Power system power flow self-adaptive regulation and control method based on deep transfer learning

Abstract

The invention discloses a power system power flow self-adaptive regulation and control method based on deep transfer learning, and the method comprises the steps: 1, constructing a power system source domain power flow analysis model, and constructing a loss function for a power flow error in combination with the power flow data characteristics of a source domain and a target domain; step 2, designing a network parameter initialization method suitable for power flow analysis, realizing transferable representation from source domain data to a target domain, and constructing an initialized target domain power flow analysis model; 3, introducing a sample selection and incremental learning mechanism, and carrying out dynamic screening and retraining on key power flow samples in a target domain; and 4, carrying out load flow calculation and optimization control on the target power system by adopting the trained target domain load flow analysis model, and realizing self-adaptive optimization of the system operation state through iterative correction and parameter updating. According to the method, rapid calculation and dynamic optimization of the power flow of the power system are realized, and the intelligent level, stability and calculation reliability of power grid operation analysis are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and in particular to a power system power flow adaptive control method based on deep transfer learning. Background Technology

[0002] With the continuous expansion of power system scale and the increasing proportion of renewable energy integration, the operating characteristics of power grids are becoming increasingly complex. Power flow analysis, as a fundamental component of power system operation and planning, has a significant impact on the safe and stable operation of the power grid due to its computational accuracy and convergence speed. The results of power flow analysis directly affect the rationality of dispatching plans, the balance of equipment load, and the accuracy of system safety margin assessment. However, with the large-scale integration of distributed generation, flexible loads, and power electronic devices, power systems exhibit significant nonlinear, time-varying, and high-dimensional coupling characteristics, gradually diminishing the adaptability and robustness of traditional power flow calculation methods in complex power grid environments.

[0003] Currently, power flow calculations in power systems primarily rely on traditional numerical algorithms such as the Newton-Raphson method and PQ decomposition. These methods perform well when system parameters are stable and model accuracy is high, but they often face challenges such as convergence difficulties, low computational efficiency, and large result deviations when parameter uncertainties, measurement errors, or sudden changes in operating conditions exist. On the other hand, deep learning technology has seen initial applications in power system state estimation and power flow prediction in recent years. Utilizing neural networks to model complex nonlinear relationships can improve computational speed and generalization ability to some extent. However, training deep learning models typically depends on large amounts of high-quality labeled data. Structural differences and operating condition distributions between different power grids lead to poor model transfer performance, making it difficult to maintain stable accuracy in new regions or operating scenarios. To address these issues, there is an urgent need to propose a power flow analysis and control method that can fully utilize existing knowledge in the source domain and quickly adapt to the characteristics of the target domain, enabling adaptive learning and accurate calculation of the model under different power grid environments. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing deep learning models, which typically rely on large amounts of high-quality labeled data for training. Furthermore, the structural and operational differences between different power grids lead to poor model transfer performance and difficulty in maintaining stable accuracy in new regions or operating scenarios. This invention provides a power system power flow adaptive control method based on deep transfer learning. By establishing a feature mapping relationship between the source and target domains and introducing a sample selection mechanism and adaptive parameter update strategy, this method effectively improves the model's transfer efficiency and generalization ability under limited sample conditions. This power flow sub-control method fully utilizes existing knowledge in the source domain and quickly adapts to the characteristics of the target domain, enabling rapid calculation and dynamic optimization of power system power flow, significantly improving the intelligence, stability, and computational reliability of power grid operation analysis.

[0005] The objective of this invention is achieved through the following technical solution: An adaptive power flow control method for power systems based on deep transfer learning includes the following steps: Step 1: Construct a power system source domain power flow analysis model, and combine the power flow data characteristics of the source domain and the target domain to construct a loss function for power flow error; Step 2: Design a network parameter initialization method suitable for power flow analysis, realize the transferable representation of source domain data to target domain, and construct an initialized target domain power flow analysis model; Step 3: Introduce a sample selection and incremental learning mechanism to dynamically screen and retrain key power flow samples in the target domain. Step 4: Use the trained target domain power flow analysis model to perform power flow calculation and optimization control on the target power system, and achieve adaptive optimization of the system operating state through iterative correction and parameter update.

[0006] Preferably, step 1 specifically includes: Step 1a: Describe the source domain power flow analysis model using polar coordinate power flow equations; Step 1b: Perform nonlinear fitting of source domain power flow data using a deep neural network, and construct a loss function minimizing the deviation between the deep neural network output and the real source domain data as the optimization objective. Step 1c: Introduce the polar coordinate power flow equation as a penalty term and add it to the minimization loss function to obtain the comprehensive loss function.

