Intelligent power supply planning method based on medium-voltage power distribution network

Abstract

The invention provides an intelligent power supply planning method based on a medium-voltage power distribution network. The method comprises the following steps: constructing a digital twinborn model according to preprocessed data; generating a self-adaptive dynamic cloning model pool, monitoring future load change and equipment state of the power grid, and coping with different power grid operation conditions and sudden accidents; according to real-time state data output by the clone model pool and the digital twin model, load prediction and power supply strategy simulation are carried out, and an optimal strategy is selected to cope with various possible power grid operation conditions and emergencies; combining a multi-modal load prediction result and carrying out priority ranking according to comprehensive score calculation; the optimal power supply strategy is executed, real-time monitoring is carried out, and strategy parameters are dynamically adjusted when the strategy is executed; according to the data monitored and fed back in real time, a fault prediction and self-healing mechanism is implemented. The invention provides an intelligent, dynamic and integrated solution, and lays a foundation for the development of a future intelligent power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of medium voltage distribution networks, and in particular relates to an intelligent power supply planning method based on a medium voltage distribution network. Background Art

[0002] The medium-voltage distribution network is an important part of the power system, responsible for transmitting electric energy from the high-voltage power grid to the low-voltage distribution network and finally to the user end. With the accelerated urbanization process and the continuous increase in electricity demand, the structure and operating environment of the medium-voltage distribution network have become more and more complex. This complexity is reflected in the following aspects: First, the medium-voltage distribution network covers a wide range, with a large number of lines and nodes, and there are significant differences in load characteristics in different regions, and load fluctuations are frequent and unpredictable. Secondly, with the access of distributed energy resources (DERs) and the increase in the proportion of renewable energy generation, the operating characteristics of the medium-voltage distribution network have become more complicated, which puts higher requirements on power supply stability and power quality. Traditional power supply planning methods for medium-voltage distribution networks mainly rely on static models and pre-set scheduling strategies, which have obvious shortcomings in dealing with load fluctuations and faults.

[0003] Existing technologies mainly use fixed scheduling strategies or static optimization methods based on historical data for power supply planning. These methods face multiple challenges in practical applications. First, traditional methods lack the ability to dynamically respond to changes in real-time load and network status, and cannot adapt to rapidly changing load demands and network topology in a timely manner. This lag can easily lead to problems such as local overload, voltage deviation, and low power supply efficiency. Secondly, due to the wide variety of equipment and complex operating environment in the medium-voltage distribution network, it is difficult for traditional static planning methods to accurately predict and respond to equipment failures and emergencies. Third, with the advancement of information and communication technology, although some systems have introduced data-driven prediction and optimization methods, these systems usually rely on a large amount of labeled data for training, with high data acquisition costs and limited model generalization capabilities, making it difficult to adapt to the ever-changing distribution network environment. In addition, existing technologies often lack a unified and intelligent power supply management platform to integrate various types of information and optimization strategies, making it difficult to achieve global optimization and intelligent management of the medium-voltage distribution network.

[0004] Therefore, under the current technological status quo, how to effectively improve the real-time, reliability and efficiency of medium-voltage distribution network power supply planning has become a difficult problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to design an intelligent power supply planning method based on a medium-voltage distribution network, through an adaptive dynamic cloning model and digital twin technology. This method adaptively generates the optimal power supply strategy model through real-time data collection and analysis, and uses digital twin technology for precise simulation and optimization, which can significantly improve the responsiveness and operating efficiency of the medium-voltage distribution network in complex environments.

