Intercut small power supply adaptive control method for intelligent substation

Abstract

The invention relates to the technical field of safety control of a power system, and discloses an intercut small power supply adaptive control method for an intelligent substation, which comprises the following steps: collecting and receiving multi-source operation data of a power grid in real time, the multi-source operation data comprises protection action information from protection equipment, real-time electric quantity of a power grid, an operation state of a small power supply, state monitoring data of primary equipment and environment data; based on the multi-source operation data, performing dynamic risk assessment through a preset intelligent control algorithm, and generating a risk index representing a power grid fault risk level and an associated basis; generating and executing an intertripping control decision based on the risk index and the association basis; wherein the intertripping control decision is used for controlling the expansion relay group to generate a hard contact signal so as to drive the corresponding conventional line protection equipment to execute small power supply splitting operation. According to the method, a set of complete processing flow and data specification is defined, and the defects of different implementation paths and different standards in the prior art are overcome.

Description

Technical Field

[0001] This invention relates to the field of power system safety control technology, specifically to an adaptive control method for switching small power sources in smart substations. Background Technology

[0002] Driven by the energy structure transformation trend, distributed energy has developed rapidly, with a large number of small power sources such as distributed photovoltaic and energy storage power stations being connected to the grid. Taking Linyi City as an example, many small power sources are connected to 110kV smart substations via 10kV lines. However, this development trend has brought many problems. When a short circuit fault occurs on the 110kV line or main transformer, if the 10kV small power source is not disconnected in time, it will supply power in reverse, making it difficult to extinguish the fault arc, which may lead to serious consequences such as primary equipment explosion and blocking of reclosing function. In 2021, a 110kV substation experienced a 10-hour power outage because the 10kV photovoltaic system was not disconnected in time, and the fault arc lasted for 3 seconds. This not only seriously affected the reliability of power supply but also caused huge economic losses and negative social impacts.

[0003] Furthermore, the current design of the interlocking small power supply circuit in 110kV smart substations has flaws. The interlocking circuits between multiple smart protection devices and conventional 10kV line protection within the substation require the use of smart terminals to relay GOOSE messages and hard contacts. However, due to the lack of a unified standard, the implementation paths differ across substations, resulting in a chaotic distribution of hard contact plates for the interlocking circuits during equipment installation and varying cable laying paths. After the substation is put into operation, maintenance personnel face difficulties locating dozens of hard contact plates, making operation and maintenance extremely inconvenient and increasing maintenance costs and safety risks. Therefore, there is an urgent need to optimize the interlocking small power supply circuits and their control methods in smart substations. Summary of the Invention

[0004] This invention aims to overcome the shortcomings of existing technologies and provide an adaptive control method for switching small power sources in intelligent substations. By introducing an intelligent dynamic risk assessment and decision-making mechanism, it improves the safety and reliability of the power grid.

[0005] This invention is achieved through the following technical solution:

[0006] An adaptive control method for switching small power sources in smart substations includes:

[0007] The system collects and receives multi-source operation data from the power grid in real time. This multi-source operation data includes protection action information from protection devices, real-time electrical quantities of the power grid, operating status of small power sources, status monitoring data of primary equipment, and environmental data.

[0008] Based on the multi-source operation data, a dynamic risk assessment is performed through a preset intelligent control algorithm to generate a risk index characterizing the power grid fault risk level and related criteria.

[0009] Based on the risk index and related criteria, a joint switching control decision is generated and executed; wherein, the joint switching control decision is used to control the extended relay group to generate hard contact signals, so as to drive the corresponding conventional line protection equipment to perform a small power supply disconnection operation.

[0010] As an optimization, based on the multi-source operational data, the specific process of performing dynamic risk assessment through a preset intelligent control algorithm to generate a risk index characterizing the power grid fault risk level is as follows:

[0011] The multi-source operational data is compared and analyzed with the pre-built risk assessment model to obtain the risk index and its correlation basis, which are used to quantitatively characterize the degree of similarity between the current state of the power grid and the fault state.

[0012] As an optimization, the risk index, used to quantify the proximity between the current state of the power grid and the fault state, is obtained by comparing and analyzing the multi-source operational data with a pre-built risk assessment model. Specifically, this involves performing time-series analysis and multimodal fusion processing on the multi-source operational data to generate the risk index and its correlation basis. The specific process is as follows:

[0013] For continuously changing real-time electrical quantities of the power grid, time series analysis is performed to extract the changing trends, abrupt change points, and fluctuation characteristics of the real-time electrical quantities of the power grid, thereby obtaining the time series characteristics of the electrical quantities;

[0014] The electrical quantity timing characteristics are analyzed in a cross-modal manner with discrete protection action information, small power supply operating status, primary equipment status monitoring data and environmental data to identify the correlation basis between different data sources and generate the risk index.

[0015] As an optimization, the specific process of performing cross-modal correlation analysis on the timing characteristics of the electrical quantities with discrete protection action information, the operating status of small power supplies, the status monitoring data of primary equipment, and environmental data is as follows:

[0016] A hybrid intelligent analysis model is used to perform fusion analysis on multimodal data composed of the timing characteristics of electrical quantities, protection action information, operating status of small power supplies, status monitoring data of primary equipment, and environmental data;

[0017] The hybrid intelligent analysis model includes an expert model and a data-driven model. The expert model performs forward reasoning on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies a first set of data sources with strong causal associations supporting the first risk value. The data-driven model performs latent space analysis on the multimodal data through representation learning technology, outputs a quantified second risk value, and identifies a second set of data sources with consistent representations supporting the second risk value.

[0018] Collaborative decision-making is performed based on the outputs of the expert model and the data-driven model: when the absolute difference between the first risk value and the second risk value does not exceed the first preset threshold, it is determined that there is spatiotemporal coupling between the first data source set and the second data source set, and the weighted average of the first risk value and the second risk value is used as the risk index. At the same time, the intersection of the first data source set and the second data source set is used as the correlation basis to characterize the spatiotemporal coupling.

