Multi-dimensional evaluation method for power network synchronous recovery AI game algorithm

Abstract

The invention relates to the field of power network recovery evaluation, in particular to a multi-dimensional evaluation method for a power network synchronous recovery AI game algorithm, which realizes systematic quantitative evaluation of a power network high-power electromagnetic pulse out-of-step AI algorithm, covers system-level and node-level indexes and supports self-adaptive test of multiple topology scenes. The scheme comprises the following steps: constructing a power grid model standard topology library containing IEEE standard nodes and derivative topologies thereof, constructing a dynamic scene, and solving a power system differential equation set through a fourth-order Runge-Kutta method to generate reference data; based on space-time alignment of the reference data and the prediction result of the AI game algorithm, multi-level indexes are calculated, and dynamic weight distribution is carried out; generating a comprehensive score by weighting and aggregating the system-level and node-level indexes; and generating a capability report, wherein the capability report comprises a comprehensive score, index decomposition and a visual chart. The method is suitable for power network recovery evaluation.

Description

Technical Field

[0001] The present invention relates to the field of power network restoration evaluation, and in particular to a multi-dimensional evaluation method for an AI game algorithm for synchronous restoration of a power network. Background Art

[0002] With the continuous expansion of the scale of power systems and the improvement of their intelligence, the safety and stability of power networks in extreme electromagnetic environments such as high-altitude strong electromagnetic pulses are becoming increasingly prominent. Strong electromagnetic pulses have extremely high field strengths and extremely fast rise times, which can induce strong transient currents and voltages in power systems, leading to serious consequences such as equipment damage, protection malfunctions, and control system failures, and even causing large-scale power outages. Therefore, how to effectively evaluate and improve the anti-interference and recovery capabilities of power systems in HEMP environments has become an important research topic in the field of power system safety.

[0003] In recent years, AI (Artificial Intelligence) technology has been increasingly used in power systems, especially in power system stability analysis, fault diagnosis and recovery control. AI algorithms, such as deep learning and reinforcement learning, are used to predict the dynamic response and recovery process of power systems under HEMP. However, there are still many problems in the performance evaluation of these AI algorithms:

[0004] (1) Single evaluation dimension: The current method mainly relies on the whole network synchronization recovery time error (such as mean square error MSE) as the core indicator, which cannot reflect local dynamic characteristics such as node phase deviation and frequency deviation. In addition, there is a lack of quantitative analysis of the differences in oscillation modes between regions, making it difficult to evaluate the adaptability of AI algorithms in complex topologies.

[0005] (2) Insufficient scenario coverage: Existing evaluation methods usually use fixed test scenarios (such as a single IEEE 30-node topology) and do not consider the dynamic changes of different grid scales and topologies. In addition, there is a lack of systematic sampling of HEMP parameters (such as pulse intensity and incident angle), resulting in insufficient test coverage.

[0006] (3) Hierarchical fragmentation problem: Existing methods do not establish a correlation analysis model between system-level macro indicators and node-level micro indicators, resulting in a lack of hierarchy in the evaluation results. For example, it is impossible to quantify the relationship between the node degree of a certain node and the synchronization recovery time of the entire network.

[0007] (4) Incomparability of results: Existing AI algorithms, such as LSTM (Long Short-Term Memory Network) and GNN (Graph Neural Network), have made certain progress in power grid synchronization recovery prediction, but lack a unified performance evaluation standard. The output data of different AI algorithms lack a standardized alignment method, making it difficult to compare horizontally. Therefore, there is a lack of targeted suggestions for the algorithm optimization direction, resulting in the evaluation results being difficult to directly guide algorithm improvement. Summary of the Invention

[0008] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a multi-dimensional evaluation method for the AI game algorithm for power grid synchronization recovery, realizing the systematic quantitative evaluation of the AI algorithm for power grid out-of-step resistance to high-power electromagnetic pulses, covering system-level and node-level indicators, and supporting adaptive testing of multiple topological scenarios.

