A distribution network resource adaptive scheduling method, system, device and medium

Abstract

The present invention relates to the field of distribution network technology, and in particular to a distribution network resource adaptive scheduling method, system, device and medium, comprising: constructing a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power element and the initial data set of the distribution network based on the initial data set of the distribution network and the real-time parameters of the new power element using a support vector machine algorithm; decomposing the multivariate interaction effect in the proxy model using a global sensitivity analysis method to obtain a sensitivity index; adaptively correcting the sampling space of the proxy model according to the sensitivity index to obtain an optimized proxy model; and generating a distribution network resource adaptive scheduling strategy based on the optimized proxy model and the real-time monitoring data of the distribution network using a particle swarm optimization algorithm, thereby optimizing the configuration of distribution network resources under the access scenario of the new power element. The present invention realizes efficient and accurate optimization and configuration of distribution network resources under the access scenario of the new power element, significantly improving the operating performance and stability of the distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network technology, and in particular to a distribution network resource adaptive scheduling method, system, equipment and medium. Background Art

[0002] With the rapid development of new power systems, the operating environment of distribution networks has changed significantly. The widespread access of new power elements such as distributed power sources, electric vehicles and new energy storage in distribution networks has transformed the operating characteristics of distribution networks from the traditional one-way power supply mode to a complex interactive power system with multi-source coordination. This transformation has greatly enriched the operating scenarios of distribution networks, but at the same time it has also posed severe challenges to the real-time regulation capabilities of the power grid, especially in scenarios such as node load balancing, new energy absorption capacity and rapid fault response.

[0003] However, most existing technologies rely on static linear models or single-variable analysis methods, which make it difficult to accurately capture the nonlinear coupling relationship between new power elements such as distributed power sources, electric vehicles and new energy storage and the multidimensional performance of distribution networks. This limitation makes it difficult to provide accurate and reliable decision-making support when facing complex and changeable operating scenarios, thereby affecting the stable operation and efficient management of the power grid. In addition, the existing dispatching system lacks the support of a multidimensional sensitivity analysis framework, making it difficult to establish a mapping relationship between the degree of factor influence and dispatching decisions. Such defects make it difficult for the existing dispatching system to achieve adaptive optimization in a dynamic environment when faced with the high uncertainty brought by new power elements, which restricts the improvement of the overall operating efficiency and reliability of the distribution network.

[0004] In summary, existing technologies are difficult to provide reliable scheduling support in complex and changeable operating scenarios. Especially under the multi-dimensional influence of new power elements, the real-time performance and accuracy of the scheduling system are severely restricted. Therefore, there is an urgent need to provide a distribution network resource adaptive scheduling method to build a quantitative analysis system that can analyze the nonlinear coupling mechanism of multiple elements and provide a refined decision-making basis for intelligent scheduling. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method, system, device and medium for adaptive scheduling of distribution network resources.

[0006] In a first aspect, the present invention provides a method for adaptively scheduling distribution network resources, the method comprising the following steps:

[0007] Collecting distribution network operating status data and preprocessing the distribution network operating status data to obtain an initial distribution network data set;

[0008] Based on the distribution network initial data set and the real-time parameters of the new power elements, a proxy model is constructed using a support vector machine algorithm to characterize the nonlinear coupling relationship between the real-time parameters of the new power elements and the distribution network initial data set;

[0009] A global sensitivity analysis method is used to decompose the multivariate interaction effects in the proxy model and quantify the sensitivity index of each new power element to the initial data set of the distribution network;

[0010] Adaptively modifying the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model;

[0011] Based on the optimization agent model and the real-time monitoring data of the distribution network, a particle swarm optimization algorithm is used to generate an adaptive scheduling strategy for distribution network resources;

[0012] The distribution network resources in the new power element access scenario are optimized and configured according to the distribution network resource adaptive scheduling strategy.

[0013] In a further embodiment, the step of constructing a proxy model representing the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network using a support vector machine algorithm based on the initial data set of the distribution network and the real-time parameters of the new power elements includes:

[0014] Aligning the distribution network initial dataset and the real-time parameters of the new power elements according to timestamps to form an initial sampling spatial dataset;

[0015] Based on the initial sampling spatial data set, a support vector machine algorithm is used to identify key distribution network features that affect the real-time parameters of the new power elements, and a low-dimensional sample point data set is generated;

[0016] The low-dimensional sample point data set is mapped and converted into a high-dimensional feature space using a radial basis kernel function to obtain a high-dimensional feature vector;

[0017] In the high-dimensional feature vector, a nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network is fitted by a support vector regression function to obtain an initial regression model in the high-dimensional feature space;

[0018] Taking minimizing the prediction error as the optimization goal, the initial regression model is solved by a sequential minimum optimization algorithm to obtain the optimal parameters of the regression model;

[0019] According to the optimal parameters of the regression model, a proxy model is constructed to characterize the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network.

[0020] In a further embodiment, the step of using a global sensitivity analysis method to decompose the multivariate interaction effects in the agent model and quantify the sensitivity index of each new power element to the initial data set of the distribution network includes:

[0021] Based on the global sensitivity analysis method, the variance of the new power elements in the proxy model is decomposed to obtain the first-order sensitivity component caused by the independent action of a single new power element, the second-order sensitivity component generated by the superposition of any two new power elements, and the high-order sensitivity component generated by the coupling of at least three new power elements;

[0022] Adding the first-order sensitivity component, the second-order sensitivity component, and the high-order sensitivity component of each new power element to obtain the total effect sensitivity of the corresponding new power element to the fluctuation of the distribution network operation state;

[0023] Calculating the variance contribution ratio of the first-order sensitivity component of each new power element to the total effect sensitivity to obtain the main effect contribution rate of a single new power element;

[0024] The main effect contribution rate is normalized to form the sensitivity index of each new power element to the initial data set of the distribution network.

[0025] In a further implementation scheme, the agent model includes a baseline probability term of the normal operation state of the distribution network under the condition of no new power element access disturbance, a first-order independent influence term that characterizes the independent influence of a single new power element on the distribution network state classification probability, a second-order superposition effect term that characterizes the influence of the superposition effect between any two new power elements on the distribution network state classification probability, and a high-order interaction term that characterizes the influence of the coupling effect of at least three new power elements on the distribution network state classification probability.

