*4.4. Other AI-Based MPPT*

This section of the paper explains other artificial intelligence methods applied in the field of tracking maximum power from the PV array along with a report of the various latest research performed concerning it in Tables 7 and 8 respectively.

#### 4.4.1. Fuzzy Logic Control

FLC converts its analog input to digital values. This technique examines the output power of PV array for every sample. If the change fraction is greater than zero, voltage is enhanced by FLC by adjusting the duty cycle and vice versa. As a result, the maximum power ratio is zero. FLC inputs error "e", and its change "*∂*e" with samples in time "ki" can be computed as

$$e = \frac{P\_{pv}(k) - P\_{pv}(k-1)}{V\_{pv}(k) - V\_{pv}(k-1)}\tag{42}$$

$$
\partial \mathcal{e} = \mathcal{e}(k) - \mathcal{e}(k-1) \tag{43}
$$

Figure 21 shows a block diagram of FLC control. The input variables are changed to linguistic variables by using different distinct membership functions. Thereafter, they are manipulated on the basis of the "if-then" rule by applying the required conduct of the scheme. Finally, they are converted to their numerical equivalent [107]. This approach shows fewer oscillations, fast response [108], and high tracking efficiency in contrast to conventional MPPT approaches. However, it suffers from high computational complexity.

**Figure 21.** Block representation of FLC-based MPPT.

#### 4.4.2. Artificial Neural Network

An ANN is a set of static learning models. For anticipating a precise output for each input, this approach simulates a biological neural system. Figure 22 shows the threelayered structure of ANN in which the neuron quantity in each layer varies depending on the situation.

**Figure 22.** Three-layer structure of ANN [109].

These networks are used as an MPP system to predict the best possible values of power or voltage that can be produced at a given time. These values act as base values in deciding the converter's duty cycle. The PV module parameters and atmospheric parameters are included in the input variables and then processed by hidden layers in the network. The procreation algorithm is retroactive and grades in a mishap. Thereafter, utilizing neurons of center layer, it feeds back the output through the input neurons. The following Equation is used to calculate the presence of hidden neurons:

$$m\_h = \frac{1}{2}(n\_i + n\_o) + \sqrt{n\_t} \tag{44}$$

A complete experimental setup assists in data collection. The dataset is then obtained by feeding atmospheric conditions and array parameters into the ANN to find output Vm and Pm. This set is then transformed into an instructional one, which moves into the premeditated ANN, where it is taught how to perform. Moreover, the functions of input data serve as instruction data for the ANN model that was created. Then, the model learns how to execute on its own. The assessment datasets examine the performance of the constructed ANN after the instruction phase, and the errors are sent back to the ANN until all of the neurons' weights are changed correctly. MPPT using ANN is more accurate and shows less oscillation around MPP [109]. These algorithms suffer from the drawback of high computational complexity.

#### 4.4.3. Evolutionary Computational Techniques

Evolutionary computation is an area of artificial intelligence and soft computing that studies a family of algorithms for global optimization inspired by biological evolution. GA and DE are ones amongst them used to track MPP.

GA is a computer model that is inspired by evolution and consists of chromosomes. These chromosomes include information on a potential solution to a problem. Each chromosome has its own set of characteristics. This algorithm is used in wide applications. In contrast to tracking MPP, it is able to boost the PV voltage, which represents the chromosomes and their fitness value that corresponds to PV power. The main idea is to make genetic changes to a population of people and discover the ideal ones corresponding to the fitness function. Figure 23 shows the flowchart of GA.

**Figure 23.** Flowchart of GA [110].

DE is another evolutionary computational algorithm applied to problems based on global optimization. It is applicable to track GMPP in PSCs due to its simpler execution and wide search freedom. The DC converter duty cycle is used as a target vector "*∂*n" by this approach. Initially, the target vector with two dimensions is initialized as "*∂*n" for each iteration and generation as the population. It chooses three random particles after one generation in order to reduce the execution time. Following that, the selected duty cycles are used to calculate the PV array's associated powers "Pn". "Pbest" is picked as the maximum power in the set of "Pn", and "*∂*best" is chosen as the corresponding "*∂*n". The weight difference between any two target vectors is then used by a mutation factor (M) and forms the mutated particle by adding this difference to the remaining target vector. The mutated particle is also called the donor vector "DVn". The mutation's way should be towards "Pbest". Following mutation, donor and target vectors are combined by a crossover procedure to create trial vector "TVn" and estimate the PV array's power.