[0006] In order to achieve the above object, the present invention provides an intelligent power supply planning method based on a medium voltage distribution network, the method comprising the following steps: Step 1: collect data in real time and preprocess it, and build a digital twin model based on the preprocessed data; specifically, the following steps are included: S11. Install smart sensors at key nodes of the medium voltage distribution network to collect multi-modal parameters of the power grid in real time; S12, using an exponentially weighted moving average algorithm to smooth the time series data of the multimodal parameters; S13, constructing a digital twin model based on the smoothed data; S14. Use the Kalman filter algorithm to correct the digital twin model by combining the predicted state and actual observation data; Step 2: Generate an adaptive dynamic clone model pool based on the output status data of the digital twin model to predict the future load changes and equipment status of the power grid, and respond to different power grid operating conditions and emergencies; specifically, the following steps are included: S21. Extract features of the state data output by the digital twin model to obtain a feature vector; S22, construct multiple prediction models, design a clone model pool containing multiple prediction models, wherein each prediction model is used for prediction under a specific load mode or state, and then define the multiple prediction models as linear regression models for initialization to make a preliminary prediction of the load of the power grid; S23. When multiple prediction models are initialized, an adaptive weight update algorithm based on real-time feedback is designed to dynamically predict the weight and bias of each model, so that it can better predict the actual grid load; S24. Design an error-based weighting strategy to select the model that best suits the current state or fuse the prediction results of multiple models; Step 3: Based on the real-time status data output by the clone model pool and the digital twin model, load forecasting and power supply strategy simulation are performed. At the same time, by combining the real-time grid status, predicting future load changes and equipment status, different power supply strategies are simulated, and the optimal strategy is selected to deal with various possible grid operating conditions and emergencies; specifically, the following steps are included: S31. Use each prediction model in the clone model pool to predict the load at a future moment according to the state feature vector output by the digital twin model to obtain a multimodal load prediction result; S32, introducing an adaptive regularization term in the prediction to control the response of each prediction model to changes in the power grid state; S33, construct a dynamic model integration strategy that introduces fluctuation adjustment coefficients, and adapt to the dynamic changes of the power grid by adjusting the weight of each prediction model; S34, integrating the future load of each prediction model, simulating multiple power supply strategies in the digital twin model and simulating in the digital twin model, calculating the simulation output of each strategy at the next moment, and calculating the comprehensive score based on multiple evaluation indicators, and selecting the strategy with the highest comprehensive score as the power supply strategy actually implemented; Step 4: Combine the multi-modal load forecast results and the priority ranking calculated based on the comprehensive score to continue optimizing the power supply strategy to ensure that the power grid operates efficiently and stably under complex load conditions; specifically, the following steps are included: S41, defining an optimization function based on the objectives of minimizing power supply loss, maximizing voltage stability, and minimizing equipment overload according to the multimodal load prediction results and the priority ranking calculated according to the comprehensive score; S42, introducing scheduling constraints into the multimodal load forecast results and the priority ranking calculated based on the comprehensive score, to ensure that the optimal scheduling scheme is considered when generating the power supply strategy; S43, combining the target optimization function and the scheduling constraint item to obtain an optimization problem; S44, design an adaptive optimization algorithm to dynamically adjust according to the real-time state of the power grid and load forecast to obtain the optimal power supply strategy; Step 5: Execute the optimal power supply strategy and conduct real-time monitoring, provide feedback on the monitoring results, and dynamically adjust the strategy parameters when executing the strategy to ensure the accuracy of strategy execution and the stable operation of the power grid; specifically, the following steps are included: S51, setting initial parameters for current real-time sensor data, and executing the initial parameters as an execution plan for the optimal power supply strategy; S52. During the execution of the strategy, the real-time monitoring system monitors the key parameters of the power grid, transmits the real-time collected data to the central control system, compares it with the predicted strategy execution results, calculates the execution error, and adjusts the initial parameters to make real-time corrections according to the size and direction of the execution error to obtain the strategy parameters; S53. During the strategy execution process, a dynamic evaluation mechanism is introduced to evaluate the overall operation status of the power grid in real time and obtain evaluation results; Step 6: Implement fault prediction and self-healing mechanisms based on real-time monitoring and feedback data; specifically, the following steps are included: S61, construct a multi-level dynamic prediction model, using the features extracted from real-time data as input to perform fault prediction; S62, adding an adaptive anomaly detection algorithm to fault prediction, comparing the historical state and current state of each node in the power grid, and identifying possible fault modes; once it is detected that the fault probability of a node or multiple nodes exceeds a preset threshold, the self-healing mechanism will be immediately started; the adaptive anomaly detection algorithm includes a set of predefined self-healing strategies, each strategy corresponding to a specific type of fault response; S63. Combining fault response time, restoration cost and grid stability, design a multi-objective optimization problem for self-healing strategies, calculate the priority score of each self-healing strategy, evaluate the comprehensive effect of each strategy, sort all self-healing strategies according to the priority score, and select the strategy with the highest score as the self-healing measure to be finally executed; S64. After selecting the optimal self-healing strategy, immediately execute the strategy to restore the normal operation of the power grid, monitor the real-time status of the power grid, evaluate the effect of the self-healing strategy, and make necessary strategy adjustments based on feedback; S65. After each fault event is handled, a closed-loop optimization process is automatically entered to update the multi-level dynamic prediction model and self-healing strategy set by analyzing the effect of each self-healing and the efficiency of the system response.

[0007] Furthermore, the digital twin model is used to describe the dynamic behavior of the medium voltage distribution network, including: Construct a state space model to represent the digital twin model. The state space model uses a state vector Indicates that the system is at time The global state of the grid, including the voltage and current information of all nodes, is simulated and predicted by constructing a state space model.

[0008] Further, the S23 is specifically expressed as follows: When new data arrives, the prediction error of each prediction model is recalculated, and the weight of the model is adjusted according to the error, as shown below:

[0009] in, is the updated weight vector, is the current weight, is the learning rate, is the actual load value, is the predicted value of the model; By continuously adjusting weights, each prediction model can be optimized under various load modes and adapt to different operating conditions.

[0010] Further, the S24 is specifically expressed as follows: First, the prediction error of each prediction model is calculated, where the prediction error is the square difference between the actual load and the predicted value; Then, based on the prediction deviation, a weight factor of each model is calculated, and the weight factor of each model is used to fuse the prediction results of multiple models; Finally, the influence of the model is dynamically adjusted through the weight factor, and the predictions of each model are combined to obtain a more accurate final prediction value.

[0011] Furthermore, the step 2 further includes: An online adaptation and expansion mechanism is introduced. When the system detects that the prediction error continues to exceed the preset threshold for a period of time, it means that the existing clone model pool cannot adapt to the current power grid status. Then a new prediction model is automatically generated and added to the clone model pool. The new prediction model is initialized based on the parameters of the current optimal model and adjusted through online learning to adapt to the new load pattern.

[0012] Furthermore, the adaptive regularization term Including penalty terms, adjustments to the current load volatility of the power grid and historical data, it is expressed as follows: , in, represents the regularization term, which is used to capture the volatility of the power grid state; Indicates the number of grid nodes; Represents the node weight, which controls the contribution of each node's fluctuation to the overall regularization; and Respectively represent The voltage and current of each node; and Respectively represent The historical average voltage and current of each node are used to calculate the deviation of the current state.

[0013] Furthermore, the objective optimization function based on minimizing power loss, maximizing voltage stability and minimizing equipment overload is expressed as follows: , in, Indicates the power supply strategy Comprehensive optimization goal; They represent the weight coefficients of the objective function, reflecting the importance of each indicator; n represents the total number of nodes; t represents the current time; Indicated in strategy Next, node Power loss; Indicated in strategy Next, node Voltage; Indicates the reference voltage value, used to measure voltage stability; Indicated in strategy Next, node Load; Indicates the maximum allowable load of the node to prevent device overload.

[0014] Further, the S44, designing an adaptive optimization algorithm to perform dynamic adjustment according to the real-time state of the power grid and load forecast, specifically includes: Combining particle swarm optimization and Lagrangian relaxation method, the relationship between the objective function and the constraints is balanced. In each iteration, the algorithm updates the position and velocity of the particles and adjusts the Lagrangian multiplier to relax the scheduling constraints to obtain the optimal power supply strategy, which is expressed as follows: , in, Indicates The power supply strategy for the next iteration; Indicates The power supply strategy for the next iteration; Represents the learning rate, which is used to control the step size of each update; Indicates the power supply strategy The gradient of represents the Lagrange multiplier, which adjusts the relaxation degree of different scheduling strategies; Indicates The optimization goal of the power supply strategy for the next iteration; represents the scheduling constraint; P represents the overall scheduling strategy; Indicates the power supply strategy Next, The scheduling result of a scheduling strategy; Indicates The expected scheduling results of a scheduling strategy.

