Technological virtual power plant voltage prediction method and system based on high-temperature scenario generation

Abstract

The application discloses a technical virtual power plant voltage prediction method and system based on a high-temperature scene, relates to the technical field of power system voltage safety and virtual power plant operation control, and establishes a main grid, distribution network and transformer area multi-level voltage coupling model, systematically quantifies a bidirectional transmission relationship of main grid voltage support, distribution network power flow distribution and transformer area load fluctuation, solves the problem of insufficient mechanism description caused by only considering a single level in the prior art, and significantly improves the physical consistency of voltage prediction under extreme high temperature; and an index system associated with temperature and voltage and a self-adaptive SMOTE oversampling algorithm are proposed, a differentiated sample enhancement strategy is used to effectively solve the data imbalance problem caused by the scarcity of extreme high-temperature scene samples, the proportion of all types of samples is balanced, and the generalization ability of the prediction model to extreme scenes is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of power system voltage safety and virtual power plant operation control technology, specifically to a technical virtual power plant voltage prediction method and system based on high-temperature scenarios. Background Technology

[0002] In recent years, global climate change has led to frequent, intense, and prolonged extreme heat events in my country during the summer. In regions like Chongqing, summer temperatures have consistently exceeded 40°C. This, coupled with a rapid increase in air conditioning load, has caused the power grid's short-term load to repeatedly break historical records. For example, in Chongqing, the peak power load reached 30.196 million kilowatts as of August 5th. Under extreme high-temperature conditions, technology-based virtual power plants, as new power system units aggregating resources such as photovoltaics, energy storage, and controllable loads, face severe challenges to operational stability. High temperatures not only directly increase the demand for cooling loads within virtual power plants, exacerbating the imbalance between supply and demand, but also cause fluctuations in photovoltaic output and distribution network voltage. Furthermore, the output characteristics of new energy sources such as photovoltaics are significantly affected by both high temperatures and strong sunlight, further amplifying the voltage safety risks within technology-based virtual power plants.

[0003] Voltage safety is a core indicator for the stable operation of technical virtual power plants. Under extreme high-temperature scenarios, the voltage safety issues of technical virtual power plants exhibit complex characteristics of multi-level coupling and multi-factor driving: On the one hand, the voltage of the main grid hub connected to the virtual power plant constrains the starting point of the distribution network voltage through the distribution transformer, and the voltage fluctuations of the main grid caused by high temperature will be directly transmitted to the virtual power plant; on the other hand, the high proportion of photovoltaic modules in the virtual power plant are affected by high-temperature radiation, exhibiting fluctuating characteristics of sudden rises and falls in power. This fluctuation propagates along the radial feeders, which will not only cause voltage disturbances on the user side of the distribution area, but also affect the voltage support of the upstream main grid in the reverse direction; at the same time, the frequent start-up and shutdown of loads such as air conditioning in the distribution area under high temperature will form local voltage disturbances and be transmitted to the upstream network. In addition, the dynamic changes of line electrical parameters under high temperature environment further exacerbate the nonlinear characteristics of multi-level voltage coupling between the main grid, distribution network and distribution area in the virtual power plant.

[0004] Existing voltage prediction methods for technical virtual power plants largely rely on historical statistical patterns, local electrical quantity fitting, or single-level equivalent models, exhibiting two major drawbacks: First, they lack a detailed characterization of the multi-level voltage coupling mechanism between the main and distribution substations in technical virtual power plants, failing to systematically integrate the bidirectional transmission relationship between the main grid's support role, distribution network power flow distribution, and substation load fluctuations, resulting in insufficient voltage prediction accuracy under extreme high-temperature scenarios. Second, extreme high temperatures are typical low-probability, high-impact events, and the extreme scenario samples available for model training during the operation of technical virtual power plants are extremely scarce, leading to significant data imbalance. This makes existing data-driven prediction models prone to fitting normal scenarios, resulting in insufficient generalization ability and difficulty in accurately capturing voltage fluctuation patterns under extreme high temperatures, severely restricting the voltage safety management and operational resilience improvement of technical virtual power plants.

[0005] In summary, to address the core voltage safety requirements of technical virtual power plants under extreme high-temperature scenarios, it is urgent to research a voltage prediction method and system for technical virtual power plants generated based on high-temperature scenarios. By finely modeling multi-level voltage coupling mechanisms and solving the data imbalance problem caused by the scarcity of samples in extreme scenarios, accurate voltage prediction of technical virtual power plants can be achieved, providing core technical support for their safe and stable operation and control decisions. Summary of the Invention

[0006] The purpose of this invention is to provide a technical virtual power plant voltage prediction method and system based on high-temperature scenarios. By establishing a multi-level voltage coupling model of main and distribution stations, proposing an adaptive extreme high-temperature scenario generation technology and an integrated optimization prediction model, this invention solves the problems of insufficient characterization of multi-level coupling mechanisms and low prediction accuracy caused by the scarcity of extreme samples in existing technologies, and achieves accurate prediction of the voltage of technical virtual power plants under extreme high temperatures.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a technical virtual power plant voltage prediction method based on high-temperature scenarios, comprising the following steps:

[0008] S1: The random forest algorithm is used for feature correlation analysis. The importance of each feature is evaluated by shuffling the values ​​of each feature on the out-of-bag data and observing the degree of decline in the model's predictive performance. This evaluation process is based on out-of-bag data not used during training, which makes the feature selection objective and anti-overfitting. Finally, the key meteorological and electrical quantities closely related to the voltage fluctuation of the internal nodes of the virtual power plant under extreme high temperature are selected.