[0026] In a further embodiment, the step of adaptively modifying the sampling space of the proxy model using an iterative update strategy based on the sensitivity index to obtain an optimized proxy model includes:

[0027] Delineating a feasible region of the new power element according to the physical constraints of the real-time parameters of the new power element, and uniformly selecting an initial sample point set within the feasible region of the new power element using a Latin hypercube sampling method;

[0028] Inputting the initial sample point set into the proxy model for iterative optimization, and extracting the current optimal solution from the output results of the proxy model during each round of iterative optimization;

[0029] Taking the current optimal solution as the center, determine the core sampling interval of the current round of iteration according to a preset shrinkage ratio;

[0030] Based on the sensitivity index of each new power element to the initial data set of the distribution network, a high-sensitivity full-volume sampling interval and a low-sensitivity sampling interval are screened out within the core sampling interval of the current iteration;

[0031] The low-sensitivity sampling interval is compressed according to the exponential decay law to form a low-sensitivity compressed sampling interval;

[0032] The high-sensitivity full sampling interval is expanded with the current optimal solution as the center to obtain the high-sensitivity extended interval, and the high-sensitivity extended interval is combined with the low-sensitivity compressed sampling interval to form a non-uniform dynamic sampling space.

[0033] The proxy model prediction error in the non-uniform dynamic sampling space is calculated, and when the proxy model prediction error reaches a preset error threshold, the proxy model is optimized and updated according to the non-uniform dynamic sampling space to obtain an optimized proxy model.

[0034] In a further embodiment, the step of generating a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and the real-time monitoring data of the distribution network includes:

[0035] Based on the sensitivity index corresponding to the optimization agent model, a high-dimensional decision space is constructed within the feasible domain of the new power element;

[0036] Randomly generating a number of particles in the high-dimensional decision space, and initializing a learning factor of each particle according to a movement rate and a current position of each particle in the high-dimensional decision space;

[0037] Based on the particle learning factor, the particle swarm optimization algorithm is used to update the spatial position of each particle in the high-dimensional decision space;

[0038] In each round of iteration, the potential solution distance index is calculated based on the average value of the spatial position difference of the particles in the real-time parameter dimensions of each new power element, and the learning factor of the particles is adaptively adjusted according to the potential solution distance index to obtain the latest learning factor;

[0039] Calculating the relative error between the predicted value output by the optimization agent model and the real-time monitoring data of the distribution network, and monitoring the rate of change of the global optimal position during multiple consecutive iterations;

[0040] When the relative error is less than a preset relative error threshold and the rate of change of the global optimal position in multiple consecutive iterations is less than a preset position change threshold, the iterative optimization process is stopped and the current global optimal parameter solution is output;

[0041] The radial basis kernel function is used to demap the global optimal parameters to a high-dimensional space for nonlinear transformation to obtain a distribution network resource adaptive scheduling strategy.

[0042] In a further embodiment, the learning factor is used to control the learning step size and direction of the particle during the search process.

[0043] In a second aspect, the present invention provides a distribution network resource adaptive scheduling system, the system comprising:

[0044] A data acquisition module is used to collect distribution network operation status data and pre-process the distribution network operation status data to obtain an initial distribution network data set;

[0045] A model construction module is used to construct a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network based on the initial data set of the distribution network and the real-time parameters of the new power elements using a support vector machine algorithm;

[0046] A variable decomposition module is used to decompose the multivariate interaction effects in the agent model using a global sensitivity analysis method, and quantify the sensitivity index of each new power element to the initial data set of the distribution network;

[0047] a space optimization module, configured to adaptively modify the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model;

[0048] A strategy generation module is used to generate a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and real-time monitoring data of the distribution network;

[0049] The optimization configuration module is used to optimize the configuration of distribution network resources in the new power element access scenario according to the distribution network resource adaptive scheduling strategy.

[0050] In a third aspect, the present invention also provides a computer device comprising a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory so that the computer device performs the steps of implementing the above method.

[0051] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0052] The present invention provides a method, system, device, and medium for adaptive scheduling of distribution network resources. The method collects distribution network operating status data and preprocesses the data to obtain an initial distribution network dataset. Based on the initial distribution network dataset and real-time parameters of new power elements, a support vector machine algorithm is used to construct a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial distribution network dataset. A global sensitivity analysis method is used to decompose the multivariate interaction effects in the proxy model and quantify the sensitivity index of each new power element to the initial distribution network dataset. Based on the sensitivity index, the sampling space of the proxy model is adaptively modified using an iterative update strategy to obtain an optimized proxy model. Based on the optimized proxy model and real-time distribution network monitoring data, a particle swarm optimization algorithm is used to generate an adaptive scheduling strategy for distribution network resources. Based on the adaptive scheduling strategy, distribution network resources are optimized for the new power element access scenario. Compared with the existing technology, this method analyzes the nonlinear coupling relationship between the new power elements and the distribution network through techniques such as sensitivity analysis and particle swarm optimization, thereby achieving efficient, accurate, and adaptive optimization of distribution network resources for the new power element access scenario, significantly improving the operating performance and stability of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for adaptively scheduling distribution network resources provided by an embodiment of the present invention;

[0054] Figure 2 This is a comparison chart of prediction results of the support vector machine algorithm training set provided by an embodiment of the present invention;

[0055] Figure 3 This is a comparison chart of prediction results of the support vector machine algorithm test set provided by an embodiment of the present invention;

[0056] Figure 4 This is a block diagram of a distribution network resource adaptive scheduling system provided by an embodiment of the present invention;

[0057] Figure 5 It is a structural diagram of a computer device provided by an embodiment of the present invention.

[0058] Explanation of reference numerals: 101, data acquisition module; 102, model building module; 103, variable decomposition module; 104, space optimization module; 105, strategy generation module; 106, optimization configuration module. DETAILED DESCRIPTION

[0059] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings. The embodiments are provided for illustrative purposes only and are not to be construed as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of protection of the present invention. Many changes may be made to the present invention without departing from the spirit and scope of the present invention.

[0060] refer to Figure 1 , the embodiment of the present invention provides a method for adaptively scheduling distribution network resources, such as Figure 1 As shown, the method includes the following steps:

[0061] S1. Collecting distribution network operating status data and preprocessing the distribution network operating status data to obtain an initial distribution network data set.

