Research On Multi-scale Electricity Consumption Prediction Using Hybrid Data-driven Model

The reform of energy structure and the strengthening of environmental protection policies pose a great challenge to the development of the energy industry.As a renewable energy,electric energy plays an indispensable role in China's modernization process.Accurate and effective power consumption prediction is of great practical significance for rational distribution of power resources and prevention of power waste.Data-driven model has been widely used in various energy consumption prediction scenarios in recent years due to its ease of use and optimization adaptability.Among them,the forecasting model based on artificial neural network(ANN)is the most popular in solving the problem of power load forecasting.This paper combines intelligent algorithm with artificial neural network model to study the multi-scale power consumption prediction method of hybrid data-driven model.The specific work is as follows:(1)The basic process and method of power consumption prediction using data-driven model are proposed based on the artificial neural network,which combining with data analysis and preprocessing methods.Firstly,selecting horizontal processing method or wavelet transform processing method(DWT)according to the characteristics of the original input variable data to delete the abnormal points;Then completing the missing data;Finally,using principal component analysis(PCA)to reduce the input redundancy.(2)In order to improve the optimization performance of neural network,hybrid prediction model based on teaching learning based optimization(TLBO)algorithm is proposed.TLBO is a new heuristic method based on swarm intelligence.The basic idea is to simulate the two stages of "teaching" and "learning".In this algorithm,students acquire knowledge in class through teachers' teaching and through mutual learning among learners.In order to further improve the optimization accuracy of TLBO,three parts of improvement are carried out on the basis of TLBO.The improved iTLBO is proved to have better optimization performance than TLBO through benchmark function comparison experiment.Combined with the characteristics of iTLBO and ANN,iTLBO-ANN hybrid forecasting model is designed for power load forecasting.At the same time,two typical swarm intelligence algorithms(particle swarm optimization(PSO)and genetic algorithm(GA))are combined with ANN to build hybrid prediction model.(3)In order to verify the multi-scale power load prediction ability of the proposed iTLBO-ANN hybrid prediction model,this paper selects two practical cases with different energy scales for simulation verification.(i)At a low scale,the power consumption of a single building is studied by using an actual case of power consumption of a university library.(ii)At a high scale,the actual energy data of a city in JiangSu province are used to study the energy consumption of city-level power.After data preprocessing,this paper applies a variety of data-driven models for power consumption prediction.The results show that iTLBO-ANN model has better prediction accuracy and shorter prediction time than other mixed prediction models(iTLBO-ANN,iPSO-ANN,GA-ANN).In addition,the model is simple and easy to implement,so it can be used for short-term online prediction of different energy scales and has a strong universal applicability.