[0009] S2: Quantify the voltage transmission process from generator to distribution transformer bus and the supporting role of the main grid; establish a recursive model of feeder voltage in a virtual power plant for photovoltaic access and analyze its topology dependence characteristics; construct a simplified voltage model of user nodes in the distribution area based on radial wiring and stability parameter assumptions, so as to systematically analyze the mutual influence and transmission mechanism of node voltages at different levels under extreme high temperatures;

[0010] S3: Temperature is selected as the meteorological factor for extreme high temperature scenarios. In order to fully consider the coupling and correlation characteristics between temperature and the voltage of key nodes inside the virtual power plant and to achieve a more realistic temperature category classification, a temperature-voltage correlation index is proposed, and a clustering method is used to classify representative extreme high temperature scenario categories.

[0011] S4: After the dataset is divided into multiple categories of scenarios by combining temperature and voltage correlation index with clustering algorithm, the distribution of each category in the dataset is measured, the sample proportion is calculated, and the sample distribution is correlated by normalized weight coefficient. An adaptive SMOTE oversampling calculation formula that is inversely proportional to the original sample quantity is established to effectively generate balanced extreme high temperature scenario data.

[0012] S5: Using the constructed dataset covering multi-level coupling features and enhanced samples as model input, a multi-layer perceptron and random forest ensemble prediction model (DOA-MLP-RF) optimized based on the jackal algorithm is adopted to achieve accurate prediction of the main and distribution station coordinated voltage in a virtual power plant under extreme high temperature.

[0013] Preferably, in S1, a random forest is used to assess the importance of features by observing the degree of performance degradation after shuffling the feature values. The formula for calculating the importance of the arrangement is as follows:

[0014] (1)

[0015] In equation (1), Features The importance score of the arrangement; is the total number of decision trees in the random forest; T is a single decision tree in the forest; The baseline prediction error of tree T on its out-of-bag (OOB) data; To feature The prediction error of tree T is calculated after the values ​​are randomly arranged in the OOB data.

[0016] Preferably, in S2, a multi-level voltage transmission model considering photovoltaic (PV) access is established based on physical mechanisms. This model includes the voltage transmission process from the generator to the distribution transformer bus, a recursive model of the feeder voltage considering PV access, and a simplified voltage model for the user node in the distribution area based on radial wiring and stability parameter assumptions.

[0017] (2)

[0018] (3)

[0019] (4)

[0020] (5)

[0021] (6)

[0022] (7)

[0023] Equations (2)-(4) are used to characterize the voltage support effect of the main grid on the distribution transformer side bus. This section models the voltage transmission process from the generator node to the target distribution transformer bus side. The generator node includes p traditional energy units and q new energy units, and its equivalent unit terminal voltage vector is denoted as... In the formula The target distribution transformer bus node voltage vector. Let be the voltage vector of the remaining transmission nodes. In steady-state analysis, it is usually assumed that there is no injected current at the target bus node and the remaining transmission nodes, i.e. and Equation (5) is the voltage at a certain node derived segment by segment from the distribution transformer bus voltage when the influence of distributed generation access is neglected in a radial distribution network. Let be the voltage amplitude at the m-th node; This refers to the voltage amplitude of the low-voltage side busbar of the distribution transformer. and These are the active power and reactive power of the k-th node, respectively. and Let n be the resistance and reactance of the i-th segment of the line, respectively; n be the total number of downstream nodes, and m be the node number under investigation (1≤m≤n); Equation (6) shows that after a distributed photovoltaic (PV) system is connected at node g (1≤g≤n), the power generated by the PV system will change the power flow distribution along the line. The voltage amplitude of node m after being connected to the distributed photovoltaic system; and Let g represent the active power and reactive power emitted by the photovoltaic system at node m (with the direction of injection into the distribution network as positive), where g ≥ m. Otherwise, it is 0; Equation (7) represents the voltage coupling relationship of the user nodes in the distribution area. For user voltage amplitude; , The measured net active and reactive power already include the influence of photovoltaic output. , The resistance and reactance per unit length of the line are given by ; L is the line length.

[0024] Preferably, in order to fully consider the coupling relationship between temperature and voltage and realize the classification of extreme high temperature scenarios that closely match the actual operating characteristics, S3 proposes a temperature and voltage correlation index including temperature index C1, temperature change index C2, voltage index V1, and voltage change index V2. Based on this index, a clustering method is used to classify representative extreme high temperature scenario categories.

[0025] (8)

[0026] (9)

[0027] Equation (8) is the correlation index between temperature and voltage, where, This refers to the daily temperature index. The average daily maximum temperature of the peak voltage; The temperature at hour h; This is the voltage data at this time; is the peak voltage threshold; Equation (9) is the voltage-temperature fusion feature vector of the i-th hourly sample under extreme high temperature scenario, and then the clustering method is used to realize the automatic division of extreme temperature range.

[0028] Preferably, in S4, the adaptive SMOTE sampling calculation formula established by the present invention correlates the category sample distribution through normalized weight coefficients. The weights are inversely proportional to the number of original samples. Specifically, the category with fewer original samples has a higher proportion of synthesized samples; conversely, the category with more original samples has a lower proportion of synthesized samples, thereby achieving differential sample enhancement.

