Statistical analysis-based power time sequence prediction model evaluation method and device, equipment, medium and product

Abstract

The invention relates to a statistical analysis-based power time sequence prediction model evaluation method and device, equipment, a medium and a product, and relates to the technical field of power grids. The method comprises the following steps: acquiring a prediction result and a prediction error output by a power time sequence prediction model after performing power prediction on a target power grid, and an operation constraint condition of the target power grid; performing feasibility evaluation on the prediction result based on the operation constraint condition to obtain application feasibility of the prediction result in the target power grid; identifying an operation scene category of the target power grid under the prediction result; respectively determining the evaluation weight of the prediction error and the evaluation weight of the application feasibility under the operation scene category; and according to each evaluation weight, carrying out weighted fusion on the prediction error and the application feasibility to obtain an evaluation result of the power time sequence prediction model. By adopting the method, the evaluation accuracy of the power time sequence prediction model can be improved.

Description

Technical Field

[0001] This application relates to the field of power grid technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for evaluating power time-series prediction models based on statistical analysis. Background Technology

[0002] With the large-scale grid connection of new energy sources and the increasing complexity of load characteristics, the power grid system has placed higher demands on the ability to accurately predict key power parameters such as load and new energy output. As a result, power time-series prediction models have become an important tool to support the safe and efficient operation of the system.

[0003] In traditional technologies, the evaluation of power time-series forecasting models is mostly limited to using conventional statistical indicators such as mean absolute error and root mean square error. While this evaluation method can reflect the numerical closeness between predicted and actual values, it fails to fully consider the inherent physical operating constraints of the power grid, resulting in inaccurate model evaluation results. Summary of the Invention

[0004] Therefore, it is necessary to provide a statistical analysis-based method, apparatus, computer equipment, computer-readable storage medium, and computer program product for evaluating power time-series forecasting models, which can improve the accuracy of power time-series forecasting model evaluation, in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for evaluating power time-series forecasting models based on statistical analysis, including:

[0006] Obtain the prediction results, prediction errors, and operating constraints of the target power grid output by the power time series prediction model after performing power prediction on the target power grid;

[0007] Based on operational constraints, a feasibility assessment is conducted on the prediction results to determine their application feasibility in the target power grid.

[0008] Identify the operational scenario category of the target power grid under the predicted results;

[0009] Determine the evaluation weights of prediction error and application feasibility under each operational scenario category;

[0010] Based on the evaluation weights, the prediction error and application feasibility are weighted and integrated to obtain the evaluation results of the power time series prediction model.

[0011] In one embodiment, the feasibility of application includes a safety margin index value used to characterize the safe operating level of the target power grid under the prediction results;

[0012] Based on operational constraints, a feasibility assessment is conducted on the prediction results to determine their application feasibility in the target power grid, including:

[0013] Extract safe operating constraints from operating constraints;

[0014] The safety operation level of the target power grid under the prediction results is evaluated based on the safety operation constraints, and the safety margin index value corresponding to the prediction results is obtained.

[0015] In one embodiment, the feasibility of application includes an energy acceptance index value used to characterize the energy acceptance level of the target power grid under the predicted results;

[0016] Based on operational constraints, a feasibility assessment is conducted on the prediction results to determine their application feasibility in the target power grid, including:

[0017] Extract energy acceptance constraints from operational constraints;

[0018] The energy acceptance level of the target power grid under the predicted results is evaluated based on the energy acceptance constraints, and the corresponding energy acceptance index value is obtained.

[0019] In one embodiment, identifying the target power grid's operating scenario category under the predicted results includes:

[0020] The prediction result is mapped to the feature space to obtain the mapping position of the prediction result in the feature space.

[0021] Determine the mapping region in the feature space corresponding to the power grid operation condition sample under at least one candidate operation scenario category;

[0022] Determine the distance between the mapping location and the mapping area;

[0023] If the distance within a candidate operating scenario category meets the proximity condition, the candidate operating scenario category is determined as the operating scenario category of the target power grid during the prediction process.

[0024] In one embodiment, the evaluation weights of prediction error and application feasibility are determined respectively under the operational scenario category, including:

[0025] With the goal of minimizing prediction error and maximizing the applicability of prediction results, evaluation weights are assigned under the category of operating scenario to obtain a weight allocation scheme.

[0026] The evaluation weights for prediction error and application feasibility are obtained from the weight allocation scheme.

[0027] In one embodiment, with the goal of minimizing prediction error and maximizing the applicability of prediction results, evaluation weights are assigned under the running scenario category to obtain a weight allocation scheme, including:

[0028] With the goal of minimizing prediction error and maximizing the applicability of prediction results, the weight allocation is evaluated to obtain an initial allocation scheme;

[0029] The distance between the mapping position of the prediction result in the feature space and the mapping region corresponding to the running scene category in the feature space is used as the weight adjustment factor for the running scene category.

[0030] The initial allocation scheme is adjusted based on the weight adjustment factor to obtain the adjusted weight allocation scheme.

[0031] Secondly, this application also provides a power time-series prediction model evaluation device based on statistical analysis, comprising:

[0032] The information acquisition module is used to acquire the prediction results, prediction errors, and operating constraints of the target power grid output by the power time series prediction model after making power predictions on the target power grid.

