Hydropower station control method and device, electronic equipment and storage medium

Abstract

The application discloses a control method and device of a hydropower station, an electronic device and a storage medium, which normalize and map multi-source heterogeneous monitoring data of the hydropower station to a unified state space, accurately calculate a state offset amount relying on a real-time working condition health reference state, realize scientific quantification of accident risks and grades in combination with a multi-dimensional coupling correction factor, instead of relying on single-point threshold alarm and artificial experience judgment, and generate optimal disposal actions under a multi-objective optimization logic in a safety boundary and action constraint, effectively breaking through the limitations of fragmented heterogeneous data and fragmented disposal processes, achieving technical effects of improving multi-source heterogeneous data fusion adaptability, accelerating accident response speed, realizing systematic and standardized disposal, strengthening accident cause and effect tracing capability, improving accident emergency disposal efficiency and safety management level, and guaranteeing safe and stable operation of the hydropower station and the power grid.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a control method, apparatus, electronic device and storage medium for a hydropower station. Background Technology

[0002] As a crucial component of the power system, the safe and stable operation of hydropower stations directly impacts the reliability of the power grid and the continuity of energy supply. Related technologies utilize the coordinated operation of computer monitoring systems, speed governor systems, excitation systems, and condition monitoring systems to construct a comprehensive monitoring system covering data acquisition, alarm notifications, and accident recording.

[0003] Existing methods for handling hydropower station accidents directly rely on alarm mechanisms based on single-point thresholds and human experience-based judgment. This can lead to problems such as poor compatibility of heterogeneous systems, delayed accident response, fragmented handling processes, and weak causal tracing capabilities, thereby affecting the efficiency of emergency response and the improvement of safety management. Summary of the Invention

[0004] This disclosure provides a control method, apparatus, electronic device, and storage medium for a hydropower station.

[0005] According to a first aspect of this disclosure, a control method for a hydroelectric power station is provided, comprising:

[0006] The multi-source heterogeneous monitoring data of the hydropower station is mapped to a unified state space to obtain the current comprehensive operating status, and the state offset is determined based on the health reference status of the real-time operating conditions. Based on the state offset and multidimensional coupling correction factor, the accident risk is calculated and the accident level is determined. When the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and the handling action, the optimal handling action is calculated and executed with the optimization objectives of suppressing accident risk, reducing action cost, and ensuring smooth action.

[0007] Optionally, the step of mapping multi-source heterogeneous monitoring data of the hydropower station to a unified state space, obtaining the current comprehensive operating status, and determining the state offset based on the real-time operating condition health reference status includes: The collected multi-source heterogeneous monitoring data is processed by protocol parsing, time synchronization and normalization to construct the current comprehensive operating status vector; Based on historical healthy operation samples and real-time operating parameters, a health reference state vector matching the current operating conditions is determined; The difference between the integrated operating state vector and the health reference state vector is calculated to obtain the state offset vector.

[0008] Optionally, the step of calculating the accident risk and determining the accident level based on the state offset and the multidimensional coupling correction factor includes: The state offset is weighted and accumulated within a sliding time window to obtain the accident energy accumulation value. The alarm mutation factor, which reflects the characteristics of alarm timing aggregation, the topological coupling factor, which reflects the intensity of abnormality propagation along the process chain, and the logic abnormality factor, which reflects the abnormal state of protection logic, are integrated to correct the accident energy accumulation value. The modified accident energy accumulation value is transformed into a standardized accident risk through a mapping function.

[0009] Optionally, when the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, the optimal response action is calculated and executed with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action, including: Establish a comprehensive cost function with multiple objectives: minimizing the potential risk of accidents in the future prediction time domain, minimizing the cost of response actions, and maximizing the smoothness of action changes. Based on the unit state evolution model, and under the constraints of equipment action limits and state safety thresholds, the sequence of actions that minimizes the comprehensive cost function is solved, and the optimal actions are executed.

[0010] Optionally, when the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, after calculating and executing the optimal response action with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action, the method further includes: Extract discrete accident event nodes and calculate the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph. The propagation contribution is recursively calculated in reverse along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident.

[0011] Optionally, the step of extracting discrete accident event nodes and calculating the causal propagation strength between event nodes based on multidimensional association features to construct an accident causal graph includes: Based on at least one of the correlation features among temporal proximity, parameter change synergy, and process topology connectivity, the causal propagation strength between event node pairs is quantified. When the causal propagation strength meets the preset association threshold, a directed causal edge is constructed between the corresponding event nodes to obtain a directed weighted accident causal graph that characterizes the accident evolution path.

