Long-term river water quality prediction method based on improved TCN model

The invention relates to a long-term river water quality prediction method based on an improved TCN model. The method comprises the steps of firstly inputting river water pollutant concentration data into an SG filter for noise reduction processing, then decomposing the processed data into a trend sequence, a seasonal sequence and a residual sequence by adopting an STL time sequence decomposition method, and respectively inputting the trend sequence and the residual sequence obtained after decomposition into an improved TCN model for training and prediction. And finally, fusing predicted values of the trend sequence and the residual sequence with an original seasonal sequence to obtain a river water quality long-term prediction result. According to the method, the defects of a basic TCN model are improved, and the long-term prediction capability of the model is further improved through a data noise reduction and data decomposition method. The effectiveness of the method is verified through specific experiment