[0033] The feasibility assessment module is used to assess the feasibility of the prediction results based on operational constraints, and to obtain the feasibility of applying the prediction results in the target power grid.

[0034] The scene recognition module is used to identify the operating scene category of the target power grid under the prediction results;

[0035] The weight allocation module is used to determine the evaluation weights of prediction error and application feasibility under the running scenario category, respectively.

[0036] The model evaluation module is used to weight and fuse prediction errors and application feasibility according to each evaluation weight to obtain the evaluation results of the power time series prediction model.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-mentioned evaluation method for the power time-series prediction model based on statistical analysis.

[0038] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described evaluation method for the power time-series prediction model based on statistical analysis.

[0039] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described statistical analysis-based power time-series prediction model evaluation method.

[0040] The aforementioned statistical analysis-based evaluation method, apparatus, computer equipment, computer-readable storage medium, and computer program product for power time-series forecasting models first acquire the forecast results, forecast errors, and operational constraints of the target power grid output by the power time-series forecasting model after forecasting power in the target power grid. Based on these operational constraints, a feasibility assessment is performed on the forecast results to determine their application feasibility in the target power grid. Next, the operational scenario categories of the target power grid during the forecasting process are identified, and the evaluation weights of forecast error and application feasibility under each operational scenario category are determined. Finally, the forecast error and application feasibility are weighted and fused according to their respective evaluation weights to obtain the evaluation result of the power time-series forecasting model. Thus, this solution evaluates the application feasibility of the model's forecast results in a real power grid environment based on the power grid's operational constraints, and intelligently adjusts the evaluation focus according to the actual operational scenarios of the power grid, significantly improving the accuracy of the power time-series forecasting model evaluation. This provides a more comprehensive, reliable, and practically relevant decision-making basis for the selection, deployment, and optimization of the model in actual business operations. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is an application environment diagram of a power time-series prediction model evaluation method based on statistical analysis in one embodiment;

[0043] Figure 2 This is a flowchart illustrating an evaluation method for a power time-series prediction model based on statistical analysis in one embodiment.

[0044] Figure 3 This is a schematic diagram of an evaluation process based on safe operation constraints in one embodiment;

[0045] Figure 4 This is a schematic diagram of an evaluation process based on energy acceptance constraints in one embodiment;

[0046] Figure 5 This is a schematic diagram of the scene recognition process in one embodiment;

[0047] Figure 6 This is a schematic diagram of the weight allocation process in one embodiment;

[0048] Figure 7 This is a flowchart illustrating the weight adjustment process in one embodiment;

[0049] Figure 8 This is a structural block diagram of a power time-series prediction model evaluation device based on statistical analysis in one embodiment;

[0050] Figure 9 This is an internal structural diagram of a computer device in one embodiment;

[0051] Figure 10 This is a diagram of the internal structure of a computer device in another embodiment. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0053] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0054] The power time-series forecasting model evaluation method based on statistical analysis provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0055] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, etc. The server 104 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0056] The method provided in this application embodiment can be executed by either server 104 or terminal 102 alone, or by server 104 and terminal 102 interactively. The following description uses server 104 executing alone as an example. Specifically, server 104 first obtains the prediction results, prediction errors, and operational constraints of the target power grid output by the power time-series prediction model after performing power prediction on the target power grid. Based on these operational constraints, server 104 performs a feasibility assessment of the prediction results to obtain the application feasibility of the prediction results in the target power grid. Next, server 104 identifies the operational scenario categories of the target power grid during the prediction process and determines the evaluation weights of prediction error and application feasibility under each operational scenario category. Finally, server 104 weights and fuses the prediction error and application feasibility according to their respective evaluation weights to obtain the evaluation result of the power time-series prediction model.

[0057] In one exemplary embodiment, such as Figure 2 As shown, a statistical analysis-based method for evaluating power time-series forecasting models is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0058] Step S202: Obtain the prediction results, prediction errors, and operating constraints of the target power grid output by the power time series prediction model after performing power prediction on the target power grid.

[0059] Among them, the power time-series forecasting model is a mathematical model used to predict the future numerical values ​​of power parameters that change over time in a target power grid, such as at least one of the following: load, renewable energy output, and node voltage. The target power grid refers to a specific power network with a defined topology, equipment parameters, operational boundaries, and control requirements; it is the practical application object of the forecast results. The forecast result is the sequence of predicted numerical values ​​output by the power time-series forecasting model for the power parameters in the target power grid at various points in time within a future period, such as the load forecast value every 15 minutes for the next 24 hours. The forecast error is the difference between the forecast result and the actual observed value. Commonly used indicators include, but are not limited to, at least one of the following: mean absolute error, root mean square error, and mean absolute percentage error, used to quantify the model's performance in numerical accuracy. Operational constraints refer to the physical limitations that the target power grid must comply with for safe and stable operation, including but not limited to at least one of the following: safe operation constraints, energy acceptance constraints, electricity market constraints, environmental constraints, and power grid equipment constraints.