[0019] When the absolute difference exceeds the first preset threshold, a contradiction is determined to exist. At this time, the degree of difference between the first data source set and the second data source set is calculated, and a risk index is calculated simultaneously. If the degree of difference is lower than the second preset threshold, the risk index is the larger of the first risk value and the second risk value. If the degree of difference is not lower than the second preset threshold, the risk index is a preset constant value representing unknown high risk. At the same time, a divergence description vector containing the first risk value, the second risk value, and the degree of difference is generated as a correlation basis for characterizing the contradiction.

[0020] As an optimization, the difference is calculated as follows:

[0021] The contribution of each data source in the set of the expert model and the data-driven model to the final risk value is obtained respectively. For each data source in the union of the first data source set and the second data source set, the absolute value of the difference in contribution of the data sources in the union set in the expert model and the data-driven model is calculated. The difference is the sum or average of the absolute values ​​of the contribution differences of all data sources.

[0022] As an optimization, the expert model performs forward inference on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies the first set of data sources with strong causal associations supporting this first risk value. The specific process is as follows:

[0023] Based on the topology and electrical connections of the power grid, a power grid causal graph is pre-constructed, and the multimodal data collected in real time is mapped to the corresponding nodes in the power grid causal graph. The power grid causal graph is a directed graph, and the nodes of the directed graph represent electrical equipment, electrical quantity timing characteristics, protection action information, small power supply operating status, primary equipment status monitoring data and environmental data in the power grid. The directed edges of the directed graph represent the fault or disturbance propagation paths and causal logical relationships between nodes.

[0024] Starting from the top-level node representing the root cause of the fault, forward propagation calculation is performed according to the causal logic relationship defined by the directed edge, activating the affected downstream nodes layer by layer; wherein, the causal logic relationship includes limit violation judgment, logical AND, logical OR and probability propagation;

[0025] The first risk value is calculated by forward propagation and the maximum confidence level when the signal is finally transmitted to one or more preset terminal nodes representing specific severe consequences. The specific severe consequences include main transformer fault tripping, 110kV line fault tripping, or substation busbar loss of voltage.

[0026] Tracing back from one or more of the terminal nodes, all upstream nodes that were activated during this forward propagation calculation and whose contribution to the final risk value exceeds the contribution threshold constitute the first data source set.

[0027] As an optimization, the data-driven model performs latent space analysis on the multimodal data using representation learning techniques, outputs a quantified second risk value, and identifies a second set of data sources that support this second risk value and exhibit consistent representations. The specific process is as follows:

[0028] The multimodal data is jointly encoded using a pre-trained multimodal encoder to generate a state representation vector; the multimodal encoder includes a timing encoder for processing the timing features of electrical quantities, and a feature encoder for processing the protection action information, the operating status of the small power supply, the status monitoring data of the primary equipment, and the environmental data.

[0029] Calculate the mathematical distance between the current time step's state representation vector and a predefined baseline vector representing the historical normal operating mode of the power grid;

[0030] The mathematical distance is mapped to the second risk value through a preset monotonically increasing function;

[0031] The gradient of the second risk value with respect to each data dimension in the original multimodal data is calculated using the backpropagation method;

[0032] The original data source corresponding to the data dimension whose absolute gradient value exceeds the sensitivity threshold is identified as the second data source set.

[0033] As an optimization, the specific process for generating and executing joint control decisions based on risk indices and correlation criteria is as follows:

[0034] A solution strategy library is built based on historical experience. The solution strategy library pre-stores multiple solution strategies, and each solution strategy defines different combinations of solution objects, solution ratios and solution orders.

[0035] Based on the risk index and correlation criteria, the optimal resolution strategy is matched from the resolution strategy library through the strategy selector;

[0036] Generate and execute the joint switching control command based on the optimal unblocking strategy.

[0037] As an optimization, if the correlation basis is the intersection of data sources that characterize spatiotemporal coupling, then the small power sources involved in the intersection of the data sources are taken as the core dismantling target, and the strategy with the core dismantling target as the minimum and the dismantling ratio positively correlated with the risk index is selected as the optimal dismantling strategy.

[0038] If the correlation is based on a divergence description vector that characterizes contradiction:

[0039] When the risk index exceeds the third preset threshold and the difference is lower than the fourth preset threshold, a fast conservative strategy is selected as the optimal unblocking strategy. The fast conservative strategy is to target the 10kV small power source in the data source set corresponding to the larger of the first risk value and the second risk value recorded in the divergence description vector as the unblocking target, and issue unblocking instructions in parallel.

[0040] When the risk index exceeds the third preset threshold and the difference is not lower than the fourth preset threshold, the disconnection strategy is not selected, and instead the highest level of manual intervention request alarm is sent to the substation monitoring system.

[0041] As an optimization, the specific process for selecting a strategy where the unwinding ratio is positively correlated with the risk index is as follows:

[0042] A basic resolution ratio is determined; wherein the basic resolution ratio is calculated using the formula: Basic resolution ratio = (Current risk index - Lower limit of risk index) / (Upper limit of risk index - Lower limit of risk index); the upper and lower limits of the risk index are preset constants;

[0043] The power sources are sorted according to their current output, type, or grid connection priority. Starting with the power source with the lowest priority, they are sequentially added to the disconnection list until the total power of the power sources in the disconnection list reaches the basic disconnection ratio of the total system output. The disconnection strategy for this disconnection list is then taken as the optimal disconnection strategy.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0045] 1. It enhances the proactive defense capabilities of power grid security, achieving a leap from passive response to proactive early warning and precise decision-making. By integrating multi-source operational data and conducting dynamic risk assessment, this invention can identify risk signs before a fault occurs or in its early stages. In particular, when the hybrid intelligent analysis model reaches contradictory conclusions, the system can issue the highest level of warning, prompting maintenance personnel to pay attention to highly uncertain potential risks, thus gaining valuable time for manual intervention and pre-emptive handling, and effectively preventing serious accidents such as GIS bursting or station-wide power outages.

[0046] 2. This invention achieves precise and optimized control over power outages, maximizing power supply reliability. Traditional power outage logic is rigid, often resulting in a blanket, complete power cutoff. This invention, by generating correlation data, can accurately pinpoint the key sources of risk. Under consensus, the system uses the intersection of data sources as its core objective and achieves a dynamic positive correlation between the power cutoff ratio and the risk index, enabling on-demand power cutoffs. During decision-making, a power cutoff list is constructed through priority ranking, prioritizing the cutoff of less important power sources. This refined decision-making—regarding which power sources to cut, how many to cut, and in what order—minimizes the scope of power outages while ensuring safety, thus guaranteeing power supply to critical loads.

