CN-122022881-A - Mid-long term-spot electricity price collaborative prediction method and system based on regional feature image

CN122022881ACN 122022881 ACN122022881 ACN 122022881ACN-122022881-A

Abstract

The invention discloses a medium-long term-spot electricity price collaborative prediction method and a system for regional characteristic images, wherein the method comprises the steps of collecting multisource data related to electricity price in a region, and carrying out fusion processing to obtain a time-space aligned multisource data set; the method comprises the steps of constructing a dynamic weighted regional feature image library based on a space-time aligned multisource data set to obtain a structural storage feature matrix, constructing a trans-regional knowledge migration model, taking the dynamic weighted regional feature image library as input of the trans-regional knowledge migration model to obtain grouped feature groups, acquiring training data based on the grouped feature groups to train a medium-long-spot collaborative prediction model, predicting electricity price intervals and hour-level node electricity prices based on the regional feature image library through the trained medium-long-spot collaborative prediction model, and optimizing the model through forward constraint and reverse correction collaborative mechanism. The invention provides a unified and self-adaptive electricity price collaborative prediction solution for different area electric power markets.

Inventors

YUAN JIE
GAO ZHUO
QIN HONGXIA
LI YUNBO

Assignees

北京四方继保自动化股份有限公司
北京四方继保工程技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (12)

