption and CO₂ emissions [21]. Eco-driving essentially belongs to multidimensional co-operation that includes driving behavior, route selection, and other choices or behaviors that are related to vehicle fuel consumption (such as use of premium fuels, air conditioning, peak hour driving, etc.) [22]. Kurani et al., consider eco-driving as anything that drivers do in a specific vehicle to improve fuel economy [23]. Sanguinetti et al. define 10 eco-driving behaviors [24], which provide a solid basis for developing policies and interventions. Recently, research on driver behavior occurring in electric vehicles has attracted wide attention [25,26,27,28,29]. Through proper information feedback and driver training, eco-driving will facilitate a reduction in fuel consumption and greenhouse gas (GHG) emissions [30]. The most commonly used strategy to promote eco-driving is on-board feedback, which conveys information about fuel efficiency to the driver and can lead to fuel economy improvement by 6.6% [31]. While the total fuel savings that are achieved by operation optimization of vehicle powertrain (including engine and transmission operation) with the most advanced technology is estimated to be about 10–12% [32]. Thus, the driving behavior that is based eco-driving optimization deserves to be investigated more in-depth.

Electric buses have become the main alternatives of urban transport because of their low energy consumption rate, zero emissions, and low noise. Compared with the technical difficulties and research cycles to improve the battery motor performance and energy management strategies, the optimization of driving behavior is an efficient way to reduce the energy consumption during electric bus operation [5]. A variety of researchers have focused on bus entry and exit [5,33,34,35], different urban road scenarios [36], predictive control of energy consumption [37,38,39,40], and energy-efficient driving strategies [41,42,43]. For bus operation, the safety and economy of driving behavior of drivers is critical. The economics of driving behavior is directly related to the economic profit of bus operators. In order to encourage drivers to save energy, “fuel saving awards” are usually suggested, however, there is a lack of an appropriate evaluation method to assess the driving economy. Moreover, without additional monitoring equipment, transit operators are unable to monitor and provide feedback on driver behaviors that are related to safety issues, such as distracted driving and fatigued driving. However, with the development of the Internet of things (IOTs) and communication technologies, massive operation data can be transmitted to the cloud monitoring center and can be leveraged to evaluate the driving behavior with more authority and fairness.

However, the current research on buses mainly focuses on fuel vehicles [5]. Due to the different energy consumption modes, the evaluation strategy of its operation economy has a certain guiding role for the evaluation of electric buses, but it cannot be directly used. In addition, the evaluation of electric buses requires comprehensive consideration of driving safety and operating economy. Studies addressing how to comprehensively evaluate the safety and economy of electric buses, to the authors’ knowledge, are relatively scarce. The challenges can be attributed to the following aspects: (1) The evaluation index selection for safety and economy evaluation of electric bus is not established; (2) The relationship between safety and economy evaluation of electric buses is not clearly presented, and in-depth discussion on the relationship between safety and economy needs to be conducted. Motivated by the above challenges, this study aims to comprehensively assess the safety and economy of electric buses, clarify the selection method and process of evaluation indexes for the given dataset, and proposes a method with a better evaluation effect. Based on a given dataset of electric buses, appropriate methods are used for data pre-processing and evaluation indicators are selected by different correlation testing methods and safety and economic analysis processes. A relatively complete set of data processing and index selection methods is proposed in this paper. Aiming at the safety and economy evaluation of passenger vehicles, considering the high computational efficiency of ELM and the excellent stability of Elman neural network, the safety evaluation model based on ELM and the economy evaluation model based on Elman neural network are constructed. After analyzing the evaluation results, the safety and economy evaluation results are compared and the relationship between the safety and economy of electric buses is further discussed.

This paper is structured as shown in Figure 1. The data sources, data pre-processing methods, and indicator selection are described in Section 2. A description of the research methodology including the construction method and specific steps of the evaluation model is presented in Section 3. Section 4 details the comparison and analysis of the safety and economy assessment results. The paper concludes with a discussion of the implications for practice and areas for further research in Section 5 and Section 6.