[0060] For example, the server first extracts the output data, i.e., the prediction results, from the executed power time-series forecasting task, typically stored in a database or results file as a time-series array or matrix. Simultaneously, the server obtains historical actual observation data for the corresponding time period from the power grid data platform. By comparing the predicted sequence with the actual sequence, it calculates one or more prediction error indices to reflect the statistical accuracy of the model. To construct operational constraints, the server can further extract a comprehensive power grid dataset from data sources such as the target power grid's SCADA (Supervisory Control and Data Acquisition) system, meteorological monitoring platforms, and power market operation systems. This dataset serves as the source of operational constraints and includes, but is not limited to, at least one of the following: multivariate time-series measurement data, meteorological data, power market data, power grid dispatching data, power grid topology, and equipment parameters. The data includes, but is not limited to, at least one type of data reflecting the real-time operating status of the power grid, such as power flow of each transmission line and voltage of each node; meteorological data, including but not limited to, at least one type of data such as temperature, humidity, and wind speed; electricity market data, which reflects the market supply and demand relationship and dispatch plan; power grid dispatch data, including but not limited to, at least one type of data such as dispatch logs and unit combinations; and equipment parameters, including but not limited to, equipment parameters of at least one key piece of equipment in the power grid, such as generators and transformers. Based on this power grid dataset, the server can construct the operating constraints of the target power grid.

[0061] In some embodiments, the operating constraints can be pre-built and stored based on a power grid dataset. This allows the server to directly read the operating constraints when evaluating a power time-series prediction model corresponding to the target power grid. In practical applications, the operating constraints can be dynamically updated based on the real-time operating status of the target power grid to ensure their accuracy and reliability.

[0062] Step S204: Based on the operational constraints, conduct a feasibility assessment of the prediction results to obtain the feasibility of applying the prediction results in the target power grid.

[0063] Feasibility assessment refers to the process of using the predicted results as a hypothetical operating mode to evaluate whether it meets the various operating constraints of the target power grid and quantifies the degree to which it meets these constraints. Its core is to verify the physical feasibility and safety of the predicted results. Application feasibility, on the other hand, is a comprehensive quantitative judgment obtained after the above feasibility assessment on whether the predicted results can be safely and reliably applied in the target power grid. It can be reflected in one or more index values, such as at least one of the following: safety margin index value, energy acceptance index value, etc. It can also be a comprehensive index value formed by the fusion of multiple index values, such as a weighted sum of the safety margin index value and the energy acceptance index value.

[0064] For example, the server inputs the predicted results as the hypothetical future operating state of the power grid, combining them with a static model of the target power grid's topology, equipment parameters, etc., to construct the predicted operating condition to be evaluated. Subsequently, based on the physical laws of the power system (such as power flow equations) and operating procedures, the server uses simulation tools (such as at least one power flow calculation and security analysis module) to analyze the predicted operating condition and check whether it meets various operating constraints. Finally, the server synthesizes and quantifies the results of the constraint checks, for example, outputting the safety margin index value and the absorption compliance index value separately, and / or outputting a comprehensive index value obtained by fusing the two index values, which serves as the evaluation conclusion for the application feasibility.

[0065] Step S206: Identify the target power grid's operating scenario category under the prediction results.

[0066] The operational scenario category refers to the types of power grid operating conditions with different risk characteristics and operational features. In this embodiment, the core categories include at least a stable operation scenario and a high-risk disturbance scenario. The former represents a stable operation of the power grid, while the latter represents a risky state caused by disturbances from at least one factor such as extreme weather, faults, or severe power fluctuations. The operational scenario category of the target power grid under the prediction results refers to the operational scenario category that the target power grid may have in the future time period corresponding to the prediction results.

[0067] For example, the server can extract the feature vector corresponding to the prediction result, and based on the feature vector, match the prediction result with multiple pre-determined candidate running scenario categories to obtain the running scenario type that matches the prediction result.

[0068] In some embodiments, the server can pre-process historical operating condition data of the target power grid using unsupervised learning algorithms to cluster or detect anomalies in the data, thereby automatically identifying and classifying several key operating scenarios, i.e., candidate operating scenario categories. Specifically, the server can employ a data compression and reconstruction network based on an encoder and decoder structure to map high-dimensional historical operating condition data containing multiple power parameter variables into a low-dimensional feature space, generating a corresponding low-dimensional feature vector for each historical moment. In this low-dimensional feature space, the algorithm calculates the probability density distribution of all low-dimensional feature vectors, i.e., the frequency of each low-dimensional feature vector appearing in the space, thereby identifying the region with the densest distribution of data points, i.e., the core mapping region. Mathematically, this region can be represented as a continuous, low-dimensional spatial structure, called the scenario baseline manifold, which centrally represents the characteristics of a typical operating state of the power grid. Typically, the core mapping region representing a stable operating state can be identified first to construct the stable operating scenario baseline manifold. Correspondingly, the regions distributed on the periphery or with lower density can be regarded as feature regions corresponding to high-risk disturbance scenarios. Furthermore, for each historical or future low-dimensional feature vector, the server quantifies its deviation from the normal state by calculating its geometric distance, such as Mahalanobis distance, between it and the benchmark manifold of the stable operating scene, and thus determines the scene category to which the vector belongs based on the magnitude of this deviation.

[0069] Step S208: Determine the evaluation weights of prediction error and application feasibility under the respective operating scenario categories.

[0070] The evaluation weights include evaluation weights corresponding to prediction error and application feasibility. These are relative importance coefficients assigned to the prediction error and application feasibility indicators. The weights are a set of non-negative real numbers, typically with a normalized sum of 1. Their allocation directly affects the calculation of the final evaluation score and the determination of model performance. In this embodiment, the weights are adaptive based on the scenario, i.e., dynamically generated for different operating scenario categories.

