Power system knowledge closed-loop hybrid enhancement method and system

Abstract

The embodiment of the invention provides an electric power system knowledge closed-loop hybrid enhancement method and system, and relates to the technical field of electric power system optimization scheduling. According to the method, after operation data of a power system is obtained, a time sequence prediction result of the operation data is calculated through a prediction model, and then the time sequence prediction result is evaluated according to an expert agent co-evolution interaction mechanism to obtain evaluation feedback information. And 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 types is effectively improved, the stability and the response speed of the power system in the face of complex disturbance are enhanced, and the control precision and the real-time performance of the power system are improved.

Description

Technical Field

[0001] This application relates to the field of power system optimization and dispatching technology, and in particular to a power system knowledge closed-loop hybrid enhancement method and system. Background Technology

[0002] Distributed energy sources, once integrated into the power system, can impact the power grid. Due to the volatility and uncertainty of renewable energy sources such as solar and wind power, key parameters of the power grid, such as voltage and frequency, are affected. Therefore, it is necessary to effectively regulate the voltage and frequency of the power grid to ensure the stable operation of the power system.

[0003] For example, in the context of microgrids and smart grids, power systems not only need to cope with the integration of large-scale distributed generation resources, but also need to consider the coordination and dispatch of multiple time scales and energy types. This places demands on the control and dispatch capabilities of the power system. Voltage and frequency control of the power system can rely on control methods such as PID control, optimal control, and robust control. These control methods are widely used in power systems, especially when the grid load is relatively balanced and there is no large-scale integration of renewable energy, and they can effectively maintain the stability of the power system.

[0004] However, with the integration of renewable energy, the complexity and uncertainty faced by power systems have increased significantly. For example, PID control cannot effectively handle the nonlinear and time-varying characteristics of the power grid; while linear control methods can achieve a certain degree of voltage and frequency control through optimization algorithms, they ignore various complex factors existing in the system; and while adaptive control methods can dynamically adjust according to the system state, they rely on accurate system models and are difficult to achieve ideal control effects when facing large-scale complex power systems. Therefore, the control accuracy and real-time performance of the above-mentioned control methods are low, and they cannot meet the needs of power systems. Summary of the Invention

[0005] In view of this, embodiments of this application provide a power system knowledge closed-loop hybrid enhancement method and system to solve the problems of low control accuracy and real-time performance in power systems.

[0006] According to a first aspect of this application, a power system knowledge closed-loop hybrid enhancement method is provided, the method comprising: Acquire operational data, including power data, voltage data, meteorological data, load data, and system status of the power system; The time-series prediction results of the operating data are calculated by a prediction model, which is a power phase-preserving complex value network model based on machine learning. The prediction model is modeled using complex domain hidden states and unitary state transition matrices, and is used to capture the dynamic phase changes and fluctuation characteristics of the power system based on the operating data. The time-series prediction results are evaluated according to the expert agent collaborative evolution interaction mechanism to obtain evaluation feedback information; the evaluation feedback information includes adjustment terms available for the prediction model. The model parameters and knowledge information of the prediction model are updated based on the evaluation feedback information; The control strategy of the power system is optimized based on the updated prediction model, and closed-loop feedback information is generated by executing the control strategy to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolution interaction mechanism through heterogeneous knowledge fusion.

[0007] In some embodiments, obtaining runtime data includes: Raw operational data is obtained from a data source device, which includes at least one of a sensor and a data acquisition device; The raw runtime data is standardized to transform all variables in the raw runtime data into the same order of magnitude; The original operational data of the same magnitude is divided into data segments according to time windows to obtain the operational data.

[0008] In some embodiments, the method further includes: Input data is extracted from the running data, and the input data is a vector containing multiple temporal features within each time step; The prediction model is constructed such that its input layer is configured to receive input data within a time window; and its output layer is configured to output time-series prediction results in complex form, and to map the complex-form time-series prediction results to the real number domain through a linear transformation to obtain predicted values ​​of the power grid state. The error function of the prediction model is set, wherein the error function is the complex mean square error; Based on the error function, the prediction model is trained using the input data; during the training of the prediction model, the Wirtinger derivative is used to calculate the gradient, and the network weights are updated using the gradient descent method.

[0009] In some embodiments, constructing the prediction model includes: Get the hidden state of the previous time step; The learning parameters are set, including a first learning parameter and a second learning parameter. The first learning parameter is the weight matrix of the input data at the current time step; the second learning parameter is the weight matrix of the hidden state at the previous time step. Calculate candidate hidden states based on the hidden states and the learning parameters; Obtain the gating parameters of the gated loop unit in the prediction model, wherein the gating parameters include the update gate and the reset gate; The hidden state update function of the prediction model is constructed based on the gating parameters, the hidden state of the previous time step, and the candidate hidden states.

[0010] In some embodiments, the time-series prediction results are evaluated according to an expert agent collaborative evolutionary interaction mechanism to obtain evaluation feedback information, including: Read the prediction parameter items corresponding to the time-series prediction results, including the prediction voltage, prediction power, and prediction system state; Obtain the actual observed values ​​corresponding to the predicted parameter items, including the observed voltage, observed power, and observed system status collected from the power system; The prediction error is calculated by comparing the time-series prediction results with the actual observed values. If the prediction error meets any preset triggering condition, expert feedback data is obtained. The expert feedback data includes power grid status assessment results, control strategy adjustment suggestions, and prediction model adjustment suggestions. The expert feedback data is mapped to the evaluation feedback information.

[0011] In some embodiments, updating the model parameters and knowledge information of the prediction model based on the evaluation feedback information includes: Knowledge base entries are generated based on the expert feedback data. These entries include triggering conditions, influencing variables, suggested actions, and priorities. Update the model knowledge base of the prediction model using the knowledge base entries; Based on the current system state vector, retrieve the set of related entries in the updated model knowledge base; The set of associated entries and the suggested operations are injected into the current model training or control module.

[0012] In some embodiments, updating the model knowledge base of the prediction model using the knowledge base entries includes: Read the target field from the expert feedback data, the target field including at least one of the first type of field, the second type of field, and the third type of field; When the expert feedback data contains the first type of field, the training set is modified and new sample weights are assigned based on the knowledge base entries. When the expert feedback data includes a second type of field, a defined weight is set according to the triggering condition, and a weighted loss is calculated based on the defined weight and the loss function of the prediction model; When the expert feedback data includes a third type of field, a parameter set update function is defined, and the model structure of the prediction model and / or the model parameters are adjusted based on the parameter set update function.

[0013] In some embodiments, optimizing the control strategy of the power system based on the updated prediction model includes: Based on the estimated feedback information, an expert feedback reward function is constructed, which is used for voltage deviation correction, power imbalance correction, and operating cost correction. Obtain the expert feedback weights, and establish a total reward update strategy function based on the expert feedback weights and the expert feedback reward function; An optimization agent is used to update the control strategy of the power system according to the total reward update strategy function; the optimization agent is used to select control actions according to the current system state, and to minimize power imbalance and voltage deviation through iterative optimization.

[0014] In some embodiments, generating closed-loop feedback information by executing the control strategy includes: Read the control objective and the output variables that affect the control strategy; By executing the control strategy, the power system modifies the influencing output variables according to the control objective; The closed-loop feedback information is generated based on the real-time status of the power system. The control strategy is updated through a closed-loop control mechanism based on the closed-loop feedback information.

