CN-122000858-A - Knowledge closed-loop hybrid enhancement method and system for power system

CN122000858ACN 122000858 ACN122000858 ACN 122000858ACN-122000858-A

Abstract

The embodiment of the application provides a power system knowledge closed-loop hybrid enhancement method and system, and relates to the technical field of power system optimal scheduling. According to the method, after the operation data of the power system are obtained, a time sequence prediction result of the operation data is calculated through a prediction model, and then evaluation is carried out on the time sequence prediction result according to an expert agent cooperative evolution interaction mechanism so as to obtain evaluation feedback information. And then updating model parameters and knowledge information of the prediction model according to the evaluation feedback information, optimizing a control strategy of the power system based on the updated prediction model, and generating closed-loop feedback information by executing the control strategy to realize continuous optimization of the control strategy. According to the method, the prediction model can be updated through expert feedback, the coping capacity of the power system in coordinated scheduling of multiple time scales and multiple energy sources is effectively improved, the stability and response speed of the power system in the face of complex disturbance are improved, and the control precision and instantaneity of the power system are improved.

Inventors

HU SHUBO
XIE CIJIAN
ZHAO QINGSONG
SUN JUNJIE
MA XINTONG
GE YANGYANG
HU XUGUANG
XIE BING
LIU XINRUI
ZHANG XIAOTONG
ZHANG ZHIHONG
LI PING

Assignees

国网辽宁省电力有限公司电力科学研究院
东北大学

Dates

Publication Date: 20260508
Application Date: 20251203

Claims (10)

1. A method for closed-loop hybrid augmentation of power system knowledge, the method comprising: acquiring operation data, wherein the operation data comprises power data, voltage data, meteorological data, load data and system states of a power system; calculating a time sequence prediction result of the operation data through a prediction model, wherein the prediction model is a power phase maintenance complex-valued network model based on machine learning, and the prediction model is modeled through a complex domain hidden state and a unitary state transition matrix and is used for capturing dynamic phase change and fluctuation characteristics in a power system according to the operation data; Performing evaluation on the time sequence prediction result according to an expert agent cooperative evolution interaction mechanism to obtain evaluation feedback information, wherein the evaluation feedback information comprises available adjustment items of the prediction model; Updating model parameters and knowledge information of the prediction model according to the evaluation feedback information; And optimizing a control strategy of the power system based on the updated prediction model, and generating closed-loop feedback information by executing the control strategy so as to continuously optimize the control strategy through the synergistic effect of heterogeneous knowledge fusion enhanced closed-loop and expert agent cooperative evolution interaction mechanism.
2. The method of claim 1, wherein obtaining operational data comprises: Acquiring original operation data from a data source device, wherein the data source device comprises at least one of a sensor and a data acquisition device; Performing a normalization process on the raw operational data to translate all variables in the raw operational data to the same magnitude; and carrying out data segmentation on the original operation data with the same magnitude according to a time window to obtain the operation data.
3. The method according to claim 1, wherein the method further comprises: Extracting input data from the operation data, wherein the input data is a vector containing a plurality of time sequence characteristics in each time step; The output layer of the prediction model is configured to output a complex form of time sequence prediction result, and map the complex form of time sequence prediction result to a real number domain through linear transformation to obtain a predicted value of a power grid state; setting an error function of the prediction model, wherein the error function is complex mean square error; And in the process of training the prediction model, gradient calculation is carried out by using a Wirtinger derivative, and network weight is updated by combining a gradient descent method.
4. A method according to claim 3, wherein constructing the predictive model comprises: acquiring the hidden state of the previous time step; Setting learning parameters, wherein the learning parameters comprise a first learning parameter and a second learning parameter, the first learning parameter is a weight matrix of input data of the current time step, and the second learning parameter is a weight matrix of a hidden state of the previous time step; calculating candidate hidden states according to the hidden states and the learning parameters; Acquiring gating parameters of a gating circulation unit in the prediction model, wherein the gating parameters comprise an updating gate and a resetting gate; And constructing a hidden state updating function of the prediction model according to the gating parameter, the hidden state of the previous time step and the candidate hidden state.
5. The method of claim 1, wherein performing an evaluation on the time series prediction result according to an expert agent co-evolution interaction mechanism to obtain evaluation feedback information comprises: reading a prediction parameter item corresponding to the time sequence prediction result, wherein the prediction parameter item comprises a prediction voltage, a prediction power and a prediction system state; acquiring an actual observed value corresponding to the prediction parameter item, wherein the actual observed value comprises an observed voltage, an observed power and an observed system state which are acquired to a power system; Calculating a prediction error by comparing the time sequence prediction result with the actual observed value; If the prediction error meets any preset triggering condition, expert feedback data are obtained, wherein the expert feedback data comprise a power grid state evaluation result, a control strategy adjustment suggestion and a prediction model adjustment suggestion; Mapping the expert feedback data into the assessment feedback information.
6. The method of claim 5, wherein updating model parameters and knowledge information of the predictive model based on the assessment feedback information comprises: generating a knowledge base entry according to the expert feedback data, wherein the knowledge base entry comprises a triggering condition, an influencing variable, a suggested operation and a priority; updating a model knowledge base of the predictive model using the knowledge base entry; Retrieving a set of associated entries in the updated model knowledge base based on a current system state vector; and injecting the association item set and the suggested operation into a model training or control module of the present round.
7. The method of claim 6, wherein updating the model knowledge base of the predictive model using the knowledge base entry comprises: Reading a target field from the expert feedback data, wherein the target field comprises at least one of a first type field, a second type field and a third type field; When the expert feedback data contains a first field, modifying a training set based on the knowledge base entry and giving a new sample weight; Setting a definition weight according to the triggering condition when the expert feedback data contains a second type field, and calculating a weighted loss based on the definition weight and a loss function of the prediction model; Defining a parameter set update function when a third type of field is included in the expert feedback data, and adjusting a model structure and/or updating model parameters of the predictive model based on the parameter set update function.
8. The method of claim 1, wherein optimizing a control strategy of a power system based on the updated predictive model comprises: Constructing an expert feedback rewarding function based on the estimated feedback information, wherein the expert feedback rewarding function is used for correcting voltage deviation, power unbalance and operation cost; acquiring expert feedback weights, and establishing a total rewards updating strategy function based on the expert feedback weights and the expert feedback rewards function; And updating the control strategy of the power system according to the total rewards updating strategy function by using an optimizing agent, wherein the optimizing agent is used for selecting control actions according to the current system state and minimizing power unbalance and voltage deviation through iterative optimization.
9. The method of claim 1, wherein generating closed-loop feedback information by executing the control strategy comprises: reading a control target and an influence output variable of the control strategy; modifying the influencing output variable by the power system according to the control target by executing the control strategy; generating the closed loop feedback information according to the real-time state of the power system; And updating the control strategy through a closed-loop control mechanism according to the closed-loop feedback information.
10. A power system knowledge closed loop hybrid enhancement system, the system comprising: The data acquisition module is used for acquiring operation data, wherein the operation data comprises power data, voltage data, meteorological data, load data and system states of the power system; The prediction module is used for calculating a time sequence prediction result of the operation data through a prediction model, wherein the prediction model is a machine learning-based power phase retention complex-valued network model, and the prediction model is modeled through a complex domain hidden state and a unitary state transition matrix and is used for capturing dynamic phase change and fluctuation characteristics in a power system according to the operation data; The prediction evaluation module is used for performing evaluation on the time sequence prediction result according to an expert agent cooperative evolution interaction mechanism so as to obtain evaluation feedback information, wherein the evaluation feedback information comprises available adjustment items of the prediction model; The updating module is used for updating the model parameters and knowledge information of the prediction model according to the evaluation feedback information; And the closed-loop feedback module is used for optimizing the control strategy of the power system based on the updated prediction model, and generating closed-loop feedback information by executing the control strategy so as to continuously optimize the control strategy through the synergistic effect of heterogeneous knowledge fusion enhanced closed loop and expert agent cooperative evolution interaction mechanism.