[0062] This embodiment determines the collection period according to the actual operation of the distribution network and the data analysis requirements, and collects multi-dimensional distribution network operation status data from the distribution network monitoring system in real time during the collection period. The distribution network operation status data includes distribution network operation indicators, distribution network carrying capacity indicators and distribution network power supply support capacity indicators. Among them, the distribution network operation indicators may include substation load rate, line load rate and line loss rate. Among them, the substation load rate represents the ratio of the actual power supply of the substation to the rated power supply within the preset collection period, which is used to measure the operation efficiency and load utilization of the substation; the line load rate represents the ratio of the actual power of the line to the rated power, which is used to measure the utilization rate and operation status of the line; the line loss rate represents the proportion of power grid transmission loss to the total power supply; the distribution network carrying capacity is the ratio of the actual power supply to the rated power of the line .... The load capacity index may include the new energy power generation absorption rate, which represents the proportion of the actual absorption of new energy power in the total power generation. It is an important indicator for measuring the effective utilization of new energy power and reflects the absorption capacity of new energy power in the power grid; the distribution network power supply support capacity index may include physical parameters such as fault load loss and power supply interruption time. The fault load loss represents the total amount of load decoupled due to the fault, and the power supply interruption time represents the cumulative time required for fault recovery. This embodiment performs a preprocessing operation on the collected distribution network operation status data to obtain an initial data set of the distribution network, wherein the preprocessing operation includes data cleaning and data standardization processing, and data cleaning includes checking whether there are missing values ​​in the data set and deleting duplicate values ​​in the data set.

[0063] At the same time, this embodiment obtains the real-time operating parameters of the new power elements to form the real-time parameters of the new power elements. The real-time parameters of the new power elements may include multiple dimensions such as distributed power output, electric vehicle charging load, energy storage charging and discharging power, and microgrid operation mode. Among them, distributed power sources can reduce the load rate of substations and lines in normal operation of the distribution network, and effectively reduce line losses; in terms of carrying capacity, the access of distributed power sources may reduce the new energy absorption rate, and in extreme cases even aggravate the phenomenon of wind and solar power abandonment; in terms of power supply support capability, distributed power sources can reduce fault load losses and shorten power outage time in the event of a fault, thereby enhancing the overall flexibility of the system.

[0064] The disordered charging of electric vehicles during normal operation of the distribution network can increase the load rate and line loss. In terms of carrying capacity, if the charging load of electric vehicles does not match the timing of new energy power generation, it will weaken the absorption of new energy and increase the peak-shaving pressure of the traditional power grid. In terms of power supply support capabilities, the vehicle-grid interaction technology of electric vehicles can help reduce fault losses and improve response speed.

[0065] During the normal operation of the distribution network, the energy storage charging and discharging power effectively reduces the load rate of substations and lines through the peak shaving and valley filling strategy, and reduces line losses caused by power flow fluctuations; in terms of carrying capacity, new energy storage can increase the penetration rate of distributed power sources and the absorption rate of new energy; in terms of power supply support capability, its rapid response characteristics can greatly reduce load losses and power outage time during faults, significantly enhancing the anti-disturbance capability of the distribution network.

[0066] The microgrid operation mode can reduce the load pressure on the upper-level substations and lines and reduce network losses through optimized scheduling of internal power sources during normal operation of the distribution network; in terms of carrying capacity, the independent operation characteristics of the microgrid may weaken the main grid's ability to control new power elements, especially when multiple microgrids simultaneously supply power to the main grid, which may cause the penetration rate in local areas to exceed the standard, resulting in voltage fluctuations or protection malfunction; in terms of power supply support capability, the microgrid's island operation capability can minimize the scope of fault impact and significantly shorten the power supply restoration time.

[0067] S2. Based on the distribution network initial data set and the real-time parameters of the new power elements, a support vector machine algorithm is used to construct an agent model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the distribution network initial data set.

[0068] In some embodiments, the step of constructing a proxy model representing the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network using a support vector machine algorithm based on the initial data set of the distribution network and the real-time parameters of the new power elements includes:

[0069] Aligning the distribution network initial dataset and the real-time parameters of the new power elements according to timestamps to form an initial sampling spatial dataset;

[0070] Based on the initial sampling spatial data set, a support vector machine algorithm is used to identify key distribution network features that affect the real-time parameters of the new power elements, and a low-dimensional sample point data set is generated;

[0071] The low-dimensional sample point data set is mapped and converted into a high-dimensional feature space using a radial basis kernel function to obtain a high-dimensional feature vector;

[0072] In the high-dimensional feature vector, a nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network is fitted by a support vector regression function to obtain an initial regression model in the high-dimensional feature space;

[0073] Taking minimizing the prediction error as the optimization goal, the initial regression model is solved by a sequential minimum optimization algorithm to obtain the optimal parameters of the regression model;

[0074] According to the optimal parameters of the regression model, a proxy model is constructed to characterize the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network.

[0075] Specifically, this embodiment aligns the distribution network initial data set and the real-time parameters of the new power elements by timestamp, and combines the aligned data into an initial sampling spatial data set containing synchronized time points to ensure data consistency in the time dimension. For missing time points, this embodiment can perform interpolation processing or ignore them according to actual conditions. After alignment, the initial sampling spatial data set contains data of the distribution network initial data set and the real-time parameters of the new power elements at the same time point. Then, this embodiment uses the recursive feature elimination method of the support vector machine to analyze the correlation between the distribution network initial data set and the real-time parameters of the new power elements, and selects key distribution network features with correlations greater than a preset correlation threshold, thereby identifying key distribution network features that have a significant impact on the real-time parameters of the new power elements. These key distribution network features may include substation load rate, line load rate, and line loss rate, etc. This embodiment uses the screened key distribution network features as input of low-dimensional sample points and the real-time parameters of the new power elements as output to generate a low-dimensional sample point data set.

[0076] Next, this embodiment adopts a radial basis kernel function as a mapping function. The radial basis kernel function can map low-dimensional sample point data to a high-dimensional feature space, solve the problem of linear inseparability in the low-dimensional space, and thus better capture the complex nonlinear relationship between the new power element and the distribution network. Specifically, this embodiment maps each sample point in the low-dimensional sample point data set to a high-dimensional feature space through a radial basis kernel function to obtain a high-dimensional feature vector. These high-dimensional feature vectors contain the feature representation of the original low-dimensional data in the high-dimensional space, and can more comprehensively reflect the relationship between the new power element and the distribution network. In this embodiment, in the high-dimensional feature space, a support vector regression function is used to fit the nonlinear coupling relationship between the real-time parameters of the new power element and the initial data set of the distribution network, wherein the support vector regression function can be expressed as:

[0077]

[0078] Where y is the support vector regression function, which is used to fit the nonlinear coupling relationship between the real-time parameters of the new power elements and the multidimensional state of the distribution network; represents the predicted output value of the support vector regression function; is the weight vector; is the nonlinear mapping function from low-dimensional space to high-dimensional space; b is the bias term.