[0012] Optionally, the step of recursively calculating the propagation contribution along the accident causality graph from the apparent node to locate the root cause node includes: The representation node that characterizes the final external result of the accident is determined, and the causal propagation strength from the upstream node to the representation node is recursively accumulated along the directed causal edge starting from the representation node to obtain the propagation contribution of each candidate node. The candidate node with the highest propagation contribution and no preceding cause node is selected as the root cause node. The abnormal parameters of the root cause node are corrected to a healthy reference state, and the accident risk potential is reassessed. If the decrease in risk potential meets the verification threshold, the root cause localization is confirmed to be correct.

[0013] According to a second aspect of this disclosure, a control device for a hydroelectric power station is provided, comprising: The state determination unit is used to map multi-source heterogeneous monitoring data of hydropower stations to a unified state space, obtain the current comprehensive operating status, and determine the state offset based on the real-time operating condition health reference status. The risk rating unit is used to calculate the accident risk and determine the accident level based on the state offset and the multidimensional coupling correction factor. The optimization and handling unit is used to calculate and execute the optimal handling action when the accident level reaches the warning condition, under the constraints of the unit operation safety boundary and handling action, with the optimization objectives of suppressing accident risk, reducing action cost and ensuring smooth action.

[0014] Optionally, the state determination unit is further configured to: The collected multi-source heterogeneous monitoring data is processed by protocol parsing, time synchronization and normalization to construct the current comprehensive operating status vector; Based on historical healthy operation samples and real-time operating parameters, a health reference state vector matching the current operating conditions is determined; The difference between the integrated operating state vector and the health reference state vector is calculated to obtain the state offset vector.

[0015] Optionally, the risk rating unit is also used for: The state offset is weighted and accumulated within a sliding time window to obtain the accident energy accumulation value. The alarm mutation factor, which reflects the characteristics of alarm timing aggregation, the topological coupling factor, which reflects the intensity of abnormality propagation along the process chain, and the logic abnormality factor, which reflects the abnormal state of protection logic, are integrated to correct the accident energy accumulation value. The modified accident energy accumulation value is transformed into a standardized accident risk through a mapping function.

[0016] Optionally, the optimization processing unit is further configured to: Establish a comprehensive cost function with multiple objectives: minimizing the potential risk of accidents in the future prediction time domain, minimizing the cost of response actions, and maximizing the smoothness of action changes. Based on the unit state evolution model, and under the constraints of equipment action limits and state safety thresholds, the sequence of actions that minimizes the comprehensive cost function is solved, and the optimal actions are executed.

[0017] Optionally, it may also include a root cause localization unit; The root cause localization unit is used for: Extract discrete accident event nodes and calculate the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph. The propagation contribution is recursively calculated in reverse along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident.

[0018] Optionally, the root cause localization unit is further configured to: Based on at least one of the correlation features among temporal proximity, parameter change synergy, and process topology connectivity, the causal propagation strength between event node pairs is quantified. When the causal propagation strength meets the preset association threshold, a directed causal edge is constructed between the corresponding event nodes to obtain a directed weighted accident causal graph that characterizes the accident evolution path.

[0019] Optionally, the root cause localization unit is further configured to: The representation node that characterizes the final external result of the accident is determined, and the causal propagation strength from the upstream node to the representation node is recursively accumulated along the directed causal edge starting from the representation node to obtain the propagation contribution of each candidate node. The candidate node with the highest propagation contribution and no preceding cause node is selected as the root cause node. The abnormal parameters of the root cause node are corrected to a healthy reference state, and the accident risk potential is reassessed. If the decrease in risk potential meets the verification threshold, the root cause localization is confirmed to be correct.