[0015] Furthermore, the dynamic evaluation mechanism is expressed as follows: , in, It represents the overall evaluation value of the current strategy execution; Indicates the weights of different operating indicators and controls the impact of each indicator on the overall evaluation; If the evaluation function If the preset target value is not reached, the strategy parameters are automatically recalculated and the next round of strategy execution and adjustment cycle is entered until the evaluation function Reach the ideal level.

[0016] Furthermore, the adaptive anomaly detection algorithm is constructed as follows: Setting a dynamic failure prediction index , evaluate the failure probability of each node, and use the Bayesian update mechanism combined with the Markov chain Monte Carlo method to calculate the failure probability, which is expressed as follows: , in, Indicates at time The probability of failure at a given moment; Represents the conditional likelihood function, which means that when executing the parameters The observed eigenvector possibility; represents the prior probability and represents the time Prior knowledge of faults at the moment; Represents the feature vector The marginal probability of The multi-objective optimization problem of self-healing strategy is designed by combining fault response time, restoration cost and grid stability, and the priority score of each self-healing strategy is calculated, which is expressed as follows: , in, Represents a self-healing strategy Priority score; Representation strategy The response time, the shorter the better; Representation strategy the cost of restoration; Representation strategy The impact function on grid stability, the higher the better; Represents the priority weight coefficient, which is used to balance the impact of different goals; Based on priority score Sort all self-healing strategies and select the one with the highest score As a final self-healing measure.

[0017] The beneficial technical effects of the present invention are at least as follows: The present invention introduces an adaptive dynamic cloning model and uses a machine learning algorithm to automatically generate and adjust load forecasting and power supply strategy models under different operating scenarios based on real-time data. The model can adaptively clone and optimize strategies, dynamically respond to the actual operating status of the distribution network, and effectively solve the problem of frequent load fluctuations and the inability of traditional methods to respond in real time. Through online learning and continuous optimization, the model can still maintain efficient and reliable power supply planning in a complex and changing environment.

[0018] The present invention reflects the physical state and operating parameters of the distribution network in real time by constructing a digital twin model of the medium-voltage distribution network, thereby realizing accurate simulation and global optimization of the entire power grid. Digital twin technology can integrate multi-source data and dynamically update the model status, so that the optimization of the power supply strategy is not only based on current data, but also can simulate and verify future states, ensuring the optimality and stability of the power supply solution under various complex conditions. Compared with traditional static models, the digital twin model provides a real-time, dynamic global perspective, which can better cope with emergencies such as equipment failures and load mutations in the medium-voltage distribution network, and achieve rapid response and self-healing.

[0019] The present invention proposes an integrated intelligent power supply management platform. The platform combines adaptive dynamic cloning models and digital twin technology to perform full-process intelligent management from data collection, load forecasting, strategy simulation to power supply optimization. Through real-time monitoring and feedback control, the platform can continuously optimize the power supply strategy and maximize power supply efficiency and safety. At the same time, the system has emergency response and self-healing functions, and can quickly execute emergency strategies in the event of failures or abnormalities, reducing the time and scope of power supply interruptions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0021] Figure 1 The present invention is a flowchart of an intelligent power supply planning method based on a medium voltage distribution network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0023] In one or more embodiments, Figure 1 As shown, the present invention discloses an intelligent power supply planning method based on a medium voltage distribution network, the method comprising steps 1 to 6, including: Step 1: Collect data in real time and preprocess it, and build a digital twin model based on the preprocessed data.

[0024] Specifically, smart sensors are installed at key nodes of the medium-voltage distribution network (such as substations, distribution lines, and switchgear) to collect real-time parameters such as voltage, current, power, and temperature of the power grid. The data of each node includes multiple dimensions: ,in Indicates the node number, Indicates time. The data collected by the sensor is updated once per second to form a time series data set.

[0025] Among them, the data of all nodes are collected once per second to form a real-time data matrix: , here, Indicates that all nodes are at time The sensor data matrix, is the number of nodes. These data are the basic input of the digital twin model and reflect the current status of the distribution network.

[0026] Furthermore, in order to ensure the accuracy of the data and eliminate noise, the data of each node needs to be preprocessed. Specifically, the sensor data may contain noise or outliers, which will affect the accuracy of the model. In order to smooth the data and eliminate short-term fluctuations, the exponentially weighted moving average (EWMA) is used to smooth the time series data: , In this formula, Representation Node In time The smoothed data, is a smoothing coefficient (e.g. 0.3), which is used to control the weight of recent data and historical data. This preprocessing can reduce short-term fluctuations in data, highlight long-term trends, and make the model input data more stable. It will be used as input for the next model.

[0027] Furthermore, the construction of the digital twin model is to use the preprocessed data to build a state space model (SSM), which is used to describe the dynamic behavior of the medium voltage distribution network. The core of the state space model is to use the state vector Indicates that the system is at time The global state of the grid includes the voltage and current information of all nodes. By constructing a state space model, the change of the grid state can be simulated and predicted. The system state update equation is: , Here, represents the state vector at the next moment (such as voltage and current), is the state transfer matrix, which describes how the current state affects the state at the next moment; is the control input matrix, which represents the impact of external control actions (such as load adjustment) on the system state; is the control input vector, is the system noise (assumed to be Gaussian white noise), which represents the unmeasurable disturbance in the power grid. This model is used to capture the dynamic changes of system status over time and predict the future power grid status.

[0028] Furthermore, in order to ensure that the digital twin model can accurately reflect the actual power grid state, the Kalman filter algorithm is used for real-time state estimation and correction. Kalman filtering is a recursive algorithm that corrects the model by combining the predicted state and actual observation data. The update formula of Kalman filtering is:

[0029] In this formula, It's time The corrected state estimate is is the prior state estimate, the predicted value based on the state update equation. is the Kalman gain, which determines how to adjust the estimate, is the actual observation vector (such as power and temperature obtained from the sensor), It is the observation matrix, which maps the state vector to the observation space. Through this step, the present invention can continuously adjust the model state according to the real-time observation data, reduce the prediction error, and improve the model accuracy.