[0079] This embodiment can find the hyperplane that can best describe the relationship between the new power element and the distribution network in the high-dimensional feature space by fitting the support vector regression function. At the same time, in order to determine the weight vector and the bias term, this embodiment minimizes the prediction error and model complexity (given by the modulus of the weight vector w). ) is the optimization goal to strike a balance between the complexity of the model and the fitting accuracy. At the same time, the penalty factor and relaxation factor are introduced to control the complexity of the model and the degree of fitting to the training data. The objective function of the optimization problem is as follows:

[0080]

[0081] Where Q is the objective function; is the weight vector; C is the penalty factor, which is used to balance the model complexity and training error; is the lower bound relaxation factor used to handle the negative direction; is the upper bound relaxation factor for processing the positive direction, which allows sample points to exceed Error tolerance range of insensitive areas; is the square norm of the weight vector, and in the objective function, by minimizing To control the complexity of the model, prevent overfitting, and reflect the overall strength of the model's sensitivity to features; m is the number of training samples.

[0082] The constraints of the optimization problem are:

[0083]

[0084] Where, is the input feature vector of the i-th sample; For samples The high-dimensional feature vector after mapping; b is the bias term; is the true output value of the i-th sample; is the insensitive region of the loss function, which defines the allowable boundary of the prediction error, i.e. Forecast errors within this range are considered acceptable and are not penalized.

[0085] This embodiment uses the sequential minimal optimization (SMO) algorithm to iteratively solve the constructed optimization problem. The SMO algorithm is a quadratic programming algorithm that can select two variables for optimization one by one, thereby gradually approaching the global optimal solution. In each iteration, the SMO algorithm selects two Lagrange multipliers for optimization until the convergence condition is met. After solving, the optimal weight vector of the regression model is obtained. and the optimal bias term b. In this embodiment, the optimal parameters of the regression model obtained are substituted into the support vector regression function to obtain the final proxy model. The proxy model can characterize the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network. Based on the initial data set of the distribution network and the real-time parameters of the new power elements, the changes in the multi-dimensional indicators of the distribution network are predicted, providing a basis for subsequent sensitivity analysis and decision-making.

[0086] In this embodiment, the proxy model is used to quantify the nonlinear coupling relationship between new power elements and the operating state of the distribution network. The mathematical terms of the proxy model include a baseline probability term for the normal operating state of the distribution network without any new power element access disturbance; a first-order independent influence term that represents the independent influence of a single new power element (such as distributed power generation output or energy storage charging and discharging power) on the distribution network state classification probability when acting independently; a second-order superposition effect term that represents the influence of the superposition effect between any two new power elements (such as electric vehicle charging load and microgrid operating mode) on the distribution network state classification probability; and a higher-order interaction term that represents the influence of the coupled effects of at least three new power elements (such as distributed power generation, energy storage, and electric vehicles) on the distribution network state classification probability. By superimposing the baseline probability term and the multi-order effect terms, this embodiment can comprehensively characterize the dynamic impact of the independent and synergistic effects of new power elements on the operating state of the distribution network.

[0087] S3. Use the global sensitivity analysis method to decompose the multivariate interaction effects in the agent model and quantify the sensitivity index of each new power element to the initial data set of the distribution network.

[0088] In some embodiments, the step of using a global sensitivity analysis method to decompose the multivariate interaction effects in the agent model and quantify the sensitivity index of each new power element to the initial data set of the distribution network includes:

[0089] Based on the global sensitivity analysis method, the variance of the new power elements in the proxy model is decomposed to obtain the first-order sensitivity component caused by the independent action of a single new power element, the second-order sensitivity component generated by the superposition of any two new power elements, and the high-order sensitivity component generated by the coupling of at least three new power elements;

[0090] Adding the first-order sensitivity component, the second-order sensitivity component, and the high-order sensitivity component of each new power element to obtain the total effect sensitivity of the corresponding new power element to the fluctuation of the distribution network operation state;

[0091] Calculating the variance contribution ratio of the first-order sensitivity component of each new power element to the total effect sensitivity to obtain the main effect contribution rate of a single new power element;

[0092] The main effect contribution rate is normalized to form the sensitivity index of each new power element to the initial data set of the distribution network.

[0093] Specifically, in order to quantify the impact of new power elements on the multidimensional capacity of the distribution network, this embodiment uses the Sobol global sensitivity analysis method based on the nonlinear fitting relationship represented by the agent model to decompose the multidimensional interactive effect of new power elements on the distribution network state, and converts the nonlinear relationship output by the agent model into It is decomposed into baseline probability term, independent influence term and interaction effect term, which can be expressed as follows:

[0094]

[0095] Where, It is a probability model for the nonlinear fitting relationship between the new power elements and the multi-dimensional distribution network, that is, it quantifies the variance contribution of the multi-dimensional state of the distribution network affected by the new power elements; is the baseline probability term, which represents the probability of normal operation of the distribution network when no new power elements are connected; is a first-order independent influence term, which represents a single new power element (such as distributed generation output) independent impact component on the probability of distribution network status classification; is the second-order superposition effect term, which represents any two new power elements and The superposition effect between them affects the probability of distribution network status classification; n is the number of new power elements; is a high-order interaction term, which represents the impact of the superposition of at least three new power elements on the probability of distribution network state classification; is the i-th new power element; is the jth new power element; It is the nth new type of power element.

[0096] In this embodiment, the total variance Decomposed into the contribution of each order term, the decomposition form of the total variance is expressed as:

[0097]

[0098] Where, is the total variance, which indicates the total contribution of all new power elements individually and in combination to the multi-dimensional state fluctuation of the distribution network; is the integral of the output of the agent model, that is, the sum of the probability contributions of all new power elements to the distribution network state; is the square value of the baseline probability, which reflects the uncertainty or fluctuation of the multi-dimensional state of the distribution network itself when there is no impact of new power elements; is an n-dimensional real number space (the value range of real-time parameters of all new power elements); dX is a differential element.