[0047] 3. This invention solves the inherent engineering challenges of existing connection / switching circuits, significantly reducing operation and maintenance costs and safety risks. Through software-defined intelligent algorithms, this invention replaces the traditional numerous, scattered, and messy hard-wired boards and complex hard-wired circuits. Maintenance personnel no longer need to deal with dozens of hard-to-find and operate hard-wired boards; they can simply configure the logic through a human-machine interface to enable or disable functions. This fundamentally solves the shortcomings of complex operation and error-prone operation in the background technology, greatly improving operation and maintenance efficiency and reducing human-caused safety risks.

[0048] 4. A highly reliable and interpretable intelligent decision-making system was constructed. A hybrid intelligent architecture combining expert models and data-driven models was adopted. The two models cross-validate each other based on different principles, avoiding the misjudgments or black-box problems that may exist with a single model. The system outputs not only a risk index but also the supporting evidence (such as the intersection of data sources or divergence description vectors), making the decision-making process transparent, traceable, and trustworthy, greatly enhancing the system's reliability in practical applications.

[0049] 5. This method provides a unified and standardized solution for the switching function of smart substations. It defines a complete set of processing procedures and data specifications, independent of the hardware implementation details of specific manufacturers. It can provide a unified and high-performance switching control paradigm for different substations, which is conducive to large-scale promotion and application, and overcomes the drawbacks of different implementation paths and inconsistent standards in existing technologies. Attached Figure Description

[0050] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a flowchart of an adaptive control method for switching small power sources in a smart substation, as described in this invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0053] The adaptive control method and system for switching small power sources in intelligent substations provided by this invention systematically solves the defects existing in the background technology from two levels: hardware architecture and control logic.

[0054] At the hardware level, by designing an integrated, interconnected small power supply cabinet, adopting a centralized hard-plate layout and optimized communication interfaces, the problems of complex circuits, scattered and messy hard-plates, and inconvenient operation and maintenance in existing technologies have been solved.

[0055] At the software and core logic level, this invention creatively introduces a dynamic risk assessment and strategic decision-making mechanism based on multi-source information fusion. It realizes flexible configuration and intelligent activation / deactivation of the connection function through software definition, replacing the traditional rigid and fixed hard-wired logic. This fundamentally improves the intelligence level and security of the system and provides a unified standard paradigm for such projects.

[0056] Next, the software and core logic will be explained through examples.

[0057] This embodiment 1 provides an adaptive control method for switching small power sources in a smart substation, such as... Figure 1 As shown, it includes:

[0058] S1. Real-time acquisition and reception of multi-source operation data of the power grid, including protection action information from protection equipment, real-time electrical quantities of the power grid, operating status of small power sources, status monitoring data of primary equipment, and environmental data;

[0059] S2. Based on the multi-source operation data, a dynamic risk assessment is performed using a preset intelligent control algorithm to generate a risk index characterizing the power grid fault risk level and related criteria.

[0060] S3. Generate and execute a joint switching control decision based on the risk index and related criteria; wherein, the joint switching control decision is used to control the extended relay group to generate hard contact signals to drive the corresponding conventional line protection equipment to perform a small power supply disconnection operation.

[0061] The core of this invention lies in a closed-loop control process. First, the system collects and receives multi-source operational data from the power grid in real time through various sensors and communication devices deployed within the station. This data transcends the electrical quantities relied upon by traditional protection systems, forming a comprehensive sensing system:

[0062] Protection action information: GOOSE messages from 110kV intelligent protection devices and 10kV conventional line protection, such as overcurrent and differential protection action signals.

[0063] Real-time electrical quantities of the power grid: including analog quantities such as voltage, current, frequency, and power, which are collected in real time by instrument transformers.

[0064] Operating status of small power sources: refers to the real-time output, power factor, and grid-connected / off-grid status of distributed photovoltaic, energy storage, and other power sources.

[0065] Primary equipment condition monitoring data: such as main transformer oil chromatography data, GIS partial discharge signals, equipment temperature, etc.

[0066] Environmental data: such as ambient temperature, humidity, thunderstorm activity, etc.

[0067] Subsequently, all this data is fed into a pre-set intelligent control algorithm for dynamic risk assessment. The output of this algorithm is not a simple "yes / no" signal, but a quantified risk index (e.g., a value between 0 and 100%, representing the probability or imminence of a failure) and a correlation basis (used to explain the source of risk or the uncertainty of the system).

[0068] Ultimately, the tripping control decision will be generated based on these two factors, and a hard contact signal will be generated by controlling the extended relay group to directly drive the 10kV conventional line protection device to perform the disconnection operation for the small power supply. This method centralizes the functions of the distributed hard-switch board into an intelligent algorithm, and greatly simplifies the circuit by defining its logic in software.

[0069] Next, we will introduce the implementation process of S2 and S3 in detail.

[0070] In some embodiments, the specific process of S2 is as follows:

[0071] The multi-source operational data is compared and analyzed with the pre-built risk assessment model to obtain the risk index and its correlation basis, which are used to quantitatively characterize the degree of similarity between the current state of the power grid and the fault state.

[0072] This is a dynamic risk assessment process, which essentially involves comparing and analyzing collected multi-source operational data with a pre-built risk assessment model. This risk assessment model is the brain of the system; by learning from historical data and the physical laws of the power grid, it can understand the complex mapping relationship between different data patterns and fault risks. The result of the comparative analysis is to generate a risk index that can quantitatively characterize the degree of proximity between the current state of the power grid and a fault state, as well as the correlation basis for explaining the causes of this index.