Description

Knowledge closed-loop hybrid enhancement method and system for power system Technical Field The application relates to the technical field of power system optimization scheduling, in particular to a power system knowledge closed-loop hybrid enhancement method and system. Background After the distributed energy is connected into the power system, impact is caused on a power supply network. Due to fluctuation and uncertainty of renewable energy sources such as photovoltaic energy and wind energy, key parameters such as voltage and frequency of a power grid are affected, and the voltage and frequency of the power grid need to be effectively regulated, so that stable operation of a power system is ensured. For example, in the context of micro-grids and smart grids, the power system needs to not only cope with the access of large-scale distributed power generation resources, but also consider coordination and scheduling of multiple time scales and multiple energy types, which puts demands on the control and scheduling capabilities of the power system. The voltage and frequency control of the power system may depend on control methods such as PID control, optimal control, robust control, etc. The control methods are widely applied to the power system, and particularly can better maintain the stability of the power system under the conditions that the power grid load is relatively balanced and no large-scale renewable energy source is connected. However, with the access of renewable energy sources, the complexity and uncertainty faced by the power system increases significantly. For example, PID control cannot effectively cope with nonlinear and time-varying characteristics in a power grid, linear control methods can realize voltage and frequency control to a certain extent through an optimization algorithm, but ignore various complex factors existing in a system, and adaptive control methods can dynamically adjust according to a system state, but rely on an accurate system model, so that ideal control effects are difficult to realize when facing a large-scale complex power system. Therefore, the control accuracy and the real-time performance of the control method are low, and the method meets the requirements of an electric power system. Disclosure of Invention In view of the above, the embodiment of the application provides a method and a system for enhancing the closed-loop hybrid knowledge of an electric power system, so as to solve the problem of low control precision and real-time performance of the electric power system. According to a first aspect of the present application, there is provided a power system knowledge closed loop hybrid enhancement method, the method comprising: acquiring operation data, wherein the operation data comprises power data, voltage data, meteorological data, load data and system states of a power system; calculating a time sequence prediction result of the operation data through a prediction model, wherein the prediction model is a power phase maintenance complex-valued network model based on machine learning, and the prediction model is modeled through a complex domain hidden state and a unitary state transition matrix and is used for capturing dynamic phase change and fluctuation characteristics in a power system according to the operation data; Performing evaluation on the time sequence prediction result according to an expert agent cooperative evolution interaction mechanism to obtain evaluation feedback information, wherein the evaluation feedback information comprises available adjustment items of the prediction model; Updating model parameters and knowledge information of the prediction model according to the evaluation feedback information; And optimizing a control strategy of the power system based on the updated prediction model, and generating closed-loop feedback information by executing the control strategy so as to continuously optimize the control strategy through the synergistic effect of heterogeneous knowledge fusion enhanced closed-loop and expert agent cooperative evolution interaction mechanism. In some embodiments, obtaining operational data includes: Acquiring original operation data from a data source device, wherein the data source device comprises at least one of a sensor and a data acquisition device; Performing a normalization process on the raw operational data to translate all variables in the raw operational data to the same magnitude; and carrying out data segmentation on the original operation data with the same magnitude according to a time window to obtain the operation data. In some embodiments, the method further comprises: Extracting input data from the operation data, wherein the input data is a vector containing a plurality of time sequence characteristics in each time step; The output layer of the prediction model is configured to output a complex form of time sequence prediction result, and map the complex form of time sequence predictio