[0020] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0021] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0022] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0023] The control method, device, electronic equipment, and storage medium for hydropower stations disclosed herein normalize and map multi-source heterogeneous monitoring data of hydropower stations to a unified state space. It accurately calculates state offsets based on real-time operating condition health reference states and scientifically quantifies accident risks and levels by combining multi-dimensional coupling correction factors, rather than relying on single-point threshold alarms and manual experience judgment. Simultaneously, under safety boundaries and action constraints, it generates optimal handling actions using multi-objective optimization logic, effectively overcoming the limitations of fragmented heterogeneous data and fragmented handling processes. Therefore, it can solve the technical problems of poor compatibility of heterogeneous systems, delayed accident response, fragmented handling processes, and weak causal traceability in existing accident handling methods, thus affecting the efficiency of accident emergency response and the improvement of safety management levels. It achieves the technical effects of improving the fusion and adaptability of multi-source heterogeneous data, accelerating accident response speed, realizing systematic and standardized handling, strengthening accident causal traceability capabilities, improving the efficiency of accident emergency response and safety management levels, and ensuring the safe and stable operation of hydropower stations and power grids.

[0024] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0025] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A schematic flowchart illustrating a control method for a hydropower station provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of a control device for a hydropower station provided in an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the structure of a control device for a hydropower station provided in an embodiment of the present disclosure; Figure 4 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] The control method, apparatus, electronic equipment, and storage medium of a hydropower station according to embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0028] Figure 1 This is a schematic flowchart of a control method for a hydropower station provided in an embodiment of the present disclosure.

[0029] like Figure 1 As shown, the method includes the following steps: Step 101: Map the multi-source heterogeneous monitoring data of the hydropower station to a unified state space, obtain the current comprehensive operating status, and determine the state offset based on the real-time operating condition health reference status. First, heterogeneous operational data distributed across different monitoring subsystems in a hydropower station are preprocessed using methods such as protocol parsing, label unification, and time synchronization, and then mapped to a unified multidimensional state space to obtain a state vector representing the current comprehensive operational status of the unit. Building upon this, the method further introduces a health reference state that matches real-time operating conditions. This health reference state reflects the expected level of normal operating characteristics of the unit under current operating conditions, avoiding misjudgments or omissions caused by using fixed thresholds or single benchmarks for anomaly detection.

[0030] By comparing the current overall operating status with the corresponding health reference status, the state offset between the two is calculated. This state offset can quantitatively characterize the overall deviation direction and magnitude of each operating parameter of the unit relative to the health status, providing a unified quantitative basis for subsequent accident identification and risk assessment. The dimensions in this unified state space can cover various operating parameters such as head, load, guide vane opening, speed, vibration, sway, pressure pulsation, temperature, control status, and protection logic status.

[0031] The health reference status can be constructed by comprehensively considering information such as historical health operation data, unit characteristic curves, or statistical characteristics of similar operating conditions, and can be dynamically updated through an online learning mechanism to adapt to factors such as unit aging and changes in operating conditions.

[0032] In one specific implementation, the above-mentioned operating quantities can be uniformly represented as a comprehensive operating state vector. A healthy reference state vector is constructed based on historical continuous accident-free operating samples and current operating condition characteristics. The state offset vector is obtained by the difference between the two, so as to achieve a unified quantitative representation of the deviation of the unit's operating state.

[0033] Step 102: Calculate the accident risk and determine the accident level based on the state offset and multidimensional coupling correction factor; Based on the aforementioned state offset, this paper further considers the cumulative effect of accident state over time and the combined influence of multidimensional coupling factors on accident evolution to construct an accident risk assessment model, thereby achieving a comprehensive quantitative assessment of accident risk. Specifically, the calculation of accident risk is first based on the continuous cumulative characteristics of state offset within the analysis time window, quantifying the temporal cumulative effect of state offset into accident cumulative energy to characterize the degree of continuous evolution of the accident from early signs to manifestation. Building upon this foundation, a multi-dimensional coupling correction factor is further introduced to comprehensively correct the accumulated energy of accidents. This multi-dimensional coupling correction factor may include, but is not limited to, the following dimensions: a time-series correction term to quantify the degree of clustering and abrupt change trends of alarm information over time; a topology coupling correction term to quantify the propagation strength of anomalies along process-related paths between equipment or subsystems; and a logic correction term to quantify the degree of abnormal activation of protection and interlocking logic. Each correction term is incorporated into the accident risk assessment through weighted fusion, thereby unifying the superposition and amplification effects of multiple factors such as alarm clustering, process propagation, and protection logic on accident risk into the evaluation system.

[0034] After obtaining the comprehensive risk assessment value of the accident, the assessment value is mapped to the accident level that can be used for engineering disposal by setting a grading threshold. Different levels correspond to different disposal and response strategies to ensure the pertinence and rationality of the accident response.