[0073] In some embodiments, comparing and analyzing the multi-source operational data with a pre-built risk assessment model to obtain the risk index used to quantify the proximity between the current state and the fault state of the power grid specifically involves performing time-series analysis and multimodal fusion processing on the multi-source operational data to generate the risk index and its correlation basis. The specific process is as follows:

[0074] S2.1 For the continuously changing real-time electrical quantities of the power grid, perform time series analysis to extract the changing trend, abrupt change points and fluctuation characteristics of the real-time electrical quantities of the power grid, and obtain the time series characteristics of the electrical quantities;

[0075] S2.2. Perform cross-modal correlation analysis on the electrical quantity timing characteristics with discrete protection action information, small power supply operating status, primary equipment status monitoring data and environmental data to identify the correlation basis between different data sources and generate the risk index.

[0076] The comparative analysis is specifically implemented through time-series analysis and multimodal fusion processing. This is a crucial data preprocessing and feature extraction step.

[0077] More specifically, time series analysis targets continuously changing real-time electrical quantities of the power grid (such as voltage and current waveforms). Algorithms (such as short-time Fourier transform, wavelet analysis, or recurrent neural networks) are used to extract their changing trends (whether they are steadily rising or falling), abrupt change points (whether there are peaks or drops), and fluctuation characteristics (whether the waveform is stable), thereby obtaining the time series characteristics of electrical quantities that reflect the dynamic process.

[0078] Multimodal fusion processing involves performing cross-modal correlation analysis on the aforementioned electrical quantity timing characteristics with discrete protection action information, small power supply operating status, primary equipment status monitoring data, and environmental data. Its purpose is to identify the inherent relationships (i.e., correlation criteria) between these different types of data. For example, it might discover a strong correlation between "main transformer oil temperature rise" and "increased harmonic current in adjacent lines," and ultimately synthesize all this information to generate the aforementioned risk index.

[0079] In some embodiments, the specific process of S2.2 is as follows:

[0080] S2.2.1 A hybrid intelligent analysis model is used to perform fusion analysis on the multimodal data composed of the electrical quantity timing characteristics, protection action information, small power supply operation status, primary equipment status monitoring data and environmental data;

[0081] The hybrid intelligent analysis model includes an expert model and a data-driven model. The expert model performs forward reasoning on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies a first set of data sources with strong causal associations supporting the first risk value. The data-driven model performs latent space analysis on the multimodal data through representation learning technology, outputs a quantified second risk value, and identifies a second set of data sources with consistent representations supporting the second risk value.

[0082] The expert model is based on a pre-built causal graph of the power grid. This graph is a directed graph where nodes represent various devices, state variables, and environmental factors, and edges represent fault propagation paths and causal logic (e.g., "overcurrent" leading to "temperature exceeding limits"). After receiving multimodal data, the model maps it to nodes in the graph and performs forward reasoning along the causal edges, starting from the root cause node of the fault. Finally, it outputs a quantified first risk value and identifies the key evidence chain leading to this risk, i.e., the first set of data sources.

[0083] In some embodiments, the process by which the expert model performs forward inference on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies a first set of data sources with strong causal associations supporting the first risk value is as follows:

[0084] A1. Based on the topology and electrical connections of the power grid, a power grid causal graph is pre-constructed, and the multimodal data collected in real time is mapped to the corresponding nodes in the power grid causal graph; wherein, the power grid causal graph is a directed graph, and the nodes of the directed graph represent electrical equipment, electrical quantity timing characteristics, protection action information, small power supply operating status, primary equipment status monitoring data and environmental data in the power grid, and the directed edges of the directed graph represent the fault or disturbance propagation paths and causal logical relationships between nodes;

[0085] A2. Starting from the top-level node representing the root cause of the fault, perform forward propagation calculations according to the causal logic relationship defined by the directed edge, and activate the affected downstream nodes layer by layer; wherein, the causal logic relationship includes limit violation judgment, logical AND, logical OR and probability propagation;

[0086] A3. The maximum confidence level when the data is finally transmitted to one or more preset terminal nodes representing specific severe consequences through forward propagation is taken as the first risk value; the specific severe consequences include main transformer fault tripping, 110kV line fault tripping, or substation busbar loss of voltage.

[0087] A4. Tracing back from one or more of the terminal nodes, all upstream nodes that were activated during this forward propagation calculation and whose contribution to the final risk value exceeds the contribution threshold constitute the first data source set.

[0088] The specific workflow of the expert model is as follows:

[0089] Modeling and Mapping: Based on the power grid topology and electrical relationships, a cause-effect graph of the power grid is pre-drawn. During runtime, real-time data (such as main transformer oil temperature: 65℃) is mapped to the corresponding nodes in the graph.

[0090] Forward reasoning: Starting from the root cause of the top-level fault (such as lightning strike), calculations are performed according to the logic defined by the directed edge (over-limit judgment: oil temperature > 75℃? Logical AND: overcurrent and protection fails to operate? Logical OR: short circuit or insulation breakdown? Probability propagation: confidence is passed based on probability) to activate downstream nodes layer by layer.

[0091] Risk value calculation: The reasoning is ultimately passed to the terminal node representing the serious consequences (such as a main transformer failure trip). The highest confidence value among all activated terminal nodes is taken as the first risk value.

[0092] Evidence tracing: By tracing back from these high-risk terminal nodes, we can find all upstream nodes that were activated in this reasoning and whose contribution exceeds the contribution threshold. These nodes constitute the first set of data sources supporting the conclusion.

[0093] Data-driven models do not rely on pre-defined rules. Instead, they project multimodal data into a latent space using representation learning techniques (such as deep learning) to learn its inherent statistical regularities. By calculating the deviation of the current state from historical normal states, it outputs a quantified second risk value and identifies the set of second data sources that contribute most to the current anomaly by analyzing the model's internal gradients.

[0094] In some embodiments, the data-driven model performs latent space analysis on the multimodal data using representation learning techniques, outputs a quantified second risk value, and identifies a second set of data sources that support the second risk value and exhibit consistent representations. The specific process is as follows:

[0095] B1. The multimodal data is jointly encoded using a pre-trained multimodal encoder to generate a state representation vector; the multimodal encoder includes a timing encoder for processing the timing features of electrical quantities, and a feature encoder for processing the protection action information, the operating status of the small power supply, the status monitoring data of the primary equipment, and the environmental data.

