Short-term Power Load Forecasting Based On CSWOA Optimized BiGRU Fusion Temporal Pattern Attention Mechanism

Short-term power load forecasting provides an important basis for the dispatching department to formulate the daily dispatching plan and power generation plan.At the same time,accurate power load forecasting is very important for the safe and reliable operation of the power grid.Due to the obvious nonstationary,nonlinear and periodic characteristics of power load,how to accurately predict power load is a very challenging problem.In order to better extract the periodicity of load data,process the nonlinearity of data and overcome the nonstationarity of data,this thesis proposes a combined forecasting method combining bidirectional gated recurrent unit（BiGRU）,temporal pattern attention mechanism（TPA）and crisscross whale optimization algorithm（CSWOA）.This thesis constructs a short-term power load forecasting model based on CSWOA-TPA-BiGRU.（1）Power load is affected by many factors and has strong time correlation,but different historical moments and different factors have different effects on load forecasting.A shortterm power load short-term forecasting model combining BiGRU and TPA is proposed.BiGRU is used to obtain the deep implicit relationship between load feature series and load,and TPA is used to further mine the time relationship between different features and load in the feature series,so as to enhance the sensitivity of key historical time nodes.Simulation results show that TPA can effectively perceive the key historical time nodes and improve the performance of short-term power load forecasting model of BiGRU.（2）Aiming at the problems of many blind areas,poor search ability and easy to fall into local optimization of whale optimization algorithm,a crisscross whale optimization algorithm based on crisscross optimization algorithm is proposed.The horizontal crossover operator is introduced into the original whale optimization algorithm to increase the way of information exchange and enhance its global search ability.By introducing the vertical crossover operator,the whale algorithm can avoid falling into local optimization and improve the optimization ability in the solution space.test function of CEC2017 is used to verify the effectiveness and superiority of CSWOA.（3）Aiming at the problem that load forecasting model of TPA-BiGRU is easy to fall into local optimization after training with gradient descent method,and the intelligent optimization algorithm cannot effectively optimize or the optimization time is too long due to too many weight coefficients and bias parameters of the forecasting model,a short-term power load forecasting model of CSWOA-TPA-BiGRU using CSWOA to optimize the weight and bias parameters of TPA-BiGRU full connection layer is proposed.In the process of optimization,in order to avoid over fitting of the model,double indexes are used together and the regular term of the weight to be optimized is combined as the objective fitness function.Two regional load data sets provided by the 9th EMCM are used for experiments are used for simulation modeling.The results show that the CSWOA-TPABiGRU short-term power load forecasting model proposed in this thesis is better than other comparison models,and has strong generalization ability.