2. Data Preparation and Indicator Selection

Based on the massive data of electric buses that was provided by the National College New Energy Vehicle Big Data Application Innovation Competition (National Big Data Alliance of New Energy Vehicles (NDANEV), 2016) [44], the safety and economic evaluation of driving behavior is attained. The data include 10 electric buses operations in Beijing, with a collection time span of 1 year and a sampling interval of 15 s. The data includes vehicle driving data and battery status data, such as vehicle speed, voltage, current, charging status, accumulated mileage, latitude and longitude, brake pedal status, gas pedal stroke, etc.

To evaluate the driving behavior, the vehicle velocity, velocity standard deviation, acceleration, acceleration standard deviation, acceleration rate of change, gas pedal travel, and brake pedal travel are initially selected as the characteristic factors to be evaluated. The selected data are pre-processed by adding and removing missing data appropriately, cleaning the data to remove noise, and deleting the useless data to mitigate redundancy. First, to maintain the variation pattern of the original data to the maximum extent possible, we compared the data processing effects of regression method, fixed value, mean, median interpolation method, and Lagrange interpolation method, as shown in Figure 2. The Lagrange interpolation method supplemented the missing data and most closely approximated the fluctuation of the original data. The Lagrange interpolation method with the best effect is used for missing data, and its calculation formula is as follows:

$$\begin{array}{l}L(x)={\displaystyle \sum _{j=0}^k{y}_i{l}_j(x)}\\ l_j(x)={\displaystyle \prod _{i=0,i\ne j}^k\frac{x-x_i}{x_j-x_i}}\end{array}$$

(1)

where $x_i$ is the independent variable, $y_i$ is the dependent variable, and $l_j(x)$ is the Lagrangian fundamental polynomial. Reasonable missing data due to vehicle stopping at night, charging, and driver shift change are removed.

Then, a wavelet denoising method is leveraged to clean the data and extract the main part of the signal with higher fidelity, as shown in Equation (2). A comparison of data completion and before and after denoising is shown in Figure 3 where the data fluctuation with large noise has been significantly improved.

$$\phi {(\mathrm{t})}_{a,b}=\phi (t-b/a)/\sqrt{\left|a\right|}$$

(2)

where a is the scale to control the scaling of the wavelet function and t is the amount of movement and corresponds to the time.

In order to make a comprehensive evaluation of driving behavior, the safety and economic evaluation indexes were first extracted. A Pearson correlation analysis and R² test were completed based on the speed, standard deviation of the speed, standard deviation of acceleration, jerk, accelerator pedal stroke, and brake pedal stroke, as shown in Equations (3) and (4). After the correlation test, as shown in Figure 4a, the most obvious information is that there is a high correlation between the speed and acceleration, a positive correlation between the acceleration and accelerator pedal stroke, a negative correlation between the accelerator pedal stroke and brake pedal stroke, and a negative correlation between the accelerator pedal stroke and brake pedal stroke. As can be seen from Figure 4b, the highest correlations are found between the standard deviation of the velocity, the percentage of positive velocity (v > 60 km/h), and the mean and standard deviation of acceleration and energy consumption. Therefore, the jerk is excluded and the acceleration can represent both the acceleration and braking characteristics, so removing acceleration pedal stroke and brake pedal stroke can greatly improve the calculation efficiency and remove the redundant characteristic factors.

Further, based on the safety analysis of the “Rapid Acceleration and Deceleration” Standard of Chinese Public Transport [45], combined with the dataset, according to the relevant regulations of the existing road transport management department on bus speed, the threshold of bus speeding is set as V, V > 60 km/h, and according to the existing study of brake hard threshold and the road transport industry-related experience, can set the bus as the definition of “snap acceleration, and deceleration threshold for A, |A| > 1.1 m/s². Considering the needs of driving behavior safety and economic analysis as well as the results of correlation analysis, a fast acceleration index (a > 1.1 m/s²), fast braking index (a < −1.1 m/s²), positive speed index (V ≥ 60 km/h), and negative speed index (V < 60 km/h) were extracted as driving behavior safety evaluation indices. In addition, speed standard deviation, positive speed index (V ≥ 60 km/h), average acceleration, and acceleration standard deviation were highly correlated with energy consumption and were selected as economic characteristic indicators.