[0096] B2. Calculate the mathematical distance between the current time step's state representation vector and the predefined baseline vector representing the historical normal operation mode of the power grid;

[0097] B3. Map the mathematical distance to the second risk value using a preset monotonically increasing function;

[0098] B4. Calculate the gradient of the second risk value with respect to each data dimension in the original multimodal data using the backpropagation method;

[0099] B5. Identify the original data source corresponding to the data dimension whose absolute gradient value exceeds the sensitivity threshold as the second data source set.

[0100] The specific workflow of the data-driven model is as follows:

[0101] Joint encoding: Using a pre-trained multimodal encoder (containing a temporal encoder for processing time-series data and a feature encoder for processing discrete data) to transform multimodal data into a unified state representation vector.

[0102] Distance calculation: Calculate the mathematical distance (such as Euclidean distance) between the state representation vector and the pre-stored reference vector representing the historical normal state.

[0103] Risk value mapping: Distance is mapped to a second risk value using a monotonically increasing function (such as the Sigmoid function). The greater the distance, the higher the risk value.

[0104] Evidence identification: The gradient of the second risk value with respect to the original input data is calculated using backpropagation. Data dimensions with larger absolute gradient values ​​contribute more to the current high risk. The original data sources corresponding to data dimensions whose gradients exceed the sensitivity threshold are identified as the second data source set.

[0105] S2.2.2, Make collaborative decisions based on the outputs of the expert model and the data-driven model:

[0106] S2.2.2.1 When the absolute difference between the first risk value and the second risk value does not exceed the first preset threshold, it is determined that there is spatiotemporal coupling between the first data source set and the second data source set. The weighted average of the first risk value and the second risk value is used as the risk index. At the same time, the intersection of the first data source set and the second data source set is used as the correlation basis to characterize the spatiotemporal coupling.

[0107] S2.2.2.2 When the absolute difference exceeds the first preset threshold, a contradiction is determined to exist. At this time, the degree of difference between the first data source set and the second data source set is calculated, and a risk index is calculated simultaneously. If the degree of difference is lower than the second preset threshold, the risk index is the larger of the first risk value and the second risk value. If the degree of difference is not lower than the second preset threshold, the risk index is a preset constant value representing unknown high risk. At the same time, a divergence description vector containing the first risk value, the second risk value, and the degree of difference is generated as a correlation basis for characterizing the contradiction.

[0108] Collaborative decision-making is the core of hybrid intelligence. The system compares the risk values ​​output by two models. If the absolute difference does not exceed a first preset threshold (e.g., 10%), the models are considered to have reached a consensus, indicating spatiotemporal coupling. In this case, the risk index is the weighted average of the two models, and the correlation basis is the intersection of their key evidence, which represents the most reliable conclusion. If the absolute difference exceeds the threshold, a contradiction is identified. In this case, the system calculates the degree of difference between the two data source sets to quantify the magnitude of the disagreement. If the degree of difference is low, it indicates similar basis but different conclusions, and the risk index is set to the larger of the two (a conservative principle). If the degree of difference is high, it indicates that the basis and conclusion are completely different, and the system falls into high uncertainty, so the risk index is set to a constant value representing unknown high risk (e.g., 95%). Simultaneously, a disagreement description vector is generated as the correlation basis, recording the viewpoints of both parties for advanced analysis.

[0109] In some embodiments, the difference is calculated as follows:

[0110] The contribution of each data source in the set of the expert model and the data-driven model to the final risk value is obtained respectively. For each data source in the union of the first data source set and the second data source set, the absolute value of the difference in contribution of the data sources in the union set in the expert model and the data-driven model is calculated. The difference is the sum or average of the absolute values ​​of the contribution differences of all data sources.

[0111] The discrepancy calculation aims to quantify the disagreement between the two models regarding their "judgment criteria." Specifically, the contribution of each data source to its respective final risk value is obtained from both the expert model and the data-driven model (e.g., causal strength in the expert model, gradient magnitude in the data-driven model). Then, for each data source in the union of the two data source sets, the absolute value of the difference in its contribution between the two models is calculated. Finally, the absolute values ​​of the contribution differences of all data sources are summed or averaged to obtain the final discrepancy. The larger this value, the more inconsistent the judgment logic of the two models.

[0112] In some embodiments, S3, the specific process of generating and executing the joint control decision based on the risk index and correlation criteria is as follows:

[0113] S3.1 Construct a solution strategy library based on historical experience, wherein the solution strategy library pre-stores multiple solution strategies, and each solution strategy defines different combinations of solution objects, solution ratios and solution order;

[0114] S3.2 Based on the risk index and correlation criteria, the optimal resolution strategy is matched from the resolution strategy library using a strategy selector;

[0115] S3.3 Generate and execute the joint switching control command according to the optimal disengagement strategy.

[0116] The process of generating joint switching control decisions is a policy-based intelligent selection. First, a disconnection policy library is pre-built based on historical experience and simulations. Each disconnection policy in the library clearly defines the disconnection object combination (which power supplies to disconnect), the disconnection ratio (how much capacity to disconnect), and the disconnection order (simultaneous or sequential disconnection). Then, a policy selector matches the optimal disconnection policy best suited for the current situation from this policy library based on the specific risk index and related criteria. Finally, the system executes the policy, generating and issuing specific joint switching control commands.

[0117] In some embodiments, the specific process of S3.2 is as follows:

[0118] S3.2.1 If the correlation basis is the intersection of data sources that characterize spatiotemporal coupling, then the small power source involved in the intersection of the data sources is taken as the core dismantling target, and the strategy that takes the core dismantling target as the minimum and whose dismantling ratio is positively correlated with the risk index is selected as the optimal dismantling strategy.

[0119] S3.2.2 If the correlation basis is a divergence description vector representing contradiction:

[0120] S3.2.2.1 When the risk index exceeds the third preset threshold and the difference is lower than the fourth preset threshold, a fast conservative strategy is selected as the optimal unblocking strategy. The fast conservative strategy is to target the 10kV small power source in the data source set corresponding to the larger of the first risk value and the second risk value recorded in the divergence description vector as the unblocking target, and issue unblocking instructions in parallel.

[0121] S3.2.2.2 When the risk index exceeds the third preset threshold and the difference is not lower than the fourth preset threshold, the disconnection strategy is not selected, and instead the highest level of manual intervention request alarm is sent to the substation monitoring system.