[0015] According to a second aspect of this application, a power system knowledge closed-loop hybrid enhancement system is provided, the system comprising: The data acquisition module is used to acquire operational data, which includes power data, voltage data, meteorological data, load data, and system status of the power system. The prediction module is used to calculate the time-series prediction results of the operating data through a prediction model, which is a power phase-preserving complex value network model based on machine learning. The prediction model is modeled through complex domain hidden states and unitary state transition matrices, and is used to capture the dynamic phase changes and fluctuation characteristics of the power system based on the operating data. The prediction and evaluation module is used to evaluate the time-series prediction results according to the expert agent collaborative evolutionary interaction mechanism to obtain evaluation feedback information; the evaluation feedback information includes adjustment terms available for the prediction model. The update module is used to update the model parameters and knowledge information of the prediction model based on the evaluation feedback information; The closed-loop feedback module is used to optimize the control strategy of the power system based on the updated prediction model, and to generate closed-loop feedback information by executing the control strategy, so as to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolution interaction mechanism through heterogeneous knowledge fusion.

[0016] According to a third aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described power system knowledge closed-loop hybrid enhancement method.

[0017] According to a fourth aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the above-described power system knowledge closed-loop hybrid enhancement method.

[0018] By employing the above technical solutions, embodiments of this application provide a power system knowledge closed-loop hybrid enhancement method and system. After acquiring the power system's operating data, the method calculates the time-series prediction results of the operating data using a prediction model, and then evaluates the time-series prediction results according to an expert intelligent agent collaborative evolution interaction mechanism to obtain evaluation feedback information. Then, based on the evaluation feedback information, the model parameters and knowledge information of the prediction model are updated, and the power system's control strategy is optimized based on the updated prediction model. Furthermore, closed-loop feedback information is generated by executing the control strategy to achieve continuous optimization of the control strategy. The method combines expert feedback with the physical model through a closed-loop feedback mechanism, ensuring that the control strategy is closer to the actual needs of the power system. Moreover, after each execution of the control strategy, continuous optimization and adjustment can be achieved through a heterogeneous knowledge fusion mechanism, and the prediction model can be updated through expert feedback to optimize voltage control and power balance, thereby ensuring the stable operation of the power system. This method can effectively improve the power system's ability to cope with multi-timescale and multi-energy-type coordinated scheduling, enhance the stability and response speed of the power system when facing complex disturbances, and improve the control accuracy and real-time performance of the power system.

[0019] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the power system topology provided in the embodiments of this application; Figure 2 A schematic diagram of the power system knowledge closed-loop hybrid enhancement method provided in the embodiments of this application; Figure 3 A schematic diagram of the power phase-preserving complex value network structure provided in the embodiments of this application; Figure 4 A schematic diagram of a closed-loop self-optimization process based on human-computer interaction feedback provided in an embodiment of this application; Figure 5 This is a schematic diagram of voltage control simulation results provided in the embodiments of this application; Figure 6 This is a schematic diagram of the closed-loop feedback mechanism provided in the embodiments of this application; Figure 7 This is a simulation diagram of the feedback reward results provided in the embodiments of this application; Figure 8 A schematic diagram of the power system knowledge closed-loop hybrid enhancement system structure provided in the embodiments of this application. Detailed Implementation

[0021] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0022] In this embodiment of the application, the power system knowledge closed-loop hybrid enhancement method can be used for continuous optimization of parameter prediction and control strategies for power systems. For example... Figure 1 As shown, a power system refers to an electrical energy production and consumption system composed of power generation, transmission, transformation, distribution, and consumption. Therefore, a power system can include power generation equipment, transmission equipment, transformation equipment, distribution equipment, and consumption equipment.

[0023] Power generation equipment can be categorized into renewable energy equipment and non-renewable energy equipment based on energy type. Renewable energy equipment includes distributed energy sources such as photovoltaics and wind power. When these distributed energy generation devices are connected to the power system, they impact the power grid. Due to the volatility and uncertainty of renewable energy sources like photovoltaics and wind power, key parameters of the power grid, such as voltage and frequency, are affected. Therefore, to meet the power demands of equipment in the power system, transmission equipment, in conjunction with substation and distribution equipment, can regulate the voltage and frequency of the power grid to ensure the stable operation of the power system.

[0024] For example, in the context of microgrids and smart grids, power systems not only need to cope with the integration of large-scale distributed generation resources, but also need to consider the coordination and dispatch of multiple time scales and multiple energy types, which places demands on the control and dispatch capabilities of power systems.

[0025] In some embodiments, voltage and frequency control of a power system can rely on control methods such as proportional-integral-derivative (PID) control, optimal control, and robust control. These control methods are widely used in power systems, especially when the grid load is relatively balanced and there is no large-scale integration of renewable energy, and can effectively maintain the stability of the power system.

[0026] However, with the integration of renewable energy, the complexity and uncertainty faced by power systems have increased significantly. For example, PID control cannot effectively handle the nonlinear and time-varying characteristics of the power grid; while linear control methods can achieve a certain degree of voltage and frequency control through optimization algorithms, they ignore various complex factors existing in the system; and while adaptive control methods can dynamically adjust according to the system state, they rely on accurate system models and are difficult to achieve ideal control effects when facing large-scale complex power systems. Therefore, the control accuracy and real-time performance of the above-mentioned control methods are low, and they cannot meet the needs of power systems.

[0027] To cope with the ever-changing and complex power systems, some implementations can also leverage artificial intelligence and machine learning technologies for power system control. For example, reinforcement learning (RL) techniques can be used to achieve power system scheduling and control. In areas such as microgrid, energy storage system scheduling, and renewable energy power regulation, reinforcement learning methods learn optimal control strategies gradually through the interaction between the agent and the environment, achieving adaptive control by relying on large amounts of data and feedback even without a clear model.

[0028] However, reinforcement learning-based power system control methods, in power systems with multiple time scales and resource scheduling, require substantial computational resources and training data during model training and lack sufficient expert experience support. This results in slow convergence and unstable performance in some complex scenarios. Furthermore, due to the complex and rapidly changing operating environment of power systems, reinforcement learning methods cannot respond quickly and effectively to sudden disturbances, reducing the accuracy and real-time performance of power system control.

[0029] To address the issues of low control accuracy and real-time performance in power systems, some embodiments of this application provide a power system knowledge closed-loop hybrid enhancement method. This method can optimize the prediction and control strategies of power systems through real-time data acquisition and expert feedback, combined with a two-way evolution mechanism of machine learning models and knowledge bases.

[0030] The method can be applied to electronic devices with data processing capabilities. These electronic devices include, but are not limited to, computers, servers, mobile terminals, smart wearable devices, and industrial control machines. For ease of description, this application embodiment uses an electronic device as the execution subject of the method. It should be understood that the method can also be applied to other types of execution subjects, which are not illustrated in this application embodiment. Figure 2 As shown, the method includes: S101. Obtain runtime data.

[0031] To optimize power system forecasting and control strategies, operational data must first be acquired. Operational data refers to data that comprehensively reflects the current state of the power system, providing a foundation for subsequent forecasting and control decisions. This operational data includes power system power data, voltage data, meteorological data, load data, and system status. Data acquisition is the starting point of the entire control process. Real-time acquisition of various operational data from the power system, including load, voltage, power, and meteorological information, ensures that the power system receives the latest operational information.