[0035] In one specific implementation, an accident risk potential function can be used to weight and fuse the accumulated accident energy value with alarm sequence mutation items, process topology coupling items, and protection and interlocking logic anomaly items, and then map it to a standardized risk value range through a Sigmoid function. Based on a preset threshold, the accident risk is divided into multiple levels such as normal monitoring, accident symptom warning, serious accident risk, and emergency accident.

[0036] Step 103: When the accident level reaches the warning condition, under the constraints of the unit operation safety boundary and the handling action, with the optimization objectives of suppressing accident risk, reducing action cost and ensuring smooth action, the optimal handling action is calculated and executed.

[0037] The accident handling problem is transformed into a multi-objective optimization problem under the constraints of safe operation boundaries and handling actions. Through mathematical modeling and optimization, the optimal handling action is calculated and executed to achieve scientific and intelligent suppression of accident risks. Specifically, this method first constructs a state prediction model based on the current accident state of the unit and its evolutionary patterns. This model describes the dynamic changing trend of the unit's accident state under the combined effects of natural evolution and control intervention, providing a predictive basis for handling optimization.

[0038] Based on this, the following optimization objectives are considered simultaneously: first, to minimize the comprehensive evaluation value of accident risk within the prediction time window, so as to effectively suppress accident risk; second, to minimize the amplitude and economic cost of the response actions, so as to avoid grid impact or economic losses caused by excessive actions; and third, to make the changes in response actions as smooth as possible, so as to prevent sudden changes in actions from introducing new shocks and disturbances. The optimization solution process must meet the unit operation safety boundary constraints, including but not limited to the safety limits of state parameters such as speed deviation, vibration amplitude, and temperature, as well as the actuator constraints such as the amplitude range and rate of change limits of the response actions, so as to ensure that the solved response actions are feasible in engineering.

[0039] The optimal response vector obtained through optimization can be directly converted into control commands recognizable by the control system, generating standardized operational guidelines and pushing them to operators or automatic control systems for execution. Furthermore, during the execution of the optimal response, the system continuously collects feedback data and dynamically updates the unit status, iteratively resolving the problem in real time to correct the response strategy, forming a closed-loop management mechanism to ensure that the response plan can adaptively adjust to changes in the accident situation and always remain optimal.

[0040] In one specific implementation, a handling action vector can be defined, which includes components such as load adjustment, guide vane limiting, speed regulation mode switching, excitation correction, auxiliary machine switching, and shutdown pre-control. The optimal action is solved using a quadratic programming method under linear constraints, and the solution speed is ensured to meet the real-time handling requirements by combining offline pre-training and online real-time solution.

[0041] In some embodiments, mapping multi-source heterogeneous monitoring data of the hydropower station to a unified state space, obtaining the current comprehensive operating status, and determining the state offset based on the real-time operating condition health reference status includes: The collected multi-source heterogeneous monitoring data is processed by protocol parsing, time synchronization and normalization to construct the current comprehensive operating status vector; Based on historical healthy operation samples and real-time operating parameters, a health reference state vector matching the current operating conditions is determined; The difference between the integrated operating state vector and the health reference state vector is calculated to obtain the state offset vector.

[0042] The data acquisition module receives real-time multi-source heterogeneous operational data from the hydropower station's computer monitoring system, governor monitoring system, excitation system, condition monitoring system, and protection devices. This data covers various operational parameters, including the head difference between the unit's inlet and outlet, active load, guide vane opening, main shaft speed, effective vibration values ​​in the X and Y directions of the main shaft, maximum runout at the main shaft flange, pressure pulsation amplitude at the volute inlet, maximum bearing temperature, stator winding temperature, and rotor temperature, control status of the speed control system, excitation system, and auxiliary equipment, as well as the operational status of various protection and interlocking logics. The data preprocessing module sequentially parses the data from the different subsystems to unify the communication format, standardizes tags to establish a one-to-one mapping between each monitored quantity and state vector component, and performs time synchronization to align the sampling timestamps of different subsystems. The sampling frequency can be selected from 1 to 10 Hz. Based on this, the 3σ criterion is used to identify outliers in the preprocessed data, remove abnormal data points caused by instantaneous sensor failures, and complete the data using the mean of adjacent time points. Then, all data are normalized to convert all operational quantities to a uniform scale, and finally, the current comprehensive operational status vector is constructed.