[0099] Among them, the decomposition items under the individual or superposition items are calculated by multidimensional integration. In this embodiment, the contribution of each new power element and its interaction to the total variance is obtained by summing the integrals of each individual and superposition effect. Assume that exist Within the scope, the decomposition items under individual or superimposed items are specifically expressed as follows:

[0100]

[0101] Where, is the multidimensional joint variance term, i.e., the variance contribution of all the new power elements considered to the distribution network status; is the set of real-time parameters of all new power elements considered, Each is any element in the real-time parameter set of all new power elements; The probability distribution of distribution network status under the coupling of multiple new power elements; are the upper and lower limits of the integration of the nth new power factor variable.

[0102] Therefore, the total variance decomposition formula and the decomposition formula under separate or superimposed terms can be further decomposed into:

[0103]

[0104] Where, is the first-order sensitivity component, representing a single new power element The degree of influence of independent actions on the multi-dimensional capabilities of the distribution network; is the second-order sensitivity component, representing and The degree of influence of the superposition of the two factors on the multi-dimensional capacity of the distribution network; It is a high-order sensitivity component, which represents the influence of multi-factor coupling on the multi-dimensional capacity of the distribution network.

[0105] Therefore, this embodiment decomposes the output variance of the proxy model into first-order sensitivity components, second-order sensitivity components and high-order sensitivity components through the above steps, wherein the first-order sensitivity component represents the variance increment caused by the independent action of a single new power element (such as distributed power output, energy storage charging and discharging power); the second-order sensitivity component represents the additional variance increment caused by the synergistic action of any two new power elements (such as electric vehicle charging load and new energy consumption rate); the high-order sensitivity component represents the complex variance increment caused by the joint action of three or more new power elements, thereby obtaining a set of independent and interactive sensitivity components of each new power element. Then, based on the variance decomposition results, this embodiment quantifies the contribution rate of each new power element to the fluctuation of the distribution network state. Specifically, For each new power element, this embodiment adds its first-order sensitivity component and all second-order and higher-order sensitivity components related to the element to obtain the total effect sensitivity of the element to the fluctuation of the distribution network operating state. This reflects the sum of the degree of influence of the new power element on the distribution network in all possible individual and interactive situations, thereby obtaining the total effect sensitivity set of each new power element. Next, this embodiment divides the first-order sensitivity component of a single new power element by its total effect sensitivity to obtain the main effect contribution rate of the single new power element. The main effect contribution rate reflects the contribution ratio of the independent action of a single new power element to the total effect. It represents the contribution ratio of the new power element to the multi-dimensional state fluctuation of the distribution network when it acts alone. The specific mathematical expression is:

[0106]

[0107] Where, It is the main effect contribution rate of a single new power factor acting independently on the total variance.

[0108] In this embodiment, all main effect contribution rates are normalized so that their values ​​are within the range of [0, 1] to eliminate dimensional differences. The normalized main effect contribution rate is defined as the sensitivity index of each new power element to the operating status of the distribution network.

[0109] S4. According to the sensitivity index, the sampling space of the proxy model is adaptively modified with an iterative update strategy to obtain an optimized proxy model.

[0110] In some embodiments, the step of adaptively modifying the sampling space of the proxy model using an iterative update strategy based on the sensitivity index to obtain an optimized proxy model includes:

[0111] Delineating a feasible region of the new power element according to the physical constraints of the real-time parameters of the new power element, and uniformly selecting an initial sample point set within the feasible region of the new power element using a Latin hypercube sampling method;

[0112] Inputting the initial sample point set into the proxy model for iterative optimization, and extracting the current optimal solution from the output results of the proxy model during each round of iterative optimization;

[0113] Taking the current optimal solution as the center, determine the core sampling interval of the current round of iteration according to a preset shrinkage ratio;

[0114] Based on the sensitivity index of each new power element to the initial data set of the distribution network, a high-sensitivity full-volume sampling interval and a low-sensitivity sampling interval are screened out within the core sampling interval of the current iteration;

[0115] The low-sensitivity sampling interval is compressed according to the exponential decay law to form a low-sensitivity compressed sampling interval;

[0116] The high-sensitivity full sampling interval is expanded with the current optimal solution as the center to obtain the high-sensitivity extended interval, and the high-sensitivity extended interval is combined with the low-sensitivity compressed sampling interval to form a non-uniform dynamic sampling space.

[0117] The proxy model prediction error in the non-uniform dynamic sampling space is calculated, and when the proxy model prediction error reaches a preset error threshold, the proxy model is optimized and updated according to the non-uniform dynamic sampling space to obtain an optimized proxy model.

[0118] Specifically, this embodiment defines a multidimensional feasible domain of real-time parameters of the feasible domain of new power elements according to the physical constraints of each new power element. For example, the physical constraints of the real-time parameters of the new power elements may include the range boundaries of parameters such as the output range of distributed power sources, the power limit of energy storage charging and discharging, and the charging load threshold of electric vehicles. Then, the Latin hypercube sampling method is used to uniformly select an initial sample point set within the feasible domain of the new power element. This method can ensure that the sample points are evenly distributed within the feasible domain and improve the representativeness of the initial sample points. The initial sample point set evenly covers the typical operating scenarios of each new power element. For example, the initial The initial sample point set can cover operating scenarios such as full charging of energy storage and centralized charging of electric vehicles. This embodiment initializes the sampling space, selects a sufficient number of sample points from the initial sample point set to fit the nonlinear relationship between the new power elements and the multidimensional capabilities of the distribution network, and inputs the initial sample point set into the agent model for iterative optimization to fit the nonlinear relationship between the new power elements and the operating status of the distribution network. In each round of iterative optimization, the current optimal solution (such as the parameter combination that minimizes the line loss rate) is extracted from the output results of the agent model as a benchmark for subsequent sampling space adjustments. For example, in the (k-1)th round of iteration, the current optimal solution is obtained. .

[0119] Next, this embodiment uses the optimal solution of the (k-1)th iteration of the agent model As the center, shrink the upper and lower limits according to the preset ratio (such as 10% of the size of the previous round of sampling space) to generate the core sampling interval of the current round, which can be expressed as:

[0120]

[0121] Where, The lower limit of the core sampling interval determined by the current optimal solution of the surrogate model at the kth iteration; The upper limit of the core sampling interval determined by the current optimal solution of the surrogate model at the kth iteration; is the current optimal solution at the (k-1)th iteration; is the sampling space size at the (k-1)th iteration.

