A power risk probability prediction method and a terminal

Abstract

The application discloses a power risk probability prediction method and a terminal, obtains historical operation data and historical extreme events of a power system, extracts risk data samples from the historical extreme events, presets a mapping rule, maps the risk data samples into operation disturbance variables acting on system operation data through the mapping rule, creates a risk probability prediction model with the power system operation data as input and the risk probability as output, trains the risk probability prediction model according to the historical operation data and the operation disturbance variables, obtains current operation data of the power system, substitutes the current operation data into the trained risk probability prediction model, and obtains an expected risk probability of the power system within a preset time in the future. The application improves the short-term risk identification accuracy and robustness, enhances the explainability of the risk output, and guarantees the safe operation of the power system.

Description

Technical Field

[0001] This invention relates to the field of power system safety early warning and risk assessment technology, and in particular to a power risk probability prediction method and terminal. Background Technology

[0002] Against the backdrop of high-proportion renewable energy integration, frequent extreme weather events, and increased volatility in the external energy market and supply chain, the power system's supply and demand security risks exhibit characteristics of "high suddenness, multiple coupling channels, and high timeliness requirements." Therefore, short-term early warning for the power system is becoming increasingly important.

[0003] However, existing short-term early warning methods mostly employ fixed thresholds or single statistical models. While these can characterize risks to some extent, they struggle to simultaneously meet engineering requirements such as cross-seasonal adaptability, robustness to extreme samples, and interpretability and auditability. Furthermore, although purely data-driven models can learn complex nonlinear relationships, they are prone to false alarms / missed alarms when there are significant abnormal disturbances in the power system, and the causes of the outputs are difficult to trace. Rule systems that rely entirely on expert experience are also unable to cover novel complex events and multi-source coupled transmission paths.

[0004] Therefore, when facing security risks such as supply and demand imbalances on an hourly to intraday scale, it is particularly important to improve the accuracy and robustness of short-term risk identification. Only by quickly and accurately completing risk warnings and outputting risk source tracing can timely response strategies be formulated to ensure the safe operation of the power system. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a power risk probability prediction method and terminal that improves the accuracy and robustness of short-term risk identification, enhances the interpretability of risk output, and ensures the safe operation of the power system.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for predicting the probability of power risk includes the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system operation data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

[0007] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A power risk probability prediction terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it performs the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system operation data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

[0008] The beneficial effects of this invention are as follows: It provides a power risk probability prediction method and terminal, which acquires historical operating data and historical extreme events of the power system, extracts risk data samples from the historical extreme events, and maps the risk data samples into operating disturbance variables that act on the system operating data through mapping rules. That is, the uncertainty of historical extreme events is quantified into operating disturbance variables that can be calculated and learned, and applied to the model input and model training of the risk probability prediction model. Then, it acquires the current operating data of the power system, substitutes the current operating data into the trained risk probability prediction model, and obtains the expected risk probability of the power system in the future within a preset time. It fully considers the influencing factors of extreme events and is suitable for hourly to intraday scale supply and demand imbalance risk early warning, supply guarantee decision support, and risk source location and disposal linkage. It improves the accuracy and robustness of short-term risk identification, enhances the interpretability of risk output, and ensures the safe operation of the power system. Attached Figure Description

[0009] Figure 1 This is a schematic diagram illustrating the steps of a power risk probability prediction method according to the present invention; Figure 2 This is a flowchart of a power risk probability prediction method according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the risk prediction output when a power risk probability prediction method according to an embodiment of the present invention is applied to a provincial power system. Figure 4 This is a system block diagram of a power risk probability prediction terminal according to an embodiment of the present invention.

[0010] Label Explanation: 1. A power risk probability prediction terminal; 2. Memory; 3. Processor. Detailed Implementation

[0011] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0012] To facilitate understanding of the technical content of this invention, some of the technical features involved in this invention are explained and described, as shown in Table 1: Table 1 Explanation of Technical Features

[0013] Please refer to Figure 1 A method for predicting the probability of power risk includes the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system operation data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

[0014] As can be seen from the above description, the beneficial effects of the present invention are as follows: It acquires historical operating data and historical extreme events of the power system, extracts risk data samples from these events, and maps these risk data samples to operating disturbance variables that affect the system's operating data through mapping rules. This quantifies the uncertainty of historical extreme events into calculable and learnable operating disturbance variables, which are then applied to the model input and training of the risk probability prediction model. Subsequently, it acquires the current operating data of the power system, substitutes this data into the trained risk probability prediction model, and obtains the expected risk probability of the power system within a preset future timeframe. This fully considers the influencing factors of extreme events and is suitable for hourly to intraday scale supply-demand imbalance risk warnings, supply guarantee decision support, and risk source location and handling linkage. It improves the accuracy and robustness of short-term risk identification while enhancing the interpretability of risk output, thus ensuring the safe operation of the power system.

