Development of a composite control strategy based on PID, multivariate regression and load forecasting for ASHPs

For variable-speed air source heat pump systems (ASHPs), using traditional PID feedback control is prone to causing oscillations and overshoot in the controlled object, while the neural network-based load forecasting feedforward control easily leads to loss of control due to limited performance in predicting the initial load of discontinuous loads. To address these problems, composite control strategies are proposed, which integrate the advantages of PID and load forecasting control. This strategy consists of three control models: PID, load forecast regression (LFR) and load forecast (LF), among which Long Short-Term Memory (LSTM) neural networks are used to develop the load forecast model and the regression model is built based on five input variables and one output variable. The composite strategy can automatically switch back and forth between PID and LFR modes to ensure smoothness and stability of control. The experimental verification of the ASHPs has also been conducted. The integration time absolute error (ITAE), Overshoot (Os) and Regulation Time of supply water temperature were chosen as the evaluation indicators for the control effect. In cooling conditions, compared to traditional PID control, the ITAE and RT of the proposed strategy have been reduced by up to 64.8% and 90.6%, even though Os is slightly higher. In heating conditions, the ITEA and Os are decreased by 9.3% and 62.5%, respectively, and the RT is close to that of traditional PID control. Comparative analyses demonstrated that the composite strategy will not lose control even when the load forecasting deviations are large and can achieve a more precise and smoother control.