[0071] For example, for a given category of operating scenarios, the server invokes a pre-trained multi-objective optimization weight decision model for that category. This model is essentially a multi-objective optimization problem, whose objective function set includes at least two objectives: one aimed at minimizing the prediction error, and another aimed at maximizing the application feasibility. By solving this multi-objective optimization problem, corresponding evaluation weights can be assigned to the prediction error and the application feasibility.

[0072] In some embodiments, the server can also apply a model attribution analysis algorithm to the trained power time series prediction model to calculate the sensitivity or contribution of the model's predicted output to each input feature, generating a set of feature importance vectors. The similarity or distribution difference between these feature importance vectors and a pre-constructed physical mechanism prior importance vector based on power system expert knowledge is measured. The measurement method is to calculate the cosine similarity between the two vectors. The resulting measurement is defined as a model interpretability index and incorporated as a new optimization objective into the multi-objective optimization framework, i.e., maximizing model interpretability.

[0073] In some embodiments, a set of adversarial perturbation operators can be designed for the input characteristics of the power time-series prediction model. These perturbation operators include at least a continuous small-amplitude offset operator simulating sensor data drift and a random data point masking operator simulating communication data packet loss. The data characteristics processed by the perturbation operators are input into the power time-series prediction model, and the performance degradation rate (e.g., prediction time, computational resources consumed, etc.) relative to the original input is calculated. The performance degradation rate is defined as a model robustness index and integrated into the multi-objective optimization problem as a constraint or an independent optimization objective.

[0074] Step S210: According to each evaluation weight, the prediction error and application feasibility are weighted and fused to obtain the evaluation result of the power time series prediction model.

[0075] The evaluation result refers to the final quantitative score of the comprehensive performance of the power time series prediction model in the current prediction task, generated after weighted fusion. This result is a scalar value, the magnitude of which directly reflects the overall performance of the model in a specific scenario, taking into account both prediction accuracy and physical feasibility.

[0076] For example, the server combines the two indicators of prediction error and application feasibility by multiplying them by their respective evaluation weights and then adding them together or by using other aggregation functions to synthesize a comprehensive evaluation value, which is the evaluation result of the power time series prediction model.

[0077] In this embodiment, the server first obtains the prediction results, prediction errors, and operational constraints of the target power grid output by the power time-series prediction model after performing power prediction on the target power grid. Based on these operational constraints, a feasibility assessment is performed on the prediction results to determine their application feasibility in the target power grid. Next, the operational scenario categories of the target power grid during the prediction process are identified, and the evaluation weights of prediction error and application feasibility under each operational scenario category are determined. Finally, the prediction error and application feasibility are weighted and fused according to their respective evaluation weights to obtain the evaluation result of the power time-series prediction model. Thus, this embodiment evaluates the application feasibility of the model's prediction results in a real power grid environment based on the power grid's operational constraints, and intelligently adjusts the evaluation focus according to the actual operational scenarios of the power grid, significantly improving the accuracy of the power time-series prediction model evaluation. This provides a more comprehensive, reliable, and practically relevant decision-making basis for the selection, deployment, and optimization of the model in actual business operations.

[0078] In one exemplary embodiment, such as Figure 3 As shown, based on operational constraints, a feasibility assessment is performed on the prediction results to determine their application feasibility in the target power grid, including:

[0079] Step S302: Extract safe operation constraints from the operation constraints.

[0080] Step S304: Based on the safety operation constraints, evaluate the safety operation level of the target power grid under the prediction results, and obtain the safety margin index value corresponding to the prediction results.

[0081] Among them, safe operation constraints refer to the boundary conditions for ensuring the safe and stable operation of the target power grid. Their core is to prevent at least one of the following situations: equipment damage, system instability, or power outages. These are usually specified in the form of explicit upper and lower numerical limits, such as: the maximum allowable long-term / short-term current carrying capacity of transmission lines or transformers, the safe operating range of bus voltages at each voltage level, the power grid frequency deviation limit, or the power angle stability limit of generator units, etc. Application feasibility includes a safety margin index value used to characterize the safe operation level of the target power grid under the predicted results. The safety margin index value refers to the distance between the predicted result and the boundaries of each safe operation constraint condition; it is a comprehensive quantitative assessment result, usually positive or negative. A positive value indicates a margin, meaning that the predicted operating condition corresponding to the predicted result has not exceeded the safety limit and there is still a distance from the limit, i.e., a margin. A negative value indicates exceeding the limit, meaning that the predicted operating condition has exceeded the safety limit, and the power grid is in a risky state. The safety margin index value can be a multi-dimensional vector containing the safety margins corresponding to each safety operation constraint, or it can be aggregated into a comprehensive safety score. This value intuitively reflects the buffer space or risk level of the prediction results in the physical safety dimension of the power grid.

[0082] For example, the server automatically identifies and extracts all constraint entries directly related to the physical safety of the power grid from the operational constraints, i.e., safe operation constraints. Specifically, the server parses the name and description of each constraint entry, classifying constraints involving line / equipment capacity, voltage amplitude, frequency, power angle stability, short-circuit current, etc., into safe operation constraints.