[0015] Furthermore, observable transmission indicators and system operation indicators corresponding to the power system are preset; From the risk data sample, the transmission disturbance variables corresponding to the observable transmission indicators are extracted. The transmission disturbance variables include disturbance intensity parameters, disturbance duration, disturbance direction, and disturbance type. Based on the observable transmission index and the transmission disturbance variable, establish the operation disturbance variable corresponding to the system operation index.

[0016] As described above, by clearly defining the correspondence between observable transmission indicators and system operation indicators, and accurately extracting transmission disturbance variables that include disturbance intensity, duration, direction, and type, the construction of operational disturbance variables forms a complete mapping chain of "event characteristics - transmission path - system impact." This not only ensures the accuracy of quantifying the uncertainty of extreme events but also clearly restores the mechanism by which extreme events affect the operation of the power system, further improving the effectiveness of model training and the accuracy of risk prediction.

[0017] Furthermore, a scenario description item corresponding to the power system is preset; Based on the content of the risk data sample, set the corresponding scenario description item.

[0018] As can be seen from the above description, the addition of the scenario description item supplements the scenario-based information of the risk data sample, making the mapping of operational disturbance variables more comprehensive and three-dimensional, and is used to cover scarce extreme combinations and narrow the scope of influencing factor investigation.

[0019] Furthermore, the risk data samples extracted from the aforementioned historical extreme events include: Pre-set reserve ratio indicators and supply-demand gap indicators; Calculate and determine whether the power system's reserve ratio in the historical extreme event meets the first preset condition or whether the supply-demand gap indicator meets the second preset condition; If so, then the historical extreme event is set as the risk data sample.

[0020] As described above, by using reserve ratio and supply-demand gap, two core indicators for the safe operation of the power system, as the screening criteria, an objective and clear admission standard for risk data samples was established. This ensures that the selected samples are all key extreme events that directly threaten the supply-demand balance and safe supply of the power system, effectively eliminating redundant extreme event data that are unrelated to the core operational risks of the system, and improving the quality and relevance of risk data samples.

[0021] Furthermore, it also includes: The characteristic contribution value of the expected risk probability is calculated using the risk probability prediction model. The feature contribution value represents the ability of each input feature of the power system operation data to influence the model output.

[0022] As described above, the calculation of feature contribution values ​​enables "reverse tracing" of risk prediction results. It can quantify the impact weight of each operational data feature on the risk probability, clearly locate the key factors that trigger system risks, and accurately trace back to specific extreme events.

[0023] Furthermore, it also includes: Pre-set target variables corresponding to the power system and risk assessment conditions corresponding to the target variables; Obtain change cases of the target variable from the historical operational data; The operating data of the power system corresponding to the change cases that meet the risk assessment conditions are selected as new risk data samples.

[0024] As described above, by pre-setting target variables and risk assessment conditions, the source of risk data samples is expanded from "historical extreme events" to "potential risk cases in regular historical operating data," which effectively makes up for the scarcity of extreme event samples and enriches the diversity and coverage of risk samples.

[0025] Furthermore, it also includes: At least two risk warning levels are preset, and each risk warning level corresponds one-to-one with a risk probability of different sizes; Based on the expected risk probability, determine the risk warning level that is suitable for the power system; Based on the risk warning level, the corresponding response strategy is triggered.

[0026] As described above, maintenance personnel can quickly match corresponding response measures according to the warning level, which can significantly shorten the decision-making time and execution cycle of risk handling, and improve the power system's ability to respond quickly to short-term risks and the efficiency of emergency response.

[0027] Furthermore, it also includes: Obtain the historical risk probabilities generated by the risk probability prediction model; Based on the historical variation pattern of risk probabilities, a range of risk probabilities is set for each risk warning level.