[0030] Furthermore, at each time step, the digital twin model uses the above formula to update the state and estimate the correction, ensuring that the state vector of the model and the estimated error covariance The model output, including the latest state estimation and error covariance matrix, provides an accurate data basis for subsequent load forecasting and power supply optimization. The model is updated every second to reflect the state changes of the distribution network in real time.

[0031] In this step, the digital twin model integrates real-time collected data, data preprocessing and state space modeling, combined with the Kalman filter algorithm, to achieve high-precision and dynamic simulation of the state of the medium-voltage distribution network. This real-time updated model provides a reliable foundation for subsequent load prediction and optimization.

[0032] Step 2: Generate an adaptive dynamic clone model pool based on the status data output by the digital twin model to predict future load changes and equipment status of the power grid, while responding to different power grid operating conditions and emergencies.

[0033] Specifically, an adaptive dynamic clone model pool is generated from the real-time status data output by the digital twin model constructed in step 1. This model pool is designed to predict future load changes and equipment status of the power grid to cope with different power grid operating conditions and emergencies. Taking into account the complexity and dynamic changes of the medium-voltage distribution network, the present invention designs a set of adaptive mechanisms that enable these models to be continuously adjusted according to real-time data to improve the accuracy and adaptability of the prediction. The core of this step is to maintain the stability and reliability of the power grid operation through online learning and dynamic updating of the model, and provide accurate input for subsequent power supply optimization strategies.

[0034] Specifically, in step 1, the present invention has constructed a digital twin model that can be updated in real time, and the output state estimation vector and the error covariance matrix Reflects the current dynamic behavior of the power grid. These data provide real-time status information of each node in the power grid (such as substations and lines), including voltage, current, etc.

[0035] In order to generate an adaptive dynamic clone model, the present invention first needs to extract key feature vectors from these outputs , which contains important state information of each node. These features are used to describe the overall state of the power grid and provide input for the prediction model. For example, if the state estimation Included The voltage and current information of each node, the present invention can define the characteristic vector as: , here, and Representation Node In time In this way, the present invention obtains the characteristic representation of the current state of the power grid, and these characteristics will be used as the input of the clone model.

[0036] Furthermore, after extracting the state features, the present invention needs to build multiple prediction models to adapt to different power grid states. Considering the diversity of power grid loads and operating states, the present invention designs a clone model pool containing multiple models. Each model Used for prediction under specific load patterns or conditions.

[0037] Initially, each model in the model pool is defined as a simple linear regression model to make a preliminary prediction of the load on the power grid. The mathematical representation of these models is: , In this formula, Representation Model Prediction of the load at the next moment, is the weight vector of the model, is a bias. Each model is set according to different initial conditions to capture different load patterns. Through this structure, the present invention can use multiple models to run in parallel to preliminarily predict different load changes.

[0038] Furthermore, once the clone model pool is initialized, the present invention needs to ensure that these models can dynamically adapt to the actual situation of the power grid. The present invention designs an adaptive weight update algorithm based on real-time feedback to adjust the weight and bias of each model so that it can better predict the actual power grid load.

[0039] Specifically, when new data arrives, the present invention calculates the prediction error of each model and adjusts the weight of the model according to the error: , here, is the updated weight vector, is the current weight, is the learning rate, is the actual load value, is the predicted value of the model. By continuously adjusting the weights, each model It can be optimized under various load modes and adapt to different operating conditions. This adaptive update mechanism enables the model to continuously learn and adapt to the dynamic changes of the power grid.

[0040] In order to improve the accuracy of the overall prediction, the present invention requires a strategy to select the model that best suits the current state or fuse the prediction results of multiple models. The present invention uses an error-based weighting strategy to achieve this.

[0041] First, calculate the prediction error of each model , which is the squared difference between the actual load and the predicted value: , Then, based on these errors, the present invention calculates the weight factor for each model , these weight factors are used to fuse the prediction results of multiple models: , in, is a smoothing parameter used to control the effect of error on weight. Through this weighting strategy, the present invention can dynamically adjust the influence of the model and combine the predictions of each model to obtain a more accurate final prediction value. : , This fusion strategy allows the system to adjust the contribution of each model based on real-time prediction performance, ensuring the best prediction results under different grid conditions.

[0042] Furthermore, the load pattern and state of the power grid may change significantly with time and environmental changes, so relying solely on the existing model pool may not be sufficient to cope with all situations. In order to improve the flexibility and adaptability of the system, the present invention introduces an online adaptation and expansion mechanism.

[0043] When the system detects that the prediction error continues to exceed the preset threshold for a period of time, it means that the existing model pool cannot adapt well to the current power grid status. At this time, the system automatically generates a new model , and add it to the model pool. The new model can be initialized based on the parameters of the current optimal model and further adjusted through online learning to adapt to the new load pattern.

[0044] This mechanism ensures the dynamic scalability and continuous optimization capabilities of the model pool, and can respond to changes in power grid status in real time, ensuring the system's prediction accuracy and reliability.

[0045] Through this scheme, the initial adaptive dynamic clone model generation process is systematically designed as a series of steps from data extraction, model construction, parameter update, model selection to online adaptive expansion. Each step is closely dependent on the output of the previous step and ensures that the model can be dynamically adjusted and expanded to adapt to the complex and changeable state of the medium-voltage distribution network.

[0046] Step 3: Based on the real-time status data output by the cloned model pool and the digital twin model, load forecasting and power supply strategy simulation are performed. At the same time, by combining the real-time grid status, predicting future load changes and equipment status, different power supply strategies are simulated, and the optimal strategy is selected to deal with various possible grid operating conditions and emergencies.