[0009] The present invention adopts the following technical solutions to achieve the above purpose. The present invention provides a multi-dimensional evaluation method for the AI game algorithm for power grid synchronization recovery, including:

[0010] S1. Construct a standard topology library of the power grid model containing IEEE standard nodes and their derived topologies, construct a dynamic scenario, and generate benchmark data by solving the differential equations of the power system using the fourth-order Runge-Kutta method;

[0011] S2. Calculate multi-level indicators based on the spatio-temporal alignment of the benchmark data and the prediction results of the AI game algorithm, and perform dynamic weight allocation;

[0012] S3. Generate a comprehensive score by weighted aggregation of system-level and node-level indicators;

[0013] S4. Generate an ability report, and the ability report includes a comprehensive score, index decomposition, and visualization charts.

[0014] Furthermore, constructing a standard topology library of the power grid model containing IEEE standard nodes and their derived topologies, and constructing a dynamic scenario specifically includes:

[0015] Construct a standard topology library, which includes IEEE standard nodes and their derived topologies. Each topology includes node coordinates, branch impedances, and generator parameters. For the generator of the i-th node, the set parameters include the initial frequency and the initial phase

[0016] Generate high-altitude strong electromagnetic pulses with different parameters. The pulse waveforms of the high-altitude strong electromagnetic pulses are all approximated by the waveforms of double-exponential functions, and their time-domain expressions are as follows:

[0017] E(t)=1.05×E0(e -αt -e -βt );

[0018] Where E0 represents the peak value of the field intensity, α and β are the parameters of the leading and trailing edges of the pulse, respectively.

[0019] Furthermore, the fourth-order Runge-Kutta method is used to solve the power system differential equations to generate benchmark data, including:

[0020] Use the fourth-order Runge-Kutta method to solve the power system differential equations and output benchmark data

[0021] in, represents the benchmark recovery time obtained by the fourth-order Runge-Kutta numerical simulation, represents the phase angle of the i-th node at time t obtained by using the fourth-order Runge-Kutta numerical simulation, represents the frequency of the i-th node at time t of the fourth-order Runge-Kutta numerical simulation.

[0022] Furthermore, the calculation of multi-level indicators specifically includes:

[0023] Calculate the error rate of the synchronization recovery time of the entire network:

[0024]

[0025] In the formula, represents the AI-predicted recovery time of the kth scenario, N test represents the number of scenarios generated by simulation, T err Indicates the error rate of synchronization recovery time of the entire network;

[0026] Calculate the node-level metric node-phase root mean square error:

[0027]

[0028] Where N T Represents the total number of time series points, which is determined by the simulation duration and step size, N node represents the total number of nodes in the power grid, represents the phase angle of the i-th node predicted by AI at time t, Δθ RMS represents the phase deviation root mean square error of all nodes;

[0029] Calculate the maximum absolute error of the node frequency deviation of the node-level indicator:

[0030]

[0031] In the formula, It represents the frequency of the i-th node predicted by AI at time t, which is used to capture the extreme deviation of the AI algorithm in frequency prediction.

[0032] Furthermore, the specific steps for dynamic weight allocation include:

[0033] The weights of system-level and node-level indicators are dynamically allocated using the following formula:

[0034]

[0035] In the formula, ω sys represents the weight of the system-level indicator, and ω node is the weight of the node-level indicator.

[0036] Furthermore, the specific steps for generating a comprehensive score by weighted aggregation of system-level and node-level indicators include:

[0037] The comprehensive score is generated by weighted aggregation of system-level and node-level indicators, and the calculation formula is:

[0038]

[0039] In the formula, S represents the comprehensive score result.