[0007] Preferably, step 2 specifically includes: Step 2a: Extract the weight parameters and bias parameters of the deep neural network, and evaluate the importance of the weight parameters and bias parameters of different network layers. By using a joint index based on gradient sensitivity and weight sparsity, select the parameter set that contributes more to the expression of power flow characteristics. Step 2b: Construct a feature mapping function to realize the transferable representation of the weight parameters and bias parameters of the source domain to the target domain, and construct an initialized target domain power flow analysis model; Step 2c: Based on the statistical characteristics of the target domain data, the weight parameters and bias parameters after migration are recalibrated using parameter normalization and perturbation correction strategies.

[0008] Preferably, the parameter normalization and perturbation correction strategy is as follows: The distribution alignment of the transfer parameters is achieved by minimizing the distance metric between the source domain output distribution and the target domain input distribution.

[0009] Preferably, step 3 specifically includes: Step 3a: For the target domain sample set, use the target domain power flow analysis model to predict the target domain samples and calculate the error between the predicted output and the actual power flow result. Step 3b: Select the top K samples to form the key sample set based on the error magnitude or the representativeness of the features; Step 3c: Input the selected key sample set into the model and perform incremental training based on the comprehensive loss function.

[0010] Preferably, step 4 specifically includes: Step 4a: Obtain dynamic operating data of the target power system, quickly output the core power flow parameters under the current operating conditions through a deep neural network, compare the calculated power flow parameters with the real-time measured data of the target system, if the error is within the preset threshold, the current calculation result is deemed valid and proceeds to the subsequent optimization stage; if the error exceeds the standard, backtrack to check whether the data acquisition is abnormal, or trigger the model emergency parameter fine-tuning. Step 4b: Based on the calculation results of the core power flow parameters and combined with the safe operation criteria of the power system, determine whether there is an optimization requirement for the current state. If there is an optimization requirement, proceed to step 4c. Step 4c: Based on the optimization requirements and combined with the trend characteristics, formulate an initial control strategy; Step 4d: The initial control strategy is converted into virtual input data and input into the target domain model. The power flow state after control is recalculated. It is determined whether there is a deviation between the calculated result after control and the preset optimization target. If there is a deviation, the model parameters and control strategy are adjusted synchronously to form an iteration until the power flow parameters meet the optimization target. The iteration process is completed, and the corrected control strategy is sent to the actual control terminal of the target power system to perform physical control.

[0011] Preferably, the method for acquiring the dynamic operation data of the target power system is periodic acquisition.

[0012] The beneficial effects of this invention are as follows: This invention fully explores the implicit correlation between power flow data in the source and target domains, overcoming the problems of insufficient transfer performance of existing deep learning models under cross-regional or changing operating conditions, reliance on large amounts of labeled data, and poor computational convergence. Through a transfer learning framework, it achieves efficient transfer from source domain knowledge to the target domain model, significantly reducing the model's dependence on sample quantity and manual annotation. The sample selection mechanism can dynamically identify high-value samples, improving the model's adaptability under complex power grid operating conditions; the parameter adaptive initialization method effectively improves the convergence speed and stability of model training. The combination of these two aspects forms an intelligent adaptive control system for power flow analysis.

[0013] Compared with traditional power flow calculation methods, this invention has advantages such as high migration efficiency, high calculation accuracy, good convergence performance, and strong robustness to changes in power grid topology and operating status. It has good engineering applicability and promotion prospects, and can provide reliable technical support for intelligent analysis and optimized scheduling of power systems. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0015] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0016] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0017] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0018] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0019] Example: An adaptive power flow control method for power systems based on deep transfer learning, such as Figure 1 As shown, it includes the following steps: Step 1: Construct a power system source domain power flow analysis model, and combine the power flow data characteristics of the source domain and the target domain to construct a loss function for power flow error; Step 2: Design a network parameter initialization method suitable for power flow analysis, realize the transferable representation of source domain data to target domain, and construct an initialized target domain power flow analysis model; Step 3: Introduce a sample selection and incremental learning mechanism to dynamically screen and retrain key power flow samples in the target domain. Step 4: Use the trained target domain power flow analysis model to perform power flow calculation and optimization control on the target power system, and achieve adaptive optimization of the system operating state through iterative correction and parameter update.