[0032] For example, the data acquisition process may involve several important variables, including: power data (including active power Pt and reactive power Qt), voltage data (including voltage amplitude Vt and voltage phase angle θt), meteorological data Wt, grid load Lt, energy storage status Et, etc.

[0033] In some embodiments, different acquisition frequencies can be set according to the nature of different data. For example, power data, voltage data, etc., can be acquired every 15 minutes or 5 minutes; meteorological data may need to be acquired hourly or more frequently to capture rapid changes in wind speed and light intensity.

[0034] Since operational data can be correlated with power grid equipment with specific functions, when acquiring operational data, the power grid equipment in the power system can be traversed and its functions labeled, and the specific content contained in the operational data can be adjusted based on the traversal and function labeling.

[0035] For example, photovoltaic modules, energy storage systems, controllable loads, and electric vehicle charging stations are key components of the system, providing electrical energy, storing electrical energy, regulating power consumption, and charging power, respectively, to support the power balance and voltage stability of the power grid. By collecting key variables such as photovoltaic power generation, the state of charge (SOC) of the energy storage system, the power of the controllable load, the charging power of electric vehicles, the grid voltage, and meteorological data in real time, the operating status of each distributed energy source in the microgrid and the power grid can be obtained.

[0036] After acquiring data through real-time data acquisition, the acquired data can be preprocessed. In some embodiments, raw operational data can be obtained first from a data source device, which includes at least one of a sensor and a data acquisition device. Then, standardization processing is performed on the raw operational data to transform all variables in the raw operational data to the same magnitude. Finally, the raw operational data of the same magnitude is segmented according to a time window to obtain the operational data.

[0037] Raw operating data from various devices can be acquired in real time through sensors and data acquisition equipment installed in the power system. Raw operating data refers to the detection data before preprocessing. Raw operating data can include photovoltaic power generation, the status of energy storage systems, the power consumption of controllable loads, electric vehicle charging power, grid voltage, and meteorological data, etc.

[0038] Photovoltaic power generation P pv The output power of a photovoltaic (PV) power generation system at a specific moment is expressed in kW. PV power generation is affected by sunlight intensity and weather conditions. The State of Charge (SOC) of an energy storage system represents its state of charge (%). SOC reflects the current battery charge level, ranging from 0% (fully discharged) to 100% (fully charged). The power consumption P of a controllable load... load This represents the power consumption of a controllable load (unit: kW). Controllable loads can be adjusted according to the needs of the power grid. Electric vehicle charging power P EV This indicates the power consumption of the electric vehicle charging station (unit: kW). The charging power of the electric vehicle will be dynamically adjusted according to the voltage and power demand of the power grid. Power grid voltage V grid This represents the voltage of the power grid (unit: V). Power grid voltage is a crucial indicator of power grid stability and must be maintained within a certain range. Meteorological data I solar Light intensity (unit: W / m²) 2 ), directly affecting the output power of the photovoltaic power generation system, v wind This indicates wind speed (unit: m / s), which affects the meteorological conditions of the system and the possible wind power generation.

[0039] After obtaining the raw running data, it can be standardized to ensure that all variables are on the same order of magnitude. For example, by normalization, the raw running data can be transformed to the range of [0, 1] to avoid the excessive influence of certain variables during model training.

[0040] To facilitate subsequent processing by the prediction model, the standardized data can be divided into time windows to form a training set. For example, data from the past N time steps can be used to predict the state at the current time step.

[0041] S102. Calculate the time-series prediction results of the running data using the prediction model.

[0042] After acquiring the operational data, it can be input into the prediction model to calculate the time-series prediction results. The prediction model is a machine learning-based power phase-preserving complex-valued network model; it is modeled using complex-domain hidden states and unitary state transition matrices to capture dynamic phase changes and fluctuation characteristics in the power system based on the operational data.

[0043] After inputting operational data into the prediction model, the state of the power system can be predicted in time series using an improved power phase-preserving complex value network model. This power phase-preserving complex value network can capture the dynamic phase changes and fluctuation characteristics in the power system, accurately predict the grid behavior at future moments, and identify potential nonlinear characteristics such as voltage fluctuations and load changes in advance, providing a reliable decision-making basis for subsequent control strategies.

[0044] Therefore, before performing time series prediction, the prediction model can be built and trained first. That is, in some embodiments, input data can be extracted from the running data. The input data is a vector containing multiple time series features within each time step.

[0045] For example, the input data for a predictive model may include Where t represents the current time. To reflect the dynamic phase changes and fluctuation characteristics of the power system, a complex mapping can be defined, namely:

[0046] in, For time difference, Representing the state phase, complex inputs help capture the amplitude and phase characteristics of the signal.

[0047] The input undergoes a linear transformation to obtain the complex hidden layer input, i.e.:

[0048] in, Indicates the length of the input sequence. Indicates the number of channels.

[0049] A predictive model is then constructed, namely, a power phase-preserving complex-valued network model. This model structure, by introducing complex-domain hidden states and a unitary state transition matrix, enhances the model's ability to model time-series characteristics of the power system, such as load fluctuations, phase drift, and periodic disturbances. Specifically, the input layer of the predictive model is configured to receive input data within a time window; the output layer is configured to output time-series prediction results in complex form, and to map these results to the real domain through a linear transformation to obtain predicted values ​​of the power grid state.

[0050] To construct a prediction model, in some embodiments, the hidden state of the previous time step can be obtained, and learning parameters can be set. The learning parameters include a first learning parameter and a second learning parameter. The first learning parameter is the weight matrix of the input data at the current time step, and the second learning parameter is the weight matrix of the hidden state at the previous time step.

[0051] Next, candidate hidden states are calculated based on the hidden states and learning parameters. Then, the gating parameters of the gated recurrent units in the prediction model are obtained, and the hidden state update function of the prediction model is constructed based on the gating parameters, the hidden state of the previous time step, and the candidate hidden states. For example... Figure 3 As shown, the gate control parameters include updating the gate and resetting the gate.

[0052] For example, the hidden state of the model is a complex vector. The rules for updating the gate are as follows:

[0053] in, This represents the sigmoid activation function, which controls the degree to which information is retained.

[0054] The door reset rules are:

[0055] in, This represents the sigmoid activation function, which controls the proportion of historical states involved in the current memory update.

[0056] The reset gate can control the influence of the previous hidden state on the current hidden state, so the candidate hidden state can be represented as:

[0057] in, and For learning parameters, It is the hyperbolic tangent activation function.

[0058] The candidate hidden state update rule is as follows:

[0059] Therefore, we get:

[0060] in Let be a unitary matrix, satisfying:

[0061] To ensure the L2 norm of the hidden state is maintained:

[0062] This property prevents gradient vanishing or exploding in long-sequence modeling and is applicable to stable training of data spanning day and night or seasons in power systems. Therefore, the hidden state update function of the prediction model can be expressed as:

[0063] In power phase-preserving complex-valued networks, the state transition matrix is ​​a unitary matrix to ensure that the L2 norm of the hidden states remains unchanged, preventing gradient explosion or vanishing problems. This matrix ensures the stability of complex hidden states during long training sequences.

[0064] Next, an error function for the prediction model is set, where the error function is the complex mean squared error. Then, based on the error function, the prediction model is trained using the input data. During the training process, the Wirtinger derivative is used to calculate the gradient, and the network weights are updated using gradient descent.

[0065] For example, the predicted output is obtained by linear decoding of the complex hidden state:

[0066] For voltage and power prediction tasks, the output can be mapped back to the real number domain:

[0067] The prediction error is expressed as the complex mean square modulus error:

[0068] in, It is the complex form of the objective value.

