基于长短期记忆网络-模糊控制的光伏最大功率点跟踪算法

Photovoltaic Maximum Power Point Tracking Algorithm Based on LSTM-FLC

光伏阵列具有非线性特点且其最大功率点会随环境变化而产生偏移现象。尽管最大功率点跟踪算法被广泛地应用于跟踪和预测光伏系统的最大功率点,但仍然面临着模糊控制的动态品质低、控制精度不佳等挑战。为解决前述问题,提出了一种集成长短期记忆网络与模糊控制的光伏最大功率点跟踪算法。首先,采用长短期记忆网络以时间序列法预测最大功率点电压。其次,将该预测电压与光伏阵列电压间偏差及其导数作为模糊控制的输入,然后直接调节Boost变换器的占空比。再者,为防止开关管常通,预先设置了变换器的最大、最小占空比。最后,在四种可变大气条件下,利用MATLAB/Simulink对所提算法进行仿真验证。实验结果表明:与长短期记忆网络、电导增量法和遗传算法相比,所提出的算法具有良好的跟踪性能、稳定精度及效率,并且具有波形更平滑、振幅较小的优点。

Photovoltaic array features nonlinear performance,and its maximum power point(MPP)will shift with environmental changes.Although the maximum power point tracking(MPPT)algorithm is widely used to track and predict the MPP of photovoltaic systems,it still faces challenges such as low dynamic quality and poor control accuracy of fuzzy logic control(FLC).To solve the problems men-tioned,a photovoltaic maximum power point tracking algorithm based on long-short term memory-flC(LSTM-FLC)is proposed.Firstly,the LSTM network predicts the MPP voltage by a time series method based on the light intensity and temperature datasets.Secondly,the deviation between the predicted voltage and the photovoltaic array voltage,as well as its derivative,are used as the input of FLC,and thus FLC is used to direct adjust the duty cycle of the boost converter.At the same time,the maximum and minimum duty ratios are pre-set to prevent the switch from being normally turned on.Simulation verification is carried out using MATLAB/Simulink under four varia-ble atmospheric conditions.Experimental results show that compared with LSTM,conductance incremental method,and genetic algo-rithm,the proposed MPPT algorithm has good tracking performance,stable accuracy,and efficiency,and takes the advantages of smoother waveform and smaller amplitude.