Metaheuristics-Based AI Techniques
Metaheuristics techniques are inspired by natural or biological processes and play a major role in solving many engineering problems. Some of the contributions of metaheuristics are being discussed in this section. In Ref. , the problem of identification of PV model parameters was solved. This problem was solved by using a learning search algorithm (LSA). Salient attributes of the proposed LSA are as follows:
- • As the iteration passes, rate of self-adjustment changes
- • On the basis of the current worst and best solution, learning patterns get modified
- • In the final stage, perturbation was allowed to ensure achieving of global optimum
The capability of the proposed LSA was compared to single-/double-diode PV models and other algorithms in terms of rate of convergence and computation time. Practical results suggest that the proposed technique performed well in finding the optimized parameters for PV systems. There is an important issue in PV systems to build a mathematical model suitable for optimized performance and reliable operation. In Ref. , a Tree Growth Algorithm (TGA) was implemented for finding unknown parameters of a PV model. TGA was able to find the precise values of parameters very quickly for the developed PV model. In TGA, there are two steps, namely exploration and exploitation. In first step, there is a competition among several trees to absorb the maximum sunlight for preparing their food. Initially, a population of trees was generated randomly; then, the fitness corresponding to every tree was determined. In the next step, some of the trees move toward the tree having the best fitness to receive abundant light. The trees having the worst fitness are removed and new trees take their places. As time passes, there will be movement of trees in the direction in which better food exists. In the next stage, most of the trees move towards best solutions and will be near to a global solution. There is a balance to be maintained between the stages of exploration and exploitation, and this is done by selection of proper parameters. The experimental values of PV systems’ parameters obtained by TGA were found comparable to the datasheet provided by the manufacturer of the PV arrays.
There is an issue of finding a global MPP for the operation of PV systems under partial shading situations where PV characteristics have multiple optima points. Metaheuristics-based AI can conveniently solve this problem. In Ref. , a modified bat algorithm (BA) was proposed to achieve MPP quickly and efficiently. The study focused on two objectives:
- • To reduce the time of MPP tracking for the case of sudden change in irradiance
- • To converge at global MPP with a high level of confidence
In order to meet these two objectives, a combination of BA and cuckoo search (CS) algorithm was proposed. Results showed that proposed methodology was able to achieve global MPP accurately with higher rate of convergence as compared to BA and PSO. Metaheuristics-based AI also find their application in grid-integrated microgrid solutions. In Ref. , BA with a rule-based concept was proposed for the management of energy in the microgrid application. The microgrid under study comprised distributed units of PV modules, fuel cells, battery banks, and a microturbine. The proper functioning of the microgrid requires a day in advance prediction of not only intermittent data of PV modules and fuel cells but also modelling of the probabilistic nature of load. The prediction of PV data and load was done by this new BA algorithm. The forecasted data were fed to an energy management system on an hourly basis. It was shown that optimum management of the state of charge and reac- tive/active power were achieved.
In Ref. , optimization-based AI was used to control the microgrid based on a combination of PMDC generator, battery bank, and solar thermal power. The controllers adopted for the microgrid were required to control bi-directional charge and buck-boost control in converter. For this purpose, separate PI and PID controllers were used. The tuning of gain parameters of controllers was done by metaheuristics algorithms such as grey wolf (GW), mine blast, and particle swarm. In this work, simulation results demonstrated the excellent performance of GW optimization from the point of view of improving the rate of convergence, DC link voltage, state of charge in energy storage, and efficiency of converters, as compared to PSO and mine blast algorithm. In Ref. , an AI scheme based on marine predators (MP) was implemented for reconfiguration of PV modules. The proposed methodology was aimed at the mitigation of the effect of partial shading and at the prevention of hotspots forming in panels while achieving MPP. The proposed AI methods optimized the fitness function and factors such as losses due to mismatching of cells, fill factor, power loss, and improvement in power. The proposed algorithm showed significant dispersion of shade, resulting in the reduction of the count of peaks characteristic of PV power.
In Ref. , use of optimization technique-based AI is proposed for solving problems of predicting the characteristics of PV. The power output of the PV array fluctuates randomly due to variation in data obtained from the meteorological centre. This adversely affects the reliable and stable operation of the electrical grid. In order to overcome this issue, a hybrid algorithm based on the combination of ant-lion optimization (ALO) and a prediction method called random forest (RF) was proposed in the presented work. ALO optimized the parameters of the RF model, which led to reduction in computation time and improved accuracy. The proposed model was compared against the other RF models. It was found that accuracy in the prediction of performance of the PV system was improved significantly. In Ref. , a clustering methodology is presented based on expansion and erosion for finding faults in the PV system. In Ref. , a review of an Al-based forecasting method was done for accurate prediction of solar irradiance. The review' also focused on key issues and future prospects for AI techniques in grid-integrated PV systems.