Research On Data Prediction Of Energy Efficiency System Based On Improved Neural Network

Energy management systems can help energy companies improve efficiency and reduce energy consumption.Energy consumption forecasting within these systems enables companies to better understand the operational status of their factories,making it a crucial component.The urgency to improve the accuracy of these forecasts cannot be overstated.Relying solely on manual data collection and energy consumption prediction can reduce the efficiency of the forecasts,making the selection of appropriate models and optimization algorithms of significant importance for enhancing prediction accuracy.This paper focuses on the energy consumption forecasting module in energy management systems and conducts research to improve its effectiveness.Specifically,it investigates and optimizes Long Short-Term Memory(LSTM)networks,addressing the challenges of determining hyperparameters and dealing with the non-linear stationarity of energy consumption data.Furthermore,the feasibility of using the model for the energy consumption forecasting module is validated.The main research contents of this paper are as follows:(1)Firstly,this paper introduces the classification of neural networks and conducts a comparative analysis of the advantages and disadvantages of Backpropagation(BP)neural networks and LSTM neural networks.After experimental comparisons,the prediction curve of the LSTM neural network is found to be closer to the true values,so the LSTM neural network is adopted as the prediction algorithm for this study.Simultaneously,the Particle Swarm Optimization(PSO)algorithm is introduced to address the selection of hyperparameters in LSTM neural networks,enabling automatic optimization of hyperparameters.This paper also improves the inertia weight and learning factors within the PSO algorithm,resulting in the design of an IPSO-LSTM model.Using actual metered electricity consumption data from a factory,the model is validated by relying on the collection intervals and peak-valley electricity time periods.The prediction results are compared with those of the PSO-LSTM and LSTM models.The results show that the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)values of the IPSO-LSTM model are smaller than those of the traditional models,proving that the improved model optimized by the PSO algorithm has higher prediction accuracy than the traditional models.(2)To address the non-linear stationarity issue in energy consumption data,this paper proposes and establishes a model combining the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)algorithm and the IPSO-LSTM algorithm,referred to as the CEEMDAN-IPSO-LSTM model.Firstly,the CEEMDAN algorithm is used to decompose the original data into several sub-sequences.Then,the IPSO-LSTM algorithm is applied to train each sub-sequence,obtaining the predicted data for each sub-sequence.The final results are obtained by summing up the predicted data of all sub-sequences.The model is compared with the IPSO-LSTM model through experimental analysis.The results show that the MAE and RMSE values of the CEEMDAN-IPSO-LSTM model are smaller than those of the IPSO-LSTM model,proving that the CEEMDAN-IPSO-LSTM energy consumption forecasting model has better prediction performance and higher accuracy.This model holds positive implications for improving the level of the energy consumption forecasting module in energy management systems.(3)Finally,this paper implements a visual display of the energy consumption forecasting model,including the data processing module,energy consumption forecasting module,and the Graphical User Interface(GUI)module.Users can import data,perform energy consumption forecasts,and export prediction results by clicking buttons.The prediction results are displayed through the GUI interface.