[0043] The determination of the health reference state vector X0(t) is based on statistical analysis of historical operating data from at least three consecutive months without accidents or abnormal alarms. Data from historical health samples are selected based on current inflow conditions and load levels, taking into account similar operating conditions. The upper limit of the 95% confidence interval of this historical health data is used as the reference threshold for each operating quantity. Simultaneously, the correspondence between the reference guide vane opening and the reference load is determined by combining the unit load-guide vane opening characteristic curve. The reference vectors for control states and protection interlocking logic are all set to normal state values, and the historical health samples are updated quarterly through an online learning mechanism to dynamically optimize the reference values. Finally, the state offset is obtained by calculating the difference between the current comprehensive operating state vector and the health reference state vector component by component. Each component of this offset vector quantitatively reflects the direction and magnitude of the deviation of the corresponding operating parameter from the health benchmark.

[0044] In some embodiments, calculating the accident risk and determining the accident level based on the state offset and the multidimensional coupling correction factor includes: The state offset is weighted and accumulated within a sliding time window to obtain the accident energy accumulation value. The alarm mutation factor, which reflects the characteristics of alarm timing aggregation, the topological coupling factor, which reflects the intensity of abnormality propagation along the process chain, and the logic abnormality factor, which reflects the abnormal state of protection logic, are integrated to correct the accident energy accumulation value. The modified accident energy accumulation value is transformed into a standardized accident risk through a mapping function.

[0045] The calculation of the accident energy accumulation value is based on the weighted accumulation of state offsets within a sliding time window. Specifically, the accident analysis time window length τ is typically set to 5 to 15 minutes. The number of sampling points M = τ × f within the sliding window is calculated based on the sampling frequency f of the hydropower station monitoring system. The state offset vector ΔX(m) at each discrete sampling moment within the time window is weighted and accumulated using a quadratic form. The diagonal elements of the weight matrix Wa are used to characterize the importance of each state offset to the accident. Among them, vibration offset ΔV(t), pressure pulsation offset ΔQ(t), and guide vane offset ΔG(t) contribute significantly to mechanical failures and hydraulic accidents, with their diagonal elements having values ​​of 0.15 to 0.2. Temperature offset ΔT(t) and head offset ΔH(t) contribute relatively less, with their diagonal elements having values ​​of 0.05 to 0.1. The off-diagonal elements are used to characterize the coupling amplification effect between different state variables. For example, when guide vane deviation and vibration increase occur simultaneously, the accident significance is greater than the simple addition of the two, with corresponding off-diagonal elements having values ​​of 0.1 to 0.15. Other variables without obvious coupling relationships have off-diagonal elements with a value of 0.

[0046] Multidimensional coupling correction is performed based on the accumulated accident energy value. The alarm mutation factor Sseq(k) is calculated by setting 20 to 50 alarm categories in the monitoring system and assigning different weights to each category. The weight of the protection action alarm is 3, the weight of the early warning alarm is 2, and the weight of the normal state alarm is 1. The alarm sequence strength is obtained by weighted summation of the activation state or intensity of each type of alarm. At the same time, a time decay weight ω(m)=e^(-(km) / M) based on the exponential decay function is introduced to weight the alarm intensity at historical moments. The principle is that the alarm closer to the current moment contributes more to the risk, and it is accumulated within the sliding time window, thereby effectively capturing the clustering trend of alarms and filtering out the interference of occasional alarms. The calculation of the topology coupling factor Stopo(k) involves constructing a hydroelectric generator process topology graph Gp = (Vp, Ep), where process nodes Vp include the guide vane system, runner system, main shaft system, bearing system, speed control system, excitation system, and auxiliary equipment system, etc. The process coupling edge Ep represents the direction of anomaly propagation and the correlation between nodes. A process coupling strength ηij is set between each node. For example, the coupling strength between the guide vane system and the runner system is 0.8, the coupling strength between the runner system and the main shaft system is 0.7, and the coupling strength between the auxiliary equipment system and other systems is 0.3. The topology coupling factor is obtained by weighted summation of the products of the local state deviations of adjacent nodes on all process coupling edges. This factor increases significantly when the anomalies of adjacent nodes on the process chain are synchronously enhanced.