[0069] Since the parameters are in complex form, the Wirtinger derivative is used for gradient updates:

[0070] The parameter update formula is:

[0071] in, ηThe learning rate is denoted by . This mechanism effectively ensures the stability of the gradient direction in the complex domain. Through this innovative network model, the system not only optimizes its ability to predict power system behavior but also provides effective early warnings for control strategy formulation and precisely optimizes countermeasures.

[0072] After constructing the prediction model based on the model parameters set above, the prediction model can be trained using the running data from the training set. In some embodiments, the output of the power phase-preserving complex network is a prediction result in complex form, which is mapped to the real domain through a linear transformation to obtain the predicted value of the power grid state. The output layer can then be represented as:

[0073] in, It is a predicted value. and The parameters are for learning, and ReLU is a linear activation function.

[0074] The error calculation for the power phase-preserving complex-valued network model uses the complex mean square error (CMSE), i.e.:

[0075] in, For predicted values, is the actual value, and N is the sample size.

[0076] During training, gradient descent is used to update the network weights. Since both the input data and hidden states are complex numbers, the Wirtinger derivative is used for gradient calculation to ensure the stability of the gradient direction in the complex domain. The parameter update formula is:

[0077] in, η For learning rate, These are the model parameters.

[0078] After constructing the prediction model, operational data can be input into the model to calculate time-series predictions for that data. During the prediction process, an improved power phase-preserving complex value network model can be used to predict future power, load, or energy storage status, yielding predicted values. .

[0079] For example, using a trained power phase-preserving complex-valued network model will output the predicted state of the power grid, including: voltage (V... grid ), power (P) pv P Load P EV The predictions will provide data support for subsequent control decisions and expert feedback, including the energy storage system state (E).

[0080] S103. Evaluate the time-series prediction results according to the expert agent collaborative evolution interaction mechanism to obtain evaluation feedback information.

[0081] After obtaining the time-series prediction results, an evaluation can be performed on the results according to the expert agent collaborative evolutionary interaction mechanism to obtain evaluation feedback information. This feedback information may include adjustment terms available to the prediction model, enabling knowledge updates and optimizations through bidirectional interaction based on the expert agent collaborative evolutionary interaction mechanism.

[0082] In this context, "expert" can broadly refer to any entity capable of providing expert-level feedback, such as an expert model embedded in the power system or an expert system connected to the power system. Experts can not only evaluate the system's predictions but also propose optimization suggestions based on the actual state of the power system.

[0083] In some embodiments, to obtain evaluation feedback information, when evaluating the time-series prediction results according to the expert agent collaborative evolution interaction mechanism, the prediction parameter items corresponding to the time-series prediction results can be read first, and the actual observed values ​​corresponding to the prediction parameter items can be obtained. The prediction parameter items include predicted voltage, predicted power, and predicted system state. Correspondingly, the actual observed values ​​include the observed voltage, observed power, and observed system state collected from the power system.

[0084] The prediction error is then calculated by comparing the time-series prediction results with the actual observations. If the prediction error meets any preset trigger condition, expert feedback data can be obtained and mapped into evaluation feedback information to update the model parameters and knowledge information of the prediction model. The expert feedback data includes power grid state assessment results, control strategy adjustment suggestions, and prediction model adjustment suggestions.

[0085] For example, such as Figure 4 As shown, during power system operation, the predicted values ​​of the power grid state can be output by the power phase-preserving complex value network prediction model and compared with the actual observed values. If the prediction errors such as grid voltage and power deviation exceed any preset trigger condition, the expert agent collaborative evolution interaction mechanism can be triggered.

[0086] The triggering conditions can be used to judge the real-time error, continuous prediction deviation, abnormal fluctuations, etc. of the prediction results. Accordingly, the preset triggering conditions can include a predetermined error threshold, a continuous deviation threshold, and a fluctuation range.

[0087] For example, regarding persistent prediction deviations in the prediction results, when the prediction error exceeds a preset error threshold ε and persists for more than K steps, it is determined that the prediction error meets a preset trigger condition, at which point expert feedback can be triggered. Similarly, for the judgment of abnormal fluctuations, when the change in grid voltage or power exceeds a certain set fluctuation range, it is also determined that the prediction error meets a preset trigger condition, thus triggering expert feedback.

[0088] When the timing prediction result meets any triggering condition, the power system can send a feedback request to the expert system to trigger information feedback. The expert system can provide feedback information through system interfaces (such as UI interfaces, log analysis, etc.). By analyzing the current grid status, such as voltage fluctuations, energy storage charging and discharging status, and photovoltaic output fluctuations, the expert system can provide expert feedback data containing specific information.

[0089] For example, expert feedback data can include assessments of grid conditions, such as excessively low or high voltage, or large power fluctuations. Expert feedback data can also include recommendations for adjusting control strategies, such as regulating the power output of energy storage systems, adjusting controllable loads, or adjusting the charging power of electric vehicles. Furthermore, expert feedback data can include recommendations for adjusting predictive models, such as adjusting model weights or increasing data samples for specific scenarios.

[0090] After obtaining expert feedback data from the expert system, the expert feedback data can be mapped into evaluation feedback information using information mapping, i.e., according to the following formula:

[0091] in, Adjustments to parameters recommended by experts. This is the mapping function for expert feedback.

[0092] Information mapping can transform expert feedback data into system-usable adjustment items to optimize power system control strategies.

[0093] S104. Update the model parameters and knowledge information of the prediction model based on the evaluation feedback.

[0094] After each expert feedback, the power system can update the model parameters and knowledge information of the prediction model based on the evaluation feedback information. That is, the prediction model structure or parameters are automatically adjusted according to the feedback content to ensure that the subsequent control process can be effectively improved.

[0095] In some embodiments, when updating the model parameters and knowledge information of the prediction model based on evaluation feedback, knowledge base entries can be generated first based on expert feedback data. These knowledge base entries include triggering conditions, influencing variables, suggested operations, and priorities.

[0096] For example, adjusting parameters based on expert suggestions provided by the expert system. The power system can generate a knowledge base entry. K j Knowledge base entries can be formally represented as follows:

[0097] in, This represents a triggering condition, which is a conditional expression, such as... . The influencing variable is the set of variables in the affected system, such as... . This indicates a suggested action, that is, an action recommended by the expert, such as "in the next model training, weight the data for this case". Increase to ≥1.2”; Indicates priority or credibility rating. .

[0098] After generating knowledge base entries, the model knowledge base of the prediction model is updated using these entries. In some embodiments, when updating the model knowledge base of the prediction model using knowledge base entries, target fields can first be read from expert feedback data. These target fields include at least one of three types: first-type fields, second-type fields, and third-type fields. First-type fields are used to trigger dataset reconstruction, such as "adding samples" or "correcting labels." Second-type fields are used to correct the loss function, such as "increasing penalties for specific situations." Third-type fields are used to trigger updates to the model structure or model parameters, such as "adjusting model gating parameters" or "introducing additional state variables z."

[0099] Depending on the type of target fields included in the expert feedback data, different methods can be used to update the model knowledge base. Specifically, when the expert feedback data includes the first type of field, the training set is modified based on the knowledge base entries, and weights are assigned to newly added samples.

[0100] For example, if O j Fields containing operations such as "add samples" or "correct labels" will trigger dataset reconstruction, i.e., modify the training set D.