[0047] Specifically, the adaptive dynamic clone model is used for load prediction, and the input is the adaptive dynamic clone model pool obtained in step 2 And the state feature vector output by the digital twin model in step 1 : First, clone each model in the model pool using Predict the load at future moments. Each model is based on the previously adjusted weights and bias To make predictions, these parameters are continuously optimized during the model update process to adapt to different grid conditions. The load prediction formula is: , in, Representation Model Prediction of the load at the next moment. Represents the weight vector of the model, which represents the weight of the impact of power grid characteristics on the load. The feature vector at the current moment includes the voltage, current and other status information of each node. Represents the bias term, which represents the basic prediction value of the model. Represents the adjustment coefficient, which dynamically adjusts the sensitivity of the model to disturbances in the prediction. Represents the regularization term, which indicates the volatility of the power grid state. This term is used to smooth the output of the model to prevent overfitting.

[0048] In order to improve the robustness of the model under complex and variable power grid conditions, the present invention introduces an adaptive regularization term , which is used to control the model’s response to changes in the power grid state. This regularization term is not just a simple penalty term, but also includes adjustments to the current load volatility and historical data of the power grid, making the model more sensitive to abnormal data or rapid changes.

[0049] , in, represents the regularization term, which is used to capture the volatility of the power grid state. Indicates the number of grid nodes. Represents the node weight, which controls the contribution of each node's fluctuation to the overall regularization. and Indicates The voltage and current at each node. and Indicates The historical average voltage and current of each node are used to calculate the deviation of the current state.

[0050] In order to further improve the accuracy of prediction, this paper adopts an innovative dynamic model integration strategy. This strategy not only takes into account the historical performance of each model, but also dynamically adjusts the weight of the model. , adapting to the dynamic changes of the power grid through the newly introduced fluctuation adjustment coefficient; , in, Representation Model In time The integration weight of . Represents the smoothing factor, which adjusts the impact of model error on weights. Representation Model In time prediction error. It represents the fluctuation adjustment coefficient, which controls the adjustment range of the power grid volatility to the weight. represents the variance of grid load fluctuation and the uncertainty of current grid status. , which can better reflect the contribution of different models to the prediction at different times, thereby improving the overall accuracy and robustness of load prediction.

[0051] Furthermore, based on the future load prediction of the integrated model , the present invention simulates multiple power supply strategies in the digital twin model. Strategy Set , each strategy Represents a different method of supplying power, such as load redistribution, transformer tap adjustment, or alternate source activation.

[0052] Simulate in the digital twin model and calculate the simulation output of each strategy at the next moment , and based on multiple evaluation indicators (such as voltage stability, power loss, equipment load rate) the comprehensive score is calculated as: , in, Representation strategy The comprehensive rating of . Indicates The weight of an evaluation indicator indicates the importance of the indicator in the overall score. Indicates The function of an evaluation index is used to calculate the performance of the strategy simulation output under this index.

[0053] Finally, choose the strategy with the highest overall score As the actual power supply strategy: , This selection ensures that the power supply strategy achieves optimal performance of grid operation under current and forecasted load conditions.

[0054] Through this detailed step, combined with the innovative load prediction model and strategy simulation method, the present invention can accurately predict the future power grid state and select the best solution among different power supply strategies. This process provides key support for the intelligent management of the power grid, ensuring efficient and stable power supply in a complex and changing power grid environment.

[0055] Step 4: Combine the multimodal load forecast results and the priority ranking calculated based on the comprehensive score to continue optimizing the power supply strategy to ensure efficient and stable operation of the power grid under complex load conditions.

[0056] Specifically, we first determine the optimization objectives and constraints, and the input is the accurate load forecast output from step 3. and scheduling policy priority These data provide the load information of the power grid at future times and the dispatching strategy with priority execution.

[0057] The present invention defines the optimization objective function , used to measure power supply strategies The optimization goal includes three core parts: minimizing power loss, maximizing voltage stability, and minimizing equipment overload. The objective function is defined as: , in, Indicates the power supply strategy comprehensive optimization goal. It represents the weight coefficient of the objective function, reflecting the importance of each indicator. Indicated in strategy Next, node of power loss. Indicated in strategy Next, node voltage. Indicates the reference voltage value, which is used to measure voltage stability. Indicated in strategy Next, node of load. Indicates the maximum allowable load of the node to prevent device overload.

[0058] In order to further optimize the flexibility and adaptability of the power supply strategy, the present invention introduces an innovative scheduling constraint term , which combines the predicted load Priority with scheduling policy It ensures that the optimal scheduling scheme is considered when generating the power supply strategy: , in, Represents the scheduling constraint item, reflecting the consistency between the power supply strategy and the scheduling strategy. Indicates the weight of scheduling policy adjustment and controls the priority between different policies. Indicates the power supply strategy Next, The scheduling result of a scheduling strategy. Indicates The expected scheduling results of a scheduling strategy.

[0059] Furthermore, combined with the objective function and scheduling constraints , define the overall optimization problem: , in, The weight coefficient representing the scheduling constraint is used to control the consistency between the power supply strategy and the scheduling strategy.

[0060] The goal of this optimization problem is to generate a power supply strategy under given load forecasting and scheduling strategy priorities. , so that the overall operation performance of the power grid is optimized, and there is a high consistency between the power supply strategy and the dispatch strategy.

[0061] Furthermore, in order to solve the above optimization problem, the present invention designs an adaptive optimization algorithm, which can be dynamically adjusted according to the real-time state of the power grid and load forecast. The core idea of ​​the algorithm is to combine particle swarm optimization (PSO) and Lagrangian relaxation to balance the relationship between the objective function and the constraints.

[0062] In each iteration, the algorithm updates the particle's position and velocity and adjusts the Lagrange multiplier To relax the scheduling constraints: , in, Indicates The power supply strategy for the next iteration. Indicates The power supply strategy for the next iteration. Represents the learning rate, which is used to control the step size of each update. Indicates the power supply strategy gradient. Represents the Lagrange multiplier, which adjusts the relaxation degree of different scheduling strategies. Through this adaptive optimization process, the present invention can effectively generate an optimal power supply strategy , ensuring the operating efficiency and stability of the power grid under complex load conditions.