[0040] Furthermore, the specific steps for generating an ability report include:

[0041] The ability report includes a comprehensive score, indicator decomposition, and a visualization chart;

[0042] Comprehensive score and indicator decomposition:

[0043] The evaluation grade is output according to the scoring criteria, where S≥0.9 indicates excellent; 0.8≤S≤0.9 indicates good; S<0.8 indicates average and needs improvement;

[0044] Output the contribution of system-level indicators:

[0045] Output the contribution of node-level indicators:

[0046] Visualization chart:

[0047] The performance in terms of time accuracy phase deviation prediction result frequency deviation prediction result is displayed through a visualization radar chart;

[0048] The waveforms of the phase angle and frequency time domain of the selected nodes are plotted, and the AI predicted values are compared with the reference values.

[0049] The beneficial effects of the present invention are:

[0050] The present invention realizes a multi-level dynamic evaluation system. Through the linkage of double-level indicators at the system level (time error) and node level (phase-frequency deviation), it breaks through the limitations of traditional single-dimensional evaluation and can simultaneously reflect the global performance and local dynamic characteristics. Combined with the dynamic weight allocation mechanism, it adapts to the evaluation needs of power grids of different scales.

[0051] The present invention provides a dynamic weight adaptive mechanism. Based on the weight formula of the logarithmic function, it realizes the smooth transition of weights with the scale of the power grid, making the evaluation results more in line with the actual needs. For example, for a 200-node power grid, the weight of the system-level index is increased to 70%, avoiding one-size-fits-all evaluation.

[0052] The present invention realizes a high coverage rate of scenarios. The present invention adopts the uniform random sampling method to generate power grid-high power electromagnetic pulse coupling test scenarios. By flexibly configuring the power grid topology, high power electromagnetic pulse parameters and operating conditions, a comprehensive test scenario library is constructed to ensure the high coverage rate and representativeness of the test scenario library. Brief Description of the Drawings

[0053] Figure 1 It is a flowchart of a multi-dimensional evaluation method for an AI game algorithm for power network synchronous restoration provided by an embodiment of the present invention. Detailed Embodiments

[0054] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0055] The present invention provides a multi-dimensional evaluation method for an AI game algorithm for power network synchronous restoration, as Figure 1 shown, specifically including:

[0056] S1. Construct a standard topology library of the power grid containing IEEE standard nodes and their derived topologies, construct dynamic scenarios, and generate reference data by solving the differential equations of the power system using the fourth-order Runge-Kutta method.

[0057] The construction of dynamic scenarios and the generation of reference data are the basic links of the present invention, aiming to generate diverse power grid-high power electromagnetic pulse coupling test scenarios and obtain reference data through high-precision numerical simulation. First, construct the standard topology library:

[0058] It contains IEEE standard node systems (such as IEEE 30-node, IEEE 57-node) and their derived topologies, and the maximum number of nodes is 200 nodes, that is, the value range of the number of nodes N is [30, 200], following a uniform distribution. Each topology contains the following data:

[0059] Node coordinates (for visualization);

[0060] Branch impedance: Z line ;

[0061] Generator parameters: For the generator of the i-th node, set the following parameters: initial frequency (Unit: Hz) The value range is [-0.1, 0.1] evenly distributed; initial phase (Unit: radian rad) The value range is [-0.5π, 0.5π] and is evenly distributed.

[0062] Then, high-altitude strong electromagnetic pulses with different parameters are generated, where the pulse waveforms of the high-altitude strong electromagnetic pulses are approximated by the waveform of a double exponential function, and their time domain expressions are as follows:

[0063] E(t)=1.05×E0(e -αt -e -βt );

[0064] Where E0 is the peak value of the field intensity, α and β are the parameters of the leading and trailing edges of the pulse, respectively. The incident field intensity E0 (unit kV / m) ranges from [10,100]; α and β are 4×10 6 and 4.76×10 8 .

[0065] Then, the benchmark data is generated, and the fourth-order Runge-Kutta method is used to solve the power system differential equations. The time step is 0.01ms to ensure high time resolution; the simulation time is 10s, covering transient and steady-state processes; output benchmark data

[0066] represents the benchmark recovery time (in seconds) obtained by using the fourth-order Runge-Kutta numerical simulation. represents the phase angle (in radians) of the i-th node at time t obtained by using the fourth-order Runge-Kutta numerical simulation. It represents the frequency of the ith node at time t of the fourth-order Runge-Kutta numerical simulation (unit: Hz).