[0020] Step 1 specifically includes: Step 1a: Describe the source domain power flow analysis model using polar coordinate power flow equations: in, and These represent the active and reactive branch power flowing from the i-th bus to the j-th bus, respectively. Let be the voltage amplitude of bus i. Let be the voltage phase angle difference between bus i and bus j. and Let be the conductance and susceptance between the i-th bus and the j-th bus, respectively. Since the voltage amplitude and phase angle can completely describe the entire system state, the voltage amplitude... and phase angle As a trend in deep transfer learning, it effectively reduces the dimensionality of the output feature vector.

[0021] Step 1b: Perform nonlinear fitting of the source domain power flow data using a deep neural network, and construct a loss function that minimizes the deviation between the deep neural network output and the actual source domain data as the optimization objective. in, Represents the loss function. This indicates the specific deviation loss item. Here, m represents the normalization coefficient, and m represents the sample size. These are the weight parameters of a deep neural network. It is the offset of the deep network, d is and Dimension To train the true values, The values ​​are calculated for deep neural networks.

[0022] Step 1c: Introduce the polar coordinate power flow equation as a penalty term and add it to the minimization loss function to obtain the comprehensive loss function: in: In the formula, L is the prediction bias loss term. The output feature vector for power flow calculation is specifically the voltage magnitude. With phase angle , This is the output value of the deep neural network. By combining physical consistency constraints with network optimization, the initial construction and pre-training of a power flow analysis model under the deep transfer learning framework are achieved, providing a foundation for subsequent target domain transfer and adaptive regulation.

[0023] Step 2 specifically includes: Step 2a: Extract the weight parameters and bias parameters of the deep neural network, and adjust the weight parameters of different network layers. and bias parameters Importance assessment is conducted, and parameter sets that contribute significantly to the representation of power flow characteristics are selected using a joint index based on gradient sensitivity and weight sparsity. ; Step 2b: Construct a feature mapping function to realize a transferable representation of the weight parameters and bias parameters of the source domain to the target domain, and construct an initialized target domain power flow analysis model: in, and Let represent the initial weights and bias matrices of the target domain model, respectively. This is a nonlinear transformation based on kernel functions. This step ensures that the model can inherit the power flow characteristics of the source domain in the early stages of the transfer, thereby improving the initial accuracy of the target domain modeling.

[0024] Step 2c: Based on the statistical characteristics of the target domain data, a parameter normalization and perturbation correction strategy is used to recalibrate the transferred weight parameters and bias parameters, specifically as follows: Alignment of the transfer parameter distribution is achieved by minimizing the distance metric between the source domain output distribution and the target domain input distribution. In the formula, ΔW represents the weight adjustment increment, and Δb represents the bias adjustment increment. D mmd Indicates the maximum difference between the means. Indicated by This represents a function that measures the difference in distribution. Input samples for the target domain.

[0025] Through the above optimizations, the network parameters maintain good trainability and gradient stability after transfer, thereby effectively avoiding gradient explosion or unstable convergence in the early stage of training in the target domain, and providing a reliable parameter basis for subsequent sample selection and incremental learning.

[0026] Step 3 specifically includes: Step 3a: For the target domain sample set, use the target domain power flow analysis model to predict the target domain samples, and calculate the error between the predicted output and the actual power flow results: in, This represents the model's predicted value. This represents the true power flow results in the target domain. The top K samples are selected as the key sample set based on error magnitude or representativeness of features. in, This is the sample error threshold. This screening mechanism ensures that the model prioritizes learning samples with high information content and high error sensitivity in the target domain, thereby improving transfer training efficiency.

[0027] Step 3b: Select the top K samples to form the key sample set based on the error magnitude or the representativeness of the features; Step 3c, select the key sample set obtained from the screening. Input model, perform incremental training based on comprehensive loss function, model parameters Update using gradient descent: in, Indicates the first k After +1 iterations, the parameter set of the target domain power flow analysis model. This represents the current set of parameters of the model at the k-th iteration. Represents the gradient operator, For learning rate, This is the k-th iteration model.

[0028] Step 4 specifically includes: Step 4a: Obtain dynamic operating data of the target power system, quickly output the core power flow parameters under the current operating conditions through a deep neural network, compare the calculated power flow parameters with the real-time measured data of the target system, if the error is within the preset threshold, the current calculation result is deemed valid and proceeds to the subsequent optimization stage; if the error exceeds the standard, backtrack to check whether the data acquisition is abnormal, or trigger the model emergency parameter fine-tuning. Step 4b: Based on the calculation results of the core power flow parameters and combined with the safe operation criteria of the power system, determine whether there is an optimization requirement for the current state. If there is an optimization requirement, proceed to step 4c. Step 4c: Based on the optimization requirements and combined with the trend characteristics, formulate an initial control strategy; Step 4d involves converting the initial control strategy into virtual input data and inputting it into the target domain model. The power flow state after control is recalculated, and it is determined whether the calculated result deviates from the preset optimization target. If a deviation exists, the model parameters and control strategy are adjusted synchronously, forming an iteration process until the power flow parameters meet the optimization target. The revised control strategy is then sent to the actual control terminal of the target power system for physical control. The method for acquiring the dynamic operating data of the target power system is periodic acquisition.