[0122] The matching rules for the strategy selector are as follows:

[0123] If the correlation is based on the intersection of data sources (consensus state), then the small power sources within the intersection are the core targets that must be removed. Based on this, a strategy is selected where the removal ratio is positively correlated with the risk index; the higher the risk, the larger the removal scope.

[0124] If the correlation is based on a divergent description vector (contradictory state):

[0125] When the risk index is high and the difference is low, a fast and conservative strategy is adopted. This strategy targets the set of data sources corresponding to larger risk values ​​and issues instructions in a simultaneous and parallel manner to eliminate the most significant risks as quickly as possible.

[0126] When the risk index is high and the degree of variation is high, the system considers the uncertainty to be too great, and automatic decision-making is too risky. Therefore, it does not select any resolution strategy, but instead sends the highest-level manual intervention request alarm to the monitoring center, leaving the final decision-making power to the operations and maintenance personnel.

[0127] In some embodiments, in S3.2.1, the specific process of selecting a strategy whose dissolution ratio is positively correlated with the risk index is as follows:

[0128] S3.2.1.1 Determine a basic unwinding ratio; wherein, the basic unwinding ratio is calculated using the formula: Basic unwinding ratio = (Current risk index - Lower limit of risk index) / (Upper limit of risk index - Lower limit of risk index); the upper and lower limits of the risk index are preset constants;

[0129] S3.2.1.2 Sort each small power source according to its current output, type, or grid connection priority, and add it to the disconnection list in sequence starting from the small power source with the lowest priority, until the total power of the small power sources included in the disconnection list reaches the basic disconnection ratio of the total system output. The disconnection strategy for this disconnection list is taken as the optimal disconnection strategy.

[0130] This process is a quantitative and fair screening mechanism:

[0131] Calculate the basic ratio: The basic ratio is calculated using the formula: Basic Ratio = (Current Risk Index - Lower Limit of Risk Index) / (Upper Limit of Risk Index - Lower Limit of Risk Index). For example, if the lower limit is 0.2 and the upper limit is 0.9, when the risk index is 0.7, the basic ratio is approximately 71.4% (0.7 - 0.2) / (0.9 - 0.2).

[0132] Generating the disconnection list: The system acquires information on all small power sources and sorts them according to current output, type, or grid connection priority (lowest priority is listed first). Then, starting from the beginning of the list (lowest priority), small power sources are added to the disconnection list sequentially, and the total power of the power sources in the list is calculated in real time. When the total power reaches or just exceeds the system's total output multiplied by the basic disconnection ratio, addition is stopped. The list obtained at this point is the final disconnection list.

[0133] Determine the optimal strategy: Determine the cut-off scheme for this list of disconnectors (including the power sources in the list and the predefined simultaneous or sequential cut-off order) as the current optimal disconnector strategy.

[0134] Suppose that a 10kV busbar of a certain 110kV smart substation is connected to four small power sources: photovoltaic A (1.5MW, low priority), photovoltaic B (1.0MW, low priority), energy storage C (2.0MW, medium priority), and small hydropower D (2.5MW, high priority), with a total output of 7MW.

[0135] Scenario 1: Model consensus, precise resection

[0136] Risk Assessment: The system detected a sudden increase in current on the 110kV line, and online oil chromatography monitoring of the main transformer showed excessive acetylene levels. The expert model deduced a path of "line fault leading to transformer insulation damage," outputting a first risk value of 85%. The data-driven model also found a significant deviation from the normal baseline in the potential space, outputting a second risk value of 80%. The difference of 5% between the two values ​​is less than the threshold of 10%, reaching a consensus.

[0137] Decision-making and execution: The risk index is taken as an average of 82.5%. The correlation is based on the intersection of the two models' data sources {110kV line current, main transformer oil chromatography}. The core target of the intersection is the main transformer, and the associated 10kV small power sources are all four. Based on this, the strategy selector selects a strategy with a disconnection ratio positively correlated with the risk index. The calculated basic disconnection ratio is (0.825-0.2) / (0.9-0.2)≈89%. The system prioritizes (PV A→PV B→Energy Storage C→Hydropower D), starting from the lowest priority, until {PV A, PV B, Energy Storage C} is selected, with a total power of 4.5MW, accounting for 64% of the total. Although it does not reach 89% (due to the limitation of the ratio levels in the strategy library, the closest level covering the core target is selected), it can effectively reduce the load on the main transformer. The system generates instructions to disconnect these three power sources simultaneously, successfully preventing the main transformer fault from escalating and saving the most important hydropower D.

[0138] Scenario 2: Model divergence, conservative warning

[0139] Risk Assessment: During thunderstorms, environmental data is abnormal, but all electrical quantities and equipment status data are normal. The expert model, based on a "lightning strike" causal path, outputs a high-risk value of 90%. The data-driven model, due to normal electrical quantities, outputs a low-risk value of 15%. The difference between the two is 75%, far exceeding the threshold, and the calculated discrepancy is extremely high (because the two rely on completely different data sources).

[0140] Decision-making and execution: The system determines that there is a contradiction. The risk index is set to a preset unknown high-risk value of 95%. The correlation is based on a divergence description vector that records two extreme risk values ​​and a high degree of difference.

[0141] Due to the high degree of discrepancy, the strategy selector did not execute automatic switching. Instead, it immediately sent a "highest level uncertainty alarm" to the monitoring center, indicating that "there is a fundamental discrepancy between the expert model and the data-driven model in their judgment of lightning strike risk; manual assessment is recommended." After the maintenance personnel intervened, they confirmed from the radar chart that the thunderstorm cloud had deviated, avoiding erroneous switching caused by misjudgment from a single model and ensuring power supply.

[0142] The specific details at the hardware level are as follows:

[0143] (I) Overall Design Concept

[0144] To address the problems existing in the current interlocking circuits, this solution adopts a centralized control and modular design concept. It designs a unified and efficient interlocking small power supply circuit scheme, simplifies the circuit structure, reduces the number of intelligent terminals used, and centrally arranges 48 interlocking circuits and output hard pressure plates, enhancing circuit flexibility. This allows for the independent selection of enabling or disabling of any upstream intelligent protection device for interlocking with downstream conventional equipment, while simultaneously reducing production costs and construction time.