$${\rho }_{X,Y}=\frac{N{\displaystyle \sum XY-{\displaystyle \sum X{\displaystyle \sum Y}}}}{\sqrt{N{\displaystyle \sum X^2-{({\displaystyle \sum X})}^2}}\sqrt{N{\displaystyle \sum Y^2-({\displaystyle \sum Y{)}^2}}}}$$

(3)

where ${\rho }_{X,Y}$ is the Pearson correlation coefficient, X and Y are variables, and N is the number of variables.

$$R^2=\frac{{\displaystyle \sum ({\widehat{Y}}_i-\overline{Y}}{)}^2}{{\displaystyle \sum {(Y_i-\overline{Y})}^2}}$$

(4)

where ${\widehat{Y}}_i$ is the estimate of the variables and $\overline{Y}$ is the mean of the variables.

3. Methodology

3.1. Safety Evaluation Model

After data pre-processing and evaluation index selection, the safety evaluation model is constructed by ELM. After cluster analysis, the safety scores of the buses are obtained, and the safety level of buses is evaluated and ranked.

(1)

Cluster analysis

First, the fuzzy C-Means (FCM) algorithm is used to cluster the sample data. The sample size of the electric bus dataset is only 10, which cannot be trained to obtain reasonable results. Therefore, in this study, the monthly data of each vehicle was used as a sample, with a total of 120 samples. Each month, 90% of the data are randomly selected as training samples and 10% as test samples. Typical training samples are extracted using the FCM algorithm, as shown in Equation (5). After FCM clustering, as shown in

Table 1. Cluster center.

Category (Security Level)	v < 60 km/h	v ≥ 60 km/h	a < −1.1 m/s2	a > 1.1 m/s2
	Percentage (%)
1	99.1332	0.8662	0.0421	0.0077
2	99.3814	0.6183	0.0323	0.005
3	99.6098	0.3903	0.0173	0.0038
4	99.869	0.1323	0.0071	0.0012

, the data are divided into different categories, and four categories of cluster centers are distinguished, and the four cluster centers show significant differences in the values of the four indicators. Further, we extract the typical samples in the dataset, as shown in Figure 5a, where the number of categories “1, 2, 3, 4” represents the 4 types of typical samples, and the number of categories “0” represents the filtered atypical samples. The affiliation of the 4 categories is shown in Figure 5b, From the intercepted sample fragments, it can be seen that the membership degrees of the four types of samples are quite different.

$$\{\begin{array}{l}U{(i)}_{1st}>U^{*}\\ U{(i)}_{1st}-U{(i)}_{2nd}>U^{\#}\end{array}$$

(5)

where U is the membership matrix after clustering by the FCM algorithm, $U{(i)}_{1st}$ is the maximum value in the sample membership matrix, $U{(i)}_{2nd}$ is the second largest value in the sample membership matrix, $U^{*}$ is the first threshold (0.5) for selecting typical samples, and $U^{\#}$ is the second threshold (0.2) for picking typical samples.

(2)

Extreme Learning Machine (ELM)

ELM is a feed-forward neural network algorithm with one-time tuning of all the parameters without iterations, which has high learning efficiency and strong generalization ability. The algorithm randomly selects the weight of the input layer and the bias of the hidden layer, and finally directly calculates the weight of the output layer by the least square method. The ELM-based evaluation model that was constructed in this study is shown in Figure 6. The input layer has 4 nodes with 4 feature indicators, and the hidden layer has 50 nodes. Among them, the activation function is $g(x)=1/(1+e^x)$, and the node is 1, which represents the number of categories.

3.2. Economic Evaluation Model

After the safety evaluation of electric buses, two economic evaluation models are constructed using LSTM and Elman neural network, respectively. First, the power consumption of 100 km is calculated based on the voltage, current, and sampling time of the bus, and the ecological score is calculated based on the power consumption. Then, the parameters and number of nodes in the input and output layers of the models are determined, and the buses are further ranked in terms of their economic performance.