[0029] (10)

[0030] (11)

[0031] (12)

[0032] Equation (10) measures the distribution of each category in the dataset and calculates its sample proportion, where R k Let $k$ be the proportion of the k-th class samples in the original dataset. To give smaller sample classes higher boosting priority, the initial weights are designed to be the reciprocal of the sample proportion. Equation (11) applies to the initial weights Perform maximum value normalization to obtain the final normalized sampling weights for each category, and the normalized sampling weights W. k The value range is (0, 1). The proportion of the original sample R... k The smaller the category, the higher its weight W. kThe closer a value is to 1, the higher the proportion of synthetic samples retained in subsequent sampling. Let the preset sample generation factor be n. Ultimately, the number of synthetic samples from class k that are retained and added to the augmented dataset is... Due to its normalized sampling weight W k The number of candidates is determined together, as shown in Equation (12).

[0033] Preferably, S5 constructs a multilayer perceptron and random forest ensemble prediction model (DOA-MLP-RF) optimized by the jackal algorithm. First, the entire dataset is divided into training, validation, and test sets in an 8:1:1 ratio. Then, the hyperparameter optimization process is initialized for both the MLP and RF models, and core parameters such as the population size and maximum number of iterations of DOA are configured simultaneously. Next, DOA automatically searches for the optimal hyperparameter combinations for both MLP and RF, aiming to minimize the validation set error, and records the validation set performance during this process. Finally, the optimal hyperparameter-optimized model M is obtained for each model. MLP With M RF .

[0034] (13)

[0035] (14)

[0036] (15)

[0037] (16)

[0038] (17)

[0039] Equation (13) minimizes the validation set error, where the characteristics of the validation set are assumed to be... The actual value is , as well as These are the model predictions obtained after validation. , The errors of MLP and RF are used; the weights for combining are determined by the errors of the validation set, as shown in Equation (14); a threshold is defined. If equation (15) is satisfied, that is, the ratio of the relative differences between the two algorithms is greater than the threshold, then they are not combined; if ,but , Conversely, , Equation (16) represents the final model prediction result, where the predicted values ​​in the test set are respectively as well as The prediction result obtained after model integration is: Equation (17) is the evaluation index for the prediction results, where S is the sample size. This is the c-th true value; This is the c-th predicted value; This is the average of the true values.

[0040] A technical virtual power plant voltage prediction system based on high-temperature scenarios includes: a system startup module, a feature selection module, a voltage mechanism analysis module, an extreme high-temperature scenario division module, a data equalization module, an integrated prediction module, a prediction result output module, and a prediction accuracy verification module.

[0041] The system startup module is used to initialize the system operating environment, load technical virtual power plant operation data, extreme high temperature meteorological data, historical voltage monitoring data and model parameters, and build the basic environment for analysis.

[0042] The feature screening module is used to evaluate the importance of features and screen key meteorological and electrical quantities that are closely related to the voltage fluctuations of internal nodes in a technical virtual power plant under extreme high temperatures.

[0043] The voltage mechanism analysis module is used to quantify the supporting role of the main grid and analyze the mutual influence and transmission mechanism of voltage at different levels of nodes in a technical virtual power plant under extreme high temperatures.

[0044] The extreme high temperature scenario classification module is used to classify representative extreme high temperature scenario categories by combining the coupling relationship between temperature and voltage.

[0045] The data equalization module is used to generate an equalized dataset of extreme high-temperature scenarios based on the distribution of scenario samples.

[0046] The integrated prediction module is used to accurately predict the main and distribution station coordinated voltage in a technical virtual power plant under extreme high temperatures, based on multi-level coupling features and enhanced samples.

[0047] The prediction result output module is used to visualize the output voltage prediction results and scenario analysis reports, providing decision support for virtual power plant scheduling and voltage control.

[0048] The prediction accuracy verification module is used to verify the prediction performance of the proposed model using multi-dimensional error indicators and to evaluate the prediction accuracy advantage under extreme high temperature scenarios.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] 1. This invention establishes a multi-level voltage coupling model of the main grid, distribution network and transformer substations, systematically quantifies the bidirectional transmission relationship between the main grid voltage support, distribution network power flow distribution and transformer substation load fluctuation, solves the problem of insufficient mechanism characterization caused by the existing technology only considering a single level, and significantly improves the physical consistency of voltage prediction under extreme high temperature.

[0051] 2. This invention proposes a temperature and voltage correlation index system and an adaptive SMOTE oversampling algorithm. Through a differentiated sample enhancement strategy, it effectively solves the data imbalance problem caused by the scarcity of samples in extreme high-temperature scenarios, making the proportion of samples of each category balanced and significantly improving the generalization ability of the prediction model to extreme scenarios.

[0052] 3. This invention constructs a DOA-MLP-RF ensemble prediction model based on the Jackal algorithm optimization. Through adaptive hyperparameter optimization and dynamic weighted ensemble, it fully leverages the nonlinear fitting capability of MLP and the anti-overfitting advantage of RF.

[0053] 4. This invention also provides a complete prediction system architecture, realizing full-process automation from data loading, feature processing, scene generation to prediction output and accuracy verification. It can be directly applied to the voltage safety management system of technical virtual power plants, providing technical support for grid collaborative scheduling under extreme weather conditions. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of the present invention;

[0056] Figure 2 This is a voltage feature optimization analysis diagram of a technical virtual power plant based on feature importance and prediction performance according to the present invention;

[0057] Figure 3 This invention analyzes the multi-level voltage coupling relationship diagram of the source-grid-load-storage virtual power plant.