[0101] It can also assign weights to newly added samples. β j ,Right now:

[0102] When the expert feedback data includes a second type of field, weights are defined based on the triggering conditions, and a weighted loss is calculated based on the defined weights and the loss function of the prediction model. For example, if Oj Including "increasing penalties for specific situations" triggers a correction to the loss function, thus allowing the definition of a weighted loss:

[0103] Among them, w t This indicates that weights can be defined based on whether the trigger condition T is met. j Set to β j Or 1.

[0104] When expert feedback data includes a third type of field, define a parameter set update function, and adjust the model structure and / or update the model parameters based on the parameter set update function. For example, if Fields such as "adjust model gating parameters" or "introduce additional state variable z" trigger updates to the model structure or model parameters. Therefore, parameter set updates can be defined, i.e.:

[0105] in, Represents the parameter set; L j Based on trigger conditions T j subset loss; η j This is the learning rate.

[0106] After updating the model knowledge base of the prediction model with knowledge base entries, the associated entry set is retrieved in the updated model knowledge base based on the current system state vector, and the associated entry set and suggested operations are injected into the current round of model training or control module.

[0107] For example, in each operating cycle of the power system (e.g., daily / hourly), it can be based on new entries. Based on the corresponding model adjustment results, the priority or credibility of existing entries in the knowledge base is dynamically updated, i.e.:

[0108] in, This represents the amount of error improvement, used to measure the improvement in prediction error after an item is executed, and can be defined as follows: , γ To adjust the rate, i.e. .

[0109] Given the current system state vector, the knowledge base's storage and retrieval mechanism is as follows:

[0110] Therefore, the power system can retrieve data that meets the conditions. Collection of entries and its recommendations Inject into the current model training or control module.

[0111] Prediction module output After combining suggestions from the knowledge base entries and model revisions, the process proceeds to the control strategy layer. For example, regarding energy storage power... The control formula can be:

[0112] in, Let the real part be the hidden state. α Basic adjustment coefficient, δ j For the entry K j The recommended weight is... Power adjustment recommendations.

[0113] Similarly, the reactive power output and voltage regulation amount It can also be incorporated into the knowledge base for suggested revisions, namely:

[0114] in, Let the real part be the hidden state. β Basic adjustment coefficient, δ j For the entry K j The recommended weight is... Voltage adjustment recommendations.

[0115] S105. Optimize the control strategy of the power system based on the updated prediction model, and generate closed-loop feedback information by executing the control strategy.

[0116] After updating the prediction model, the control strategy of the power system can be optimized based on the updated prediction model, realizing reinforcement learning control based on expert feedback information.

[0117] For example, the state of the system can be defined as:

[0118] in, It is the system load power (kW). It is the system reactive power (kVAR). It is the system voltage (V). It is the system phase angle (°). It is the energy storage system status (SOC, %). It is the load forecasting error (kW). It refers to the power generation capacity of renewable energy (kW). It is the radiation temperature (°C).

[0119] Accordingly, the actions of the optimized agent are:

[0120] in, It refers to the power dispatch (kW) of the energy storage system. It is the inverter reactive power regulation (kVAR). This refers to load dispatch adjustments (kW). These actions determine the system's dynamic response and influence the evolution of its subsequent state.

[0121] The strategy for optimizing the agent is:

[0122] The strategy is a parameterized function that maps the current state. To action .

[0123] Therefore, the initial reward function can be... Designed as follows:

[0124] in, It is the system power imbalance. This is the reference voltage value. These are operating costs, such as fuel consumption and equipment wear and tear; α 1, α 2, α 3 represents the weighting factor. This reward function aims to minimize power imbalance, voltage deviation, and operating costs.

[0125] Based on expert feedback data, the power system performed well at critical moments during each operation. Triggering expert evaluation, i.e., experts observing the system status. intelligent agent actions Its effects, such as voltage fluctuations and energy storage dispatch results, are then used to provide feedback signals. Feedback signal This includes: real number scores (such as "-2 to +2"), preference comparisons (option A is better than option B), and suggestions for correction, such as reducing the energy storage discharge rate.

[0126] Map expert feedback to additional rewards or corrective measures in reinforcement learning:

[0127] in, β It is the weight of expert feedback.

[0128] In action selection or strategy updates, feedback intervention points can also be considered to introduce a dynamic "manual / automatic" control switching mechanism, namely:

[0129] Furthermore, optimizing the agent can also use "experience when experts intervene" as training data.

[0130] The time steps including expert feedback are recorded as follows: Add a priority sampling mechanism during experience revisiting or training, namely:

[0131] in, It is a moment i TD error, It is a weighting factor.

[0132] The corresponding training loss is:

[0133] And add mandatory correction items based on expert feedback:

[0134] in, η For feedback correction weights.

[0135] When experts intervene and provide feedback, the system automatically generates or updates the corresponding knowledge base entries:

[0136] And update credibility:

[0137] in, This indicates the improvement in system performance after using this entry, such as the reduction rate of power imbalance.

[0138] When selecting a strategy, prioritize triggering strategies that satisfy certain conditions. The entries in the condition, in the current state satisfy Strategies can be added to item suggestions. As a correction to the action:

[0139] in, k j It is the item weight.

[0140] To ensure the stability and efficiency of policy updates, the Proximal Policy Optimization (PPO) algorithm is employed. PPO avoids performance crashes caused by excessively large policy updates by limiting the magnitude of policy updates, while maintaining high sample efficiency.

[0141] The PPO algorithm updates the policy by maximizing the following objective function: ; in, The dominant function; Is the current strategy in the state? Select action The probability of; It's an old strategy; This is a hyperparameter used to control the magnitude of policy updates.

[0142] The advantage function is calculated using generalized advantage (GAE) estimation, which improves the stability of policy updates by reducing the variance of the estimation. The advantage function formula is:

[0143] in, It is the TD residual. It is a discount factor. These are hyperparameters of GAE.

[0144] In some embodiments, when implementing a control strategy to optimize the power system based on an updated prediction model, an expert feedback reward function can also be constructed based on estimated feedback information. This expert feedback reward function is used for voltage deviation correction, power imbalance correction, and operating cost correction.

[0145] Next, expert feedback weights are obtained, and a total reward update strategy function is established based on these weights and the expert feedback reward function. Then, an optimization agent is used to update the power system's control strategy according to the total reward update strategy function. The optimization agent selects control actions based on the current system state and minimizes power imbalance and voltage deviation through iterative optimization.

[0146] For example, expert feedback can be incorporated into a reward function to help the agent optimize its control strategy. Specifically, the reward function is designed as follows:

[0147] in, For the total reward, The original reward for the agent. Rewards for expert feedback Weighting based on expert feedback.

[0148] Expert feedback reward function This can include voltage deviation correction, power imbalance correction, and operating cost correction. Among these, expert recommendations for voltage deviation correction can adjust voltage control strategies to reduce voltage fluctuations, such as… Figure 5 As shown. Power imbalance correction is achieved by adjusting energy storage, photovoltaic output, or load to balance power. Operating cost correction, on the other hand, weights the rewards based on operating costs such as equipment wear and tear and energy consumption, as reported by experts.

[0149] The expert reward function can be expressed as:

[0150] in, For voltage deviation, Due to power imbalance, For operating costs, , , These are the weighting coefficients.

[0151] The agent is based on the new total reward Update the strategy. The total reward function includes expert suggestions, enabling the agent to adjust its behavior under expert guidance. Optimizing the reward function helps the agent learn more reasonable control strategies based on expert feedback.