[0028] As can be seen from the above description, dynamically setting the warning level threshold based on the changing patterns of historical risk probabilities replaces the subjective experience-based setting method, making the probability division of each warning level more in line with the risk evolution characteristics of the actual operation of the power system, and avoiding fluctuating warnings.

[0029] Furthermore, it also includes: Create a manual rule base, which contains probability constraint rules; The expected risk probability is adjusted according to the probability constraint rules.

[0030] As described above, the introduction of probabilistic constraint rules into the manual rule base enables the organic integration of model prediction results with domain expert experience and power system operation standards. This allows for the calibration and correction of the expected risk probability output by the model, thereby improving the reliability and compliance of the expected risk probability.

[0031] Please refer to Figure 4 A power risk probability prediction terminal includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor runs the computer program, it performs the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system operation data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

[0032] Please refer to Figure 1 and Figure 2 Embodiment 1 of the present invention is as follows: A power risk probability prediction method 100, such as Figure 1 As shown, it includes: S102. Obtain historical operating data and historical extreme events of the power system, and extract risk data samples from historical extreme events.

[0033] In this embodiment, the core of the historical operation data of the power system covers multi-source heterogeneous data such as power grid flow, status of power generation, transmission and distribution equipment, supply and demand balance indicators, and external conditions such as weather. The historical extreme events include, but are not limited to, data on various events caused by new energy fluctuations, equipment failures, and extreme weather, taking into full account the characteristics of the working environment and natural environment of the target area where the power system is located.

[0034] This process involves risk identification of historical extreme events; structuring these events and the risk identification results into a typical event database; establishing unified fields for each event segment record, including but not limited to event identifier, start and end times, duration, event type, and key event factors; and simultaneously constructing a distinctive historical feature database and field dictionary to ensure the traceability of subsequent risk data samples and interpretation outputs. When extracting risk data samples from historical extreme events, core indicators for measuring power system risk can be selected, such as the reserve ratio. and supply and demand gap indicators All types of historical extreme events can affect these two indicators, as expressed below: .

[0035] .

[0036] Among them, among them, This indicates a discrete-time index (consistent with the data's time granularity, such as hourly or finer granularity). express Total system load demand at any time (in MW or power units consistent with capacity). express The available supply capacity of the system at any time (in MW) can be calculated from the available capacity of conventional power sources, the effective output of renewable energy, and the available external transmission / mutual aid capacity to meet the upper limit of dispatchable supply to the load. The reserve rate is used to characterize the system's performance. The level of supply leeway at any given time, when When its value is positive, it indicates that there is a margin. When its value is negative, it indicates insufficient supply; To standardize the supply-demand gap, used to characterize the degree of supply shortage, it is achieved through... The constraint is non-negative, such that when hour ,when hour (Or in extreme cases it may be greater than 1, depending on the criteria), thereby unifying the statistical criteria for risk event judgment and facilitating cross-period comparisons.

[0037] Then, preset reserve ratio and supply-demand gap indicators are established; the reserve ratio of the power system in historical extreme events is calculated and judged to see if it meets the first preset condition or the supply-demand gap indicator meets the second preset condition; if so, the historical extreme event is set as a risk data sample. In this embodiment, a short-term monitoring label is constructed using a future window determination method: the forward window is set to H, preferably H=6 hours; the first preset condition is that the reserve ratio is lower than the threshold α, and the second preset condition is that the supply-demand gap is higher than the threshold β. When the reserve ratio is lower than the threshold α or the supply-demand gap is higher than the threshold β within the future window, it is marked as a risk data sample. Its expression is as follows: .

[0038] Where k represents the time step index within the future window, and L represents the starting index of the future time window.

[0039] Furthermore, to expand to more scenarios, target variables corresponding to the power system and risk assessment conditions corresponding to the target variables are preset; change cases of the target variables are obtained from historical operating data; and the operating data of the power system corresponding to the change cases that meet the risk assessment conditions are selected as new risk data samples, taking Latin hypercube sampling as an example: .