[0058] Figure 4 This is a schematic diagram illustrating the adaptive SMOTE principle of the present invention for technology-based virtual power plants;

[0059] Figure 5 This is a structural diagram of the technical virtual power plant DOA-MLP-RF collaborative voltage prediction model proposed in this invention;

[0060] Figure 6This is a node diagram of the control area of ​​a virtual power plant in Chongqing in an embodiment of the present invention;

[0061] Figure 7 This is a comparison chart of voltage prediction error indicators for technical virtual power plants under different operating scenarios in this invention embodiment;

[0062] Figure 8 This is a comparison diagram of sample distribution before and after adaptive SMOTE oversampling in an embodiment of the present invention;

[0063] Figure 9 This is a comparison chart of the joint distribution of the voltage prediction value and the actual value proposed in the embodiments of the present invention;

[0064] Figure 10 This is a block diagram of a technical virtual power plant voltage prediction system based on high-temperature scenarios according to the present invention. Detailed Implementation

[0065] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0066] This invention employs the random forest algorithm for feature correlation analysis to screen key meteorological and electrical quantities closely related to voltage fluctuations under extreme high temperatures, systematically quantifying the multi-level voltage coupling relationship within a technical virtual power plant (transmission / maintenance / distribution / station). Considering the dynamic coupling characteristics of temperature and voltage, a temperature-voltage correlation index system is constructed, and representative extreme high-temperature scenario categories are divided using clustering methods. Addressing the uneven distribution of samples across categories, a normalized weighting coefficient is used to correlate sample distribution, employing an adaptive SMOTE oversampling calculation formula inversely proportional to the original sample size to generate balanced extreme high-temperature scenario data. Finally, using a dataset encompassing multi-level coupling characteristics and enhanced samples as input, a multilayer perceptron and random forest ensemble prediction model (DOA-MLP-RF) optimized based on the jackal algorithm is used to achieve accurate voltage prediction. The aforementioned prediction algorithm and functional modules are systematically encapsulated to construct a complete prediction system, providing technical reference for voltage safety support and optimized control of technical virtual power plants.

[0067] Specifically as follows:

[0068] Example 1:

[0069] Please see Figure 1 A technically-based virtual power plant voltage prediction method based on high-temperature scenarios includes the following:

[0070] S1: The random forest algorithm is used for feature correlation analysis. The importance of each feature is evaluated by shuffling the values ​​of each feature on the out-of-bag data and observing the degree of decline in the model's predictive performance. This evaluation process is based on out-of-bag data not used during training, which makes the feature selection objective and anti-overfitting. Finally, the key meteorological and electrical quantities closely related to the voltage fluctuation of the internal nodes of the virtual power plant under extreme high temperature are selected.

[0071] S2: Quantify the voltage transmission process from generator to distribution transformer bus and the supporting role of the main grid; establish a recursive model of feeder voltage in a virtual power plant for photovoltaic access and analyze its topology dependence characteristics; construct a simplified voltage model of user nodes in the distribution area based on radial wiring and stability parameter assumptions, so as to systematically analyze the mutual influence and transmission mechanism of node voltages at different levels under extreme high temperatures;

[0072] S3: Temperature is selected as the meteorological factor for extreme high temperature scenarios. In order to fully consider the coupling and correlation characteristics between temperature and the voltage of key nodes inside the virtual power plant and to achieve a more realistic temperature category classification, a temperature-voltage correlation index is proposed, and a clustering method is used to classify representative extreme high temperature scenario categories.

[0073] S4: After the dataset is divided into multiple categories of scenarios by combining temperature and voltage correlation index with clustering algorithm, the distribution of each category in the dataset is measured, the sample proportion is calculated, and the sample distribution is correlated by normalized weight coefficient. An adaptive SMOTE oversampling calculation formula that is inversely proportional to the original sample quantity is established to effectively generate balanced extreme high temperature scenario data.

[0074] S5: Using the constructed dataset covering multi-level coupling features and enhanced samples as model input, a multi-layer perceptron and random forest ensemble prediction model (DOA-MLP-RF) optimized based on the jackal algorithm is adopted to achieve accurate prediction of the main and distribution station coordinated voltage in a virtual power plant under extreme high temperature.

[0075] In S1, a random forest is used to assess importance by observing the degree of performance degradation of the model after shuffling the feature values. In the random forest, features... The formula for calculating the importance of the arrangement is as follows:

[0076] (1)

[0077] In equation (1), Features The importance score of the arrangement; is the total number of decision trees in the random forest; T is a single decision tree in the forest; The baseline prediction error of tree T on its out-of-bag (OOB) data; To feature The prediction error of tree T is calculated after the values ​​are randomly arranged in the OOB data.

[0078] S2 establishes a multi-level voltage transmission model considering photovoltaic (PV) integration, starting from the physical mechanism. This model covers the voltage transmission process from the generator to the distribution transformer bus, a recursive model of the feeder voltage considering PV integration, and a simplified voltage model for the user node in the distribution area based on radial wiring and stability parameter assumptions.