1. The medium-long term-spot electricity price collaborative prediction method based on the regional characteristic image is characterized by comprising the following steps of: acquiring multisource data related to electricity prices in the area, and carrying out fusion processing on the multisource data to obtain a multisource data set with time-space alignment; Constructing a dynamic weighted regional feature image library based on the space-time aligned multi-source data set to obtain a feature matrix of the structured storage; Constructing a cross-regional knowledge migration model, taking a dynamic weighted regional feature image library as input of the cross-regional knowledge migration model to obtain grouped feature groups, and acquiring training data based on the grouped feature groups to train the middle-long-term-spot collaborative prediction model; And predicting the electricity price interval and the hour-level node electricity price based on the regional feature portrait library by using the trained medium-long term-spot collaborative prediction model, and optimizing the model by using a forward constraint and reverse correction collaborative mechanism.
2. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 1, characterized in that, The multi-source data comprise electric power market transaction data, new energy output data, power grid basic data and auxiliary data.
3. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 1, characterized in that, The construction of the dynamic weighted regional characteristic image library specifically comprises the following steps: Constructing space-time characteristic three-dimensional vectors based on space-time aligned multi-source data sets, and quantizing to form a regional power characteristic image library; Calculating feature weights of the time-space feature three-dimensional vectors by adopting a double-layer attention mechanism to obtain a model input feature set; And (3) constructing a dynamic weighted regional feature image library, namely fusing the feature importance weight and the time attenuation weight to form dynamic weight configuration of each feature of each region, storing the dynamic weight configuration in association with feature vectors in the regional power feature image library, and updating to obtain the dynamic weighted regional feature image library.
4. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 3, characterized in that, The construction of the space-time characteristic three-dimensional vector specifically comprises the following steps: Calculating according to the multi-source data to obtain characteristic class data, wherein the characteristic class data comprises market class data, new energy class data, load class data, power grid class data, weather class data, policy class data and power source class data; based on the time-space aligned multisource data set and the feature class data, a time-space feature three-dimensional vector form is obtained.
5. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 3, characterized in that, The feature weight calculation method adopting the double-layer attention mechanism specifically comprises the following steps: The attention weight is calculated through the feature importance attention layer, the time attenuation weight is calculated through the time attenuation attention, and the two weights are combined to obtain a weighted feature matrix obtained through weighted aggregation, specifically as follows: The characteristic matrix at the historical moment is subjected to weighted aggregation according to the time attenuation weight to obtain a time attenuation weighted comprehensive characteristic matrix: wherein F represents a time decay weighted composite feature matrix, Is a feature matrix weighted by the first layer of attention, And T is the time step number, and is the time decay weight.
6. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 3, characterized in that, The acquisition of the dynamic weighting characteristic configuration set is specifically as follows: obtaining the feature importance weight corresponding to the feature f of the node n at the time step t through the feature importance attention layer ; The attenuation weight of each historical time step T relative to the current moment T is obtained through a time attenuation attention layer ; Combining the feature importance weights and decay weights to form dynamic weights for each feature f of each node n : Wherein, the Is an importance decay coefficient, and gamma is [0,1]; The representation is The weight value of the vector f-th dimension; Storing the dynamic weighted feature representation of each node n as a feature vector and a corresponding dynamic weight vector And (3) forming a dynamically weighted regional feature image library.
7. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 1, characterized in that, The method for obtaining the grouped feature groups by taking the dynamic weighted regional feature image library as the input of the cross-regional knowledge migration model specifically comprises the following steps: Acquiring three-dimensional feature vectors of all areas from a dynamic weighted area feature image library; Carrying out data preprocessing on the three-dimensional feature vector, wherein the data preprocessing comprises standardization processing, so that features of different dimensions are in the same dimension; Carrying out fusion clustering on the preprocessed data, namely carrying out clustering grouping by adopting a K-means and hierarchical clustering fusion algorithm; And analyzing the characteristics of each group according to the clustering result, and marking the characteristics as corresponding regional characteristic groups to obtain grouped characteristic groups.
8. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 1, characterized in that, The group-based feature group acquisition training data trains the middle-long-term-spot collaborative prediction model, and specifically comprises the following steps: Constructing a medium-long-term-spot collaborative prediction model, for each regional characteristic group, selecting data of one region as a reference to train the medium-long-term-spot collaborative prediction model, and taking the trained model as a sharing basic model of the group; When the target area needs to construct a prediction model but has insufficient data, matching the target area to the feature group with highest similarity by calculating the similarity between the target province feature vector and the mass centers of all groups according to the three-dimensional feature vector; and migrating the reference prediction model corresponding to the matched group to a target area, and performing model fine adjustment by utilizing the data of the target area to obtain a trained medium-long term-spot collaborative prediction model for prediction.
9. The method for collaborative prediction of medium-long term-spot price of electricity based on regional feature images according to claim 1, characterized in that, The medium-long term-spot collaborative prediction model after training predicts the electricity price interval and the hour level node electricity price based on the regional feature portrait library, and specifically comprises the following steps: the medium-long term-spot collaborative prediction model comprises a time sequence deep learning network and a gradient lifting decision tree framework with constraint; The regional characteristic image library is used as the input of a medium-long term-spot cooperative prediction model, the medium-long term prediction is carried out based on a time sequence deep learning network, the regional characteristic image library is output as a predicted electricity price interval, the spot prediction is carried out based on a constrained gradient lifting decision tree frame, and the regional characteristic image library is output as a predicted hour-level node electricity price.
10. A mid-long term-spot electricity price collaborative prediction system based on regional feature images, for implementing the mid-long term-spot electricity price collaborative prediction method based on regional feature images according to any one of claims 1-9, comprising: The acquisition module is used for acquiring multisource data related to electricity prices in the area and carrying out fusion processing on the multisource data to obtain a space-time aligned multisource data set; The data processing module is used for constructing a dynamic weighted regional characteristic image library based on the multi-source data set aligned in time and space to obtain a structural storage characteristic matrix; the grouping training module is used for constructing a cross-regional knowledge migration model, taking a dynamic weighted regional feature image library as input of the cross-regional knowledge migration model, obtaining grouping feature groups, and acquiring training data based on the grouping feature groups to train the long-term and spot collaborative prediction model; The prediction module predicts the electricity price interval and the hour-level node electricity price based on the regional feature portrait library through the trained medium-long term-spot collaborative prediction model, and optimizes the model through a forward constraint and reverse correction collaborative mechanism.
11. A terminal comprises a processor and a storage medium, and is characterized in that: The storage medium is used for storing instructions; The processor is operative according to the instructions to perform the steps of the regional feature representation-based medium-long term-spot price co-prediction method according to any one of claims 1-9.
12. A computer-readable storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the regional feature image-based medium-long term-spot price collaborative prediction method according to any of claims 1-9.