[0063] Through this detailed step, combined with innovative optimization objectives and scheduling constraint design, the present invention provides a complete and innovative solution for power supply strategy generation, ensuring that the power grid can dynamically adapt to various complex operating conditions and maintain efficient and stable power supply. Each step fully utilizes the output of the previous step and provides a solid foundation for subsequent practical execution.

[0064] Step 5: Execute the optimal power supply strategy and conduct real-time monitoring, provide feedback on the monitoring results, and dynamically adjust the strategy parameters when executing the strategy to ensure the accuracy of strategy execution and the stable operation of the power grid.

[0065] Specifically, the optimized power supply strategy output in step 4 is The strategy includes specific operation instructions for each node (such as load transfer, transformer adjustment, switch operation, etc.), and is the optimal solution generated based on the current grid status and predicted load.

[0066] Before the strategy is executed, based on the current real-time data Set initial parameters. Execution parameters As a strategy The execution plan is in the form of: , in, Representation strategy The execution parameters are the instruction sets for actual operations. Represents the adjustment step size, which is used to ensure the stability and flexibility of execution. Represents the optimization objective function The gradient of the strategy is used to correct the execution direction of the initial strategy.

[0067] When executing the initial strategy, by adjusting the step size and gradient correction, so that the strategy execution is more in line with the needs of the actual power grid and avoids strategy deviation caused by inaccurate initial settings.

[0068] Furthermore, during the execution of the strategy, the operating state of the power grid will continue to change. Therefore, the present invention needs to monitor the key parameters of the power grid (such as voltage , Current ,load Real-time monitoring data It is represented by the following matrix: , The monitoring system collects data in real time Transmitted to the central control system for comparison with the expected policy execution results.

[0069] Furthermore, in order to ensure the accuracy of strategy execution, the present invention designs an adjustment mechanism based on feedback control. Specifically, the present invention monitors data in real time. Expected results of strategy execution Compare and calculate the execution error ; , in, Represents the execution error matrix, which reflects the difference between the actual execution results and the expected ones. Representation strategy The expected execution result is based on the predicted data during strategy optimization.

[0070] According to the error matrix The size and direction of To make real-time corrections. The adjustment formula is as follows: , in, Represents the feedback control coefficient, which controls the sensitivity of error adjustment. Represents the revised execution parameters, which are the actual instructions for the strategy execution at the next time step.

[0071] Through this adjustment mechanism, the system can dynamically correct the strategy during the strategy execution process to ensure that the actual execution effect is consistent with the optimization result, thereby ensuring the stable operation of the power grid.

[0072] Furthermore, during the strategy execution process, the present invention introduces a dynamic evaluation mechanism to evaluate the overall operation status of the power grid in real time. The present invention uses the following evaluation function To judge the effectiveness of the current strategy execution: , in, Indicates the overall evaluation value of the current policy execution. It represents the weight of different operating indicators and controls the impact of each indicator on the overall evaluation.

[0073] If the evaluation function If the preset target value is not reached, the system will automatically recalculate the strategy parameters , and enter the next round of strategy execution and adjustment cycle until the evaluation function Reach the ideal level, or the system runs stably.

[0074] Furthermore, during the strategy execution and adjustment process, the present invention sets a variety of termination conditions, such as: the grid operation state reaches stability, the error Reduce to an acceptable range and evaluate the function Achieve preset goals, etc. Once any termination condition is met, the strategy execution and adjustment process ends, and the power grid enters a stable operation state.

[0075] At the same time, the system will save the strategy parameters for this execution and related data to provide reference data for the next round of optimization and execution, ensuring continuous optimization and improvement of the system.

[0076] Step 6: Implement fault prediction and self-healing mechanisms based on real-time monitoring and feedback data.

[0077] Specifically, according to the real-time monitoring feedback data in step 5 and the corrected execution parameters The data includes real-time status information of key grid parameters such as voltage, current, and load.

[0078] Then, in order to achieve high-precision fault prediction, the present invention constructs a multi-level dynamic prediction model , using real-time data to continuously evaluate the state of the power grid. First, extract key features For failure prediction: , in, Represents the fault prediction feature vector, which contains the voltage of all nodes , Current ,load and its rate of change ; in, , Representation Node At the moment The rate of change of voltage and current.

[0079] Furthermore, in order to achieve high-precision prediction of power grid faults, the present invention adopts an adaptive anomaly detection algorithm The algorithm is based on a step-by-step learning approach, comparing the historical and current states of each node in the power grid to identify possible failure modes.

[0080] Setting a dynamic failure prediction index , which is used to evaluate the failure probability of each node. The present invention uses a Bayesian update mechanism combined with a Markov chain Monte Carlo (MCMC) method to calculate the failure probability representation , in, Indicates at time The probability of failure at a given time. Represents the conditional likelihood function, which means that when executing the parameters The observed eigenvector possibility. represents the prior probability and represents the time Prior knowledge of faults at the moment. Represents the feature vector The marginal probability of .

[0081] The conditional likelihood and marginal probability are estimated through the MCMC sampling method, and the failure probability of each node is gradually updated to achieve dynamic monitoring of potential failures.

[0082] Furthermore, once the failure probability of a node or multiple nodes is detected Exceeding the preset threshold , the system will immediately start the self-healing mechanism The mechanism includes a set of predefined self-healing strategies , each strategy corresponds to a specific type of failure response.

[0083] In order to determine the optimal self-healing strategy, this paper designs a multi-objective optimization problem, which comprehensively considers fault response time, restoration cost and grid stability. Define the self-healing priority score function , the comprehensive effect of each strategy is evaluated as follows: , in, Represents a self-healing strategy The priority score of . Representation strategy The shorter the response time, the better. Representation strategy The lower the recovery cost, the better. Representation strategy The impact function on grid stability, the higher the better. Represents the priority weight coefficient, which is used to balance the impact of different objectives.