(1)

Ecological score calculation

Since there is no energy consumption data in the dataset, the ecological score is further calculated on the basis of the calculated electricity consumption per 100 km for the subsequent energy consumption characterization. The calculation of electricity consumption per 100 km is shown in Equation (6), and the calculation of ecological score is shown in Equation (7).

$$E_i={\displaystyle \sum _{s=0}^{s=100}\frac{U_i{I}_{i}T}{1000\times 3600}}$$

(6)

where $E_i$ is the power consumption per 100 km of the sample i (kWh), $U_i$ is the instantaneous voltage (V) of the sample i, $I_i$ is the instantaneous current (A) of the sample i, and T is the sampling time (s).

$$EC{O}_i=1+3\times (1-\frac{E_i-E_{\min }}{E_{\max }-E_{\min }})$$

(7)

where $EC{O}_i$ is the ecological score of the sample i, $E_i$ is the power consumption per 100 km (kWh) of the sample i, $E_{\min }$ is the minimum power consumption per 100 km (kWh) of the sample I, and $E_{\max }$ is the maximum power consumption per 100 km (kWh) of the sample i.

(2)

Elman neural network

The Elman neural network is a typical local regression network. Its main structure is a feedforward connection, including input layer, hidden layer, and output layer, which can be regarded as an RNN with local memory units and local feedback connections, as shown in Equation (8):

$$\{\begin{array}{l}y(k)=g(w_{3}x(k))\\ x(k)=f(w_1{x}_c(k)+w_{2}u(k-1))\\ x_c(k)=x_c(k-1)\end{array}$$

(8)

where y is the m-dimensional output node vector; x is the n-dimensional intermediate layer node unit vector; u is the r-dimensional input vector; $x_c$ is the n-dimensional feedback state vector; $w_3$ is the connection weight from the middle layer to the output layer; $w_2$ is the connection weight from the middle layer to the output layer; $w_1$ is the connection weight from the successor layer to the middle layer; g() is the transfer function of the output neuron, which is a linear combination of the output of the middle layer; and f() is the transfer function of the middle layer neuron, which is often the S function.

In this study, the Elman-based evaluation model that was constructed is shown in Figure 7. The number of nodes in the input layer is 4, which is the index of economic characteristics of driving behavior, and the number of nodes in the output layer is 1, which is the ecological score that is calculated based on the electricity consumption per 100 km.

4. Results

4.1. Results of Safety and Economy Evaluation of Driving Behavior

In this study, an ELM-based evaluation model is constructed to rank the safety levels of 10 buses. After being assessed by the ELM safety evaluation model, the safety scores and grades of the 10 buses in the dataset are shown in

Table 2. Results of ELM Evaluation.

Evaluation Object	J	H	A	C	F	E	B	D	G	I
Safety score	1.287	1.477	1.484	1.898	2.252	2.528	2.653	3.316	3.473	3.902
Sorting result	1	2	3	4	5	6	7	8	9	10

. According to the safety score, the safety level of each electric bus increases from 1 to 10, among which the safety score of D, G, and I is between 3.3 and 4, which is relatively safe, while the safety score of A, H, and J is between 1.2 and 1.5, which is relatively unsafe.

In this study, an LSTM-based evaluation model is constructed to economically rank 10 electric buses that were lettered A–J. After constructing the ELM-based economic evaluation model, the results are shown in

Table 3. Results of the Elman neural network evaluation.

Evaluation Object	J	H	A	C	G	E	B	I	D	F
Safety score	2.369	2.528	2.987	3.057	3.182	3.189	3.2	3.277	3.355	3.418
Sorting result	1	2	3	4	5	6	7	8	9	10

. In the order of 1–10, the economy gradually increases. It can be found that among the 10 buses, buses I, D, and F are more economical, with scores ranging from 3.2 to 3.5, while buses A, H, and J are less economical, with scores ranging from 2.3 to 3.

The results of the economic evaluation model are analyzed, as shown in Figure 8, and it is found that the Elman-based economic evaluation score is close to the reference score, and the relative error is also small. The accuracy of the evaluation results is then analyzed by calculating the root mean square error (RMSE), as shown in Equation (9). The RMSE of the Elman neural network-based evaluation result is 0.1329, indicating that the mean error of the buses’ economic evaluation result is small.

$$R_e=\sqrt{{\displaystyle \sum d_i{}^{\frac{2}{n}}}}$$

(9)

where $R_e$ is the RMSE, n is the number of observations, and $d_i$ is the deviation between the measured value and the real value.