[0047] The calculation of the logic anomaly factor Slogic(k) involves judging the action status of each rule in the hydropower station's protection and interlocking system one by one to see if it is consistent with the expectation. When the protection fails to act when it should, or acts when it should not, the anomaly activation level is assigned a value of 1. When the action is consistent with the expectation, it is assigned a value of 0. When there is partial anomaly or action delay, it is assigned a value of 0.5. Weights are also assigned according to the importance of each protection rule. For example, the weight of the shutdown protection rule is 3 and the weight of the load adjustment interlocking rule is 1. Finally, the weighted anomaly levels of all rules are summed. Finally, the accident energy accumulation value Ea(k) and the above three correction factors are weighted and fused using adjustment coefficients ρ1, ρ2, and ρ3. The comprehensive accident quantity is then mapped to a standardized accident risk value R(k) with a range of 0 to 1 using the Sigmoid mapping function σ(x) = 1 / (1+e^(-x)). Based on the preset classification thresholds θ1 = 0.2, θ2 = 0.5, and θ3 = 0.8, the accident risk is divided into different levels such as normal monitoring, accident symptom warning, serious accident risk, and emergency accident.

[0048] In some embodiments, when the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, calculating and executing the optimal response action with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action includes: Establish a comprehensive cost function with multiple objectives: minimizing the potential risk of accidents in the future prediction time domain, minimizing the cost of response actions, and maximizing the smoothness of action changes. Based on the unit state evolution model, and under the constraints of equipment action limits and state safety thresholds, the sequence of actions that minimizes the comprehensive cost function is solved, and the optimal actions are executed.

[0049] A unit state evolution model is established as the predictive basis for treatment optimization. This model uses the discrete state equation X(k+1)=AX(k)+BU(k)+Dw(k) to describe the dynamic changes of the unit state under the combined effects of natural evolution and control intervention. The state evolution matrix A is 10×10 dimensional, and its matrix elements are obtained by fitting historical accident data. It is used to characterize the natural evolution trend between various state variables under no intervention conditions. For example, when the guide vane opening deviation will lead to an increase in vibration deviation at the next moment, the corresponding element in A is positive. The intervention influence matrix B is 10×6 dimensional, and its matrix elements are determined by unit characteristic tests. It is used to characterize the regulatory effect of treatment actions on various state variables. For example, when the load reduction action is negative, it will lead to a decrease in speed deviation. The corresponding element in B is negative. The external disturbance input matrix D is 10×3 dimensional, and its external disturbance vector w(k) contains three components: head fluctuation, inflow change, and load disturbance. Gaussian process modeling is used to describe the impact of uncontrollable factors on the unit state.

[0050] A comprehensive cost function is established, which includes three optimization objective terms: the first term is the sum of the accident risk potential R(m) at each prediction time within the prediction step size Np, where Np is typically taken as 3 to 10 sampling points depending on the accident evolution speed, and the accident risk potential is gradually predicted by the state evolution equation, used to minimize the future accident risk; the second term is the action cost term U. (m)ΛU(m), where Λ is a 6×6 dimensional action cost weight diagonal matrix. Each diagonal element is determined according to the economic cost of the action. For example, the cost weight of a shutdown action is 10, and the cost weight of a load adjustment action is 2. This is used to constrain the magnitude of the action to avoid excessive action leading to grid impact and economic losses. The third term is the action smoothing penalty term ΔU. (m)ΓΔU(m), where ΔU(m)=U(m)-U(m-1) is the change in action between adjacent time points, and Γ is a 6×6 dimensional action smoothing weight diagonal matrix. The larger the value of each diagonal element, the more severe the penalty for sudden action changes. For example, the smoothing weight of the guide vane opening action is 5 to prevent hydraulic impact caused by sudden guide vane changes. The handling action vector U(k) contains six components, namely the load adjustment action amount u1, the guide vane limiting action amount u2, the speed regulation mode switching variable u3, the excitation correction action amount u4, the auxiliary machine switching variable u5, and the shutdown pre-control variable u6. Regarding the constraints, the action constraints include the load adjustment amount satisfying -5≤u1(k)≤5MW, the guide vane opening change rate satisfying |u2(k)-u2(k-1)|≤5% / s and the opening range being 0 to 100%, and the excitation adjustment range being 10 to 15kV; the condition safety constraints include vibration not exceeding 4.5mm / s, speed deviation not exceeding ±5% of rated speed, and temperature not exceeding 80℃.

