Power grid regulation and control model training method and system, computer device, and storage medium

Abstract

The present invention relates to the field of power automation. Disclosed are a power grid regulation and control model training method and system, a computer device, and a storage medium. The method comprises: acquiring training data of a power grid regulation and control model, and grouping the training data to obtain a plurality of training data groups; training a plurality of preset learning bodies in parallel by means of the plurality of training data groups on the basis of a one-to-one correspondence between the training data groups and the learning bodies to obtain a plurality of execution bodies, wherein when the learning bodies are trained by means of the training data groups, the training data groups are divided into a plurality of training data units, a training effect of each training data unit is acquired to obtain a shared training data unit, and the shared training data unit is shared with other training data groups than the training data group to which the shared training data unit belongs; and finally, combining the plurality of execution bodies to obtain a power grid regulation and control model. By means of parallel training and parallel prediction, quality is improved on the basis of quantity, and shared learning of high-quality sample data is achieved, thereby effectively improving a training effect, and being capable of effectively reducing the training time of the power grid regulation and control model under a big data condition.

Description

Power grid control model training method, system, computer equipment and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is based on the Chinese patent application with application number 202310803693.7 and application date of July 3, 2023, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby introduced into this application in its entirety. Technical Field

[0003] The present invention belongs to the field of electric power automation and relates to a power grid control model training method, system, computer equipment and storage medium. Background Art

[0004] In the context of the development of new power systems, power grid control faces the challenge of aggregated analysis and unified computation of massive amounts of data. From a technological perspective, new technologies such as big data, cloud computing, and artificial intelligence (AI) have entered a period of rapid development. Data-driven approaches, leveraging massive amounts of data input, offset the shortcomings of model-driven approaches and effectively address issues such as nonlinearity, discontinuity, uncertainty, and randomness in power grids. However, the current application of AI in power grids still faces limitations, including poor model interpretability and unstable analysis and prediction results. These limitations hinder the stable, reliable, and secure operation of power grid control systems. This is because, without accurate and complete data, the large-scale application of AI in power grids presents risks. Consequently, data-centric artificial intelligence (AI) has become a trend across various industries. This data-centric approach argues that AI has reached a point where data is more important than models. If AI is viewed as a system with moving parts, the model should remain relatively fixed, focusing on fine-tuning it with high-quality data rather than continually pushing for marginal improvements.

[0005] Although power grid operations have accumulated vast amounts of data resources, the value density of grid control big data is low, resulting in a shortage of high-quality sample data suitable for training specific business models. In particular, data primarily focuses on normal operation, while data with unique characteristics, such as anomalies or faults, is scarce, resulting in low data quality. This leads to sample imbalance or small sample sizes in some machine learning tasks, which in turn impacts the training process and post-release prediction performance of AI-based grid control models. While data quality is improved after validation and cleansing of grid control big data, data validation is often based on simplified validation rules, and data cleansing often employs equivalence or approximate calculation methods, failing to ensure data quality meets the training requirements of all AI business models. Furthermore, big data generally utilizes distributed parallel computing technologies, while machine learning typically relies on in-memory computing. Therefore, large data cannot be directly loaded into memory for machine learning model training. Therefore, given the inability to fully guarantee data quality and balance, how to leverage massive control data resources to maximize the training and post-release prediction performance of grid control models remains an urgent challenge.

[0006] Summary of the Invention

[0007] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a power grid control model training method, system, computer equipment and storage medium.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] In a first aspect, the present invention provides a power grid regulation model training method, comprising: obtaining power grid regulation model training data and grouping them to obtain a number of training data groups; according to the one-to-one correspondence between the training data groups and the learning bodies, training a number of preset learning bodies in parallel through the several training data groups to obtain a number of execution bodies; wherein, when training the learning body through the training data group, the current training data group is divided into a number of training data units, and the current learning body is trained in sequence through the several training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups except the training data group to which it belongs; wherein the several learning bodies are a number of power grid regulation business AI models constructed according to the power grid regulation needs; and combining a number of execution bodies to obtain a power grid regulation model; wherein the combination includes taking the inputs of the several execution bodies together as the input of the power grid regulation model, and fusing the outputs of the several execution bodies as the output of the power grid regulation model.

[0010] Optionally, the method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: according to the one-to-one correspondence between the training data groups and the learning bodies, training a plurality of preset learning bodies in parallel through a plurality of training data groups in combination with preset evaluation bodies corresponding one-to-one to each learning body; and after the training is completed, obtaining the preset prediction indicators of each learning body through each evaluation body, and taking the learning body whose preset prediction indicators meet the preset prediction indicator training requirements as the execution body.

[0011] Optionally, the several power grid regulation business AI models are constructed using two or more machine learning algorithms or artificial intelligence algorithms.