[0040] Where, x i,j This represents the sampled value of the j-th uncertainty variable (i.e., the target variable) in the i-th sample. π represents the inverse function of the cumulative distribution function of the j-th variable, used to map uniformly distributed samples to the true distribution of the variables; j (i) represents the i-th permutation index of the j-th variable; u i,j U(0,1) represents a uniformly distributed random number in the interval [0,1] for the i-th sample and the j-th variable, used to perform random perturbation within each equally divided segment; U(0,1) represents a standard uniform distribution; N is the total number of samples.

[0041] Furthermore, the expectation of the Monte Carlo sample can be approximated by the sample mean: .

[0042] Where, x i Let represent the vector of uncertain variables in the i-th scenario.

[0043] S104. Preset mapping rules: Map risk data samples to operational disturbance variables that affect system operation data through mapping rules.

[0044] In this embodiment, the mapping rules are defined using a hierarchical system, namely, establishing a hierarchical structure of "Tier-3 (exogenous events / policy scenarios) — Tier-2 (observable transmission indicators) — Tier-1 (bottom model input features)", including the following: Tier-3 includes a scenario description item for scenario explanation.

[0045] Tier-2 includes observable transmission indicators such as energy prices, fuel costs, trade sentiment indices, policy intensity indices, or supply chain constraints.

[0046] Tier-1 includes transmitted disturbance variables, such as weather, typhoons, available capacity, fuel inventory, gas margin, renewable energy share, and mutual aid capacity, which are directly input into the risk probability prediction model.

[0047] The mapping process of the mapping rules includes: First, extracting the transmission perturbation variables corresponding to the observable transmission indicators from the risk data samples. Specifically, for each type of extreme event that can be used as a risk data sample, setting the intensity parameter θ_E and the duration parameter T_E, defining the perturbation direction, perturbation type (additive / multiplicative / piecewise), and corresponding formulas for the observable transmission indicators, forming a mapping table from events to observable transmission indicators. Transmission perturbation variables It can be uniformly represented as: .

[0048] in, Indicates the baseline scenario The propagation perturbation variable at time 1 / 2; is a multiplicative perturbation function used to characterize proportional shock effects; E represents historical extreme events; It is an additive perturbation function used to characterize the impact effect of absolute quantities; This parameter represents the intensity of the event and is used to characterize the magnitude of the impact. The duration window represents the event duration; the disturbance function can be a piecewise linear function, an exponentially decaying function, or a rectangular window function to characterize the initiation, duration, and recovery process of the shock.

[0049] It is understandable that there is more than a one-to-one relationship between historical extreme events and risk data samples. When risk data samples correspond to the superposition of multiple historical extreme events, commutative weighted superposition rules or sequential superposition rules can be used for synthesis. The mapping results must meet physical feasibility constraints and statistical boundary constraints to ensure the continuity of observable transmission indicators and the rationality of engineering.

[0050] Then, based on the observable transmission indicators and transmission disturbance variables, the operating disturbance variables corresponding to the system operating indicators are established, and their expressions are as follows: .

[0051] in, Indicates the baseline scenario The running perturbation variable of Tier-2 at time step 1.

[0052] At the implementation level, relative shock measures and linear / piecewise linear transfer functions can be used to convert operational disturbance variables into Tier-1 increments: .

[0053] in, This indicates the value of the perturbation variable under impact scenarios; This represents the baseline value of the disturbance variable under the impact scenario; ε represents a very small positive number; shock(x) represents the relative impact degree; This represents the conduction function.

[0054] Finally, the mapping rules are checked for directional consistency, magnitude boundaries, and superposition constraints. The mapping version number, applicable scope, and revision description are recorded to ensure that scenario generation and risk interpretation are traceable.

[0055] S106. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on historical operation data and operation disturbance variables.

[0056] In this embodiment, the system state feature vector at time t is used as the reference. To input a risk probability prediction model, its output is the risk probability. Its expression is as follows: .

[0057] in, Forward-looking risk labels; This represents a probability model.

[0058] In this embodiment, the feature contribution value of the expected risk probability is calculated using a risk probability prediction model; the feature contribution value represents the influence of each input feature of the power system operation data on the model output. The expression is as follows: .

[0059] in, Feature groups representing input features; This represents the feature contribution value of feature j at time t.

[0060] S108. Obtain the current operating data of the power system, substitute the current operating data into the trained risk probability prediction model, and obtain the expected risk probability of the power system within a preset time period in the future.