[0083] Based on safe operation constraints, the server uses simulation tools to calculate the differences between the actual values ​​and limits of each safety constraint under the predicted operating conditions. For example, it calculates at least one of the following: the percentage of actual power flow to the thermal stability limit of a certain line, or the magnitude of voltage deviation from the rated value at a certain node. Subsequently, the server normalizes and quantifies these differences to generate the margin for each safety constraint. Finally, through at least one method, such as weighted aggregation, taking the minimum margin, or constructing a multi-dimensional vector, a safety margin index value is formed to characterize the safe operation level of the target power grid under the predicted results.

[0084] In this embodiment, by quantifying the degree to which the prediction results meet the constraints of safe operation into specific safety margin index values, the physical feasibility assessment of the prediction results is transformed from a qualitative judgment into an objective and comparable quantitative analysis, providing a core basis for the accurate assessment of the model in the safety dimension.

[0085] In one exemplary embodiment, such as Figure 4 As shown, based on operational constraints, a feasibility assessment is performed on the prediction results to determine their application feasibility in the target power grid, including:

[0086] Step S402: Extract energy acceptance constraints from the operating constraints.

[0087] Step S404: Evaluate the energy acceptance level of the target power grid under the prediction results based on the energy acceptance constraints, and obtain the energy acceptance index value corresponding to the prediction results.

[0088] Energy acceptance constraints refer to the set of conditions used to measure and limit the maximum permissible absorption capacity of renewable energy sources, such as wind and solar power, in the power grid. The core of these conditions is to maximize the utilization of renewable energy, and may include, but is not limited to, at least one of the following: regional / node renewable energy output caps, minimum grid spinning reserve requirements, market- or policy-based renewable energy quotas, and absorption rate targets. Application feasibility includes energy acceptance index values ​​used to characterize the energy acceptance level of the target power grid under the predicted results. The energy acceptance index value refers to the degree to which the renewable energy output portion of the predicted results meets the energy acceptance constraints. It is a key performance quantification result, typically a percentage ratio between the actual absorption rate and the target absorption rate, or a margin value. A positive margin value indicates that the predicted output is within the absorption capacity range and has a margin, while a negative margin value indicates that the predicted output has exceeded the current grid absorption capacity. This index directly reflects the feasibility of the predicted results in promoting cleaner energy.

[0089] For example, the server automatically identifies and extracts all constraint items directly related to the renewable energy acceptance capacity from the operational constraints, i.e., energy acceptance constraints. Specifically, the server parses the name and description of each constraint item, classifying constraints involving renewable energy, energy quotas, absorption rates, energy output, etc., into energy acceptance constraints.

[0090] Based on energy acceptance constraints, the server uses simulation tools to calculate the differences between the actual values ​​and limits of each energy acceptance constraint under the predicted operating conditions. For example, it calculates the percentage of the total predicted renewable energy output in the region relative to the region's maximum capacity, or assesses whether the grid can still meet the specified minimum reserve requirements under the predicted output. Subsequently, the server normalizes and quantifies these differences to generate a margin for each energy acceptance constraint. Finally, through at least one method—weighted aggregation, taking the margin of key constraints, or constructing a multi-dimensional vector—an energy acceptance index value characterizing the target grid's energy acceptance level under the predicted results can be formed.

[0091] In this embodiment, the degree to which the prediction results meet the energy acceptance constraints is quantified into specific energy acceptance index values, so as to intuitively reflect the contribution or feasibility of the prediction results to the grid's clean energy consumption, thereby guiding and optimizing the prediction model in a direction that is more conducive to supporting the high proportion of new energy access and the green transformation of the grid.

[0092] In one exemplary embodiment, such as Figure 5 As shown, the target power grid's operating scenario categories under the prediction results are identified, including:

[0093] Step S502: Map the prediction result to the feature space to obtain the mapping position of the prediction result in the feature space.

[0094] The feature space is a low-dimensional feature space, where each dimension corresponds to a learned intrinsic feature of power grid operation, capable of capturing the dynamic patterns and essential attributes of power grid operation. The mapping location refers to the coordinate point in the low-dimensional feature space corresponding to the input of the prediction result into the trained unsupervised learning model.

[0095] For example, the server invokes a pre-trained data compression and reconstruction network to organize the predicted result sequence, along with covariate data such as weather and market conditions associated with the predicted period, into an input sample in the same format as historical operating data. Subsequently, this sample is forward-propagated through an encoder, where its high-dimensional information is compressed and transformed, ultimately outputting a fixed-length low-dimensional feature vector. The position of this vector in space is the mapping location.

[0096] Step S504: Determine the mapping region in the feature space corresponding to the power grid operation condition sample under at least one candidate operation scenario category.

[0097] In this context, the mapping region refers to the spatial range formed by the clustering of mapping locations of historical operating condition data belonging to the same candidate operating scenario category (such as a stable operating scenario) in the feature space. This region is typically defined and described by its core scenario baseline manifold.

[0098] For example, the server loads the baseline models for each candidate runtime scenario category that were built during the historical learning phase. For each candidate scenario, such as a stable runtime scenario, its mapping region in the feature space is defined by the baseline manifold corresponding to that scenario and a distance threshold around that manifold.

[0099] Step S506: Determine the distance between the mapping location and the mapping area.

