Search

CN-121981770-A - Intelligent electricity price prediction cloud platform integrating multisource heterogeneous data and dynamic model update

CN121981770ACN 121981770 ACN121981770 ACN 121981770ACN-121981770-A

Abstract

The invention discloses an electricity price intelligent prediction cloud platform integrating multisource heterogeneous data and dynamic model updating, which comprises a multisource heterogeneous data acquisition and integration module, a dynamic model updating and collaborative training module, an electricity price intelligent prediction and credibility assessment module and a cloud platform service interface module, wherein the multisource heterogeneous data acquisition and integration module is configured to acquire multisource heterogeneous data from an electric power market operation system, a meteorological monitoring system, a new energy power generation system, a macroscopic economic database and a user electricity behavior system and perform unified data management and feature integration, the dynamic model updating and collaborative training module is configured to realize online updating and multi-model collaborative optimization of a prediction model based on incremental learning and transfer learning technology, the electricity price intelligent prediction and credibility assessment module is configured to generate an electricity price prediction result by adopting a multi-model integration prediction method and perform prediction credibility assessment based on the integrated feature data, and the cloud platform service interface module is configured to provide a standardized API interface and support multi-tenant access and personalized prediction service customization. The invention improves the accuracy and the robustness of electricity price prediction.

Inventors

  • ZHANG YUNFEI
  • PEI XUDONG
  • ZHANG XUBO
  • ZHANG RUIYUN

Assignees

  • 清能(西安)能源科技有限公司

Dates

Publication Date
20260505
Application Date
20260126

Claims (10)

  1. 1. An intelligent electricity price prediction cloud platform integrating multisource heterogeneous data and dynamic model update is characterized by comprising: The multi-source heterogeneous data acquisition and fusion module (1) is configured to acquire multi-source heterogeneous data from a plurality of data source systems, and perform unified data management and feature fusion on the acquired data to form a fusion feature data set; the dynamic model updating and collaborative training module (2) is configured to realize online updating and multi-model collaborative optimization of the prediction model based on an incremental learning and transfer learning technology; The electricity price intelligent prediction and credibility evaluation module (3) is configured to generate an electricity price prediction result by adopting a multi-model integrated prediction method based on the fusion characteristic data set, and perform credibility evaluation on the prediction result; The cloud platform service interface module (4) is configured to provide a standardized API interface, support multi-tenant access and personalized predictive service customization.
  2. 2. The intelligent electricity price prediction cloud platform for fusing multi-source heterogeneous data and dynamic model updating according to claim 1, wherein the multi-source heterogeneous data acquisition and fusion module (1) comprises: A multi-source data access unit (11) configured to collect real-time data and historical data from the electric power market operation system, the weather monitoring system, the new energy power generation system, the macro economic database and the user electricity behavior system through the standardized data interface; a data management and quality assessment unit (12) configured to perform cleaning, denoising, normalization and missing value processing on the acquired multi-source heterogeneous data, and to assess data quality; and the space-time feature fusion unit (13) is configured to extract space-time features of the multi-source data and realize deep fusion of the multi-source features through an attention mechanism and a feature crossing network.
  3. 3. The intelligent electricity price prediction cloud platform for fusing multi-source heterogeneous data and dynamic model updating according to claim 2, wherein the space-time feature fusion unit (13) is specifically configured to: decomposing the electricity price sequence into a plurality of intrinsic mode components by adopting a variation mode decomposition and sample entropy method; decomposing the original electricity price signal into high, medium and low frequency reconstruction sub-signals and a secondary decomposition sub-signal based on the reconstructed secondary decomposition-integration framework; and fusing the electricity price sequence characteristics with external influence factor characteristics through a cross attention mechanism to construct a multidimensional characteristic representation.
  4. 4. The electricity price intelligent prediction cloud platform integrating multi-source heterogeneous data and dynamic model update according to claim 1, wherein the dynamic model update and co-training module (2) comprises: a model dynamic updating unit (21) configured to update prediction model parameters in real time from newly arrived data based on online learning and incremental learning techniques; A multi-model collaborative training unit (22) configured to train a plurality of heterogeneous prediction models and fuse prediction results of the models by an adaptive weighted integration method; and the model performance monitoring unit (23) is configured to monitor the performance index of each prediction model in real time and trigger the model retraining and optimizing process.
  5. 5. The intelligent electricity price prediction cloud platform integrating multi-source heterogeneous data and dynamic model updating according to claim 4, wherein the multi-model co-training unit (22) is specifically configured to: training a heterogeneous model set comprising a Kerr Mo Ge Luov-Alord network, a long and short time memory network, a transformer network and a time sequence convolution network; Adopting an adaptive weighted regression method, and dynamically adjusting the integration weight according to the historical performance of each model in different market states; model hyper-parameters and integrated weight configuration are optimized based on a genetic algorithm.
  6. 6. The electricity price intelligent prediction cloud platform integrating multi-source heterogeneous data and dynamic model updating according to claim 1, wherein the electricity price intelligent prediction and credibility assessment module (3) comprises: a multi-model integrated prediction unit (31) configured to generate prediction results of a plurality of electricity price intervals using a prediction framework based on a classification network; A prediction reliability evaluation unit (32) configured to evaluate reliability of the prediction result based on consistency of the model prediction, the historical prediction accuracy, and the market volatility; And a prediction result visualization unit (33) configured to display the electricity price prediction result, the credibility evaluation, and the market analysis in a visual form.
  7. 7. The intelligent electricity price prediction cloud platform integrating multi-source heterogeneous data and dynamic model updating according to claim 6, wherein the prediction credibility assessment unit (32) is specifically configured to: triggering a space weight dynamic correction mechanism when the deviation between the grid cell predicted value and the adjacent cell actual electricity price exceeds a set threshold value; Judging the blocking state of the power network based on the blocking recognition model, and selecting different prediction models according to a blocking scene and a non-blocking scene; and calculating confidence intervals and probability distribution of the prediction results, and providing risk quantification indexes.
  8. 8. The intelligent electricity price prediction cloud platform integrating multi-source heterogeneous data and dynamic model updating according to claim 1, wherein the cloud platform service interface module (4) comprises: a multi-tenant management unit (41) configured to manage access rights, data isolation, and resource quotas of different users; A personalized service customization unit (42) configured to provide a customized predictive model and service configuration according to user type and demand; and the real-time data pushing unit (43) is configured to push electricity price prediction updating and market early warning information to a user in real time through a WebSocket and message queue technology.
  9. 9. An electricity price prediction method applied to the intelligent electricity price prediction cloud platform integrating multisource heterogeneous data and dynamic model update as claimed in any one of claims 1 to 8, which is characterized by comprising the following steps: collecting and fusing multi-source heterogeneous data to form a fused characteristic data set; Training and optimizing a multi-model prediction system based on a dynamic model updating mechanism; Generating an electricity price prediction result and carrying out credibility assessment; And providing prediction services and decision support for users through the cloud platform service interface.
  10. 10. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the electricity price prediction method of claim 9.