[0012] Optionally, each of the several power grid regulation business AI models is a power grid regulation business AI model that performs the same type of power grid regulation business, or a power grid regulation business AI model that performs complementary power grid regulation business.

[0013] Optionally, it also includes: periodically obtaining the first preset prediction index value of each executor in the power grid control model for the real-time data of the power grid control model; when the first preset prediction index value of an executor in the power grid control model fails to meet the preset first preset prediction index value requirement for a preset number of consecutive times, the executor is removed from the power grid control model, or the executor is retrained using the power grid control model training data.

[0014] Optionally, it also includes: obtaining real-time data of the power grid control model in which the second preset prediction index value of the execution body is lower than the preset second preset prediction index value prediction requirement, and obtaining data with key characteristics; and sharing the data with key characteristics as power grid control model training data to each training data group to obtain several updated training data groups; according to the one-to-one correspondence between the updated training data group and the learning body, training several preset learning bodies in parallel through several updated training data groups to obtain several updated execution bodies; and using several updated execution bodies to replace several execution bodies in the power grid control model.

[0015] Optionally, the parallel training of the preset several learning bodies through several updated training data groups also includes: obtaining relevant data of the data with key features, and sharing the relevant data of the data with key features as power grid control model training data to each updated training data group; wherein, the relevant data includes: data that has a time relationship with the data with key features, data of devices that have a topological connection relationship with the devices corresponding to the data with key features, and data that has a business logic relationship with the data with key features.

[0016] Optionally, the parallel training of several preset learning bodies through several updated training data groups also includes: using a data feature analysis method to extract features of data with key features to obtain extracted features; obtaining an optimization method, and optimizing the power grid control model training data or several learning bodies in each updated training data group according to the optimization method; wherein the optimization method adopts a manual analysis method to obtain the test evaluation effect of each learning body and the reasoning effect of each execution body, and obtains it based on the extracted features, combined with the test evaluation effect of each learning body, the reasoning effect of each execution body and business domain knowledge.

[0017] Optionally, the optimization method includes one or more of the following: changing the data range of the grid control model training data, changing the data dimension of the grid control model training data, changing the data verification rules of the grid control model training data, changing the data cleaning rules of the grid control model training data, and changing the number of learning bodies.

[0018] Optionally, the fusion of the outputs of several executors as the output of the power grid control model includes: weighting and superimposing the outputs of several executors according to the preset weighting coefficients of each executor as the output of the power grid control model; or processing the outputs of several executors according to a scoring method to obtain the output of the power grid control model.

[0019] Optionally, it also includes: dynamically adjusting the preset weighting coefficients of each executor at a dynamic frequency; the preset weighting coefficients of each executor are adjusted by: obtaining the prediction accuracy or prediction precision of each executor; and adjusting the weighting coefficients of each executor based on the prediction accuracy or prediction precision of each executor using factor analysis, principal component analysis, AHP hierarchical method, priority diagram method, entropy method, CRITIC weight method, independence weight method or information weight method.

[0020] According to a second aspect of the present invention, a power grid regulation model training system is provided, comprising: a data acquisition module for acquiring power grid regulation model training data and grouping them to obtain a number of training data groups; a model training module for training a number of preset learning bodies in parallel through a number of training data groups in accordance with a one-to-one correspondence between training data groups and learning bodies to obtain a number of execution bodies; wherein, when training a learning body through a training data group, the current training data group is divided into a number of training data units, and the current learning body is trained in sequence through a number of training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups other than the training data group to which it belongs; wherein, the number of learning bodies are a number of power grid regulation business AI models constructed according to power grid regulation requirements; a model assembly module for combining a number of execution bodies to obtain a power grid regulation model; wherein, the combination includes taking the inputs of the number of execution bodies together as the input of the power grid regulation model, and fusing the outputs of the number of execution bodies as the output of the power grid regulation model.

[0021] Optionally, the method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: according to the one-to-one correspondence between the training data groups and the learning bodies, training a plurality of preset learning bodies in parallel through a plurality of training data groups in combination with preset evaluation bodies corresponding one-to-one to each learning body; and after the training is completed, obtaining the preset prediction indicators of each learning body through each evaluation body, and taking the learning body whose preset prediction indicators meet the preset prediction indicator training requirements as the execution body.

[0022] Optionally, it also includes a retraining module, which is used to obtain real-time data of the power grid control model in which the second preset prediction index value of the execution body is lower than the preset second preset prediction index value prediction requirement, and obtain data with key characteristics; and share the data with key characteristics as power grid control model training data to each training data group to obtain several updated training data groups; according to the one-to-one correspondence between the updated training data group and the learning body, train several preset learning bodies in parallel through several updated training data groups to obtain several updated execution bodies; and use several updated execution bodies to replace several execution bodies in the power grid control model.