[0122] In this embodiment, within the core sampling interval, a high-sensitivity full sampling interval with a sensitivity index higher than a preset sensitivity threshold is expanded with the current optimal solution as the center to obtain a high-sensitivity expansion interval. For a low-sensitivity sampling interval with a sensitivity index lower than the preset sensitivity threshold, the sampling range of the low-sensitivity sampling interval of the previous iteration is compressed using a preset exponential decay coefficient (e.g., compression by 20% per iteration) to form a low-sensitivity compressed sampling interval. In this embodiment, the high-sensitivity expansion interval and the low-sensitivity compressed sampling interval are combined to form a non-uniform dynamic sampling space. Finally, a new set of sample points is regenerated within the non-uniform dynamic sampling space, preferentially covering the high-sensitivity parameter area. The proxy model parameters are updated based on the new sample points. The prediction error of the proxy model within the non-uniform dynamic sampling space is calculated using the root mean square error. If the prediction error of the proxy model within the non-uniform dynamic sampling space reaches the preset error threshold and the error change rate for three consecutive iterations is less than 1%, convergence is determined. The proxy model is retrained using the sample point set within the non-uniform dynamic sampling space, and the model parameters are updated to obtain an optimized proxy model.

[0123] S5. Based on the optimization agent model and the real-time monitoring data of the distribution network, a particle swarm optimization algorithm is used to generate an adaptive scheduling strategy for distribution network resources.

[0124] S6. Optimize the configuration of distribution network resources in the new power element access scenario according to the distribution network resource adaptive scheduling strategy.

[0125] In some embodiments, the step of generating a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and real-time distribution network monitoring data includes:

[0126] Based on the sensitivity index corresponding to the optimization agent model, a high-dimensional decision space is constructed within the feasible domain of the new power element;

[0127] Randomly generating a number of particles in the high-dimensional decision space, and initializing a learning factor of each particle according to a movement rate and a current position of each particle in the high-dimensional decision space;

[0128] Based on the particle learning factor, the particle swarm optimization algorithm is used to update the spatial position of each particle in the high-dimensional decision space;

[0129] In each round of iteration, the potential solution distance index is calculated based on the average value of the spatial position difference of the particles in the real-time parameter dimensions of each new power element, and the learning factor of the particles is adaptively adjusted according to the potential solution distance index to obtain the latest learning factor;

[0130] Calculating the relative error between the predicted value output by the optimization agent model and the real-time monitoring data of the distribution network, and monitoring the rate of change of the global optimal position during multiple consecutive iterations;

[0131] When the relative error is less than a preset relative error threshold and the rate of change of the global optimal position in multiple consecutive iterations is less than a preset position change threshold, the iterative optimization process is stopped and the current global optimal parameter solution is output;

[0132] The radial basis kernel function is used to demap the global optimal parameters to a high-dimensional space for nonlinear transformation to obtain a distribution network resource adaptive scheduling strategy.

[0133] Specifically, this embodiment expands the parameter sampling range for highly sensitive factors based on the sensitivity index, constructs a high-dimensional decision space, and randomly generates a particle swarm in the high-dimensional decision space using the Latin hypercube sampling method. Each particle represents a potential distribution network resource scheduling strategy. At the same time, this embodiment initializes the individual learning factor (exploring historical optimality) and social learning factor (tracking global optimality) of the particle based on the particle's movement rate and current position in space. For example, the initial values ​​of the individual learning factor and the social learning factor are 1.5 and 2.0, respectively. The learning factors are used to control the learning step size and direction of the particle during the search process. The individual learning item is used to control the particle to accelerate toward its own historical optimal position, and the social learning item is used to control the particle to accelerate toward the current global optimal position. In this embodiment, the current speed is updated by a weighted combination of the inertia weight, the individual learning item, and the social learning item, and the coordinates of the particle in the parameter space are adjusted according to the updated speed. For each particle, this embodiment uses the following formula to update its speed and position:

[0134]

[0135]

[0136] Where, is the velocity of the qth particle in the gth iteration in the dth dimension; is the inertia weight, which is used to balance the global and local search capabilities of particles; is the velocity of the qth particle in the dth dimension space at the (g-1)th iteration; is the individual learning factor, which reflects the particle’s ability to learn its own historical optimal position; is the social learning factor, which reflects the particle’s ability to learn the global optimal position; and are mutually independent random functions between [0, 1]; is the historical optimal position of the qth particle; is the position of the qth particle in the dth dimensional space at the (g-1)th iteration; is the global optimal position; is the position of the qth particle in the gth iteration in the dth dimensional space.

[0137] In each round of iteration, this embodiment calculates the average value of the position difference of the particle swarm in each parameter dimension to obtain the potential solution distance index. The calculation formula of the potential solution distance is:

[0138]

[0139] Where, is the potential solution distance; e is the number of particles; h is the spatial dimension; is the value of the qth particle in the dth dimension; is the average value of all particles in the d-th dimension; is the number of samples in the sampling space.

[0140] This embodiment adaptively adjusts the individual learning factor and social learning factor of the particle based on the potential solution spacing index. When the potential solution spacing index is too large, the social learning factor is increased to accelerate the global search; when the potential solution spacing index is too small, the individual learning factor is reduced to enhance the local optimization capability, thereby obtaining a dynamically adjusted learning factor parameter set. This embodiment can statistically calculate the relative error between the agent model prediction value and the actual monitoring value. If the relative error is less than 3% and the rate of change of the global optimal position is less than 1% in three consecutive iterations, a radial basis kernel function is used to map the global optimal parameter solution to a high-dimensional space for nonlinear transformation, thereby enhancing the model's ability to analyze complex coupling effects, thereby obtaining a high-precision parameter solution set that meets the convergence conditions and a corrected nonlinear relationship model. The corrected nonlinear relationship model can be expressed as:

[0141]

[0142] Where, is the nonlinear relationship after correction; is the Lagrange multiplier, which is the coefficient obtained by solving the optimization problem during the training of the support vector machine algorithm, reflecting the contribution of each support vector to the decision function; For samples The optimal solution of the corresponding Lagrange multiplier; is the input feature vector of the jth sample; is the input feature vector of the i-th sample.