[0061] In this embodiment, the expected risk probability can quantify the operational risk level of the power system on an hourly or daily scale in the future, providing core quantitative basis for risk warning, supply guarantee decision-making and risk disposal linkage, and ensuring the safe and stable operation of the power system.

[0062] Please refer to Figure 2 Embodiment two of the present invention is as follows: A method for predicting the probability of power risks, based on the above embodiment one, involves creating an artificial rule base containing probability constraint rules; and adjusting the expected risk probability according to the probability constraint rules.

[0063] In this embodiment, under the expected risk probability Based on this, a manually coded rule base is invoked to apply threshold constraints, anomaly stabilization, and engineering feasibility checks to the probability output. For consistent representation, the probability can be modified in the logit space. .

[0064] in, Represents the original probability. Let be the set of rules triggered at time t. For the rule trigger function, Rule weight / credibility; (t) represents the scene information / environmental state variable at time t.

[0065] Adjusting the expected risk probability yields the mixed risk probability: .

[0066] Where σ(z) represents the output value of the Sigmoid function; z represents the input variable of the Sigmoid function.

[0067] In this embodiment, the probability constraint rules include at least the following: (a) Threshold constraint rule, which takes the expected risk probability as input and combines it with the preset engineering availability range. By performing a truncation or compression mapping on the expected risk probability, the constrained probability is obtained, thereby avoiding the probability output from exceeding the engineering interpretability range.

[0068] (b) Anomaly stabilization rules take the constrained probability and its time series neighborhood as input, and suppress instantaneous spikes and jitters through stabilization operators such as sliding window smoothing, amplitude limiting or hysteresis to obtain the stabilized probability, thereby improving the continuity and readability of the warning signal.

[0069] (c) Engineering feasibility verification rules, taking the stabilization probability, key operating variables (including at least one or a combination of reserve rate, supply and demand gap, spot electricity price and demand response level) and the explanation attribution results as inputs, verify the abnormal high / low probabilities according to the consistency criteria and output the verification correction amount, thereby realizing the audit and correction of results under extreme samples, abnormal data or distribution drift conditions; and after applying the correction amount to the stabilization probability in the logit space, output the mixed risk probability through the sigmoid function.

[0070] In the above content, the threshold constraint rule is used to limit the probability output to the range of engineering availability; the anomaly stabilization rule is used to suppress instantaneous spike fluctuations and reduce warning jitter; the engineering feasibility verification rule is used to review and audit the abnormal output by combining key variables such as reserve rate, supply and demand gap, spot electricity price and demand response level, thereby improving the readability and executability of the output.

[0071] Please refer to Figure 2 Embodiment 3 of the present invention is as follows: A power risk probability prediction method, based on the above embodiment one or two, obtains the historical risk probability generated by the risk probability prediction model; and sets the value range of the risk probability corresponding to each risk warning level according to the variation law of the magnitude of the historical risk probability.

[0072] In this embodiment, multiple mixed risk probabilities are combined to estimate an empirical distribution based on the mixed risk probability sequence within a historical rolling window, and multiple quantiles are calculated as early warning thresholds: .

[0073] in, Let be the q-quantile function, L be the length of the scroll window, and ( );τ Y (t) represents the yellow warning threshold (low-level risk warning) at time t, determined by the lower quantile Q. qY Calculated; τ O (t) represents the orange alert threshold (medium-level risk alert) at time t, determined by the median q. O Calculated; τ R (t) represents the red alert threshold (high-level risk warning) at time t, determined by the high quantile q. R Calculated.

[0074] Furthermore, at least two risk warning levels are preset, each corresponding one-to-one with a risk probability of different magnitudes. The appropriate risk warning level for the power system is determined based on the expected risk probability. The corresponding response strategy is then triggered based on the risk warning level. An example of setting risk warning levels is shown below: .

[0075] Meanwhile, to enhance the ability to identify persistent risks, an escalation is triggered when an orange alert occurs consecutively for a number of times K or for a duration threshold; a downgrade is triggered when the probability of a mixed risk continues to decline and falls below a downgrade threshold, thereby avoiding frequent fluctuations. .

[0076] This embodiment outputs the corresponding response strategy number / text based on the warning level, forming a structured result that includes predicted probability, dynamic threshold, warning level, dominant risk source labeling, and response suggestions, to support closed-loop processing on the scheduling side.