[0145] (II) Hardware Architecture Design

[0146] Interlocking Mini Power Supply Cabinet: A new type of interlocking mini power supply cabinet is designed as the core equipment, integrating a high-performance intelligent control unit, an expansion relay group, and a centralized hard-plate module. The intelligent control unit has powerful data processing and logic judgment capabilities, and can quickly receive and process signals sent by protection equipment; the expansion relay group is used to realize reliable signal conversion and transmission; the centralized hard-plate module centrally arranges the output hard-plates of 48 interlocking circuits for convenient operation by maintenance personnel.

[0147] Communication interface optimization: The communication interface between intelligent protection devices, interlocking small power supply cabinets, and conventional 10kV line protection is optimized. A standardized communication protocol is adopted to reduce the number of relay links in intelligent terminals, directly achieving the conversion between GOOSE messages and hard contacts, reducing loop complexity, and improving the accuracy and reliability of signal transmission.

[0148] 1. Intelligent control unit

[0149] Central Processing Unit (CPU): A high-performance industrial-grade ARM processor, STM32H7 series, is selected. It has high-speed data processing capabilities and rich peripheral interfaces, which can quickly process the signals sent by the protection device and execute intelligent control algorithms to meet real-time requirements.

[0150] Memory: Includes flash memory and random access memory (RAM). Flash memory is used to store intelligent control algorithm programs, system configuration parameters, etc., with a capacity of not less than 128MB; RAM is used for data storage and processing during program execution, with a capacity of not less than 64MB.

[0151] Communication interface circuit: integrates fiber optic interface and Ethernet interface. The fiber optic interface uses an optical module conforming to the IEC 61850 standard to achieve high-speed and reliable communication with 110kV intelligent protection equipment and 10kV conventional line protection; the Ethernet interface is used for data exchange with substation monitoring systems or other equipment.

[0152] Signal acquisition circuit: High-precision voltage and current transformers are used in conjunction with A / D conversion chips to acquire voltage, current and other signals of the power grid operation status in real time, convert them into digital signals and then process them by the CPU.

[0153] 2. Extended relay group

[0154] Electromagnetic relays: High-reliability electromagnetic relays are selected, which have high contact capacity and good electrical isolation performance. They are used to realize the conversion and transmission of high current and high voltage signals, and ensure the reliable execution of switching signals.

[0155] Solid-state relays: Solid-state relays enable contactless control and have advantages such as fast response speed, long lifespan, and strong anti-interference ability. They are suitable for high-frequency signal switching and control scenarios.

[0156] 3. Centralized hard plate module

[0157] Pressure plate base: Made of flame-retardant and insulating engineering plastic material, with a standardized installation structure for easy connection to the internal circuit of the integrated power supply cabinet, ensuring stable installation of the hard pressure plates at the outlets of the 48 integrated circuits.

[0158] Pressure plate connecting piece: Made of high-quality metal material, it has good conductivity and mechanical strength, and realizes the connection and disconnection control of the circuit through plugging and unplugging operation.

[0159] 3. Communication interface devices

[0160] Fiber optic interface: Uses a single-mode fiber optic transceiver module that conforms to power industry standards, with a transmission rate of no less than 1Gbps, to meet the high-speed transmission requirements of GOOSE messages and ensure the accuracy and real-time performance of signal transmission.

[0161] Ethernet interface: Employs industrial-grade Ethernet chips and network transformers, supporting TCP / IP protocol to enable network communication with other equipment within the substation, facilitating data exchange and remote management.

[0162] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for adaptive control of switching small power sources in intelligent substations, characterized in that, include: The system collects and receives multi-source operation data from the power grid in real time. This multi-source operation data includes protection action information from protection devices, real-time electrical quantities of the power grid, operating status of small power sources, status monitoring data of primary equipment, and environmental data. Based on the multi-source operation data, a dynamic risk assessment is performed through a preset intelligent control algorithm to generate a risk index characterizing the power grid fault risk level and related criteria. Based on the risk index and related criteria, a joint switching control decision is generated and executed; wherein, the joint switching control decision is used to control the extended relay group to generate hard contact signals, so as to drive the corresponding conventional line protection equipment to perform a small power supply disconnection operation.

2. The adaptive control method for switching small power supplies in a smart substation according to claim 1, characterized in that, Based on the aforementioned multi-source operational data, the specific process of generating a risk index characterizing the power grid fault risk level through dynamic risk assessment using a preset intelligent control algorithm is as follows: The multi-source operational data is compared and analyzed with the pre-built risk assessment model to obtain the risk index and its correlation basis, which are used to quantitatively characterize the degree of similarity between the current state of the power grid and the fault state.

3. The adaptive control method for switching small power sources in a smart substation according to claim 2, characterized in that, The risk index, used to quantify the proximity between the current state of the power grid and a fault state, is obtained by comparing and analyzing the multi-source operational data with a pre-built risk assessment model. Specifically, this involves performing time-series analysis and multimodal fusion processing on the multi-source operational data to generate the risk index and its correlation basis. For continuously changing real-time electrical quantities of the power grid, time series analysis is performed to extract the changing trends, abrupt change points, and fluctuation characteristics of the real-time electrical quantities of the power grid, thereby obtaining the time series characteristics of the electrical quantities; The electrical quantity timing characteristics are analyzed in a cross-modal manner with discrete protection action information, small power supply operating status, primary equipment status monitoring data and environmental data to identify the correlation basis between different data sources and generate the risk index.