[0143] This embodiment can generate corresponding control instructions based on the global optimal parameter solution. For example, the energy storage system can smooth load fluctuations according to the optimal charging and discharging timing; electric vehicles dynamically adjust the charging power upper limit to guide the charging load to match the peak and valley of new energy output; microgrids switch and disconnect the grid mode to alleviate the voltage fluctuation of the main grid; distributed power sources adjust the output priority according to the sensitivity weight to improve the absorption efficiency. This embodiment transmits the control instructions to each new power element controller through the distribution automation system, monitors the grid status after the instruction is executed in real time, and triggers the iterative update of the strategy. In summary, this embodiment uses dynamic particle swarm optimization and kernel function correction to accurately generate a scheduling strategy that adapts to the new power element access scenario, significantly improving the operating efficiency and stability of the distribution network. Figure 2 and Figure 3 The figure shows a comparison of the prediction results of the support vector machine algorithm training set and test set. It can be seen from the results that the support vector machine algorithm has a good fitting effect on the nonlinear relationship between the new power elements and the multi-dimensional operating status of the distribution network. Among them, the root mean square error of the training set is 0.9253, and the root mean square error of the test set is 0.9355.

[0144] An embodiment of the present invention provides a method for adaptively scheduling distribution network resources. The method collects distribution network operating status data and preprocesses the data to obtain an initial distribution network dataset. Based on the initial distribution network dataset and real-time parameters of new power elements, a support vector machine algorithm is used to construct a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial distribution network dataset. A global sensitivity analysis method is used to decompose the multivariate interaction effects in the proxy model and quantify the sensitivity index of each new power element to the initial distribution network dataset. Based on the sensitivity index, the sampling space of the proxy model is adaptively modified using an iterative update strategy to obtain an optimized proxy model. Based on the optimized proxy model and real-time distribution network monitoring data, a particle swarm optimization algorithm is used to generate an adaptive distribution network resource scheduling strategy. Based on the adaptive distribution network resource scheduling strategy, distribution network resources are optimized for the new power element access scenario. Compared with the existing technology, this method analyzes the nonlinear coupling relationship between the new power elements and the distribution network through techniques such as sensitivity analysis and particle swarm optimization, thereby achieving efficient, accurate, and adaptive optimization of distribution network resources for the new power element access scenario, significantly improving the operating performance and stability of the distribution network.

[0145] It should be noted that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of this application.

[0146] In one embodiment, Figure 4As shown, an embodiment of the present invention provides a distribution network resource adaptive scheduling system, the system comprising:

[0147] The data acquisition module 101 is used to collect the distribution network operation status data and pre-process the distribution network operation status data to obtain the distribution network initial data set;

[0148] A model building module 102 is configured to build, based on the distribution network initial data set and the new power element real-time parameters, a proxy model that characterizes the nonlinear coupling relationship between the new power element real-time parameters and the distribution network initial data set using a support vector machine algorithm;

[0149] A variable decomposition module 103 is configured to decompose the multivariate interaction effects in the agent model using a global sensitivity analysis method, and quantify the sensitivity index of each new power element to the initial data set of the distribution network;

[0150] A space optimization module 104 is configured to adaptively modify the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model;

[0151] A strategy generation module 105 is configured to generate a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and real-time monitoring data of the distribution network;

[0152] The optimization configuration module 106 is used to optimize the configuration of distribution network resources in the new power element access scenario according to the distribution network resource adaptive scheduling strategy.

[0153] For the specific definition of a distribution network resource adaptive scheduling system, please refer to the above-mentioned definition of a distribution network resource adaptive scheduling method, which will not be repeated here. A person of ordinary skill in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in this application can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0154] An embodiment of the present invention provides a distribution network resource adaptive scheduling system. The system collects distribution network operating status data through a data acquisition module and pre-processes the distribution network operating status data to obtain an initial distribution network data set; a model construction module uses a support vector machine algorithm to construct a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial distribution network data set based on the initial distribution network data set and the real-time parameters of the new power elements; a variable decomposition module uses a global sensitivity analysis method to decompose the multivariate interaction effect in the proxy model and quantify the sensitivity index of each new power element to the initial distribution network data set; a space optimization module uses an iterative update strategy to adaptively correct the sampling space of the proxy model based on the sensitivity index to obtain an optimized proxy model; a strategy generation module uses a particle swarm optimization algorithm to generate a distribution network resource adaptive scheduling strategy based on the optimized proxy model and the real-time monitoring data of the distribution network; and an optimization configuration module optimizes the configuration of distribution network resources in the new power element access scenario according to the distribution network resource adaptive scheduling strategy. Compared with existing technologies, this system analyzes the nonlinear coupling relationship between new power elements and distribution networks through sensitivity analysis and particle swarm optimization algorithms, and realizes efficient, accurate and adaptive optimization configuration of distribution network resources under the access scenarios of new power elements, significantly improving the operating performance and stability of the distribution network.

[0155] Figure 5 A computer device provided in an embodiment of the present invention includes a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and can transmit the stored data to the processor, and the processor can execute the program instructions stored in the memory to perform the steps of the above method.

[0156] The memory may include volatile memory or non-volatile memory, or may include both volatile and non-volatile memory; the processor may be a central processing unit, a microprocessor, an application-specific integrated circuit, a programmable logic device, or a combination thereof. By way of example and not limitation, the programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0157] Additionally, the memory may be a physically separate unit or integrated with the processor.

[0158] It can be understood by those skilled in the art that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have the same component arrangement.

[0159] In one embodiment, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above method are implemented.

[0160] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., an SSD).

[0161] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0162] The above-described embodiments merely represent several preferred implementations of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art could make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be based on the scope of protection of the claims.

Claims

1. A method for adaptively scheduling distribution network resources, characterized in that: The following steps are involved: Collecting distribution network operating status data and preprocessing the distribution network operating status data to obtain an initial distribution network data set; Based on the distribution network initial data set and the real-time parameters of the new power elements, a support vector machine algorithm is used to construct an agent model that represents the nonlinear coupling relationship between the real-time parameters of the new power elements and the distribution network initial data set; the new power elements include distributed power sources, electric vehicles and new energy storage in the distribution network; Real-time parameters of new power elements include distributed power output, electric vehicle charging load, energy storage charging and discharging power, and microgrid operation mode; A global sensitivity analysis method is used to decompose the multivariate interaction effects in the proxy model and quantify the sensitivity index of each new power element to the initial data set of the distribution network; Adaptively modifying the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model; Based on the optimization agent model and the real-time monitoring data of the distribution network, a particle swarm optimization algorithm is used to generate an adaptive scheduling strategy for distribution network resources; The distribution network resources in the new power element access scenario are optimized and configured according to the distribution network resource adaptive scheduling strategy.