[0077] To better illustrate the application process and achievable technical effects of the above technical solutions, the following examples are provided: Combined with appendix Figures 1 to 3 This embodiment uses hourly historical data from a certain province as a basis to construct a short-term risk probability prediction model for the day. The forward window H is 6 hours, and the reserve ratio and supply-demand gap are used as the criteria for risk assessment.

[0078] Reference Figure 2 A power system risk probability prediction method based on human-machine hybrid augmented intelligence, comprising the following four parts: (1) Historical data collection; (2) Mapping relationship design; (3) Risk probability prediction; (4) Early warning level classification.

[0079] (1) Historical data collection. First, construct a dictionary of fields for a specific province's historical dataset, as shown in Table 2.

[0080] Table 2. Selected Historical Datasets of a Certain Province

[0081] Regarding the construction of risk event segments, risk data sample sequences are generated based on hourly outcome variables according to the aforementioned criteria, and adjacent risk data samples are merged to obtain risk event segments. To facilitate annual statistics and selection of typical events, event segments are aggregated by calendar year to obtain annual statistical results, as shown in Table 3. Table 3. Annual Statistics of Short-Cycle Risk Events (yearly_summary)

[0082] The percentage statistics for each event segment are summarized by key event factors (risk sources), as shown in Table 4: Table 4. Statistics on the proportion of different risk sources (source_share)

[0083] Select event segments with longer durations on an annual basis to form a list of typical risk events. An example is shown in Table 5: Table 5 List of Typical Risk Events (Selecting Event Segments with Longer Durations on an Annual Basis)

[0084] (2)Mapping relationship design. Treat exogenous events / policy scenarios as Tier-3 scenario knobs, slow variables such as price, trade, policy intensity, and supply chain as Tier-2 observable conduction indicators, and available capacity, available days of fuel inventory, gas supply-demand margin, renewable proportion, typhoon distance and intensity, duration of heavy precipitation and high temperature, etc. as Tier-1 features that can be directly incorporated into the model. Define the perturbation direction, type, and formula through the mapping tables from events to Tier-2 and from Tier-2 to Tier-1, and conduct boundary constraint checking on the superimposed scenarios. Details are shown in Tables 6 and 7: Table 6 List of Mapping Relationships from Events to Tier-2 / 3 (Excerpt)

[0085] Table 7 Mapping Relationships from Tier2 / 3 to Tier1 (Excerpt)

[0086] (3)Risk probability prediction. Use as the input and as the output to train a probability prediction model to obtain the expected risk probability. Aggregate the SHAP contributions within the risk event segment during the interpretation phase and output the dominant risk source category. Subsequently, call the artificial rule base to perform threshold constraint, abnormal stabilization, and engineering feasibility checking on the expected risk probability to obtain the hybrid risk probability for early warning triggering and linkage output. The results of the hybrid risk probability are as Figure 3 shown. Among them, Short-cycle riskprobability represents the short-cycle risk probability; Model probability represents the expected risk probability; Hybridprobability represents the hybrid risk probability.

[0087] (4)Early warning level classification. Construct dynamic thresholds based on the historical rolling distribution of the hybrid risk probability, and map the probability to green, yellow, orange, and red early warning levels. An example of the risk-response linkage matrix is shown in Table 7, and examples of early warning classification and response linkage results at some time points are shown in Table 8 (where the daily-scale examples can directly correspond to the rolling intraday prediction results of the short-term window).

[0088] Table 7 Example of Risk-Response Linkage Matrix

[0089] Table 8 Examples of Risk Warning Classification and Response Linkage Results at Selected Time Points

[0090] Through the above implementation methods, continuous risk probability output can be transformed into early warning level signals that can be executed on the scheduling side, and the dominant risk source labeling and response strategy suggestions can be provided simultaneously on the output side, thereby forming an engineering closed loop of "risk probability assessment - dynamic threshold triggering - early warning classification - strategy matching".

[0091] Please refer to Figure 4 Embodiment four of the present invention is as follows: A power risk probability prediction terminal 1 includes a memory 2, a processor 3, and a computer program stored in the memory 2 and run on the processor 3. When the processor 3 runs the computer program, it implements a power risk probability prediction method as described in Embodiments 1, 2, or 3.