4. The adaptive control method for switching small power supplies in a smart substation according to claim 3, characterized in that, The specific process of performing cross-modal correlation analysis on the aforementioned electrical quantity timing characteristics with discrete protection action information, small power supply operating status, primary equipment status monitoring data, and environmental data is as follows: A hybrid intelligent analysis model is used to perform fusion analysis on multimodal data composed of the timing characteristics of electrical quantities, protection action information, operating status of small power supplies, status monitoring data of primary equipment, and environmental data; The hybrid intelligent analysis model includes an expert model and a data-driven model. The expert model performs forward reasoning on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies a first set of data sources with strong causal associations supporting the first risk value. The data-driven model performs latent space analysis on the multimodal data through representation learning technology, outputs a quantified second risk value, and identifies a second set of data sources with consistent representations supporting the second risk value. Collaborative decision-making is performed based on the outputs of the expert model and the data-driven model: when the absolute difference between the first risk value and the second risk value does not exceed the first preset threshold, it is determined that there is spatiotemporal coupling between the first data source set and the second data source set, and the weighted average of the first risk value and the second risk value is used as the risk index. At the same time, the intersection of the first data source set and the second data source set is used as the correlation basis to characterize the spatiotemporal coupling. When the absolute difference exceeds the first preset threshold, a contradiction is determined to exist. At this time, the degree of difference between the first data source set and the second data source set is calculated, and a risk index is calculated simultaneously. If the degree of difference is lower than the second preset threshold, the risk index is the larger of the first risk value and the second risk value. If the degree of difference is not lower than the second preset threshold, the risk index is a preset constant value representing unknown high risk. At the same time, a divergence description vector containing the first risk value, the second risk value, and the degree of difference is generated as a correlation basis for characterizing the contradiction.

5. The adaptive control method for switching small power supplies in a smart substation according to claim 4, characterized in that, The difference is calculated as follows: The contribution of each data source in the set of the expert model and the data-driven model to the final risk value is obtained respectively. For each data source in the union of the first data source set and the second data source set, the absolute value of the difference in contribution of the data sources in the union set in the expert model and the data-driven model is calculated. The difference is the sum or average of the absolute values ​​of the contribution differences of all data sources.

6. The adaptive control method for switching small power sources in a smart substation according to claim 4, characterized in that, The expert model performs forward inference on the multimodal data based on a pre-built power grid causal graph, outputs a quantified first risk value, and identifies the first set of data sources with strong causal associations supporting this first risk value. The specific process is as follows: Based on the topology and electrical connections of the power grid, a power grid causal graph is pre-constructed, and the multimodal data collected in real time is mapped to the corresponding nodes in the power grid causal graph. The power grid causal graph is a directed graph, and the nodes of the directed graph represent electrical equipment, electrical quantity timing characteristics, protection action information, small power supply operating status, primary equipment status monitoring data and environmental data in the power grid. The directed edges of the directed graph represent the fault or disturbance propagation paths and causal logical relationships between nodes. Starting from the top-level node representing the root cause of the fault, forward propagation calculation is performed according to the causal logic relationship defined by the directed edge, activating the affected downstream nodes layer by layer; wherein, the causal logic relationship includes limit violation judgment, logical AND, logical OR and probability propagation; The first risk value is calculated by forward propagation and the maximum confidence level when the signal is finally transmitted to one or more preset terminal nodes representing specific severe consequences. The specific severe consequences include main transformer fault tripping, 110kV line fault tripping, or substation busbar loss of voltage. Tracing back from one or more of the terminal nodes, all upstream nodes that were activated during this forward propagation calculation and whose contribution to the final risk value exceeds the contribution threshold constitute the first data source set.

7. The adaptive control method for switching small power sources in a smart substation according to claim 4, characterized in that, The data-driven model uses representation learning techniques to perform latent space analysis on the multimodal data, outputs a quantified second risk value, and identifies a second set of data sources that support this second risk value and exhibit consistent representations. The specific process is as follows: The multimodal data is jointly encoded using a pre-trained multimodal encoder to generate a state representation vector; the multimodal encoder includes a timing encoder for processing the timing features of electrical quantities, and a feature encoder for processing the protection action information, the operating status of the small power supply, the status monitoring data of the primary equipment, and the environmental data. Calculate the mathematical distance between the current time step's state representation vector and a predefined baseline vector representing the historical normal operating mode of the power grid; The mathematical distance is mapped to the second risk value through a preset monotonically increasing function; The gradient of the second risk value with respect to each data dimension in the original multimodal data is calculated using the backpropagation method; The original data source corresponding to the data dimension whose absolute gradient value exceeds the sensitivity threshold is identified as the second data source set.

8. The adaptive control method for switching small power sources in a smart substation according to claim 1, characterized in that, The specific process of generating and executing joint control decisions based on risk indices and correlation criteria is as follows: A solution strategy library is built based on historical experience. The solution strategy library pre-stores multiple solution strategies, and each solution strategy defines different combinations of solution objects, solution ratios and solution orders. Based on the risk index and correlation criteria, the optimal resolution strategy is matched from the resolution strategy library through the strategy selector; Generate and execute the joint switching control command based on the optimal unblocking strategy.

9. The adaptive control method for switching small power sources in a smart substation according to claim 8, characterized in that, If the correlation basis is the intersection of data sources that characterize spatiotemporal coupling, then the small power sources involved in the intersection of the data sources are taken as the core dismantling target, and the strategy that takes the core dismantling target as the minimum and whose dismantling ratio is positively correlated with the risk index is selected as the optimal dismantling strategy. If the correlation is based on a divergence description vector that characterizes contradiction: When the risk index exceeds the third preset threshold and the difference is lower than the fourth preset threshold, a fast conservative strategy is selected as the optimal unblocking strategy. The fast conservative strategy is to target the 10kV small power source in the data source set corresponding to the larger of the first risk value and the second risk value recorded in the divergence description vector as the unblocking target, and issue unblocking instructions in parallel. When the risk index exceeds the third preset threshold and the difference is not lower than the fourth preset threshold, the disconnection strategy is not selected, and instead the highest level of manual intervention request alarm is sent to the substation monitoring system.

10. The adaptive control method for switching small power supplies in a smart substation according to claim 9, characterized in that, The specific process for selecting a strategy where the dissolution ratio is positively correlated with the risk index is as follows: A basic resolution ratio is determined; wherein the basic resolution ratio is calculated using the formula: Basic resolution ratio = (Current risk index - Lower limit of risk index) / (Upper limit of risk index - Lower limit of risk index); the upper and lower limits of the risk index are preset constants; The power sources are sorted according to their current output, type, or grid connection priority. Starting with the power source with the lowest priority, they are sequentially added to the disconnection list until the total power of the power sources in the disconnection list reaches the basic disconnection ratio of the total system output. The disconnection strategy for this disconnection list is then taken as the optimal disconnection strategy.

Images