[0152] The goal of the intelligent agent is to select actions such as energy storage charging and discharging, and load regulation based on the current grid voltage, power and other system states, in order to minimize power imbalance and voltage deviation.

[0153] The state space of an intelligent agent can include grid voltage. Energy storage system status Photovoltaic power generation Controllable load power Electric vehicle charging power wait.

[0154] The state vector of the agent is represented as:

[0155] The action space of an intelligent agent includes energy storage power regulation. Photovoltaic output regulation Load regulation wait.

[0156] The corresponding action vector is then represented as:

[0157] The agent, based on the current state Choose an action And execute it, then based on environmental feedback, including rewards. and the state at the next moment To update the policy, reinforcement learning uses the Proximal Policy Optimization (PPO) algorithm to ensure that each update is not too large, thus maintaining the stability of the policy update.

[0158] To combine expert feedback with automated decision-making, in some embodiments, the power system can also optimize the predictive model and control strategy based on a dynamic human-machine control switching mechanism. For example, in certain situations, expert intervention may be more effective than automated systems; therefore, the system dynamically decides whether to enter expert intervention mode based on the current state and the influence of expert feedback.

[0159] Automatic control mode allows the power system to rely entirely on a reinforcement learning agent for control when expert intervention is unnecessary. Expert control mode, on the other hand, prioritizes adjustments based on expert suggestions when strong expert feedback is received, and incorporates this experience into subsequent reinforcement learning training.

[0160] Based on the current state The influence of expert feedback This determines whether to enter expert control mode. The core of the switching lies in threshold judgment, that is, dynamically adjusting the control mode based on the current state and feedback signals. The expression for the switching rule is:

[0161] in, The influence of expert feedback. This is a preset feedback intensity threshold. If the expert feedback intensity exceeds this threshold, the system will enter expert control mode.

[0162] In expert-controlled mode, the system adjusts the current reward function based on expert feedback. The reward items are determined by expert feedback. This will be incorporated into the overall reward function, influencing the agent's control decisions. The reward function can be expressed as:

[0163] in, To reinforce the regular rewards of the learning agent, objectives such as voltage deviation and power imbalance are included; The reward is for expert feedback, reflecting the experts' assessment and recommendations on the state of the power grid; The weighting coefficient for expert feedback determines its influence on the reward.

[0164] Under expert control mode, the adjustment formula for the control strategy is:

[0165] in, For the adjusted power output, This is the current power output. The target power output recommended by experts based on the current power grid conditions. The adjustment factor represents the adjustment range recommended by experts.

[0166] If the system enters expert control mode, the adjusted power output The system will be dynamically adjusted based on expert advice to restore grid voltage stability or balance power.

[0167] After switching to expert control mode, the system will make control decisions based on new expert feedback, rather than relying entirely on reinforcement learning control by the agent. The control strategy will combine expert adjustment suggestions with the system's real-time status to optimize power grid operation.

[0168] After optimizing the power system control strategy based on the updated prediction model, closed-loop feedback information can be generated by executing the control strategy. This closed-loop feedback information is used to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolutionary interaction mechanism through heterogeneous knowledge fusion.

[0169] By applying the technical solutions of the above embodiments, the power system knowledge closed-loop hybrid enhancement method described in the above embodiments can be based on a novel power system knowledge closed-loop hybrid enhancement driven by human-machine interaction feedback. By collecting power system data in real time and combining it with a power phase-preserving complex value network model for dynamic prediction, it can accurately identify potential system problems and provide decision support, improving the flexibility and response speed of the control process. Simultaneously, by introducing an expert intelligent agent collaborative evolution interaction mechanism, the power system can self-optimize in each round of operation, further improving the accuracy and stability of the control strategy. Through heterogeneous knowledge fusion enhancement closed loop, the power system can combine multi-source information from expert feedback and machine learning models to form a bidirectional evolutionary knowledge update mechanism, continuously improving the system's adaptive capability, thereby optimizing decision-making and timely adjusting to adapt to the dynamic changes of the power system. Furthermore, the bidirectional evolutionary mechanism of the knowledge base enables the system to continuously optimize control decisions as the power system changes, especially when dealing with complex problems such as renewable energy fluctuations and energy storage dispatch, significantly improving grid stability and dispatch efficiency. This method not only enhances the response speed and flexibility of the grid but also has broad application prospects and practical value, especially suitable for variable power system environments.

[0170] In some embodiments, as a refinement and extension of the specific implementation of the above embodiments, and in order to fully illustrate the specific implementation process of this embodiment, some embodiments of this application also provide a power system knowledge closed-loop hybrid enhancement method, such as... Figure 6As shown, the difference between this method and the above embodiment is that, when generating closed-loop feedback information by executing the control strategy, it can also perform self-adjustment again based on the feedback information after execution, forming a closed-loop feedback mechanism. The method includes: S201, Read the control objective and output variables of the control strategy; S202. By executing control strategies, the power system modifies the output variables that affect the control objectives. S203. Generate closed-loop feedback information based on the real-time status of the power system; S204. Update the control strategy through the closed-loop control mechanism based on the closed-loop feedback information.

[0171] After optimizing the power system's control strategy, the control objectives and influencing output variables of the strategy can be identified. By executing the control strategy, the power system can modify the influencing output variables according to the control objectives. Then, closed-loop feedback information is generated based on the real-time state of the power system, and the control strategy is updated through a closed-loop control mechanism based on this feedback information.

[0172] For example, by implementing optimized control strategies, voltage stability and power balance in the power grid can ultimately be achieved. This can include steps such as generating and executing control strategies, as well as feedback and self-adjustment. The power system ensures stable grid operation by precisely regulating the output of distributed energy resources, such as photovoltaics, energy storage, controllable loads, and electric vehicle charging stations.

[0173] The main control objectives during the generation and execution of the control strategy are voltage stability, power balance, and minimization of operating costs. Voltage stability is achieved by regulating the output of distributed energy sources to ensure the grid voltage remains within a set safe range. Power balance ensures a balance between active and reactive power in the grid, preventing overload or power shortages. Minimizing operating costs involves optimizing control decisions to reduce unnecessary equipment operation and energy consumption.

[0174] To achieve these objectives, the control system will generate adaptive control strategies based on the current grid conditions, including voltage, power, and energy storage status. These control strategies can be used to adjust key variables such as the power of the energy storage system, the output of photovoltaic power generation, the power of controllable loads, and the charging power of electric vehicles. When regulating the power of the energy storage system... When performing photovoltaic power generation output regulation, the charging and discharging power of the energy storage system can be dynamically adjusted according to changes in grid load and battery SOC (state of charge). During the process, the output power of the photovoltaic system can be adjusted when photovoltaic power generation fluctuates significantly. This is relevant to power regulation of controllable loads. During the process, the power consumption of controllable loads can be adjusted according to the grid load demand. This applies to the adjustment of electric vehicle charging power. During the process, the power consumption of the charging pile can be adjusted to reduce the charging power under high load and increase the charging power under low load.

[0175] The control objective of the control strategy can be expressed as the following optimization problem:

[0176] in, This indicates the operating cost of the equipment. For voltage deviation, This is due to power imbalance. and This is a weighting coefficient used to balance the priority between voltage control and power balance.