[0023] Optionally, the parallel training of several preset learning bodies through several updated training data groups also includes: using a data feature analysis method to extract features of data with key features to obtain extracted features; obtaining an optimization method, and optimizing the power grid control model training data or several learning bodies in each updated training data group according to the optimization method; wherein the optimization method adopts a manual analysis method, which is obtained based on the extracted features and combined with business domain knowledge.

[0024] In a third aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned power grid control model training method when executing the computer program.

[0025] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned power grid control model training method are implemented.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The grid control model training method of the present invention obtains grid control model training data and groups it, then performs group training on several learning bodies to obtain several execution bodies, and then combines several execution bodies to obtain a grid control model. It fully considers the characteristics of the grid control model training data in a big data environment, such as large data volume, multiple data types, and long-period historical data. Through parallel training and parallel prediction, it achieves quantity-for-quality, greatly reduces the time required for model training, and improves the training effect and prediction effect of the grid control model. At the same time, in order to address the problems of few high-quality samples, unbalanced samples, or small samples in the grid control model training data, the training data group is divided into several training data units, and high-quality samples are screened based on the training effect of each training data unit and the preset training effect requirements, that is, the training data units are shared and shared to each training data group, so as to achieve shared learning of high-quality sample data, effectively improve the training effect, and further reduce the training time of the grid control model under big data conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings herein are incorporated into the specification and constitute a part of this specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solutions of the present application.

[0029] FIG1 is a flow chart of a grid control model training method according to an embodiment of the present invention.

[0030] FIG2 is a schematic diagram showing the principles of a power grid control model training method according to an embodiment of the present invention.

[0031] FIG3 is a schematic diagram of the detailed flow of the power grid control model training method according to an embodiment of the present invention.

[0032] FIG4 is a schematic diagram of the principle of parallel training of learning bodies according to an embodiment of the present invention.

[0033] FIG5 is a schematic diagram of the principle of modifying the weight coefficient of the executive body according to an embodiment of the present invention.

[0034] FIG6 is a structural block diagram of a power grid control model training system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0037] The present invention is described in further detail below with reference to the accompanying drawings:

[0038] Referring to Figures 1 to 4, an embodiment of the present invention provides a power grid control model training method. This method, through parallel training and parallel prediction, enables parallel training and prediction of power grid control models based on control big data, improves the training and prediction effects of power grid control models under data uncertainty conditions, and achieves a quality-for-quantity approach in a big data environment. Specifically, the power grid control model training method includes the following steps:

[0039] S1: Obtain and group the training data of the power grid control model to obtain several training data groups.

[0040] S2: According to the one-to-one correspondence between the training data groups and the learning bodies, a plurality of preset learning bodies are trained in parallel through a plurality of training data groups to obtain a plurality of execution bodies.

[0041] Among them, when training a learning body through a training data group, the current training data group is divided into several training data units, and the current learning body is trained in sequence through several training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups except the training data group to which it belongs; among them, several learning bodies are several power grid control business AI models constructed according to power grid control needs.

[0042] S3: Combine several executive bodies to obtain a power grid control model.

[0043] The combination includes taking the inputs of several executive bodies as the input of the power grid control model, and fusing the outputs of several executive bodies as the output of the power grid control model.

[0044] Control big data generally refers to the collection of data related to power grid operations compiled from the power grid control big data platform. This includes all information records related to power grid dispatch and operation, including power grid model data, monitoring data, alarm data, and operation records. Its data scale meets the 5V characteristics of big data, and its collection and processing processes utilize big data platform technologies and related big data algorithms. Power grid control model training data can be obtained from control big data based on different control business requirements. Specifically, business data and related data can be filtered according to the requirements of specific power grid control business AI models, and then data fusion technology can be used to form the power grid control model training data set required for power grid control model training.

[0045] Specifically, because the historical operational data volume of power grid control big data is relatively large, its quality and balance are difficult to accurately assess. To fully leverage the massive and diverse characteristics of big data, the massive power grid control model training data is divided into subsets and fed into different learning bodies for learning. Furthermore, crossover between different training data groups is permitted.

[0046] Specifically, the number of learning bodies can be dynamically expanded based on the amount of data to be learned, the time required for learning, etc. Containerized deployment can be used to enable rapid deployment and elastic scaling of learning bodies.

[0047] Specifically, in the specific training process of the learning body, the learning process of the learning body can be carried out in stages, that is, the training data group is divided into several training data units for stage-by-stage learning. For example, the simplest division is based on the time of the data, and each training data unit is learned separately. At the same time, the learning effect of each training data unit is determined based on the training effect of each training data unit on the current learning body. The training data units with better learning effects can be shared with other training data groups to improve the learning effects of other learning bodies.