[0100] In this embodiment, distance refers to the quantized interval between the mapped position of the prediction result and the mapped region of a candidate running scenario category. Commonly used metrics include the Euclidean distance to the nearest point or center point on the baseline manifold of the scenario, or the more statistically significant Mahalanobis distance. This distance value directly measures the similarity or deviation between the prediction result and a typical running scenario category.

[0101] For example, for the mapped location of the predicted result, the server calculates the distance to the mapped region for each candidate running scenario category. Taking a stationary running scenario as an example, the server specifically calculates the minimum distance from the mapped location to the baseline manifold of the stationary running scenario. If the baseline manifold is represented by a set of representative center points, the minimum distance to these center points is calculated. If it is represented by a probability distribution model, the negative log-likelihood or Mahalanobis distance is calculated.

[0102] Step S508: If the distance under the candidate operating scenario category meets the proximity condition, the candidate operating scenario category is determined as the operating scenario category of the target power grid in the prediction process.

[0103] The proximity condition refers to pre-defined rules used to determine whether a mapped location is sufficiently close to a certain mapped area to be considered part of that scene. It typically manifests as one or more distance thresholds; when the calculated distance is less than or equal to a preset distance threshold for that scene, the proximity condition is met. This threshold is usually adaptively determined based on the data distribution during the historical learning phase.

[0104] For example, the server compares the distance from the predicted result to each candidate scenario with the proximity condition, i.e., the distance threshold, corresponding to each scenario. Taking a stable operation scenario as an example, if the distance is less than or equal to the threshold, the target power grid is considered to be in a stable operation scenario under the predicted result. If the distance is greater than the threshold, the target power grid is considered to be in a high-risk disturbance operation scenario under the predicted result.

[0105] In this embodiment, by mapping the prediction results to a low-dimensional feature space and calculating the quantized distance between it and known typical scenario regions, the automatic, objective, and interpretable identification of the operational scenario category is achieved, providing a reliable decision-making basis for subsequent determination of evaluation weights, thereby significantly improving the fit and accuracy between the model evaluation and the actual power grid operation status.

[0106] In one exemplary embodiment, such as Figure 6 As shown, the evaluation weights of prediction error and application feasibility are determined respectively under the operational scenario category, including:

[0107] Step S602: With the goal of minimizing prediction error and maximizing the application feasibility of prediction results, evaluation weights are assigned under the running scenario category to obtain a weight allocation scheme.

[0108] Step S604: Obtain the evaluation weights for prediction error and application feasibility from the weight allocation scheme.

[0109] The evaluation weight allocation refers to the process of optimizing and determining the weight of prediction error and application feasibility in the comprehensive evaluation to achieve the two objectives. A weight allocation scheme refers to one or more possible combinations of weights obtained by solving a multi-objective optimization problem; for example, the prediction error weight is 0.7, and the application feasibility weight is 0.3. Each scheme corresponds to a Pareto optimal solution.

[0110] For example, the server filters all samples belonging to a specific identified operational scenario category from the historical dataset. Based on these samples, a bi-objective optimization problem is constructed. The first objective function is to minimize the statistical value of prediction error, such as the mean root mean square error; the second objective function is to maximize the comprehensive value of application feasibility, such as the average safety margin. Subsequently, a multi-objective genetic algorithm, such as NSGA-II, is used to iteratively solve this problem. The algorithm searches in the solution space, i.e., the set of all weight combinations, by simulating the evolutionary process of selection, crossover, and mutation, and finally outputs a Pareto optimal solution set. Each solution in this set corresponds to a weight allocation scheme, and all these schemes are connected in the objective space to form a Pareto front curve. It should be noted that prediction accuracy and application feasibility are inherently conflicting objectives and are difficult to achieve simultaneously at their optimal levels. Therefore, multi-objective optimization techniques are used to automatically explore all possible and optimal trade-offs between the two.

[0111] After obtaining the Pareto front, representing the Pareto optimal set of weight allocation schemes, the server needs to select a final scheme from it. First, a critical region called the knee point is automatically identified on the Pareto front curve. This region typically corresponds to the point of maximum curvature, meaning that a slight sacrifice in accuracy at this point can yield a significant improvement in feasibility, or vice versa—representing the optimal balance point for overall benefit. The server initially uses the scheme corresponding to the knee point as a benchmark, then introduces a risk adjustment factor. That is, if the current scenario is a high-risk disturbance scenario, the server commands the weight selection point to shift from the knee point along the Pareto front in a direction that better maximizes application feasibility, thus ultimately determining a set of evaluation weights that prioritize safety and feasibility. For stable operation scenarios, the knee point weights may be used directly, or a slight shift towards accuracy may be made.

[0112] In some embodiments, the server can mathematically adjust the first objective function—the objective function that minimizes the prediction error—based on an accuracy confidence decay factor to incorporate the impact of the prediction time span into the multi-objective optimization problem. Specifically, the accuracy confidence decay factor is a quantification coefficient that monotonically decreases as the prediction time span increases, typically ranging from (0,1). It reflects the decrease in prediction accuracy confidence caused by the inherent increase in uncertainty due to the extended prediction time. In multi-objective optimization, the server uses this factor as an adjustment coefficient on prediction error indicators such as mean squared error, thereby assigning a relatively low accuracy weight to long-term predictions and making the optimization process more aligned with the actual business logic of prioritizing accuracy in the near term and physical feasibility in the long term.