[0067] S2. Based on the spatiotemporal alignment of benchmark data and AI game algorithm prediction results, multi-level indicators are calculated and dynamic weight allocation is performed.

[0068] The calculation of multi-level indicators specifically includes:

[0069] Calculate the error rate of the synchronization recovery time of the entire network:

[0070]

[0071] In the formula, Indicates the AI-predicted recovery time for the kth scenario (in seconds); N testIndicates the number of scenarios generated by simulation, with a default value of 500; T err Indicates the time error rate of the whole network synchronous recovery, T err The smaller it is, the better the system index of the AI algorithm in time prediction.

[0072] Calculate the root mean square error of node phase deviation for node-level indicators:

[0073]

[0074] In the formula, N T Indicates the total number of time series points, which is determined by the simulation duration (default 10 seconds) and the step size (0.01 ms); N node Indicates the total number of grid nodes, which ranges from 30 to 200 depending on the generated scenarios; Indicates the phase angle (unit: radian) of the i-th node predicted by AI at time t. Δθ RMS Indicates the root mean square error of phase deviation of all nodes, Δθ RMS The smaller it is, the more accurate the AI algorithm is in phase angle prediction. Usually, it is required that Δθ RMS <0.1 rad.

[0075] Calculate the maximum absolute error of node frequency deviation for node-level indicators:

[0076]

[0077] In the formula, Indicates the frequency (unit: Hz) of the i-th node predicted by AI at time t; this indicator is used to capture the extreme deviation of the AI algorithm in frequency prediction. Usually, it is required that Δf max <1 Hz.

[0078] The specific methods of dynamic weight allocation include:

[0079] To adapt to the change of the number of generated grid nodes from 30 to 200, the following formula is used to dynamically allocate the weights of system-level and node-level indicators:

[0080]

[0081] In the formula, ω sys Indicates the system-level indicator weight, with a range of 0 to 1; ω nodeis the node - level index weight, with a range of 0 to 1; ln(·) represents the natural logarithm function, ensuring a smooth transition of the weight as the number of nodes increases. This design of dynamic weight generation is based on complex network theory. As the number of nodes increases, the dominant factors of the system's dynamic behavior shift from local characteristics to global characteristics. For a small - scale power grid (such as a 30 - node grid), the node - level index weight is higher, and the evaluation focuses more on local dynamic characteristics; for a large - scale power grid (such as a 200 - node grid), the system - level index weight is higher, and the evaluation focuses more on the overall network performance.

[0082] S3. Generate a comprehensive score by weighted aggregation of system - level and node - level indicators.

[0083] The comprehensive score is generated by weighted aggregation of system - level and node - level indicators, and the calculation formula is:

[0084]

[0085] In the formula, S represents the comprehensive score result, and the score range is S ∈ [0,1]. The closer it is to 1, the better the performance of the tested AI algorithm.

[0086] S4. Generate an ability report, and the ability report includes the comprehensive score, index decomposition, and visualization charts.

[0087] Comprehensive score and index decomposition: Output the evaluation grade according to the scoring criteria, where S≥0.9 indicates excellent; 0.8≤S≤0.9 indicates good; S < 0.8 indicates general and needs improvement. Output the contribution of system - level indicators: Output the contribution of node - level indicators:

[0088] Visualization radar chart: Show the performance in the following dimensions: time accuracy Phase deviation prediction result Frequency deviation prediction result The value range of each dimension is [0,1]. The closer it is to 1, the better the performance.

[0089] Time - domain waveform comparison chart: Plot the waveforms of the phase angle and frequency in the time domain of key nodes, and compare the AI prediction values with the reference values.

[0090] The above - mentioned is only the preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in related fields. And the changes and alterations made by those skilled in the art without departing from the spirit and scope of the present invention should all be within the protection scope of the appended claims of the present invention.