[0079] (2)

[0080] (3)

[0081] (4)

[0082] (5)

[0083] (6)

[0084] (7)

[0085] Equations (2)-(4) are used to characterize the voltage support effect of the main grid on the distribution transformer side bus. This section models the voltage transmission process from the generator node to the target distribution transformer bus side. The generator node includes p traditional energy units and q new energy units, and its equivalent unit terminal voltage vector is denoted as... In the formula The target distribution transformer bus node voltage vector. Let be the voltage vector of the remaining transmission nodes. In steady-state analysis, it is usually assumed that there is no injected current at the target bus node and the remaining transmission nodes, i.e. and Equation (5) is the voltage at a certain node derived segment by segment from the distribution transformer bus voltage when the influence of distributed generation access is neglected in a radial distribution network. Let be the voltage amplitude at the m-th node; This refers to the voltage amplitude of the low-voltage side busbar of the distribution transformer. and These are the active power and reactive power of the k-th node, respectively. and Let n be the resistance and reactance of the i-th segment of the line, respectively; n be the total number of downstream nodes, and m be the node number under investigation (1≤m≤n); Equation (6) shows that after a distributed photovoltaic (PV) system is connected at node g (1≤g≤n), the power generated by the PV system will change the power flow distribution along the line. The voltage amplitude of node m after being connected to the distributed photovoltaic system; and Let g represent the active power and reactive power emitted by the photovoltaic system at node m (with the direction of injection into the distribution network as positive), where g ≥ m. Otherwise, it is 0; Equation (7) represents the voltage coupling relationship of the user nodes in the distribution area. For user voltage amplitude; , The measured net active and reactive power already include the influence of photovoltaic output. , The resistance and reactance per unit length of the line are given by ; L is the line length.

[0086] In S3, to fully consider the coupling relationship between temperature and voltage and to achieve the classification of extreme high-temperature scenarios that closely match actual operating characteristics, a temperature-voltage correlation index is proposed, which includes temperature index C1, temperature change index C2, voltage index V1, and voltage change index V2. Based on this index, a clustering method is used to classify representative extreme high-temperature scenario categories.

[0087] (8)

[0088] (9)

[0089] Equation (8) is the correlation index between temperature and voltage, where, This refers to the daily temperature index. The average daily maximum temperature of the peak voltage; The temperature at hour h; This is the voltage data at this time; is the peak voltage threshold; Equation (9) is the voltage-temperature fusion feature vector of the i-th hourly sample under extreme high temperature scenario, and then the clustering method is used to realize the automatic division of extreme temperature range.

[0090] In S4, the present invention establishes an adaptive SMOTE sampling calculation formula. This formula associates the category sample distribution with normalized weight coefficients. The weights are inversely proportional to the number of original samples. Specifically, the category with fewer original samples has a higher proportion of synthesized samples, and vice versa, thus achieving differential sample enhancement.

[0091] (10)

[0092] (11)

[0093] (12)

[0094] Equation (10) measures the distribution of each category in the dataset and calculates its sample proportion, where R kLet $k$ be the proportion of the k-th class samples in the original dataset. To give smaller sample classes higher boosting priority, the initial weights are designed to be the reciprocal of the sample proportion. Equation (11) applies to the initial weights Perform maximum value normalization to obtain the final normalized sampling weights for each category, and the normalized sampling weights W. k The value range is (0, 1). The proportion of the original sample R... k The smaller the category, the higher its weight W. k The closer a value is to 1, the higher the proportion of synthetic samples retained in subsequent sampling. Let the preset sample generation factor be n. Ultimately, the number of synthetic samples from class k that are retained and added to the augmented dataset is... Due to its normalized sampling weight W k The number of candidates is determined together, as shown in Equation (12).

[0095] S5 constructs an ensemble prediction model of multilayer perceptron and random forest (DOA-MLP-RF) optimized by the Jackal algorithm. First, the entire dataset is divided into training, validation, and test sets in an 8:1:1 ratio. Then, the hyperparameter optimization process is initialized for both the MLP and RF models, while simultaneously configuring core parameters of DOA such as population size and maximum number of iterations. Next, DOA automatically searches for the optimal hyperparameter combinations for both MLP and RF, aiming to minimize the validation set error, and records the validation set performance during this process. Finally, the optimal hyperparameter-optimized model M is obtained for each model. MLP With M RF .

[0096] (13)

[0097] (14)

[0098] (15)

[0099] (16)

[0100] (17)

[0101] Equation (13) minimizes the validation set error, where the characteristics of the validation set are assumed to be... The actual value is , as well as These are the model predictions obtained after validation. , The errors of MLP and RF are used; the weights for combining are determined by the errors of the validation set, as shown in Equation (14); a threshold is defined. If equation (15) is satisfied, that is, the ratio of the relative differences between the two algorithms is greater than the threshold, then they are not combined; if ,but , Conversely, , Equation (16) represents the final model prediction result, where the predicted values ​​in the test set are respectively as well as The prediction result obtained after model integration is: Equation (17) is the evaluation index for the prediction results, where S is the sample size. This is the c-th true value; This is the c-th predicted value; This is the average of the true values.

[0102] See Figure 2 As shown, the left axis represents feature importance, and the right axis represents model prediction accuracy, with the coefficient of determination (R²) used as the evaluation index. The performance change trends indicate that distribution transformer voltage and bus voltage are the most critical parameters contributing to model performance. Among meteorological features, temperature is the most important, indicating that it is the meteorological factor most strongly correlated with user voltage data.

[0103] See Figure 3 As shown, the voltage transmission process from the generator to the distribution transformer bus—including the feeder voltage recursive model considering photovoltaic access—is constructed based on the radial connection and stability parameter assumptions to build a simplified voltage model of the user node in the distribution area, thereby systematically analyzing the mutual influence and transmission mechanism of node voltages at different levels under extreme high temperatures.