[0177] Therefore, the generated control strategy will affect the following output variables: Energy storage power regulation:

[0178] in, It is the power regulation amount of energy storage. It is the basic regulating power. It is the power of the control strategy adjustment.

[0179] Photovoltaic power regulation:

[0180] in, It is the output power regulation of photovoltaics. It is the base power output of photovoltaics. It is the adjustment amount of the control strategy.

[0181] Controllable load power regulation:

[0182] in, It is the power regulation amount of the controllable load. It is the base power consumption of a controllable load. It is the power of the control strategy adjustment.

[0183] Electric vehicle charging power adjustment:

[0184] in, It refers to the power adjustment of electric vehicle charging stations. This is the basic power for charging electric vehicles. It is the adjustment amount of the control strategy.

[0185] The generated control strategies are executed in real time and affect the entire power system. For example, adjusting the charging and discharging power of energy storage systems helps the grid cope with load fluctuations and voltage instability. Adjusting the operating status of photovoltaic inverters based on photovoltaic power generation output optimizes photovoltaic power distribution. Dynamically adjusting controllable loads reduces power consumption during high loads or increases power consumption during low loads to maintain power balance. Regulating the power of electric vehicle charging stations ensures stable grid operation, etc.

[0186] After implementing the control strategy, the power system generates new feedback, or closed-loop feedback information, based on the real-time status of the grid, such as voltage, power, and energy storage. The power system analyzes this closed-loop feedback information to determine the control effect and adjust the control strategy accordingly.

[0187] Feedback mechanisms for adjusting control strategies can include voltage and power feedback, energy storage status feedback, and expert feedback. Voltage and power feedback monitors grid voltage and power imbalance in real time. When the voltage deviation exceeds a set threshold, the system automatically adjusts energy storage, load, and photovoltaic output. Energy storage status feedback periodically updates the SOC of the energy storage system and adjusts the control strategy based on the charging status of the energy storage. Expert feedback triggers the system to adjust the control strategy if it cannot restore voltage or power balance within a short time.

[0188] The control adjustment formula is as follows:

[0189] in, For the adjusted power output, For the current power, Target power, This is for adjusting the coefficient.

[0190] By continuously updating the control strategy through a closed-loop control mechanism, the operating state of the power grid is optimized after each round of execution. The feedback after each execution will affect the next control decision, thereby gradually optimizing the stability and efficiency of the power grid.

[0191] After each control strategy is executed, the strategy can be adjusted based on the difference (error) between the actual operating state of the power grid and the expected target state. The control error can be expressed as:

[0192] in, For the target power, This represents the current actual power output. This represents the error. Based on the error, the system can determine how to adjust the output of distributed energy sources such as energy storage, photovoltaics, and loads to optimize grid performance.

[0193] Error feedback is incorporated into the adjustment of each round of control decisions, ensuring that each adjustment is based not only on the current error but also on past error accumulation and the system's historical response. A weighted averaging strategy is introduced to calculate the feedback adjustment, making each feedback more stable and gradual. The feedback adjustment coefficient is defined as:

[0194] Among them, α and β are weighting coefficients that control the degree of influence of the current error and the historical error; For the current error, This is the accumulation of all past errors.

[0195] Optimizing the control strategy relies not only on error feedback but also on reinforcement learning algorithms to continuously adjust the system's policy. In this case, the strength of the feedback is related to the system's historical experience and accumulated feedback. Based on error feedback, the system continuously adjusts the reward function to reinforce or suppress certain control behaviors. The reward function of reinforcement learning can be adjusted using the following formula:

[0196] in, The traditional reinforcement learning reward reflects the direct effects of power grid control, such as voltage stability and power balance; λ and γ are adjustment parameters that control the influence of current and historical errors on the reward function. and These are the current and historical control errors, respectively. By introducing weights for these errors, the system can balance short-term and long-term feedback during strategy optimization, ensuring that the system finds the optimal control strategy through continuous adjustments.

[0197] As the time step increases, the system adjusts its control strategy at each step based on updated feedback. To better address control decision-making problems over long time spans, we can introduce a multi-step feedback model to measure state changes at each time step. Multi-step feedback is defined as:

[0198] in, For the adjustment amount of multi-step feedback, γ is the discount factor, which represents the degree of attention the control system pays to future errors; For the first The error at each step, where T is the feedback time span. Through a multi-step feedback model, the system can predict potential future errors and make adjustments in advance, further improving the foresight and stability of the control strategy.

[0199] After each execution, the system updates the control strategy based on the current error, historical feedback, and multi-step feedback. This process is described by the following formula:

[0200] in, To achieve the adjusted power output, various feedback adjustments are weighted and summed to comprehensively consider the current and historical errors. Based on the feedback after each execution, the system adjusts the grid control strategy using these complex formulas, ensuring that each step optimizes towards the goals of voltage stability and power balance. As the control process progresses, the accumulated feedback further optimizes the grid's operating state, enabling the system to adapt to long-term changes and sudden disturbances, such as… Figure 7 As shown.

[0201] By applying the technical solutions of the above embodiments, the power system knowledge closed-loop hybrid enhancement method described in the above embodiments can continuously strengthen and optimize the system's control strategy through each round of execution and feedback, thereby effectively improving the overall stability and efficiency of the power system. This process ensures that the system can respond quickly and continuously optimize in complex power environments, and achieve efficient grid dispatch and control, through the synergistic effect of heterogeneous knowledge fusion enhancement closed loop and expert intelligent agent collaborative evolution interaction mechanism.

[0202] In some embodiments, as a specific implementation of the power system knowledge closed-loop hybrid enhancement method described in the above embodiments, some embodiments of this application also provide a power system knowledge closed-loop hybrid enhancement system, such as... Figure 8 As shown, the system includes: The data acquisition module is used to acquire operational data, which includes power data, voltage data, meteorological data, load data, and system status of the power system. The prediction module is used to calculate the time-series prediction results of the operating data through a prediction model, which is a power phase-preserving complex value network model based on machine learning. The prediction model is modeled through complex domain hidden states and unitary state transition matrices, and is used to capture the dynamic phase changes and fluctuation characteristics of the power system based on the operating data. The prediction and evaluation module is used to evaluate the time-series prediction results according to the expert agent collaborative evolutionary interaction mechanism to obtain evaluation feedback information; the evaluation feedback information includes adjustment terms available for the prediction model. The update module is used to update the model parameters and knowledge information of the prediction model based on the evaluation feedback information; The closed-loop feedback module is used to optimize the control strategy of the power system based on the updated prediction model, and to generate closed-loop feedback information by executing the control strategy, so as to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolution interaction mechanism through heterogeneous knowledge fusion.

[0203] By applying the technical solutions of the above embodiments, the power system knowledge closed-loop hybrid enhancement system described in the above embodiments, after the data acquisition module acquires the operating data of the power system, the prediction module calculates the time-series prediction results of the operating data through the prediction model. The prediction evaluation module then evaluates the time-series prediction results according to the expert intelligent agent collaborative evolution interaction mechanism to obtain evaluation feedback information. Then, the update module updates the model parameters and knowledge information of the prediction model according to the evaluation feedback information, optimizes the control strategy of the power system based on the updated prediction model, and generates closed-loop feedback information by executing the control strategy to achieve continuous optimization of the control strategy. The system combines expert feedback with the physical model through the closed-loop feedback mechanism to ensure that the control strategy is closer to the actual needs of the power system. Furthermore, after each execution of the control strategy, it can be continuously optimized and adjusted through the heterogeneous knowledge fusion mechanism, and the prediction model can be updated through expert feedback to optimize voltage control and power balance, thereby ensuring the stable operation of the power system. The system can effectively improve the power system's ability to cope with the coordinated scheduling of multiple time scales and multiple energy types, enhance the stability and response speed of the power system when facing complex disturbances, and improve the control accuracy and real-time performance of the power system.