[0092] In summary, the present invention provides a power risk probability prediction method and terminal that acquires historical operating data and historical extreme events of the power system, extracts risk data samples from these extreme events, and maps these risk data samples to operating disturbance variables that affect the system's operating data through mapping rules. This quantifies the uncertainty of historical extreme events into calculable and learnable operating disturbance variables, which are then applied to the model input and training of the risk probability prediction model. Subsequently, the current operating data of the power system is acquired and substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset future timeframe. This method fully considers the influencing factors of extreme events and is suitable for hourly to intraday scale supply-demand imbalance risk early warning, supply guarantee decision support, and risk source location and response coordination. It improves the accuracy and robustness of short-term risk identification while enhancing the interpretability of risk output, ensuring the safe operation of the power system. Furthermore, the introduction of probability constraint rules from the manual rule base organically combines the model prediction results with domain expert experience and power system operating standards, allowing for calibration and correction of the expected risk probability output by the model, thus improving the reliability and compliance of the expected risk probability.

[0093] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting the probability of power risk, characterized in that, Includes the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system's operational data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

2. The power risk probability prediction method according to claim 1, characterized in that, The process of mapping the risk data samples into operational disturbance variables that affect system operation data through the mapping rule includes: Pre-set observable transmission indicators and system operation indicators corresponding to the power system; From the risk data sample, the transmission disturbance variables corresponding to the observable transmission indicators are extracted. The transmission disturbance variables include disturbance intensity parameters, disturbance duration, disturbance direction, and disturbance type. Based on the observable transmission index and the transmission disturbance variable, establish the operation disturbance variable corresponding to the system operation index.

3. The power risk probability prediction method according to claim 2, characterized in that, The step of mapping the risk data samples into operational disturbance variables that affect system operation data through the mapping rule also includes: Pre-defined scenario description items corresponding to the power system; Based on the content of the risk data sample, set the scenario description item corresponding to the power system.

4. The power risk probability prediction method according to claim 1, characterized in that, Risk data samples extracted from the aforementioned historical extreme events include: Pre-set reserve ratio indicators and supply-demand gap indicators; Calculate and determine whether the power system's reserve ratio in the historical extreme event meets the first preset condition or whether the supply-demand gap indicator meets the second preset condition; If so, then the historical extreme event is set as the risk data sample.

5. The power risk probability prediction method according to claim 1, characterized in that, Also includes: The characteristic contribution value of the expected risk probability is calculated using the risk probability prediction model. The feature contribution value represents the ability of each input feature of the power system operation data to influence the model output.

6. The power risk probability prediction method according to claim 1, characterized in that, Also includes: Pre-set target variables corresponding to the power system and risk assessment conditions corresponding to the target variables; Obtain change cases of the target variable from the historical operational data; The operating data of the power system corresponding to the change cases that meet the risk assessment conditions are selected as new risk data samples.

7. The method for predicting the probability of power risk according to claim 1, characterized in that, Also includes: At least two risk warning levels are preset, and each risk warning level corresponds one-to-one with a risk probability of different sizes; Based on the expected risk probability, determine the risk warning level that is suitable for the power system; Based on the risk warning level, the corresponding response strategy is triggered.

8. The power risk probability prediction method according to claim 7, characterized in that, Also includes: Obtain the historical risk probabilities generated by the risk probability prediction model; Based on the historical variation pattern of risk probabilities, a range of risk probabilities is set for each risk warning level.

9. The power risk probability prediction method according to claim 1, characterized in that, Also includes: Create a manual rule base, which contains probability constraint rules; The expected risk probability is adjusted according to the probability constraint rules.

10. A power risk probability prediction terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor runs the computer program, it performs the following steps: Acquire historical operating data and historical extreme events of the power system, and extract risk data samples from the historical extreme events; A preset mapping rule is used to map the risk data sample into operational disturbance variables that affect the system's operational data. Create a risk probability prediction model that takes power system operation data as input and risk probability as output, and train the risk probability prediction model based on the historical operation data and the operation disturbance variables; The current operating data of the power system is obtained, and the current operating data is substituted into the trained risk probability prediction model to obtain the expected risk probability of the power system within a preset time period.

Images