Research On Power Energy Prediction Based On Deep Learning LSTM Algorithm

The power forecasting problem is a key issue in the operation management of power companies.Its main task is to predict the electric energy value in a certain time period in the future based on relevant historical information data and real-time production operation data.Accurate electric power forecasting can improve the safety of powery enterprises,and formulate reasonable production plans and schedules,while meeting the power supply demand.In this way,we can manage to reduce production operation costs,protect the ecological environment and improve the economic efficiency of enterprises.Therefore,the study of the power forecasting problem has important theoretical and practical significance.In this thesis,the prediction problem in power system planning and operation is studied.Sequence to sequence(Seq2Seq)model based on long short term memory(LSTM)of deep learning is applied to power energy prediction for the first time.For the power load forecasting of iron and steel enterprises,considering the characteristics of timing and randomness of power energy forecasting,an improved Seq2Seq model is proposed.In the wind power prediction problem,considering the lack of actual wind power acquisition data,the mean method and BP neural network training are proposed to complete the data in order to ensure the time-sequence of the data.A wind power prediction model built by Seq2Seq is proposed to compare the influence of the two complementary data methods on the model.The main research contents are as follows:1)In the short-term power forecasting problem,the traditional forecasting methods are difficult to deal with the challen ges of non-linearity,randomness and large amount of data in power generation and production operation.The Seq2Seq model based on LSTM in deep learning is proposed for the first time to solve the power forecasting problem.2)In view of the problem of power load forecasting in iron and steel enterprises,a load forecasting model based on LSTM is designed.Considering the incomplete learning of time-sequence information,a Seq2Seq load forecasting model based on LSTM is proposed.Numerical experimen are carried out on the actual production data of iron and steel enterprises.The experimental results verify the effectiveness of the algorithm.3)Aiming at solving the problem of historical wind power data missing in wind power forecasting,two data completion strategies,i.e.the mean value compensation and the BP neural network,are proposed.The completed data is input into the wind power prediction model of Seq2Seq for training.Numerical experiments were carried out using the actual operating data of the wind field.The results show the effectiveness of the algorithm.4)The electric power enterprise load forecasting system is developed embedding with the above algorithms.The software system functions include basic data query,power load forecasting,maintenance planning and other modules to provide multi-model prediction scheme.The system realizes flexible manual interaction.