[0104] See Figure 4 As shown, the adaptive SMOTE method proposed in this invention differs from the traditional SMOTE method. The traditional SMOTE method only samples and generates data for classes with fewer samples. However, the adaptive SMOTE method proposed in this invention takes into account both the augmentation of small sample data and the balance of the overall data distribution. The smaller the original sample size of a class, the higher the proportion of synthesized samples; conversely, the larger the original sample size, the lower the proportion of synthesized samples, thereby achieving differentiated sample enhancement.

[0105] See Figure 5 As shown, the Jackal optimization algorithm is introduced to adaptively optimize the ensemble weights of the multilayer perceptron and random forest based on its efficient search mechanism, and finally an adaptive weighted ensemble prediction model, namely the DOA-MLP-RF model, is constructed.

[0106] Example 2:

[0107] This embodiment proposes a specific application based on the above embodiment one;

[0108] Topology information reference Figure 6 As shown, the data used in August 2024 for a virtual power plant control area in Chongqing, China, including regional meteorological data, transmission network bus voltage, pole-mounted transformer voltage, distribution network end-user voltage, and active and reactive power of transformers and users, including extreme high temperature data, are from the region.

[0109] pass Figure 1 The distribution network voltage prediction process based on extreme high temperature scenarios under the coordination of main and distribution stations is proposed. Starting from the physical mechanism, a multi-level voltage transmission model considering photovoltaic access is established, and the necessity of main and distribution station coordinated prediction is demonstrated on this basis. Figure 7 The progressively decreasing error histogram visually illustrates this gradual performance improvement pattern, fully revealing the cumulative enhancement effect of information collaboration. This indicates that distribution network-level information mainly contributes to the coupling interaction of loads in the modeling area, while transmission network-level information accurately depicts the fluctuations of the system voltage reference and its transmission process to downstream networks.

[0110] Figure 8 The paper presents a comparison of sample distributions for each scenario category after oversampling the total data. Experimental results show that the proposed adaptive SMOTE oversampling strategy effectively solves the sample imbalance problem in extreme high-temperature scenarios. On the one hand, this strategy specifically enhances the categories with scarce samples, significantly alleviating the small sample dilemma. On the other hand, after oversampling, the sample size of each category tends to be balanced, with their sample proportions stabilizing between 15% and 20%. This provides a balanced and reliable data foundation for subsequent construction of prediction models for extreme high-temperature scenarios.

[0111] Figure 9 A direct comparison was made between the joint distribution of the voltage predicted by the proposed model and the measured values. The voltage data exhibited the following distribution patterns and fluctuation characteristics: the voltage for most users was concentrated between 215-220V, while some data, due to factors such as increased ambient temperature and enhanced photovoltaic output, were dispersed to around 270V. The training and test set data points closely surrounded the fitted line, and the distribution patterns of the predicted and actual values ​​showed a high degree of agreement. This indicates that the proposed method effectively improves the prediction accuracy and stability of distribution network voltage under extreme high-temperature scenarios.

[0112] Example 3:

[0113] See Figure 10 This embodiment proposes a technical virtual power plant voltage prediction system based on high-temperature scenarios, based on the above embodiment one.

[0114] The technical virtual power plant voltage prediction system based on high-temperature scenarios is modularly designed.

[0115] Specifically, it includes: a startup module, a feature selection module, a voltage mechanism analysis module, an extreme high temperature scenario classification module, a data equalization module, an integrated prediction module, a prediction result output module, and a prediction accuracy verification module;

[0116] The system startup module is used to initialize the system operating environment, load technical virtual power plant operation data, extreme high temperature meteorological data, historical voltage monitoring data and model parameters, and build the basic analysis environment;

[0117] The feature filtering module is used to evaluate the importance of features and filter key meteorological and electrical quantities that are closely related to the voltage fluctuations of internal nodes in a technical virtual power plant under extreme high temperatures.

[0118] The voltage mechanism analysis module is used to quantify the supporting role of the main grid and analyze the mutual influence and transmission mechanism of voltage at different levels of nodes in a technical virtual power plant under extreme high temperatures.

[0119] The extreme high temperature scenario classification module is used to classify representative extreme high temperature scenario categories by combining the coupling relationship between temperature and voltage.

[0120] The data equalization module is used to generate an equalized dataset of extreme high-temperature scenarios based on the distribution of scenario samples.

[0121] The integrated prediction module is used to accurately predict the main and distribution station coordinated voltage in a technical virtual power plant under extreme high temperatures, based on multi-level coupling features and enhanced samples.

[0122] The prediction result output module is used to visualize the output voltage prediction results and scenario analysis reports, providing decision support for virtual power plant scheduling and voltage control.

[0123] The prediction accuracy verification module is used to verify the prediction performance of the proposed model using multi-dimensional error indicators and to evaluate the prediction accuracy advantage under extreme high temperature scenarios.