Claims

1. A multi-dimensional evaluation method for an AI game algorithm for power network synchronous restoration, characterized in that include: S1. Build a standard topology library for power grid models containing IEEE standard nodes and their derivative topologies, construct dynamic scenarios, and generate benchmark data by solving the power system differential equations using the fourth-order Runge-Kutta method; S2. Based on the spatiotemporal alignment of benchmark data and AI game algorithm prediction results, multi-level indicators are calculated and dynamically weighted; S3, generate a comprehensive score by weighted aggregation of system-level and node-level indicators; S4. Generate a capability report, which includes a comprehensive score, indicator decomposition, and a visualization chart.

2. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 1, characterized in that, Build a standard topology library for power grid models that includes IEEE standard nodes and their derivative topologies. Build dynamic scenarios that include: Build a standard topology library, which includes IEEE standard nodes and their derived topologies. Each topology contains node coordinates, branch impedances, and generator parameters. For the generator of the i-th node, the set parameters include the initial frequency and the initial phase Then, high-altitude strong electromagnetic pulses with different parameters are generated. The pulse waveforms of high-altitude strong electromagnetic pulses are approximated by the waveform of a double exponential function, and their time domain expressions are as follows: E(t) = 1.05×E0(e -αt -e -βt ); Where E0 represents the peak value of the field intensity, α and β are the parameters of the leading and trailing edges of the pulse, respectively.

3. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 1, wherein The benchmark data generated by solving the power system differential equations by the fourth-order Runge-Kutta method include: The fourth-order Runge-Kutta method is used to solve the differential equations of the power system, and the reference data is output Among them, represents the reference recovery time obtained by solving using the fourth-order Runge-Kutta numerical simulation, represents the phase angle of the i-th node at time t obtained by solving using the fourth-order Runge-Kutta numerical simulation, represents the frequency of the i-th node at time t in the fourth-order Runge-Kutta numerical simulation.

4. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 3, characterized in that The calculation of multi-level indicators specifically includes: Calculate the error rate of the synchronization recovery time of the entire network: Wherein, represents the recovery time of the k-th scenario predicted by AI, and N test represents the number of scenarios generated by simulation, and T err represents the error rate of the whole network synchronous recovery time; Calculate the node-level metric node-phase root mean square error: Where, N T represents the total number of points in the time series, which is determined by the simulation duration and the step size. N node represents the total number of nodes in the power grid, represents the phase angle of the i-th node predicted by AI at time t, and Δθ RMS represents the root mean square error of the phase deviation of all nodes; Calculate the maximum absolute error of the node frequency deviation of the node-level indicator: In the formula, represents the frequency of the i-th node predicted by AI at time t, which is used to capture the extreme deviation of the AI algorithm in frequency prediction.

5. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 4, characterized in that, Dynamic weight allocation specifically includes: The weights of system-level and node-level indicators are dynamically allocated using the following formula: where ω sys represents the weight of system-level indicators, and ω node is the weight of node-level indicators.

6. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 5, characterized in that, The comprehensive score is generated by weighted aggregation of system-level and node-level indicators, including: The comprehensive score is generated by weighted aggregation of system-level and node-level indicators, and the calculation formula is: In the formula, S represents the comprehensive scoring result.

7. The multi-dimensional evaluation method for the AI game algorithm for power network synchronous restoration according to claim 6, characterized in that Capability report generation specifically includes: The capability report includes comprehensive scores, indicator breakdown, and visual charts; Comprehensive score and indicator breakdown: Output the rating level according to the scoring criteria, where S≥0.9 means excellent; 0.8≤S≤0.9 means good; S<0.8 means average and needs improvement; Output system-level metric contribution: Output node-level metric contribution: Visualization chart: Show the performance in terms of time accuracy through a visual radar chart Phase deviation prediction result Frequency deviation prediction result Performance in the dimension; Draw the waveform of the phase angle and frequency time domain of the selected node, and compare the AI predicted value with the benchmark value.

Images