[0051] This optimization problem is a quadratic programming problem under linear constraints. It is solved using the gradient descent method, and the solution time is ensured to be no more than 1 second through a combination of offline pre-training and online real-time solving. The optimal action vector U obtained from the solution is... (k) It can be directly converted into control commands that the control system can recognize, generate standardized operation guidelines and execute them. At the same time, during the execution process, the closed-loop execution module continuously collects feedback data and dynamically updates the unit status, and iterates in real time to correct the handling strategy until the accident risk is reduced to below the normal range.

[0052] In some embodiments, when the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, after calculating and executing the optimal response action with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action, the method further includes: Extract discrete accident event nodes and calculate the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph. The propagation contribution is recursively calculated in reverse along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident.

[0053] During the accident event node extraction phase, after unifying the timeline of the entire accident process data, the system establishes three types of event extraction models. The state anomaly event extraction model is based on threshold triggering and duration constraints. When the offset ΔXi(k) of a certain state parameter exceeds the anomaly triggering threshold ΔXth and the duration τi exceeding the threshold reaches the minimum duration threshold τth, the corresponding state anomaly event node eXi is triggered. This node includes attributes such as trigger time, anomaly parameter type, anomaly amplitude, and duration, used to transform the continuous offset process into a discrete anomaly event. The handling action event extraction model is based on non-zero triggering of the action vector. When the j-th component uj(k) in the handling action vector U(k) ≠ 0, this type of handling action is considered to be triggered at time k, forming a handling action event node eUj. The protection and interlocking event extraction model is based on logic state recognition. When the r-th protection or interlocking logic state lr(k) = 1, the protection or interlocking event node eLr is triggered. After the three types of events are extracted, all events are deduplicated, clustered, and time-series unified to form an accident event node set E = {e1,e2,…,en}.

[0054] In the causal propagation intensity calculation stage, a three-dimensional fusion modeling method integrating temporal correlation, parameter coupling, and process topology correlation is adopted to quantify the degree of causal influence between any two event nodes in the event node set. The temporal correlation sub-model measures the proximity of the causal event and the result event in the time dimension using the exponential decay function Tij = exp(-Δt / T0), where Δt = tj - ti is the time interval between the two events, and T0 is the time decay constant, which is assigned a value only when tj > ti and 0 < Δt ≤ Tw, i.e., within the propagation time window; otherwise, it is zero, to ensure that the causal edge direction conforms to the temporal sequence of accident evolution. The parameter coupling sub-model measures the synergistic relationship between the two events in terms of data change patterns by calculating the absolute value of the Pearson correlation coefficient Cij = |ρ(ΔXi,ΔXj)| of the state parameter offset sequences corresponding to the two events, in order to verify the rationality of causality from the parameter level. The process topology correlation sub-model is based on a pre-constructed hydropower unit process topology graph Gp = (Vp, Ep). When there is a process coupling edge between the process nodes mapped to two events, the corresponding coupling strength ηij is directly assigned as the process topology correlation coefficient Pij; otherwise, it is assigned a value of zero. The three terms are weighted and fused according to the weight coefficients α = 0.4, β = 0.3, and γ = 0.3 to obtain the comprehensive causal propagation strength cij = αTij + βCij + γPij. When cij ≥ 0.3, it is determined that there is a valid causal correlation between the two event nodes, and a directed causal edge ei → ej is constructed, finally forming the accident causal graph Gc = (E, C), where C is the set of causal edges that meet the threshold condition.

[0055] In the root cause inversion phase, the system first extracts the apparent event nodes that directly trigger accident alarms, protection actions, or shutdowns from the accident level determination results, forming a set of apparent nodes Es = {es1, es2, ..., esm}, and initializes their own propagation contribution to 1, i.e., fss = 1. Then, it recursively calculates the propagation contribution of each node to the apparent nodes along the accident causality graph, using the recursive formula fis = Σ{j:(ei,ej)∈C} cij·fjs, iterating until the difference between two adjacent iterations is less than 10. - Convergence occurs within 3 seconds. After obtaining the propagation contribution of each node, the node with the largest propagation contribution and no preceding cause node in the causal graph is selected as the root cause node. If the difference in propagation contribution among multiple candidate nodes is less than 10, the root cause node is considered. - ² is then identified as a joint root cause node. To verify the correctness of the root cause localization results, the abnormal parameters corresponding to the candidate root cause node are corrected to the healthy reference state and then resubmitted into the accident state evolution equation. If the corrected accident risk potential decreases by more than 80% compared to the original value, the root cause localization is determined to be correct.