[0113] In this embodiment, multi-objective optimization technology is used to dynamically solve the optimal trade-off weight between prediction accuracy and physical application feasibility for different operating scenarios. This enables the evaluation criteria to intelligently meet the real needs of the power grid under different states, thereby ensuring that the comprehensive evaluation results can better reflect the effectiveness and reliability of the model in practical applications. Ultimately, this significantly improves the accuracy and decision value of the power time series prediction model evaluation.

[0114] In one exemplary embodiment, such as Figure 7 As shown, with the goal of minimizing prediction error and maximizing the applicability of prediction results, evaluation weights are assigned under the running scenario categories, resulting in a weight allocation scheme, including:

[0115] Step S702: With the goal of minimizing prediction error and maximizing the applicability of prediction results, evaluate the weight allocation to obtain an initial allocation scheme.

[0116] The initial allocation scheme refers to the set of benchmark weights obtained through multi-objective optimization for the category of operating scenarios, which represents a general trade-off between accuracy and feasibility. It is usually the scheme corresponding to the knee region of the Pareto front in that scenario.

[0117] For example, the server runs a multi-objective genetic algorithm with the goal of minimizing prediction error and maximizing the applicability of the prediction results. The algorithm explores the search space of all weight combinations and finally outputs the Pareto optimal solution set and the corresponding Pareto front for the scenario. Subsequently, the system identifies the knee region on the Pareto front and saves a representative weight combination corresponding to the region, such as taking the center point of the region, as the initial allocation scheme for the scenario.

[0118] Step S704: The distance between the mapping position of the prediction result in the feature space and the mapping region corresponding to the running scene category in the feature space is used as the weight adjustment factor corresponding to the running scene category.

[0119] The weight adjustment factor is a quantified parameter used to dynamically adjust the initial allocation scheme based on the degree of deviation of the current prediction result from its corresponding scene category. In this embodiment, the factor is specifically defined as the distance between the mapping position of the prediction result in the feature space and the mapping region of its corresponding running scene category.

[0120] Step S706: Adjust the initial allocation scheme based on the weight adjustment factor to obtain the adjusted weight allocation scheme.

[0121] The adjusted weight allocation scheme refers to the final weight allocation scheme generated after applying the weight adjustment factor to the initial allocation scheme. This scheme takes into account both the general requirements of the scenario category and the specific risk characteristics of the current prediction.

[0122] For example, the server pre-defines a weight adjustment function, which takes the initial allocation scheme and a weight adjustment factor (distance d) as input. Its core rule is: the larger the adjustment factor d, the lower the weight corresponding to the prediction error and the higher the weight of application feasibility. This can be understood as follows: the more extreme the predicted state, i.e., the greater the distance, the more the evaluation leans towards ensuring safety and physical feasibility, thereby achieving refined dynamic adaptation of the evaluation weights and making the evaluation results more targeted and risk-sensitive.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0124] Based on the same inventive concept, this application also provides a statistical analysis-based power time-series forecasting model evaluation device for implementing the aforementioned statistical analysis-based power time-series forecasting model evaluation method. The solution provided by this device is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more statistical analysis-based power time-series forecasting model evaluation device embodiments provided below can be found in the limitations of the statistical analysis-based power time-series forecasting model evaluation method described above, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 8 As shown, a power time-series forecasting model evaluation device based on statistical analysis is provided, comprising:

[0126] The information acquisition module 802 is used to acquire the prediction results, prediction errors, and operating constraints of the target power grid output by the power time series prediction model after making power predictions on the target power grid.

[0127] The feasibility assessment module 804 is used to assess the feasibility of the prediction results based on operational constraints, and to obtain the feasibility of the application of the prediction results in the target power grid.

[0128] Scene recognition module 806 is used to identify the operating scenario category of the target power grid under the prediction results;

[0129] The weight allocation module 808 is used to determine the evaluation weights of prediction error and application feasibility under the running scenario category, respectively.

[0130] The model evaluation module 810 is used to weight and fuse the prediction error and application feasibility according to each evaluation weight to obtain the evaluation result of the power time series prediction model.

[0131] In one embodiment, application feasibility includes a safety margin index value characterizing the safe operating level of the target power grid under the predicted results; the feasibility assessment module 804 further includes a safety assessment unit for:

[0132] Extract safe operating constraints from operating constraints;

[0133] The safety operation level of the target power grid under the prediction results is evaluated based on the safety operation constraints, and the safety margin index value corresponding to the prediction results is obtained.

[0134] In one embodiment, application feasibility includes energy acceptance index values ​​characterizing the energy acceptance level of the target power grid under the predicted results; the feasibility assessment module 804 further includes an energy assessment unit for:

[0135] Extract energy acceptance constraints from operational constraints;

[0136] The energy acceptance level of the target power grid under the predicted results is evaluated based on the energy acceptance constraints, and the corresponding energy acceptance index value is obtained.

[0137] In one embodiment, the scene recognition module 806 is further configured to:

[0138] The prediction result is mapped to the feature space to obtain the mapping position of the prediction result in the feature space.

[0139] Determine the mapping region in the feature space corresponding to the power grid operation condition sample under at least one candidate operation scenario category;

[0140] Determine the distance between the mapping location and the mapping area;

[0141] If the distance within a candidate operating scenario category meets the proximity condition, the candidate operating scenario category is determined as the operating scenario category of the target power grid during the prediction process.