[0029] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0030] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0031] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A power system power flow adaptive control method based on deep transfer learning, characterized in that, Includes the following steps: Step 1: Construct a power system source domain power flow analysis model, and combine the power flow data characteristics of the source domain and the target domain to construct a loss function for power flow error; Step 2: Design a network parameter initialization method suitable for power flow analysis, realize the transferable representation of source domain data to target domain, and construct an initialized target domain power flow analysis model; Step 3: Introduce a sample selection and incremental learning mechanism to dynamically screen and retrain key power flow samples in the target domain. Step 4: Use the trained target domain power flow analysis model to perform power flow calculation and optimization control on the target power system, and achieve adaptive optimization of the system operating state through iterative correction and parameter update.

2. The power system power flow adaptive control method based on deep transfer learning according to claim 1, characterized in that, Step 1 specifically includes: Step 1a: Describe the source domain power flow analysis model using polar coordinate power flow equations; Step 1b: Perform nonlinear fitting of source domain power flow data using a deep neural network, and construct a loss function minimizing the deviation between the deep neural network output and the real source domain data as the optimization objective. Step 1c: Introduce the polar coordinate power flow equation as a penalty term and add it to the minimization loss function to obtain the comprehensive loss function.

3. The power system power flow adaptive control method based on deep transfer learning according to claim 2, characterized in that, Step 2 specifically refers to: Step 2a: Extract the weight parameters and bias parameters of the deep neural network, and evaluate the importance of the weight parameters and bias parameters of different network layers. By using a joint index based on gradient sensitivity and weight sparsity, select the parameter set that contributes more to the expression of power flow characteristics. Step 2b: Construct a feature mapping function to realize the transferable representation of the weight parameters and bias parameters of the source domain to the target domain, and construct an initialized target domain power flow analysis model; Step 2c: Based on the statistical characteristics of the target domain data, the weight parameters and bias parameters after migration are recalibrated using parameter normalization and perturbation correction strategies.

4. The power system power flow adaptive control method based on deep transfer learning according to claim 3, characterized in that, The aforementioned parameter normalization and perturbation correction strategy specifically includes: The distribution alignment of the transfer parameters is achieved by minimizing the distance metric between the source domain output distribution and the target domain input distribution.

5. The power system power flow adaptive control method based on deep transfer learning according to claim 3, characterized in that, Step 3 specifically includes: Step 3a: For the target domain sample set, use the target domain power flow analysis model to predict the target domain samples and calculate the error between the predicted output and the actual power flow result. Step 3b: Select the top K samples to form the key sample set based on the error magnitude or the representativeness of the features; Step 3c: Input the selected key sample set into the model and perform incremental training based on the comprehensive loss function.

6. The power system power flow adaptive control method based on deep transfer learning according to claim 1, characterized in that, Step 4 specifically includes: Step 4a: Obtain dynamic operating data of the target power system, quickly output the core power flow parameters under the current operating conditions through a deep neural network, compare the calculated power flow parameters with the real-time measured data of the target system, and if the error is within the preset threshold, determine that the current calculation result is valid and proceed to the subsequent optimization stage. If the error exceeds the standard, backtrack to check whether the data collection is abnormal, or trigger emergency parameter fine-tuning of the model; Step 4b: Based on the calculation results of the core power flow parameters and combined with the safe operation criteria of the power system, determine whether there is an optimization requirement for the current state. If there is an optimization requirement, proceed to step 4c. Step 4c: Based on the optimization requirements and combined with the trend characteristics, formulate an initial control strategy; Step 4d: The initial control strategy is converted into virtual input data and input into the target domain model. The power flow state after control is recalculated. It is determined whether there is a deviation between the calculated result after control and the preset optimization target. If there is a deviation, the model parameters and control strategy are adjusted synchronously to form an iteration until the power flow parameters meet the optimization target. The iteration process is completed, and the corrected control strategy is sent to the actual control terminal of the target power system to perform physical control.

7. The power system power flow adaptive control method based on deep transfer learning according to claim 6, characterized in that, The method for obtaining dynamic operating data of the target power system is periodic acquisition.

Images