[0048] In summary, the grid control model training method of the present invention obtains grid control model training data and groups them, and then performs group training on several learning bodies to obtain several execution bodies, and then combines several execution bodies to obtain a grid control model. It fully considers the characteristics of the grid control model training data in a big data environment, such as large data volume, multiple data types, and long-period historical data. Through parallel training and parallel prediction, it achieves quantity for quality, greatly reduces the time required for model training, and improves the training effect and prediction effect of the grid control model. At the same time, in order to address the problems of few high-quality samples, unbalanced samples, or small samples in the grid control model training data, the training data group is divided into several training data units, and high-quality samples are screened based on the training effect of each training data unit and the preset training effect requirements, that is, the training data units are shared and shared to each training data group, so as to achieve shared learning of high-quality sample data, effectively improve the training effect, and further reduce the training time of the grid control model under big data conditions.

[0049] In one possible embodiment, the method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: according to a one-to-one correspondence between the training data groups and the learning bodies, training a plurality of preset learning bodies in parallel through a plurality of training data groups in combination with preset evaluation bodies corresponding one-to-one to each learning body; and after the training is completed, obtaining a preset prediction indicator for each learning body through each evaluation body, and using the learning body whose preset prediction indicator meets the preset prediction indicator training requirements as the execution body.

[0050] Specifically, after the training is completed, the preset prediction indicators of each learning body are obtained. The preset prediction indicators can be the prediction accuracy or the prediction precision. The preset prediction indicators of each learning body can be obtained by setting a test set. Optionally, a separate evaluation body can be set for each learning body to evaluate the learning effect of each learning body. The evaluation results of the evaluation body affect the output fusion process of the execution body. Optionally, learning and evaluation are a cyclic feedback process, that is, training and testing can be iterated until a certain number of learning bodies meet the needs of the regulation business, that is, the preset prediction indicators of these learning bodies reach the preset threshold requirements, and then the parameters in the learning body are solidified and the required environment and dependencies are packaged together and encapsulated as an execution body.

[0051] In addition, training can be a continuous process. For example, after the power grid control model is released, the training of the learning body can continue. The management of various versions of the learning body can be completed through version management, and the power grid control model can be upgraded and released based on the forecast of the control business.

[0052] Among them, the evaluation body is a module that evaluates the learning effect of the learning body. The evaluation body and the learning body serve as learning units together. The evaluation results of the evaluation body will affect the subsequent application of the execution body trained based on the learning body, such as affecting the weight of the corresponding execution body.

[0053] When training a learning entity, there is a cyclic feedback process between the learning entity and the evaluation entity, and the learning effect of the learning entity is improved based on the evaluation of the evaluation entity.

[0054] In one possible implementation, the plurality of power grid control business AI models are constructed using two or more machine learning algorithms or artificial intelligence algorithms. Specifically, using different algorithms to model the learning body can effectively reduce the impact of the algorithm itself on the learning effect of the learning body.

[0055] In a possible implementation, each of the several grid regulation business AI models may be a grid regulation business AI model that performs the same type of grid regulation business, or a grid regulation business AI model that performs complementary grid regulation business.

[0056] Specifically, the executor is obtained by encapsulating and publishing the learning body, and multiple executors are combined to perform parallel prediction to obtain the power grid control model. Based on the design of the power grid control business AI model, if each power grid control business AI model is a power grid control business AI model that performs the same power grid control business, then the executor obtained after learning and encapsulation is an executor that performs the same power grid control business. This is to improve the accuracy of prediction through group decision-making. If each power grid control business AI model is a power grid control business AI model that performs complementary power grid control business, then the executor obtained after learning and encapsulation is an executor that performs complementary power grid control business. This is to perform predictive analysis from multiple angles and improve the accuracy of prediction.

[0057] For similar grid control services, such as bus load forecasting, one grid control AI model is trained using the LSTM algorithm, while the other is trained using the BP neural network algorithm. These two grid control AI models differ only in their algorithms, but both perform the same task of bus load forecasting. Therefore, they are called grid control AI models performing similar grid control services. Alternatively, using the same algorithm but different training data results in different executables, but they still perform similar tasks. The resulting executable combinations are flexible and can be adjusted dynamically based on the grid control service.

[0058] For complementary grid control services, one AI model can predict which equipment on the grid will fail, along with the probability and timing of failure. Another predicts the impact of a failure. Another predicts recovery time after a failure. These three services, each with their own focus and overlapping capabilities, can be combined to form complementary subtasks within "grid equipment failure impact analysis."

[0059] During actual prediction, the real-time data of the power grid control model provided by the big data platform is input into each execution body respectively, and the execution results of the execution bodies are integrated to obtain the final prediction result and output it.

[0060] In one possible embodiment, the grid control model training method further includes: periodically obtaining a first preset prediction index value of each executor in the grid control model for the real-time data of the grid control model; when there is an executor in the grid control model whose first preset prediction value fails to meet the preset first preset prediction index value requirement for a preset number of consecutive times, the executor is removed from the grid control model, or the executor is retrained using the grid control model training data.