Description

Intelligent electricity price prediction cloud platform integrating multisource heterogeneous data and dynamic model update Technical Field The invention relates to the technical field of electric power market analysis and prediction, in particular to an intelligent electricity price prediction cloud platform integrating multisource heterogeneous data and dynamic model update. Background The electricity market price prediction is a core link of the operation of an electric power system and the trade of the electric power market, and the accurate electricity price prediction has important significance for power generation enterprises, electricity selling companies, power users and power grid operators. With the deep development of the electric power market and the large-scale access of new energy, the electricity price forming mechanism is more and more complex, and the prediction difficulty is obviously increased. The current electricity price prediction method is mainly divided into three types, namely a method based on traditional time series analysis, a method based on machine learning and a method based on deep learning. Traditional time series analysis methods such as autoregressive integral moving average (ARIMA) model, seasonal autoregressive integral moving average (SARIMA) model and the like are mainly suitable for predicting electricity price sequences with obvious regularity, but have limited prediction capability on nonlinear and non-stable electricity price sequences. The machine learning method such as a Support Vector Machine (SVM), random Forest (RF) and the like can capture nonlinear characteristics of electricity price through characteristic engineering and model training, but the characteristic extraction is highly dependent on expert experience, and the fusion capability of multidimensional characteristics is insufficient. Deep learning methods such as long and short time memory networks (LSTM), convolutional Neural Networks (CNN), transformers (transformers) and the like can automatically extract deep features of electricity price sequences, and show superior performance in electricity price prediction. The electricity market price is affected by a number of factors including electricity supply and demand relationships, fuel prices, weather conditions, new energy output, grid blockage conditions, market policies, and user behavior. These influencing factors come from different fields, and the data format and the update frequency are different, so that a typical multi-source heterogeneous data environment is formed. The multi-source heterogeneous data are effectively fused, and the complex relation between the multi-source heterogeneous data and the electricity price is excavated, so that the key for improving the electricity price prediction precision is realized. In the prior art, there have been some studies attempting to apply multi-source data to electricity price prediction. For example, a real-time electricity price prediction system based on a classification network, which is proposed by a medium energy company, acquires multi-source data comprising electricity price time sequence characteristics, external influence factor state information and electricity price fluctuation characteristics by monitoring an electricity market in real time. The method comprises the steps of respectively obtaining a first characteristic data set (comprising full-grid wind power output, full-grid photovoltaic output, urban wind speed and urban irradiation) and a second characteristic data set (comprising target node bidding space, target node tie line plan, target node wind power output, target node photovoltaic output and target node thermal power output) of a target node, and selecting different prediction models according to blocking identification results. However, the prior art still has the following limitations in the aspect of multi-source data fusion, namely firstly, multi-source data are often processed independently, a unified fusion frame is lacking, association relations among different data sources are difficult to fully mine, secondly, a data fusion method is simpler, spatial-temporal heterogeneity of the multi-source data cannot be processed effectively in a characteristic splicing or weighted average mode, and finally, a system evaluation and treatment mechanism for multi-source data quality is lacking, and low-quality data can reduce prediction performance. The electric power market environment has high dynamic property and uncertainty, and factors such as market rule adjustment, new energy permeability improvement, user behavior change and the like can cause the change of an electricity price forming mechanism, so that a static prediction model trained based on historical data is gradually invalid. Therefore, the dynamic update of the prediction model is realized, so that the prediction model can adapt to the change of the market environment, and the prediction model is an importan