[0204] It should be noted that other corresponding descriptions of the functional units involved in the power system knowledge closed-loop hybrid enhancement system provided in the embodiments of this application can be found in the corresponding descriptions in the power system knowledge closed-loop hybrid enhancement method provided in the above embodiments, and will not be repeated here.

[0205] This application also provides a computer device, specifically a personal computer, server, network device, etc. The computer device includes a bus, processor, memory, and communication interface, and may also include input / output interfaces and a display device. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores location information. The network interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0206] Those skilled in the art will understand that the structure of the computer device described above is only a partial structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components, or combine certain components, or have different component arrangements.

[0207] In one embodiment, a computer-readable storage medium is also provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0208] In one embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0209] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0210] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0211] Any references to memory, database, or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0212] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0213] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0214] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0215] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A knowledge-closed-loop hybrid enhancement method for power systems, characterized in that, The method includes: Acquire operational data, including power data, voltage data, meteorological data, load data, and system status of the power system; The time-series prediction results of the operating data are calculated by a prediction model, which is a power phase-preserving complex value network model based on machine learning. The prediction model is modeled using complex domain hidden states and unitary state transition matrices, and is used to capture the dynamic phase changes and fluctuation characteristics of the power system based on the operating data. The time-series prediction results are evaluated according to the expert agent collaborative evolution interaction mechanism to obtain evaluation feedback information; the evaluation feedback information includes adjustment terms available for the prediction model. The model parameters and knowledge information of the prediction model are updated based on the evaluation feedback information; The control strategy of the power system is optimized based on the updated prediction model, and closed-loop feedback information is generated by executing the control strategy to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolution interaction mechanism through heterogeneous knowledge fusion.

2. The method according to claim 1, characterized in that, Obtain runtime data, including: Raw operational data is obtained from a data source device, which includes at least one of a sensor and a data acquisition device; The raw runtime data is standardized to transform all variables in the raw runtime data into the same order of magnitude; The original operational data of the same magnitude is divided into data segments according to time windows to obtain the operational data.

3. The method according to claim 1, characterized in that, The method further includes: Input data is extracted from the running data, and the input data is a vector containing multiple temporal features within each time step; The prediction model is constructed such that its input layer is configured to receive input data within a time window; and its output layer is configured to output time-series prediction results in complex form, and to map the complex-form time-series prediction results to the real number domain through a linear transformation to obtain predicted values ​​of the power grid state. The error function of the prediction model is set, wherein the error function is the complex mean square error; Based on the error function, the prediction model is trained using the input data; during the training of the prediction model, the Wirtinger derivative is used to calculate the gradient, and the network weights are updated using the gradient descent method.

4. The method according to claim 3, characterized in that, Constructing the prediction model includes: Get the hidden state of the previous time step; The learning parameters are set, including a first learning parameter and a second learning parameter. The first learning parameter is the weight matrix of the input data at the current time step; the second learning parameter is the weight matrix of the hidden state at the previous time step. Calculate candidate hidden states based on the hidden states and the learning parameters; Obtain the gating parameters of the gated loop unit in the prediction model, wherein the gating parameters include the update gate and the reset gate; The hidden state update function of the prediction model is constructed based on the gating parameters, the hidden state of the previous time step, and the candidate hidden states.

5. The method according to claim 1, characterized in that, The time-series prediction results are evaluated according to the expert agent collaborative evolution interaction mechanism to obtain evaluation feedback information, including: Read the prediction parameter items corresponding to the time-series prediction results, including the prediction voltage, prediction power, and prediction system state; Obtain the actual observed values ​​corresponding to the predicted parameter items, including the observed voltage, observed power, and observed system status collected from the power system; The prediction error is calculated by comparing the time-series prediction results with the actual observed values. If the prediction error meets any preset triggering condition, expert feedback data is obtained. The expert feedback data includes power grid status assessment results, control strategy adjustment suggestions, and prediction model adjustment suggestions. The expert feedback data is mapped to the evaluation feedback information.

6. The method according to claim 5, characterized in that, The model parameters and knowledge information of the prediction model are updated based on the evaluation feedback information, including: Knowledge base entries are generated based on the expert feedback data. These entries include triggering conditions, influencing variables, suggested actions, and priorities. Update the model knowledge base of the prediction model using the knowledge base entries; Based on the current system state vector, retrieve the set of related entries in the updated model knowledge base; The set of associated entries and the suggested operations are injected into the current model training or control module.

7. The method according to claim 6, characterized in that, Updating the model knowledge base of the prediction model using the knowledge base entries includes: Read the target field from the expert feedback data, the target field including at least one of the first type of field, the second type of field, and the third type of field; When the expert feedback data contains the first type of field, the training set is modified and new sample weights are assigned based on the knowledge base entries. When the expert feedback data includes a second type of field, a defined weight is set according to the triggering condition, and a weighted loss is calculated based on the defined weight and the loss function of the prediction model; When the expert feedback data includes a third type of field, a parameter set update function is defined, and the model structure of the prediction model and / or the model parameters are adjusted based on the parameter set update function.

8. The method according to claim 1, characterized in that, Optimizing the power system control strategy based on the updated prediction model includes: Based on the estimated feedback information, an expert feedback reward function is constructed, which is used for voltage deviation correction, power imbalance correction, and operating cost correction. Obtain the expert feedback weights, and establish a total reward update strategy function based on the expert feedback weights and the expert feedback reward function; An optimization agent is used to update the control strategy of the power system according to the total reward update strategy function; the optimization agent is used to select control actions according to the current system state, and to minimize power imbalance and voltage deviation through iterative optimization.

9. The method according to claim 1, characterized in that, The closed-loop feedback information is generated by executing the control strategy, including: Read the control objective and the output variables that affect the control strategy; By executing the control strategy, the power system modifies the influencing output variables according to the control objective; The closed-loop feedback information is generated based on the real-time status of the power system. The control strategy is updated through a closed-loop control mechanism based on the closed-loop feedback information.

10. A knowledge-closed-loop hybrid enhancement system for power systems, characterized in that, The system includes: The data acquisition module is used to acquire operational data, which includes power data, voltage data, meteorological data, load data, and system status of the power system. The prediction module is used to calculate the time-series prediction results of the operating data through a prediction model, which is a power phase-preserving complex value network model based on machine learning. The prediction model is modeled through complex domain hidden states and unitary state transition matrices, and is used to capture the dynamic phase changes and fluctuation characteristics of the power system based on the operating data. The prediction and evaluation module is used to evaluate the time-series prediction results according to the expert agent collaborative evolutionary interaction mechanism to obtain evaluation feedback information; the evaluation feedback information includes adjustment terms available for the prediction model. The update module is used to update the model parameters and knowledge information of the prediction model based on the evaluation feedback information; The closed-loop feedback module is used to optimize the control strategy of the power system based on the updated prediction model, and to generate closed-loop feedback information by executing the control strategy, so as to continuously optimize the control strategy by enhancing the synergistic effect of the closed-loop and expert agent collaborative evolution interaction mechanism through heterogeneous knowledge fusion.

Images