[0084] Based on priority score Sort all self-healing strategies and select the one with the highest score As a final self-healing measure: , Furthermore, the optimal self-healing strategy is selected After the self-healing strategy is implemented, the system immediately executes the strategy and quickly restores the normal operation of the power grid by adjusting switches, transferring loads, activating backup power sources, etc. During the execution process, the system continuously monitors the real-time status of the power grid, evaluates the effectiveness of the self-healing strategy, and makes necessary strategy adjustments based on feedback.

[0085] In order to further improve the self-healing efficiency, the system adopts an adaptive update mechanism. and failure probability update , dynamically adjust the self-healing strategy parameters , ensuring that the power grid recovers quickly and operates stably after a fault occurs.

[0086] Furthermore, in order to continuously improve the performance of fault prediction and self-healing mechanisms, the system automatically enters a closed-loop optimization process after each fault event is handled. By analyzing the effect of each self-healing event and the efficiency of the system response, the fault prediction model is updated. and self-healing strategy library , achieving more accurate predictions and faster responses to future failures.

[0087] During each optimization process, the system adjusts the model parameters and strategy parameters to ensure that the fault prediction and self-healing mechanisms can improve themselves and gradually improve the stability and security of the power grid.

[0088] These are just some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and making equivalent changes according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A smart power supply planning method based on a medium voltage distribution network, characterized in that: The method comprises the following steps: Step 1: Collect data in real time and perform preprocessing, and build a digital twin model based on the preprocessed data; specifically, the following steps are included: S11. Install smart sensors at key nodes of the medium voltage distribution network to collect multi-modal parameters of the power grid in real time; S12, using an exponentially weighted moving average algorithm to smooth the time series data of the multimodal parameters; S13, constructing a digital twin model based on the smoothed data; S14. Use the Kalman filter algorithm to correct the digital twin model by combining the predicted state and actual observation data; Step 2: Generate an adaptive dynamic clone model pool based on the output status data of the digital twin model to predict the future load changes and equipment status of the power grid, and respond to different power grid operating conditions and emergencies; specifically, the following steps are included: S21. Extract features from the state data output by the digital twin model to obtain a feature vector; S22, construct multiple prediction models, design a clone model pool containing multiple prediction models, wherein each prediction model is used for prediction under a specific load mode or state, and then define the multiple prediction models as linear regression models for initialization to make a preliminary prediction of the load of the power grid; S23. When multiple prediction models are initialized, an adaptive weight update algorithm based on real-time feedback is designed to dynamically predict the weight and bias of each model, so that it can better predict the actual grid load; S24. Design an error-based weighting strategy to select the model that best suits the current state or fuse the prediction results of multiple models; Step 3: Based on the real-time status data output by the clone model pool and the digital twin model, load forecasting and power supply strategy simulation are performed. At the same time, by combining the real-time grid status, predicting future load changes and equipment status, different power supply strategies are simulated, and the optimal strategy is selected to deal with various possible grid operating conditions and emergencies; specifically, the following steps are included: S31. Use each prediction model in the clone model pool to predict the load at a future moment according to the state feature vector output by the digital twin model to obtain a multimodal load prediction result; S32, introducing an adaptive regularization term in the prediction to control the response of each prediction model to changes in the power grid state; S33, construct a dynamic model integration strategy that introduces fluctuation adjustment coefficients, and adapt to the dynamic changes of the power grid by adjusting the weight of each prediction model; S34, integrating the future load of each prediction model, simulating multiple power supply strategies in the digital twin model and simulating in the digital twin model, calculating the simulation output of each strategy at the next moment, and calculating the comprehensive score based on multiple evaluation indicators, and selecting the strategy with the highest comprehensive score as the power supply strategy actually implemented; Step 4: Combine the multi-modal load forecast results and the priority ranking calculated based on the comprehensive score to continue optimizing the power supply strategy to ensure that the power grid operates efficiently and stably under complex load conditions; specifically, the following steps are included: S41, defining an optimization function based on the objectives of minimizing power supply loss, maximizing voltage stability, and minimizing equipment overload according to the multimodal load prediction results and the priority ranking calculated according to the comprehensive score; S42, introducing scheduling constraints into the multimodal load forecast results and the priority ranking calculated based on the comprehensive score, to ensure that the optimal scheduling scheme is considered when generating the power supply strategy; S43, combining the target optimization function and the scheduling constraint item to obtain an optimization problem; S44, design an adaptive optimization algorithm to dynamically adjust according to the real-time state of the power grid and load forecast to obtain the optimal power supply strategy; Step 5: Execute the optimal power supply strategy and conduct real-time monitoring, provide feedback on the monitoring results, and dynamically adjust the strategy parameters when executing the strategy to ensure the accuracy of strategy execution and the stable operation of the power grid; specifically, the following steps are included: S51, setting initial parameters for current real-time sensor data, and executing the initial parameters as an execution plan for the optimal power supply strategy; S52. During the strategy execution process, the real-time monitoring system monitors the key parameters of the power grid, transmits the real-time collected data to the central control system, compares it with the predicted strategy execution results, calculates the execution error, and adjusts the initial parameters to make real-time corrections according to the size and direction of the execution error to obtain the strategy parameters; S53. During the strategy execution process, a dynamic evaluation mechanism is introduced to evaluate the overall operation status of the power grid in real time and obtain evaluation results; Step 6: Implement fault prediction and self-healing mechanisms based on real-time monitoring and feedback data; specifically, the following steps are included: S61, construct a multi-level dynamic prediction model, using the features extracted from real-time data as input to perform fault prediction; S62, adding an adaptive anomaly detection algorithm to fault prediction, comparing the historical state and current state of each node in the power grid, and identifying possible fault modes; once it is detected that the fault probability of a node or multiple nodes exceeds a preset threshold, the self-healing mechanism will be immediately started; the adaptive anomaly detection algorithm includes a set of predefined self-healing strategies, each strategy corresponding to a specific type of fault response; S63. Combining fault response time, restoration cost and grid stability, design a multi-objective optimization problem for self-healing strategies, calculate the priority score of each self-healing strategy, evaluate the comprehensive effect of each strategy, sort all self-healing strategies according to the priority score, and select the strategy with the highest score as the self-healing measure to be finally executed; S64. After selecting the optimal self-healing strategy, immediately execute the strategy to restore the normal operation of the power grid, monitor the real-time status of the power grid, evaluate the effect of the self-healing strategy, and make necessary strategy adjustments based on feedback; S65. After each fault event is handled, a closed-loop optimization process is automatically entered to update the multi-level dynamic prediction model and self-healing strategy set by analyzing the effect of each self-healing and the efficiency of the system response.