After evaluating the economy of the bus driving behavior, the bus safety score and the economy score are determined simultaneously. The economic score of the bus A to J bus that was based on the Elman neural network was analyzed in the order from highest to lowest safety level that was obtained based on the ELM evaluation model. As shown in Figure 9, it can be found that the bus drivers with higher safety scores also tend to have relatively higher economy, and the change trend of safety score and economy score is generally consistent, indicating that there is a correlation between safety and the economy of driving behavior to a certain extent.

4.2. Comprehensive Analysis of Driving Safety and Economy

The characteristic map of driving operation behavior can realize information visualization, effectively reflect the safety and economic characteristics of driving behavior, and serve as an auxiliary tool to improve the optimization effect of economical driving behavior. The consistency and difference of driving operation behaviors with the highest and lowest safety and economy are obtained by mapping similarity comparison, and the characteristics of driving operation behaviors corresponding to high safety and high economy behaviors are clarified. Acceleration, speed, accelerator pedal stroke, and brake pedal stroke are selected to construct a driving operation atlas. The characterization symbols and the range of variation of the data values are shown in Figure 10.

In this study, we mainly compare and analyze the characteristics of drivers’ operation behaviors of the buses with the highest and lowest safety, the buses with the highest and lowest economy, and the buses with high safety and high economy, so as to clarify the operating state characteristics of buses with different safety and economy. The established map is shown in Figure 11. On the road section with the same driving distance, the selected three buses control the vehicle speed and acceleration by trampling the brake pedal and the accelerator pedal to ensure the driving safety. It can be found that there are differences in the driving operation changes of the three buses. From Figure 11a,c, it can be seen that buses with high safety have less speed variation, fewer sharp accelerations and decelerations, and smoother driving than buses with low safety; from Figure 11b,c, it can be found that buses with high economy have more braking and more energy recovery on average over the same mileage than buses with low economy (but more sudden braking also resulted in lower safety). To analyze the driver operating behavior characteristics of the safest and most economical buses, Figure 11a,b are further compared. It can be found that the speed of buses with high safety and economy is relatively stable with little fluctuation, and the driving operation behavior of buses is relatively stable.

5. Discussion

The safety and economy of driving behavior is a vague concept, and it is intractable to evaluate the behavior criteria of the “safest and most economical” drivers in an intrinsic manner. However, information on relatively safe and unsafe, economical, and uneconomical drivers can be obtained by the relative classification of the sample set. By selecting evaluation indicators of driving behavior, different evaluation models are constructed to determine the relative ranking of the buses in terms of safety and economy.

According to the selected safety evaluation index, the data ratio of 10 buses in the dataset (a > 1.1 m/s², a < −1.1 m/s², v ≥ 60 km/h, v < 60 km/h) is calculated. The sample data are then clustered using the fuzzy C-mean (FCM) algorithm and four categories of clustering centers are identified, as shown in Table 1, with the sample data from the four categories showing variability in the values of each indicator. An ELM-based evaluation model is constructed to obtain the final safety ranking of the A–J 10 electric bus, and the evaluation results are shown in Table 2. Among them, buses that were numbered D, G, and I are safer, while buses numbered A, H, and J are less safe.

The velocity standard deviation, positive velocity index (v ≥ 60 km/h), average acceleration, and acceleration standard deviation are selected as economic characteristic indexes. After calculating the electricity consumption per 100 km and economical score, the Elman neural network-based economic evaluation model is constructed. The economic ranking of the A–J 10 electric bus is obtained, as shown in Table 3, where the buses that were numbered I, D, and F had higher economic performance and those that were numbered A, H, and J had lower economic performance. Further, by calculating the electricity consumption per 100 km to obtain the ecological score of each bus, we find that the evaluation results based on the Elman model is close to the true economical score, as shown in Figure 4. In addition, by calculating the RMSE, it is found that the Elman-based evaluation model has a small RMSE, which indicates that its evaluation results are more accurate.