[0142] In one embodiment, the weight allocation module 808 is further configured to:

[0143] With the goal of minimizing prediction error and maximizing the applicability of prediction results, evaluation weights are assigned under the category of operating scenario to obtain a weight allocation scheme.

[0144] The evaluation weights for prediction error and application feasibility are obtained from the weight allocation scheme.

[0145] In one embodiment, the weight allocation module 808 is further configured to:

[0146] With the goal of minimizing prediction error and maximizing the applicability of prediction results, the weight allocation is evaluated to obtain an initial allocation scheme;

[0147] The distance between the mapping position of the prediction result in the feature space and the mapping region corresponding to the running scene category in the feature space is used as the weight adjustment factor for the running scene category.

[0148] The initial allocation scheme is adjusted based on the weight adjustment factor to obtain the adjusted weight allocation scheme.

[0149] The modules in the aforementioned power time-series prediction model evaluation device based on statistical analysis can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0150] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores evaluation data for a power time-series forecasting model based on statistical analysis. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power time-series forecasting model evaluation method based on statistical analysis.

[0151] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a statistical analysis-based power time-series prediction model evaluation method.

[0152] Those skilled in the art will understand that Figure 9 or Figure 10The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0153] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the embodiments described above.

[0154] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the various embodiments of the above methods.

[0155] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the embodiments described above.

[0156] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0157] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0158] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0159] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for evaluating a power time-series forecasting model based on statistical analysis, characterized in that, The method includes: The prediction results, prediction errors, and operating constraints of the target power grid are obtained from the power time series prediction model after it performs power prediction on the target power grid. Based on the operational constraints, a feasibility assessment is performed on the prediction results to determine the feasibility of applying the prediction results in the target power grid. Identify the operating scenario category of the target power grid under the prediction results; Determine the evaluation weights of the prediction error and the application feasibility respectively under the respective operating scenario categories; The prediction error and the application feasibility are weighted and fused according to the evaluation weights to obtain the evaluation result of the power time series prediction model.

2. The method according to claim 1, characterized in that, The application feasibility includes a safety margin index value used to characterize the safe operation level of the target power grid under the prediction results; The feasibility assessment of the prediction results based on the operational constraints, to determine the application feasibility of the prediction results in the target power grid, includes: Extract safe operating constraints from the operating constraints; Based on the aforementioned safe operation constraints, the safe operation level of the target power grid under the predicted results is evaluated to obtain the safety margin index value corresponding to the predicted results.

3. The method according to claim 1, characterized in that, The application feasibility includes energy acceptance index values ​​used to characterize the energy acceptance level of the target power grid under the prediction results; The feasibility assessment of the prediction results based on the operational constraints, to determine the application feasibility of the prediction results in the target power grid, includes: Extract energy acceptance constraints from the operational constraints; Based on the energy acceptance constraints, the energy acceptance level of the target power grid under the predicted results is evaluated to obtain the energy acceptance index value corresponding to the predicted results.

4. The method according to claim 1, characterized in that, The identification of the target power grid's operating scenario category under the prediction results includes: The prediction result is mapped to the feature space to obtain the mapping position of the prediction result in the feature space. Determine the mapping region in the feature space corresponding to the power grid operating condition sample under at least one candidate operating scenario category; Determine the distance between the mapped location and the mapped region; If the distance under the candidate operating scenario category meets the proximity condition, the candidate operating scenario category is determined as the operating scenario category of the target power grid in the prediction process.

5. The method according to claim 1, characterized in that, The process of determining the evaluation weights of the prediction error and the application feasibility under the respective operating scenario categories includes: With the goal of minimizing prediction error and maximizing the applicability of prediction results, an evaluation weight allocation scheme is obtained by performing evaluation weight allocation under the aforementioned operational scenario category. The evaluation weights for the prediction error and the application feasibility are obtained from the weight allocation scheme.

6. The method according to claim 5, characterized in that, The evaluation weight allocation scheme, which aims to minimize prediction error and maximize the application feasibility of prediction results, is performed under the category of operating scenarios to obtain a weight allocation scheme, including: With the goal of minimizing prediction error and maximizing the applicability of prediction results, the weight allocation is evaluated to obtain an initial allocation scheme; The distance between the mapping position of the prediction result in the feature space and the mapping region corresponding to the running scene category in the feature space is used as the weight adjustment factor corresponding to the running scene category. The initial allocation scheme is adjusted based on the weight adjustment factor to obtain the adjusted weight allocation scheme.

7. A power time-series prediction model evaluation device based on statistical analysis, characterized in that, The device includes: The information acquisition module is used to acquire the prediction results, prediction errors, and operating constraints of the target power grid output by the power time series prediction model after making power predictions on the target power grid. The feasibility assessment module is used to assess the feasibility of the prediction results based on the operational constraints, and to obtain the feasibility of applying the prediction results in the target power grid. The scene recognition module is used to identify the operating scene category of the target power grid under the prediction result; The weight allocation module is used to determine the evaluation weights of the prediction error and the application feasibility respectively under the running scenario category; The model evaluation module is used to weight and fuse the prediction error and the application feasibility according to the evaluation weights, so as to obtain the evaluation result of the power time series prediction model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images