2. According to claim 1, a smart power supply planning method based on a medium voltage distribution network is characterized in that: The digital twin model is used to describe the dynamic behavior of the medium voltage distribution network, including: A state space model is constructed to represent the digital twin model. The state space model uses a state vector to represent the global state of the system at a moment, including the voltage and current information of all nodes. By constructing the state space model, the changes in the power grid state can be simulated and predicted.

3. According to claim 1, a smart power supply planning method based on a medium voltage distribution network is characterized in that: The S23 is specifically expressed as follows: When new data arrives, the prediction error of each prediction model is recalculated, and the weight of the model is adjusted according to the error, as shown below: ， in, is the updated weight vector, is the current weight, is the learning rate, is the actual load value, is the predicted value of the model; By continuously adjusting weights, each prediction model can be optimized under various load modes and adapt to different operating conditions.

4. The intelligent power supply planning method based on the medium voltage distribution network according to claim 3 is characterized in that: The S24 is specifically expressed as follows: First, the prediction error of each prediction model is calculated, where the prediction error is the square difference between the actual load and the predicted value; Then, based on the prediction deviation, a weight factor of each model is calculated, and the weight factor of each model is used to fuse the prediction results of multiple models; Finally, the influence of the model is dynamically adjusted through the weight factor, and the predictions of each model are combined to obtain a more accurate final prediction value.

5. The intelligent power supply planning method based on the medium voltage distribution network according to claim 1 is characterized in that: The step 2 further includes: An online adaptation and expansion mechanism is introduced. When the system detects that the prediction error continues to exceed the preset threshold for a period of time, it means that the existing clone model pool cannot adapt to the current power grid status. Then a new prediction model is automatically generated and added to the clone model pool. The new prediction model is initialized based on the parameters of the current optimal model and adjusted through online learning to adapt to the new load pattern.

6. The intelligent power supply planning method based on the medium voltage distribution network according to claim 1 is characterized in that: The adaptive regularization term Including penalty terms, adjustments to the current load volatility of the power grid and historical data, it is expressed as follows: ， in, represents the regularization term, which is used to capture the volatility of the power grid state; Indicates the number of grid nodes; Represents the node weight, which controls the contribution of each node's fluctuation to the overall regularization; and Respectively represent The voltage and current of each node; and Respectively represent The historical average voltage and current of each node are used to calculate the deviation of the current state.

7. The intelligent power supply planning method based on the medium voltage distribution network according to claim 1 is characterized in that: The objective optimization function based on minimizing power loss, maximizing voltage stability and minimizing equipment overload is expressed as follows: ， in, Indicates the power supply strategy Comprehensive optimization goal; They represent the weight coefficients of the objective function, reflecting the importance of each indicator; n represents the total number of nodes; t represents the current time; Indicated in strategy Next, node Power loss; Indicated in strategy Next, node Voltage; Indicates the reference voltage value, used to measure voltage stability; Indicated in strategy Next, node Load; Indicates the maximum allowable load of the node to prevent device overload.

8. The intelligent power supply planning method based on the medium voltage distribution network according to claim 7 is characterized in that: The S44, designing an adaptive optimization algorithm to dynamically adjust according to the real-time state of the power grid and load forecast, specifically includes: Combining particle swarm optimization and Lagrangian relaxation method, the relationship between the objective function and the constraints is balanced. In each iteration, the algorithm updates the position and velocity of the particles and adjusts the Lagrangian multiplier to relax the scheduling constraints to obtain the optimal power supply strategy, which is expressed as follows: ， in, Indicates The power supply strategy for the next iteration; Indicates The power supply strategy for the next iteration; Represents the learning rate, which is used to control the step size of each update; Indicates the power supply strategy The gradient of represents the Lagrange multiplier, which adjusts the relaxation degree of different scheduling strategies; Indicates The optimization goal of the power supply strategy for the next iteration; represents the scheduling constraint; P represents the overall scheduling strategy; Indicates the power supply strategy Next, The scheduling results of the scheduling strategies; Indicates The expected scheduling results of a scheduling strategy.

9. The intelligent power supply planning method based on medium voltage distribution network according to claim 1 is characterized in that: The dynamic evaluation mechanism is expressed as follows: ， Among them, represents the overall evaluation value of the current strategy execution; Indicates the weights of different operating indicators and controls the impact of each indicator on the overall evaluation; If the evaluation function If the preset target value is not reached, the strategy parameters are automatically recalculated and the next round of strategy execution and adjustment cycle is entered until the evaluation function is Reach the ideal level.

10. The intelligent power supply planning method based on medium voltage distribution network according to claim 1, characterized in that: The adaptive anomaly detection algorithm is constructed as follows: Setting a dynamic failure prediction index , evaluate the failure probability of each node, and use the Bayesian update mechanism combined with the Markov chain Monte Carlo method to calculate the failure probability, which is expressed as follows: ， in, Indicates at time The probability of failure at a given moment; Represents the conditional likelihood function, which means that when executing the parameters The observed eigenvector possibility; represents the prior probability and represents the time Prior knowledge of faults at the moment; Represents the feature vector The marginal probability of The multi-objective optimization problem of self-healing strategy is designed by combining fault response time, restoration cost and grid stability, and the priority score of each self-healing strategy is calculated, which is expressed as follows: ， in, Represents a self-healing strategy Priority score; Representation strategy The response time, the shorter the better; Representation strategy the cost of restoration; Representation strategy The impact function on grid stability, the higher the better; Represents the priority weight coefficient, which is used to balance the impact of different goals; All self-healing strategies are sorted according to the priority scores, and the strategy with the highest score is selected as the self-healing measure to be finally executed.

Images