Electricity Consumption Data Mining Based On LSTM

The time series data generated at any time can reflect the inner law of the phenomenon,which is a powerful weapon waiting to be unlocked in many areas such as anomaly detection,personalized differentiated recommendation,stock trading,load forecasting,quantum physics research,sales and so on.The data of electric energy is typical time series data.The effective data mining of electricity plays an important role in saving energy and reducing consumption,and green production and life.But it is difficult for mining electricity data because of the features such as large amount of data,many influential factors and unfixed feature length.In order to solve the above problems,this paper combines deep learning ideas,constructs LSTM(Long Short Term Memory)network models with reasonable value as the main research algorithm for power data mining.The activation function has been optimized and improved;a power energy monitoring platform based on cloud computing and storage architecture has been built,and the IEB and data mining engine used to service power can work in various network environments;Comprehensive communications solution that combines the MQTT IoT transport protocol,Zigbee,wireless and socket communications to ensure network address translation,unified access across LAN(Local Area Network)and cross-NAT WAN(Wide Area Network)Power data;The successful application of LSTM network in power data mining can be used for non-stationary data in any time period,providing very accurate load forecasting and providing a comparison between XGBoost and SVR famous algorithms for machine learning.The article compares the advantages and disadvantages of different network models,taking into account the prediction accuracy and the running speed,designing the most appropriate method for each part of the work,and constantly optimizing the network parameters during the training process.The proposed algorithm is superior to other existing algorithms in prediction accuracy.And the platform satisfies the requirements of practical application scenarios.