[0124] In summary:

[0125] This invention first employs the random forest algorithm to analyze feature correlations, screening key meteorological and virtual power plant electrical quantities closely related to voltage fluctuations. It systematically establishes and quantifies a multi-level voltage coupling model from the transmission and distribution network, the technical virtual power plant, to the user side. Based on this, a temperature-voltage correlation index system is proposed, and clustering methods are used to classify representative extreme high-temperature scenario categories. Addressing the issue of uneven sample distribution, a normalized weight coefficient is used to correlate sample distribution, establishing an adaptive SMOTE oversampling calculation formula inversely proportional to the original sample size, effectively generating balanced extreme high-temperature scenario data. Finally, using the constructed dataset encompassing multi-level coupling features and enhanced samples as input, an integrated prediction model of a multilayer perceptron and random forest based on the jackal algorithm is employed to achieve accurate voltage prediction of the technical virtual power plant under extreme high temperatures. This objectively reflects the multi-level coupling state of the virtual power plant voltage under extreme high-temperature conditions. Especially in extremely hot scenarios with scarce samples, it avoids the decrease in prediction accuracy caused by ignoring coupling characteristics and data imbalance, improving the voltage control capability, grid-connected operation stability, and grid collaborative dispatch security of the virtual power plant under extreme climatic conditions.

[0126] This addresses the dual challenges of existing technology-based virtual power plant voltage prediction methods failing to adequately consider the multi-level coupling mechanism of source-grid-load-storage systems and the scarcity of samples in extreme high-temperature scenarios.

[0127] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A technical virtual power plant voltage prediction method based on high-temperature scenario generation, characterized by: Includes the following steps: S1: The random forest algorithm is used to perform feature correlation analysis. The importance of features is evaluated by shuffling the values ​​of each feature in the out-of-bag data and observing the degree of decline in the model's prediction performance. Key meteorological and electrical quantities that are closely related to the voltage fluctuations of nodes inside the virtual power plant under extreme high temperatures are screened out. S2: Starting from the physical mechanism, a multi-level voltage transmission model is established. The main grid support role is quantified through the voltage transmission process from the generator to the distribution transformer bus. A feeder voltage recursive model considering photovoltaic access is established to analyze the topology dependence characteristics. Based on the radial connection and stability parameter assumptions, a simplified voltage model of the user node in the distribution area is constructed. The mutual influence and transmission mechanism of the voltage of different levels of nodes under extreme high temperature are analyzed. S3: A temperature-voltage correlation index system is proposed, which includes temperature index, temperature change index, voltage index and voltage change index. Based on this index, a clustering algorithm is used to classify representative extreme high temperature scenario categories. S4: Calculate the sample proportion of each category of scene, associate the sample distribution through normalized weight coefficients, establish an adaptive SMOTE oversampling calculation formula that is inversely proportional to the original sample quantity, and generate a balanced extreme high temperature scene dataset. S5: The dataset, which includes multi-level coupling features and enhanced samples, is divided into training, validation, and test sets. The Jackal algorithm is used to optimize the hyperparameters of the multilayer perceptron and random forest respectively. The ensemble weights are adaptively determined based on the validation set error, and the DOA-MLP-RF ensemble prediction model is constructed to achieve accurate prediction of the main and distribution station coordinated voltage of the virtual power plant under extreme high temperature.

2. The technical virtual power plant voltage prediction method based on high-temperature scenario generation according to claim 1, characterized in that: The random forest algorithm in S1 assesses importance by observing the degree of performance degradation of the model after shuffling feature values. In the random forest, features... The formula for calculating the importance of the arrangement is as follows: （1） in, Features The importance score of the arrangement; is the total number of decision trees in the random forest; T is a single decision tree in the forest; The baseline prediction error of tree T on its out-of-bag data; To feature The prediction error of tree T is calculated after the values ​​are randomly arranged in the OOB data.

3. The technical virtual power plant voltage prediction method based on high-temperature scenario generation according to claim 2, characterized in that: The multi-level voltage transmission model in S2 starts from the physical mechanism, considers photovoltaic access, and describes the voltage transmission process from the generator to the distribution transformer bus - a feeder voltage recursive model considering photovoltaic access - and a simplified voltage model for user nodes in the distribution area based on radial wiring and stability parameter assumptions. （2） （3） （4） （5） （6） （7） Equations (2)-(4) are used to model the voltage transmission process from the generator node to the target distribution transformer bus to characterize the voltage support effect of the main grid on the distribution transformer side bus. Here, the generator node contains p traditional energy units and q new energy units, and its equivalent unit-end voltage vector is denoted as... ; The target distribution transformer bus node voltage vector; Let be the voltage vector of the remaining transmission nodes; in steady-state analysis, it is usually assumed that there is no injected current at the target bus node and the remaining transmission nodes, i.e. and ; Equation (5) is the voltage at a certain node derived segment by segment from the distribution transformer bus voltage when the influence of distributed generation access is neglected in a radial distribution network, where: Let be the voltage amplitude at the m-th node; This refers to the voltage amplitude of the low-voltage side busbar of the distribution transformer. and These are the active power and reactive power of the k-th node, respectively. and These are the resistance and reactance values ​​of the i-th segment of the line, respectively; n is the total number of downstream nodes, and m is the node number under investigation. Equation (6) shows that after a distributed photovoltaic system is connected at node g, the power generated by the photovoltaic system will change the power flow distribution along the line, where: The voltage amplitude of node m after being connected to the distributed photovoltaic system; and Let g represent the active power and reactive power emitted by the photovoltaic system at node m (with the direction of injection into the distribution network as positive), where g ≥ m. Otherwise, it is 0; Equation (7) represents the voltage coupling relationship of the user nodes in the distribution area, where: For user voltage amplitude; , These are the measured net active power and the measured net reactive power, respectively, which already include the impact of photovoltaic power output; , These represent the resistance per unit length of the line and the reactance per unit length of the line, respectively; L is the line length.

