Prediction Of Short-term Electricity Price Based On Deep Learning Algorithm

In a market-oriented environment,electricity price signals are the focus of market participants.It is not only a key reference factor for trading decisions,but also an important regulatory indicator for market managers.Therefore,the prediction of electricity price is also a hot topic of academic research.With the successive establishment of two national power spot markets in Beijing and Guangzhou,short-term electricity price research has become an effective means for traders to maximize profits and avoid market risks.Because electricity commodities are non-storage commodities,the price elasticity is weak.The long-term and short-term factors that affect electricity prices are complex and changeable,which makes the historical data of trading electricity prices have the features,as highly volatile,highfrequency,nonlinear,and price spikes.These features cause the difficulties to predict the short-term electricity price.The features are difficult to express in a regular way.Machine learning,which originated in the last century,has the ability of algorithms to acquire knowledge by themselves,that is,the ability to extract patterns from raw data.The study of this ability is the inherent motivation of deep learning.In recent years,with the improvement of computing power of computers,deep learning technology has achieved outstanding results in processing time series with nonlinearity and fluctuation aggregation.This technology has been applied by researchers of the power industry in load forecasting,anti-theft identification,transformer fault identification,and more.Based on the latest development results of deep learning technology,this paper makes new attempts and innovations on the related issues in the short-term electricity price prediction field based on actual transaction electricity price data.The main contents include:(1)Analyze the long-term and short-term factors that affect electricity price fluctuations,analyze the characteristics of electricity price fluctuations,and study the reasons for the spike characteristics of electricity prices and the characteristics of holidays.Compare and analyze the fluctuations of electricity prices and the characteristics of traditional financial assets to explore their similarities and differences.In terms of electricity price data processing,combined with the characteristics of the electricity market itself,the traditional standardized method was improved.In terms of improving peak electricity price forecasting capabilities,an asset jump algorithm in the field of quantified finance was introduced to identify jump price sampling points,and based on this,the proportion of jump samples in the total sample was increased.Then,in the process of feature construction and extraction,feature selection and construction are performed based on the maximum number of correlations and the Hibbert-Huang transform algorithm.(2)A short-term electricity price prediction model based on improved particle swarm algorithm to optimize random forest is established.This combination model combines improved particle swarm algorithm and random forest model.With the global search capability of improved particle swarm algorithm,the framework of random forest algorithm and model hyperparameters can be performed Finding the best.The validity of the model is verified based on actual hourly transaction electricity price data.(3)For the traditional machine learning algorithms,it is difficult for the model to learn the abstract high-dimensional features.The latest research results of deep learning technology are introduced to establish a multilayer long-term and short-term memory neural network model and a long-term and short-term memory neural network model with Attention distribution mechanism.The actual hourly transaction electricity price data is compared with the optimized random forest model based on the improved particle swarm algorithm for experimental comparison.The research shows that the deep learning model has obvious advantages in the generalization ability of the model and has stronger prediction accuracy.