[0061] Specifically, over time, the accuracy of previous predictions, or their accuracy, can be verified through actual grid operation. Taking bus load forecasting as an example, by comparing predicted data with measured grid operation data, metrics such as the accuracy and precision of each actuator's predictions can be calculated. Based on these calculated metrics, each actuator can be dynamically adjusted. For example, an actuator that fails to meet a preset prediction accuracy threshold three times in a row can be removed from the grid control model to ensure that the model's predictions are not affected by actuators with lower accuracy, thus ensuring accurate predictions. Furthermore, such actuators can be adjusted through retraining.

[0062] In a possible embodiment, the power grid control model training method also includes: obtaining real-time data of the power grid control model in which the second preset prediction index value of the execution body is lower than the preset second preset prediction index value prediction requirement, to obtain data with key characteristics; and sharing the data with key characteristics as power grid control model training data to each training data group to obtain several updated training data groups; according to the one-to-one correspondence between the updated training data group and the learning body, training several preset learning bodies in parallel through several updated training data groups to obtain several updated execution bodies; and using several updated execution bodies to replace several execution bodies in the power grid control model.

[0063] Specifically, during the parallel prediction process of the executor, the thresholds of indicators such as accuracy are set as the feedback thresholds for collecting real-time data of the power grid control model. If the accuracy or precision of the executor does not reach the corresponding threshold, the corresponding real-time data of the power grid control model will be automatically collected and fed back to the learning body as new training data for the power grid control model, thereby further improving the learning effect of the learning body.

[0064] In a possible embodiment, the parallel training of the preset several learning bodies through several updated training data groups also includes: obtaining relevant data of the data with key features, and sharing the relevant data of the data with key features as power grid control model training data to each updated training data group.

[0065] Specifically, according to different business scenarios, in addition to collecting the data with key features, related data of the data with key features can also be collected, such as related data that has a time logical relationship or other relationship with the data with key features. Taking the day-ahead bus load forecast as an example, the data of the day, the previous day and the next day can be collected and fed back to the learning body as training data for the new power grid control model.

[0066] The relevant data includes: data that has a time relationship with the data with key features, data of devices that have a topological connection relationship with the devices corresponding to the data with key features, and data that has a business logic relationship with the data with key features.

[0067] In a possible embodiment, the parallel training of several preset learning bodies through several updated training data groups also includes: using a data feature analysis method to extract features of data with key features to obtain extracted features; obtaining an optimization method, and optimizing the power grid control model training data or several learning bodies in each updated training data group according to the optimization method; wherein the optimization method adopts a manual analysis method to obtain the test evaluation effect of each learning body and the reasoning effect of each execution body, and obtains it based on the extracted features, combined with the test evaluation effect of each learning body, the reasoning effect of each execution body and business domain knowledge.

[0068] Among them, domain knowledge refers to knowledge in a specific, specialized discipline or field. In the present invention, it refers to professional knowledge related to power grid regulation, including constraints on power grid operation and related business rules.

[0069] Specifically, a data feature analysis method is used to extract the features of data with key features and feed them back to manual auxiliary analysis. Through manual analysis, the reasons why the data with key features causes a decline in indicators such as the execution body prediction accuracy are confirmed. Based on this, further analysis is conducted on whether it is necessary to optimize the learning body model or expand the training data range of the learning body.

[0070] The optimization method includes one or more of the following: changing the data range of the power grid regulation model training data, changing the data dimension of the power grid regulation model training data, changing the data verification rules of the power grid regulation model training data, changing the data cleaning rules of the power grid regulation model training data, and changing the number of learning bodies.

[0071] Specifically, for changing the data verification and data cleaning rules for grid control model training data, taking bus load forecasting as an example, the bus balance and plant balance verification rules can be expanded. Or, in the case of a lack of measurement data for low-voltage buses or poor measurement data accuracy, the accuracy of the collected measurement data can be improved by using line peers or using electricity data to calculate bus power. For changing the data dimension of grid control model training data, the data dimension of grid control model training data can be increased. Taking bus load forecasting as an example, based on the grid operation measurement data, meteorological data, holiday information, load type and other data can be further expanded.

[0072] In one possible embodiment, the fusion of the outputs of several executors as the output of the power grid control model includes: weighting and superimposing the outputs of several executors according to the preset weighting coefficients of each executor as the output of the power grid control model; or processing the outputs of several executors according to a scoring method to obtain the output of the power grid control model.

