Application Research Of Gene Expression Programming In Electricity Load Prediction

With the development of national economic and deepening of industrial system’sreform, the demand of electricity quality is getting higher and higher. Electricity loadis a fundamental tool to ensure the quality of electric energy, especially important tothe short-term load prediction in next one or several days. Electricity load predictionis the chief component of plan and research of electric power system, and also thebasis of economic operation of electricity system, it is very important to electricitysystem and its plan. At present, the electricity load prediction methods emergecontinuously, but the single model application often limited within a certain rangewhile its accuracy is low. So it is crucially for the electricity system operation todevelop a electricity load prediction model which is universal and accurate, whoseapplication value is inestimable.The build of electricity load prediction model needs to take some related loadproblems into consideration which are usually stochastic or nonlinear problems, wecan separate properties of a random variable to build mathematical model to expressthe statistical regularity of electricity load in the changing process, and then setreasonable load prediction mathematical expressions based on mathematical, use thehistoric data as input to predict the electricity load in future. In recent years, themethods based on soft computing and intelligent model becomes the researchemphasis. Although the soft technology theory such as the GA and BP model isprominent but still cannot break away from the inherent imperfection which affectedthe prediction accuracy of electricity load prediction to some extent. This paper takesin-depth study on GEP(Gene Expression Programming) which is forefront incomputing intelligence theory, it is an efficient evolutionary algorithm which is brandnew and better than traditional expression and information disposition. GEP isapplicable to solve the problem of high complexity of nonlinear system, especiallysuitable for solving complex unknown system problems. Differ from the artificialneural network, GEP do not need to divide into layers and take the multi-sampletraining, thus it is easier to solve the actual problem. In this paper, the research workof electricity load prediction based on GEP can be divided into the follow parts:This paper first expounds the concept of electricity load, the influence factors and mathematical description, and then summarizes several classic models’establishment according to the requirement and principle of modeling. After that, weintroduced a single model’s imperfection when applied soft computing such as theneural network model and genetic algorithm model into the electricity load prediction,and the research trends that combining the models to predict.Secondly, we introduced the basic principle of GEP, clarifies the root cause ofhigher efficiency when compared to of the GA and GP algorithms. Because of thelimitations of GEP that it exist immature convergence, this paper proposed animproved GEP algorithm based on segmentation strategy(Gene ExpressionProgramming based on Multi-Strategy, MS-GEP) from the perspective of populationdiversity. The theory and practice show that this improved algorithm can quicklyconverge to the optimal solution, which laid a solid foundation for short-termprediction model of the electricity system load. This algorithm adopt individual outbreeding to avoid generating similar individuals and retains higher species diversitybased on the phase diversity strategy. In addition, this algorithm dynamically adjuststhe genetic operators and fitness function to prevent the algorithm falls into localconvergence. And then build the Markov chain model of MS-GEP algorithm toanalysis and verify the convergence and diversity of the algorithm.At last, aiming at the problem of slow convergence speed of neural network, thispaper utilizes MS-GEP to predict the electricity load value in short-term future whileconsidering climate sensitive factors. We adopt the optimization training networkparameters to speed up the network learning and improve the prediction accuracy, andset up GEP-ANN short-term electricity load prediction model. The simulation resultsshow that the combination predict modeling method based on the MS-GEP and theBP neural network have important breakthrough in improving the prediction accuracyand the algorithm convergence speed under the condition that reduces the requirementfor historical data collection. It is more efficient in finding a electricity load predictionmodel which has higher degree of fitting and higher predictive accuracy, this paperobtains certain achievement in electric electricity prediction modeling theory research.