A comprehensive analysis of the safety and economic evaluation results of electric buses revealed that buses with higher safety also have higher economic performance. Based on the evaluation results of the safety and economy of electric buses, the driving operation behavior graph of the bus with the highest and lowest safety and the bus with the highest and lowest economy is constructed. The differences in driving operation behavior between the high and low safety buses, and high and low economy buses are compared and analyzed, and the similarities in driving operation behavior between the high safety and high economy buses are analyzed. We found that the bus with higher safety and economy has a relatively stable speed and less fluctuation, and the driving operation behavior is relatively stable.

Wu et al. [46] investigated the characterization, evaluation, discrimination, and feedback optimization of ecological driving behavior. The study illustrated that there are many factors that influence driving behaviors that lead to high fuel consumption, but in order to obtain lower vehicle fuel consumption, driving behaviors are relatively fixed and, therefore, require drivers to implement appropriate driving operations at the right time. In contrast, there are a variety of vehicle operating conditions that can lead to higher fuel consumption, but in order to obtain economical fuel consumption levels, the vehicle operating conditions are relatively unchanged and fixed. Usually, when the vehicle is running smoothly, the driver can take appropriate driving actions at the right time, and the driving safety is high. The feedback optimization method of eco-driving behavior that is based on driver characteristics can effectively improve the driver’s acceptance of eco-driving behavior, thus enhancing the effect of eco-driving behavior training and helping to reduce vehicle energy consumption to a greater extent.

In this study, the relative safety and economic performance of 10 electric buses are evaluated. A safety evaluation model is developed using ELM, and an economic evaluation model is developed using the Elman neural network. The results illustrate that the proposed driving behavior evaluation model that is based on ELM and the Elman neural network can effectively assess the safety and economy of driving behavior. The main innovation of this study is to evaluate driving behavior by detailed quantitative indicators. The fuzzy C-means method is used to select typical samples by combining the ideas of fuzzy and clustering, and the ELM is trained to improve the convergence speed of model learning. Considering these results, the following recommendations are made to improve driving safety and economy.

• The optimization model and method of “characterization-evaluation-discrimination-feedback optimization” of driving behavior can effectively improve the safety and economy of bus operation. In fact, the index selection method and evaluation model that were provided in this study can be used as a reference.
• The feedback on the driver’s driving behaviors such as starting, stopping, and following, as well as the ranking of safe and ecological driving behaviors in the overall situation, is conducive to optimizing and improving driving behaviors. The proposed method system can rank a certain number of buses for safety and economy.

There are differences in drivers’ driving behavior in the summer and winter, and seasonal factors have a significant impact on driving behavior; however, this paper does not consider the effect of seasonal factors on the safety and economy of driving behavior. Subsequent studies will verify the scientific validity and feasibility of the model by incorporating the climate of the data collection site. Another limitation of this study is that the latitude and longitude data in the dataset were not combined, and the location coordinates of the bus were not considered when analyzing the safety and economy evaluation of driving behavior. In future research, we can combine the location coordinates of the bus to further investigate the safety and economy of starting and stopping the bus during inbound and outbound stops.

6. Conclusions

In this paper, we evaluated the safety and economy of 10 buses in the electric bus dataset, evaluation indices were extracted, and two types of safety evaluation models and two types of economic evaluation models were constructed, respectively. Through comparative analysis, the following conclusions are drawn.

• There is a correlation between the safety and economy of bus driving behavior, and buses with high safety tend to be more economically inclined.
• The results of this study reveal that bus drivers with higher safety and economy can take appropriate driving operations at the right time, and the running state of their buses is relatively stable.
• This study uses the ideas of fuzzy C-means and fuzzy and clustering to select typical samples for ELM training, which improves the convergence speed of model learning.
• The proposed ELM-based safety evaluation model and the Elman neural network-based economy evaluation model have high classification accuracy and can effectively evaluate the safety and economy of bus driving behavior.

The research can provide theoretical support for the research on the safety and economy evaluation of bus drivers’ driving behavior and can also provide relevant organizations with the basis for the evaluation and retraining of driver safety and energy saving.