[0073] Specifically, when the outputs of several executors are processed according to the scoring method, the scoring methods commonly used include mean scoring method, clustering algorithm (such as K-Means, GMM, etc.) and proximity measurement method (such as distance-based (KNN) and density-based (LOF) abnormal data detection method) and similar methods. Among them, the mean scoring method generally removes a maximum value and a minimum value, and then calculates the average value. When there are many executors, clustering algorithms and proximity measurement methods are generally used. Among them, the clustering method includes: dividing the predicted values ​​of all executors into multiple clusters through a clustering algorithm (such as the K-Means algorithm), and then calculating the cluster size, compactness and spacing, so as to obtain a higher intra-cluster similarity and a lower inter-cluster similarity, and further obtaining the cluster center of the largest and most compact cluster as the output of the power grid control model.

[0074] Optionally, referring to FIG5 , the preset weighting coefficients for each executor are set by obtaining the prediction accuracy or precision of each executor; and setting the weighting coefficients for each executor based on the prediction accuracy or precision of each executor using factor analysis, principal component analysis, AHP, pecking order diagram, entropy, CRITIC weighting, independence weighting, or information weighting. Optionally, the sum of the preset weighting coefficients for each executor is 1.

[0075] In a possible implementation, the grid control model training method further includes: dynamically adjusting a preset weighting coefficient of each executor under a dynamic frequency.

[0076] Specifically, the preset weighting coefficients for each executor are a dynamically changing process. These weighting coefficients can first be determined based on the prediction accuracy or precision of each executor on a set test set. Then, in the subsequent parallel prediction process, they can be dynamically adjusted based on the prediction accuracy or precision of each executor for the real-time data of the power grid control model within a certain period of time. For example, the weighting coefficients of executors with higher prediction accuracy can be increased, while those with lower prediction accuracy can be decreased. The acquisition of prediction accuracy or precision and the adjustment of the weighting coefficients of each executor can be set with corresponding acquisition and adjustment frequencies based on specific business application requirements.

[0077] The following are device embodiments of the present invention, which can be used to perform the method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0078] Referring to FIG6 , in another embodiment of the present invention, a power grid control model training system is provided, which can be used to implement the above-mentioned power grid control model training method. Specifically, the power grid control model training system includes a data acquisition module, a model training module, and a model assembly module. Among them, the data acquisition module is used to obtain the grid control model training data and group it to obtain several training data groups; the model training module is used to train several preset learning bodies in parallel through several training data groups according to the one-to-one correspondence between the training data groups and the learning bodies to obtain several execution bodies; wherein, when training the learning body through the training data group, the current training data group is divided into several training data units, and the current learning body is trained in turn through several training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups except the training data group to which it belongs; wherein, several learning bodies are several grid control business AI models constructed according to the grid control needs; the model assembly module is used to combine several execution bodies to obtain a grid control model; wherein, the combination includes taking the inputs of several execution bodies together as the input of the grid control model, and fusing the outputs of several execution bodies as the output of the grid control model.

[0079] In one possible embodiment, the method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: according to the one-to-one correspondence between the training data groups and the learning bodies, combined with the preset evaluation bodies corresponding one-to-one to each learning body, training the plurality of preset learning bodies in parallel through a plurality of training data groups; and after the training is completed, obtaining the preset prediction indicators of each learning body through each evaluation body, and using the learning body whose preset prediction indicators meet the preset prediction indicator training requirements as the execution body.

[0080] In one possible implementation, the plurality of power grid regulation business AI models are constructed using two or more machine learning algorithms or artificial intelligence algorithms.

[0081] In a possible implementation, each of the several power grid regulation business AI models is a power grid regulation business AI model that performs the same type of power grid regulation business, or a power grid regulation business AI model that performs complementary power grid regulation business.

[0082] In a possible embodiment, a monitoring module is further included, which is used to periodically obtain a first preset prediction index value of each executor in the power grid control model for the real-time data of the power grid control model; when the first preset prediction value of an executor in the power grid control model fails to meet the preset first preset prediction index value requirement for a preset number of consecutive times, the executor is removed from the power grid control model, or the executor is retrained using the power grid control model training data.

[0083] In a possible embodiment, it also includes a retraining module, which is used to obtain real-time data of the power grid control model in which the second preset prediction index value of the execution body is lower than the preset second preset prediction index value prediction requirement, and obtain data with key characteristics; and share the data with key characteristics as power grid control model training data to each training data group to obtain several updated training data groups; according to the one-to-one correspondence between the updated training data group and the learning body, train several preset learning bodies in parallel through the several updated training data groups to obtain several updated execution bodies; and use the several updated execution bodies to replace several execution bodies in the power grid control model.

[0084] In a possible embodiment, the parallel training of the preset several learning bodies through several updated training data groups also includes: obtaining relevant data of the data with key features, and sharing the relevant data of the data with key features as power grid control model training data to each updated training data group.

