Research On Power Load Prediction And Cooperative Control Strategy Based On Data Mining

With the goal of carbon peaking and carbon neutrality,the proportion of renewable energy in primary energy consumption continues to increase,and the new power system will develop into a more flexible,open,and highly intelligent energy Internet system.However,with the characteristics of uncertainty and volatility of renewable energy and the disadvantages of high cost,low efficiency and slow response speed of traditional power generation regulation making the real-time supply and demand balance of the new power system face severe challenges,the power grid needs to exploit the potential of demand-side resource,implement electricity market reforms and demand response,to supplement the insufficiency of traditional regulation methods,and improve its own flexibility to cope with the uncertainty of renewable energy.Aiming at the problem that it is difficult to effectively extract potential high-dimensional features in historical sequences and important information is easily lost,this paper firstly decomposes the daily electricity consumption sequence into trend items,periodic items,and random items,and analyzes its dynamic change law.Then a CNN architecture composed of one-dimensional convolution layers and pooling layers is built to extract the high-dimensional features reflecting complex dynamic changes,which can be constructed into timeseries as GRU input to model the internal dynamic change of the feature.Then the attention mechanism is introduced to assign different weights to GRU hidden states through mapping weight and learning parameter matrix,to reduce the loss of historical information and enhance the impact of important information.The prediction accuracy achieved 97.15%and 97.44%respectively,when tested the model with electric load dataset from a US nonprofit orgnization and a Power Grid Company in Northwest China.The results show that the method can effectively improve the accurary of electric load prediction.Aiming at the problem of multi-point error accumulation and complex model caused by existing sequence decomposition algorithms,this paper proposes an integrated learning prediction based on differential decomposition and error compensation.First,an integrated learning framework based on ANN is proposed to reduce the prediction errors of a single learner.Then the firstorder difference of the original sequence is used as an input feature to transform the load prediction problem into the variation prediction problem.And multiple sets of fake sequences are introduced based on a set of actual load sequences,then the integrated network is used to construct a multi-objective iterative prediction network.Last considering the changing trend and stationarity of the iterative errors of each sequence,an error compensation network based on the sequence similarity and artificial neural network integrated model is constructed.Lastly,we tested the model with electric load dataset from a US nonprofit orgnization and a Power Grid Company in Southeast China.The results show that the method can effectively improve the accurary of electric load prediction.Aiming at the sharp decline of the adjustable capacity caused by the load overshoot,this paper designs a multi-objective regulation and optimization strategy of temperature-controlled load considering load regulation margin balance.According to the load prediction,the new energy consumption target is formulated,and response priority is calculated by quantifying the impact of different load behaviors on the load tunable margin of the next time slot.Then,combined with the multi-level incentive mechanism,an optimization objective function is established to minimize incentive cost and the influence of the control behavior on the adjustable margin.And the multi-objective optimization problem is transformed into a single-objective programming problem by using the method of combined weights,and the optimal solution is obtained,which can suppress the rapid decline of the adjustable capacity and the insufficient consumption of renewable energy caused by overshoot.Finally,simulation experiments under different strategies,different system scales,different durations,and different weights of the proposed strategy can guarantee the adjustable margin of reliable equalization in the time slot.Aiming at the problem that it is difficulty to meet the differentiated comfort and economic needs at the same time due to that traditional load control straregies are mostly oriented to group optimization and ignore the preference of individuas.Therefore,this paper proposes a cluster air-conditioning load collaborative control strategy considering the differentiated energy consumption needs.For summer peak power consumption under load forecasting,taking the air-conditioning load as the research object,a LSTM-based differential energy demand simulation is established firstly.Then this paper introduces the similarity of energy consumption behavior to measure the degree of compliance with the differentiated needs,and adopts DQN to formulate personalized energy consumption strategies,which can reduce energy costs and meet the comfort requirements of individuals,and effectively reduce the peak-to-valley difference.Finally,the effectiveness and advantages are verified by simulation analysis.