Description

Mid-long term-spot electricity price collaborative prediction method and system based on regional feature image Technical Field The invention relates to the technical field of electricity price prediction, in particular to a medium-long term-spot electricity price collaborative prediction method and system based on regional characteristic images. Background Current electric market price prediction research is focused mainly on a single province or a single market time scale. On the one hand, the mainstream method often builds a nationally oriented prediction model but in units of independent provinces, and on the other hand, research is mainly focused on single time scale prediction only for medium-long term markets or only for spot markets. However, this paradigm of research suffers from significant drawbacks: on the one hand, the existing system is lack of cross-province feature fusion and dynamic weighting mechanisms, and the existing system is generally lack of systematic extraction and quantitative characterization of 'provincial difference core influencing factors' (such as new energy output features, load forms, market policy guidance and the like of the provincial/temporary provincial). More importantly, the failure to dynamically assign differential weights to these features of different provinces according to a specific scenario results in a model with limited ability to characterize cross-power-saving price linkage and provincial domain heterogeneity. On the other hand, the multi-time scale collaborative prediction path is split, medium-long term price and spot price have profound association and mutual influence, but the existing method mostly regards the medium-long term price and the spot price as independent prediction problems. The lack of a cooperative mechanism for effectively restraining spot prediction fluctuation of the medium-long term result and simultaneously utilizing spot high-frequency characteristics to back feed the medium-long term model limits the consistency and the accuracy of the prediction result. Finally, it is difficult to adapt to the needs of complex market bodies, and along with the activity of trans-provincial power trading, diversified bodies such as power generators, electric sellers, energy storage users, large users and the like with trans-provincial trading qualification are needed, and unified prediction support capable of covering multiple provinces and medium-long-term-spot double markets simultaneously is needed. The existing method is focused on a single-province single-scale prediction model, and the requirements of the single-province single-scale prediction model on the consistency and the adaptability of electricity price prediction when a trans-regional trans-time-scale transaction strategy is specified are difficult to meet. Disclosure of Invention In order to solve the defects in the prior art, the invention provides a medium-long term-spot electricity price collaborative prediction method and a system based on regional characteristic images, which realize unified high-precision prediction of electricity prices with multiple provinces and multiple time scales by constructing a provincial image quantization index library and establishing a two-way feedback mechanism. The invention adopts the following technical scheme. A medium-long term-spot electricity price collaborative prediction method based on regional characteristic images comprises the following steps: acquiring multisource data related to electricity prices in the area, and carrying out fusion processing on the multisource data to obtain a multisource data set with time-space alignment; Constructing a dynamic weighted regional feature image library based on the space-time aligned multi-source data set to obtain a feature matrix of the structured storage; Constructing a cross-regional knowledge migration model, taking a dynamic weighted regional feature image library as input of the cross-regional knowledge migration model to obtain grouped feature groups, and acquiring training data based on the grouped feature groups to train the middle-long-term-spot collaborative prediction model; And predicting the electricity price interval and the hour-level node electricity price based on the regional feature portrait library by using the trained medium-long term-spot collaborative prediction model, and optimizing the model by using a forward constraint and reverse correction collaborative mechanism. Preferably, the multi-source data comprises electric power market transaction data, new energy output data, power grid basic data and auxiliary data. Preferably, the constructing a dynamic weighted regional feature image library specifically includes: Constructing space-time characteristic three-dimensional vectors based on space-time aligned multi-source data sets, and quantizing to form a regional power characteristic image library; Calculating feature weights of the time-space feature three-dimensional