[0085] In a possible embodiment, the parallel training of several preset learning bodies through several updated training data groups also includes: using a data feature analysis method to extract features of data with key features to obtain extracted features; obtaining an optimization method, and optimizing the power grid control model training data or several learning bodies in each updated training data group according to the optimization method; wherein the optimization method adopts a manual analysis method, obtained according to the extracted features and combined with business domain knowledge.

[0086] In one possible implementation, the optimization method includes one or more of the following: changing the data range of the grid control model training data, changing the data dimension of the grid control model training data, changing the data verification rules of the grid control model training data, changing the data cleaning rules of the grid control model training data, and changing the number of learning bodies.

[0087] In one possible embodiment, the fusion of the outputs of several executors as the output of the power grid control model includes: weighting and superimposing the outputs of several executors according to the preset weighting coefficients of each executor as the output of the power grid control model; or processing the outputs of several executors according to a scoring method to obtain the output of the power grid control model.

[0088] In one possible embodiment, the power grid control model training system also includes: dynamically adjusting the preset weighting coefficient of each executor under dynamic frequency; the preset weighting coefficient of each executor is adjusted by: obtaining the prediction accuracy or prediction precision of each executor; according to the prediction accuracy or prediction precision of each executor, using factor analysis method, principal component analysis method, AHP hierarchical method, priority diagram method, entropy method, CRITIC weight method, independence weight method or information weight method to adjust the weighting coefficient of each executor.

[0089] All relevant contents of each step involved in the embodiment of the aforementioned power grid control model training method can be referred to the functional description of the functional modules corresponding to the power grid control model training system in the embodiment of the present invention, and will not be repeated here.

[0090] The module division in the embodiments of the present invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in various embodiments of the present invention may be integrated into a single processor, exist physically as separate modules, or two or more modules may be integrated into a single module. The integrated modules may be implemented in either hardware or software functional modules.

[0091] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the power grid control model training method.

[0092] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the power grid control model training method in the above embodiment.

[0093] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0095] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0096] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A power grid control model training method, comprising: The training data of the power grid control model is obtained and grouped to obtain a number of training data groups; wherein the training data of the power grid control model is obtained from the control big data based on different control business needs; the control big data refers to a data set related to the power grid operation, including power grid model data, monitoring data, alarm data and operation records; According to the one-to-one correspondence between the training data group and the learning body, several preset learning bodies are trained in parallel through several training data groups to obtain several execution bodies; wherein, when training the learning body through the training data group, the current training data group is divided into several training data units, and the current learning body is trained in sequence through several training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups except the training data group to which it belongs; wherein the several learning bodies are several power grid control business AI models constructed according to the power grid control needs; Combining several actuators to obtain a power grid control model; wherein the combination includes taking the inputs of several actuators together as the input of the power grid control model, and fusing the outputs of several actuators as the output of the power grid control model; Among them, the power grid control business is bus load forecasting or power grid equipment failure forecasting.

2. The power grid control model training method according to claim 1, wherein: The method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: According to the one-to-one correspondence between training data groups and learning bodies, several preset learning bodies are trained in parallel through several training data groups in combination with preset evaluation bodies corresponding one-to-one to each learning body; and after the training is completed, the preset prediction indicators of each learning body are obtained through each evaluation body, and the learning body whose preset prediction indicators meet the preset prediction indicator training requirements is used as the execution body.

3. The power grid control model training method according to claim 1, wherein: The several power grid control business AI models are constructed through two or more machine learning algorithms or artificial intelligence algorithms.

4. The power grid control model training method according to claim 1, wherein: Each of the several power grid regulation business AI models is a power grid regulation business AI model that performs the same type of power grid regulation business, or a power grid regulation business AI model that performs complementary power grid regulation business.

5. The power grid control model training method according to claim 1, wherein: Also includes: Periodically obtaining a first preset prediction index value of each execution body in the power grid control model for real-time data of the power grid control model; When a first preset prediction index value of an execution body in the power grid control model fails to meet the preset first preset prediction index value requirement for a preset number of consecutive times, the execution body is removed from the power grid control model, or the execution body is retrained using the power grid control model training data.

6. The power grid control model training method according to claim 1, wherein: Also includes: Acquire real-time data of a power grid control model in which a second preset prediction index value of the execution body is lower than a preset second preset prediction index value prediction requirement, and obtain data with key features; and sharing the data with key features as power grid control model training data to each training data group to obtain a number of updated training data groups; According to the one-to-one correspondence between the update training data group and the learning body, a plurality of preset learning bodies are trained in parallel through a plurality of update training data groups to obtain a plurality of update execution bodies; Several update executives are used to replace several executives in the power grid control model.

7. The power grid control model training method according to claim 6, wherein: The step of training the preset multiple learning bodies in parallel through the multiple updated training data groups also includes: Acquire relevant data of the data with key features, and share the relevant data of the data with key features as power grid control model training data to each updated training data group; The relevant data includes: data having a time relationship with the data having key characteristics, data of devices having a topological connection relationship with the devices corresponding to the data having key characteristics, and data having a business logic relationship with the data having key characteristics.

