Short-term power load prediction method and device based on hyper-parameter optimization

Abstract

The invention specifically discloses a short-term power load prediction method and device based on hyper-parameter optimization, and the method comprises the steps: carrying out the normalization processing of original power load data, taking the data as the input of an LSTM (Long Short-Term Memory) network, taking the three hyper-parameters of the LSTM as to-be-optimized parameters of an IPSO (Improved Particle Swarm Optimization), setting an average absolute percentage error (MAPE) as a fitness function of the IPSO, and carrying out the optimization of the IPSO, changing the inertia weight from a fixed value to nonlinear change, and setting the inertia weight as an independent variable of a learning factor, so that the beneficial effect of improving the short-term power load prediction precision can be achieved.

Description

technical field [0001] The present invention relates to the field of short-term power load forecasting, in particular to a short-term power load forecasting method and device based on hyperparameter optimization. Background technique [0002] With the rapid development of social economy and technology, the accuracy of short-term power load forecasting has become very important. Short-term power load forecasting refers to the forecasting of power loads in the next few days or even hours. It plays an irreplaceable role in the safe and economical operation of the power system, and also plays an important role in reducing power generation costs, improving power quality, and optimizing power market dispatch planning. However, the power load is affected by various factors such as weather and holidays, which itself is a time series data, and traditional forecasting methods are difficult to find out the characteristics of the time series, so there is an urgent need for time series d...

AI Technical Summary

Problems solved by technology

However, when dealing with time series forecasting, the problem of gradient explosion is prone to occur, resulting in a decrease in prediction accuracy. The long-term short-term memory (LSTM) algorithm is used to solve the problem of gradient explosion, but the learning rate, the number of hidden layer neurons and the number of cycles, etc. Hyperparameters have a strong impact on the prediction results of the model and are difficult to determine; existing technologies have the following problems:

[0004] 1) In the conventional PSO, the inertia weight changes linearly, which may cause the particles to fall into the local optimum and skip the optimal solution in the later stage;

[0005] 2) In the conventional PSO, the inertia weight changes linearly, and the later weight is larger, which affects the convergence speed;

[0007] (1) Although the inertia weight is optimized nonlinearly, the learning factor is set to a fixed value, which ignores the effective effects of learning ability and social learning ability on particles in different stages of optimization, and reduces the optimization ability of particle swarm;

[0008] (2) Due to the fixed or simple processing of hyperparameter selection, it will affect the prediction results, resulting in insufficient short-term power load prediction accuracy, affecting power grid planning and power system security, and unable to maximize the benefits of the power supply system, resulting in economic waste

Images