4. The technical virtual power plant voltage prediction method based on high-temperature scenario generation according to claim 3, characterized in that: The S3 fully considers the coupling relationship between temperature and voltage and realizes the division of extreme high temperature scenarios that closely match the actual operating characteristics. Furthermore, a temperature-voltage correlation index is proposed, which includes temperature index C1, temperature change index C2, voltage index V1, and voltage change index V2. Based on this index, a clustering method is used to classify representative extreme high-temperature scenarios. （8） （9） Equation (8) is the correlation index between temperature and voltage, where, This refers to the daily temperature index. The average daily maximum temperature of the peak voltage; The temperature at hour h; This is the voltage data at this time; Peak voltage threshold; Equation (9) is the voltage-temperature fusion feature vector of the i-th hourly sample under extreme high temperature scenario. Subsequently, clustering method is used to realize the automatic division of extreme temperature range.

5. The technical virtual power plant voltage prediction method based on high-temperature scenario generation according to claim 4, characterized in that: S4 establishes an adaptive SMOTE sampling calculation formula. The adaptive SMOTE sampling calculation formula associates the category sample distribution with normalized weight coefficients. The weights are inversely proportional to the original number of samples. The smaller the original sample size of a category, the higher the proportion of synthesized samples; Conversely, the lower the proportion of synthetic samples, the better the enhancement of differentiated samples. （10） （11） （12） Equation (10) measures the distribution of each category in the dataset and calculates its sample proportion, where R k Let $k$ be the proportion of the k-th class samples in the original dataset. To give smaller sample classes higher boosting priority, the initial weights are designed to be the reciprocal of the sample proportion. ; Equation (11) applies to the initial weights Perform maximum value normalization to obtain the final normalized sampling weights for each category, and the normalized sampling weights W. k The range of its value is (0, 1]; Original sample proportion R k The smaller the category, the weight W k The closer to 1, that is, the higher the proportion of its synthetic samples retained in subsequent sampling; Let the preset sample generation factor be n. Finally, what is the number of synthetic samples from class k that are retained and added to the augmented dataset? Due to its normalized sampling weight W k The number of candidates is determined together, as shown in equation (12).

6. The technical virtual power plant voltage prediction method based on high-temperature scenario generation according to claim 5, characterized in that: The DOA-MLP-RF ensemble prediction model proposed in S5 is an ensemble prediction model of multilayer perceptron and random forest optimized by the jackal algorithm. The S5 includes at least the following steps: First, the entire dataset is divided into training, validation, and test sets in a ratio of 8:1:1; Subsequently, the hyperparameter optimization process was initialized for both the MLP and RF models, and core parameters such as the population size and maximum number of iterations of the DOA were configured simultaneously. Then, the DOA is used to automatically search the best hyperparameter combination for the MLP and RF respectively with the objective of minimizing the validation set error, and record the validation set performance in this process, and finally obtain the best model M of hyperparameter optimization respectively MLP With M RF; （13） （14） （15） （16） （17） Equation (13) minimizes the validation set error, where: let the characteristics of the validation set be... The actual value is , as well as These are the model predictions obtained after validation. , For the errors of MLP and RF; The weights of the combination are determined by the error of the validation set, as shown in Equation (14); Define a threshold When equation (15) is satisfied, that is, the ratio of the relative differences between the two algorithms is greater than the threshold, they are not combined; when ,but , Conversely, , ; Equation (16) represents the final model prediction result, where the predicted values ​​in the test set are respectively as well as The prediction result obtained after model integration is: ; Equation (17) is the evaluation index for the prediction results, where S is the sample size. This is the c-th true value; This is the c-th predicted value; This is the average of the true values.

7. A technical virtual power plant voltage prediction system based on high-temperature scenarios, used to execute the technical virtual power plant voltage prediction method based on high-temperature scenarios as described in any one of claims 1-6, characterized in that: It includes a system startup module, a feature filtering module, a voltage mechanism analysis module, an extreme high temperature scenario classification module, a data equalization module, an integrated prediction module, a prediction result output module, and a prediction accuracy verification module; The system startup module is used to initialize the system operating environment, load technical virtual power plant operation data, extreme high temperature meteorological data, historical voltage monitoring data and model parameters, and build the basic environment for analysis. The feature screening module is used to evaluate the importance of features and screen key meteorological and electrical quantities that are closely related to the voltage fluctuations of internal nodes in a technical virtual power plant under extreme high temperatures. The voltage mechanism analysis module is used to quantify the supporting role of the main grid and analyze the mutual influence and transmission mechanism of voltage at different levels of nodes in a technical virtual power plant under extreme high temperatures. The extreme high temperature scenario classification module is used to classify representative extreme high temperature scenario categories by combining the coupling relationship between temperature and voltage. The data equalization module is used to generate an equalized dataset of extreme high-temperature scenarios based on the distribution of scenario samples. The integrated prediction module is used to accurately predict the main and distribution station coordinated voltage in a technical virtual power plant under extreme high temperatures, based on multi-level coupling features and enhanced samples. The prediction result output module is used to visualize the output voltage prediction results and scenario analysis reports, providing decision support for virtual power plant scheduling and voltage control. The prediction accuracy verification module is used to verify the prediction performance of the proposed model using multi-dimensional error indicators and to evaluate the prediction accuracy advantage under extreme high temperature scenarios.

Images