8. The power grid control model training method according to claim 6, wherein: The step of training the preset multiple learning bodies in parallel through the multiple updated training data groups also includes: Using data feature analysis method, the features of data with key features are extracted to obtain extracted features; Obtain an optimization method, and optimize the power grid control model training data or several learning bodies in each updated training data group according to the optimization method; wherein the optimization method adopts a manual analysis method to obtain the test evaluation effect of each learning body and the reasoning effect of each execution body, and obtains it according to the extracted features, combined with the test evaluation effect of each learning body, the reasoning effect of each execution body and the business domain knowledge.

9. The power grid control model training method according to claim 8, wherein: The optimization method includes one or more of the following: Changing the data range of the power grid control model training data, changing the data dimension of the power grid control model training data, changing the data verification rule of the power grid control model training data, changing the data cleaning rule of the power grid control model training data, and changing the number of learning bodies.

10. The power grid control model training method according to claim 1, wherein: The fusing of the outputs of several actuators as the output of the power grid control model includes: The outputs of several actuators are weighted and superimposed according to the preset weighting coefficients of each actuator as the output of the power grid control model; or, the outputs of several actuators are processed according to the scoring method to obtain the output of the power grid control model.

11. The power grid control model training method according to claim 10, wherein: It also includes: dynamically adjusting the preset weighting coefficients of each actuator at a dynamic frequency; The weighting coefficients preset by each actuator are adjusted in the following manner: Obtain the prediction accuracy or prediction precision of each executor; According to the prediction accuracy or prediction precision of each executor, the weight coefficient of each executor is adjusted by using factor analysis, principal component analysis, AHP hierarchy method, priority diagram method, entropy method, CRITIC weight method, independence weight method or information weight method.

12. A power grid control model training system, comprising: The data acquisition module is used to acquire and group the training data of the power grid control model to obtain a number of training data groups; wherein the training data of the power grid control model is acquired from the control big data based on different control business requirements; the control big data refers to a data set related to the power grid operation, including power grid model data, monitoring data, alarm data and operation records; The model training module is used to train a plurality of preset learning bodies in parallel through a plurality of training data groups according to the one-to-one correspondence between the training data groups and the learning bodies, and obtain a plurality of execution bodies; wherein, when training the learning body through the training data groups, the current The training data group is divided into a number of training data units, and the current learning body is trained in sequence through the several training data units, and the training effect of each training data unit on the current learning body is obtained; and the training data unit whose training effect meets the preset training effect requirements is used as a shared training data unit, and the shared training data unit is shared to other training data groups except the training data group to which it belongs; wherein the several learning bodies are several power grid control business AI models constructed according to the power grid control requirements; A model assembly module, used to combine several actuators to obtain a power grid control model; wherein the combination includes taking the inputs of several actuators as the input of the power grid control model, and fusing the outputs of several actuators as the output of the power grid control model; Among them, the power grid control business is bus load forecasting or power grid equipment failure forecasting.

13. The power grid control model training system according to claim 12, wherein: The method of training a plurality of preset learning bodies in parallel through a plurality of training data groups to obtain a plurality of execution bodies includes: According to the one-to-one correspondence between training data groups and learning bodies, several preset learning bodies are trained in parallel through several training data groups in combination with preset evaluation bodies corresponding one-to-one to each learning body; and after the training is completed, the preset prediction indicators of each learning body are obtained through each evaluation body, and the learning body whose preset prediction indicators meet the preset prediction indicator training requirements is used as the execution body.

14. The power grid control model training system according to claim 12, wherein: It also includes a retraining module, which is used to obtain real-time data of a power grid control model in which the second preset prediction index value of the executive body is lower than the prediction requirement of the preset second preset prediction index value, and obtain data with key characteristics; And the data with key features are shared as the training data of the power grid control model to each training data group to obtain a number of updated training data groups; according to the one-to-one correspondence between the updated training data groups and the learning bodies, a number of preset learning bodies are trained in parallel through the number of updated training data groups to obtain a number of updated execution bodies; and the number of updated execution bodies are used to replace the number of execution bodies in the power grid control model.

15. The power grid control model training system according to claim 14, wherein: The method of training a plurality of preset learning bodies in parallel through a plurality of updated training data groups also includes: using a data feature analysis method to extract features of data with key features to obtain extracted features; obtaining an optimization method, and optimizing the power grid control model training data or a plurality of learning bodies in each updated training data group according to the optimization method; wherein the optimization method is obtained by manual analysis based on the extracted features and combined with business domain knowledge.

16. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the power grid control model training method according to any one of claims 1 to 11 are implemented.

17. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, the steps of the power grid control model training method according to any one of claims 1 to 11 are implemented.

Images