2. A method for adaptively scheduling distribution network resources according to claim 1, characterized in that: The step of constructing a proxy model representing the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network using a support vector machine algorithm based on the initial data set of the distribution network and the real-time parameters of the new power elements includes: Aligning the distribution network initial dataset and the real-time parameters of the new power elements according to timestamps to form an initial sampling spatial dataset; Based on the initial sampling spatial data set, a support vector machine algorithm is used to identify key distribution network features that affect the real-time parameters of the new power elements, and a low-dimensional sample point data set is generated; The low-dimensional sample point data set is mapped and converted into a high-dimensional feature space using a radial basis kernel function to obtain a high-dimensional feature vector; In the high-dimensional feature vector, a nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network is fitted by a support vector regression function to obtain an initial regression model in the high-dimensional feature space; Taking minimizing the prediction error as the optimization goal, the initial regression model is solved by a sequential minimum optimization algorithm to obtain the optimal parameters of the regression model; According to the optimal parameters of the regression model, a proxy model is constructed to characterize the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network.

3. A method for adaptively scheduling distribution network resources according to claim 1, characterized in that: The step of using a global sensitivity analysis method to decompose the multivariate interaction effects in the agent model and quantify the sensitivity index of each new power element to the initial data set of the distribution network includes: Based on the global sensitivity analysis method, the variance of the new power elements in the proxy model is decomposed to obtain the first-order sensitivity component caused by the independent action of a single new power element, the second-order sensitivity component generated by the superposition of any two new power elements, and the high-order sensitivity component generated by the coupling of at least three new power elements; Adding the first-order sensitivity component, the second-order sensitivity component, and the high-order sensitivity component of each new power element to obtain the total effect sensitivity of the corresponding new power element to the fluctuation of the distribution network operation state; Calculating the variance contribution ratio of the first-order sensitivity component of each new power element to the total effect sensitivity to obtain the main effect contribution rate of a single new power element; The main effect contribution rate is normalized to form the sensitivity index of each new power element to the initial data set of the distribution network.

4. A method for adaptively scheduling distribution network resources according to claim 3, characterized in that: The agent model includes a baseline probability term for the normal operation of the distribution network under the condition that there is no disturbance caused by the access of any new power elements, a first-order independent influence term that characterizes the independent influence of a single new power element on the probability of distribution network state classification, a second-order superposition effect term that characterizes the influence of the superposition effect between any two new power elements on the probability of distribution network state classification, and a high-order interaction term that characterizes the influence of the coupling effect of at least three new power elements on the probability of distribution network state classification.

5. A method for adaptively scheduling distribution network resources according to claim 1, characterized in that: The step of adaptively correcting the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model includes: Delineating a feasible region of the new power element according to the physical constraints of the real-time parameters of the new power element, and uniformly selecting an initial sample point set within the feasible region of the new power element using a Latin hypercube sampling method; Inputting the initial sample point set into the proxy model for iterative optimization, and extracting the current optimal solution from the output results of the proxy model during each round of iterative optimization; Taking the current optimal solution as the center, determine the core sampling interval of the current round of iteration according to a preset shrinkage ratio; Based on the sensitivity index of each new power element to the initial data set of the distribution network, a high-sensitivity full-volume sampling interval and a low-sensitivity sampling interval are screened out within the core sampling interval of the current iteration; The low-sensitivity sampling interval is compressed according to the exponential decay law to form a low-sensitivity compressed sampling interval; The high-sensitivity full sampling interval is expanded with the current optimal solution as the center to obtain the high-sensitivity extended interval, and the high-sensitivity extended interval is combined with the low-sensitivity compressed sampling interval to form a non-uniform dynamic sampling space. The proxy model prediction error in the non-uniform dynamic sampling space is calculated, and when the proxy model prediction error reaches a preset error threshold, the proxy model is optimized and updated according to the non-uniform dynamic sampling space to obtain an optimized proxy model.

6. A method for adaptively scheduling distribution network resources according to claim 5, characterized in that: The step of generating a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and the real-time monitoring data of the distribution network includes: Based on the sensitivity index corresponding to the optimization agent model, a high-dimensional decision space is constructed within the feasible domain of the new power element; Randomly generating a number of particles in the high-dimensional decision space, and initializing a learning factor of each particle according to a movement rate and a current position of each particle in the high-dimensional decision space; Based on the particle learning factor, the particle swarm optimization algorithm is used to update the spatial position of each particle in the high-dimensional decision space; In each round of iteration, the potential solution distance index is calculated based on the average value of the spatial position difference of the particles in the real-time parameter dimensions of each new power element, and the learning factor of the particles is adaptively adjusted according to the potential solution distance index to obtain the latest learning factor; Calculating the relative error between the predicted value output by the optimization agent model and the real-time monitoring data of the distribution network, and monitoring the rate of change of the global optimal position during multiple consecutive iterations; When the relative error is less than a preset relative error threshold and the rate of change of the global optimal position in multiple consecutive iterations is less than a preset position change threshold, the iterative optimization process is stopped and the current global optimal parameter solution is output; The radial basis kernel function is used to demap the global optimal parameters to a high-dimensional space for nonlinear transformation to obtain a distribution network resource adaptive scheduling strategy.

7. A method for adaptively scheduling distribution network resources according to claim 6, characterized in that: The learning factor is used to control the learning step size and direction of the particle during the search process.

8. A distribution network resource adaptive scheduling system, characterized in that: The system comprises: A data acquisition module is used to collect distribution network operation status data and pre-process the distribution network operation status data to obtain an initial distribution network data set; A model construction module is used to construct a proxy model that characterizes the nonlinear coupling relationship between the real-time parameters of the new power elements and the initial data set of the distribution network based on the initial data set of the distribution network and the real-time parameters of the new power elements using a support vector machine algorithm; A variable decomposition module is used to decompose the multivariate interaction effects in the agent model using a global sensitivity analysis method, and quantify the sensitivity index of each new power element to the initial data set of the distribution network; a space optimization module, configured to adaptively modify the sampling space of the proxy model using an iterative update strategy according to the sensitivity index to obtain an optimized proxy model; A strategy generation module is used to generate a distribution network resource adaptive scheduling strategy using a particle swarm optimization algorithm based on the optimization agent model and real-time monitoring data of the distribution network; The optimization configuration module is used to optimize the configuration of distribution network resources in the new power element access scenario according to the distribution network resource adaptive scheduling strategy.

9. A computer device, characterized in that: The computer device comprises a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the computer device performs the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images