[0056] In some embodiments, extracting discrete accident event nodes and calculating the causal propagation strength between event nodes based on multidimensional association features to construct an accident causal graph includes: Based on at least one of the correlation features among temporal proximity, parameter change synergy, and process topology connectivity, the causal propagation strength between event node pairs is quantified. When the causal propagation strength meets the preset association threshold, a directed causal edge is constructed between the corresponding event nodes to obtain a directed weighted accident causal graph that characterizes the accident evolution path.

[0057] After obtaining the set of event nodes E = {e1, e2, ..., en}, the system performs quantitative calculations of the three-dimensional correlation features for any two event nodes ei and ej in the set. Regarding temporal proximity, the system first determines the trigger time ti of event ei and the trigger time tj of event ej, and calculates the time difference Δt = tj - ti. The temporal correlation coefficient is calculated only when tj > ti and Δt is within the propagation time window Tw. An exponential decay function Tij = exp(-Δt / T0) is used to assign the temporal correlation degree, where T0 is a time decay constant used to control the decay rate. The shorter the time interval, the closer Tij is to 1, indicating that the two events have a stronger causal propagation order in the time dimension. When the time interval exceeds the propagation time window or the event order is reversed, Tij is directly assigned a value of zero to exclude pseudo-causal connections that do not meet the temporal sequence condition.

[0058] Regarding the degree of coordination in parameter changes, the offset sequences ΔXi and ΔXj of the corresponding state parameters of events ei and ej within the analysis time window are extracted, and the Pearson correlation coefficient ρ(ΔXi,ΔXj) between them is calculated. The absolute value of this coefficient is taken as the parameter coupling coefficient Cij=|ρ(ΔXi,ΔXj)|. This coefficient reflects the degree of coordination between the two events in the data change patterns. For example, when the guide vane opening offset sequence and the vibration offset sequence show a synchronous upward trend, the Pearson correlation coefficient is close to 1, indicating that there is a significant coupling relationship between the two at the parameter change level.

[0059] Regarding the connectivity of the process topology, the judgment is based on the pre-constructed process topology graph Gp = (Vp, Ep) of the hydropower unit. First, event nodes ei and ej are mapped to their corresponding process nodes vi and vj, respectively. If there is a process coupling edge between vi and vj in the process topology graph, i.e., (vi, vj) ∈ Ep, then the process topology correlation coefficient Pij is assigned equal to the corresponding coupling strength ηij. This coupling strength is pre-calibrated according to the tightness of the process correlation between equipment. For example, the coupling strength between the guide vane system and the runner system is 0.8, the coupling strength between the runner system and the main shaft system is 0.7, and the coupling strength between the auxiliary system and other systems is 0.3. If there is no coupling edge between two process nodes, i.e., (vi, vj) If Ep is set to zero, then Pij is assigned a value of zero, indicating that there is no reasonable path for abnormal propagation from the perspective of the structural mechanism of the hydropower unit.

[0060] After obtaining the quantified values ​​of the three correlation features mentioned above, a weighted fusion is performed according to preset weight coefficients α = 0.4, β = 0.3, and γ = 0.3 to obtain the comprehensive causal propagation strength cij = αTij + βCij + γPij of event node ei to event node ej. This comprehensive propagation strength characterizes the credibility of the causal relationship between the two events from three dimensions: temporal orderliness, parameter synergy, and mechanistic rationality. Event pairs that are correlated only in one dimension but not in other dimensions will be effectively suppressed by reducing their propagation strength due to weighting. Subsequently, the system compares the calculated comprehensive causal propagation strength with the preset correlation threshold cth = 0.3. When cij ≥ 0.3, it is determined that there is a valid causal relationship between event node ei and event node ej, and a directed causal edge from ei to ej is constructed between them, with cij as the edge weight of the causal edge; when cij < 0.3, it is considered that there is no significant causal relationship between the two events, and no causal edge is constructed.

[0061] After traversing all event node pairs, all directed causal edges that satisfy the threshold condition are gathered into a causal edge set C = {(ei,ej)|cij≥cth}, which together with the event node set E constitutes a directed weighted accident causal graph Gc = (E,C). In this causal graph, nodes represent discrete events in the accident evolution process, directed edges represent the causal propagation direction between events, and edge weights represent the strength of causal propagation, thus completely depicting the entire chain evolution path of an accident from initial triggering through intermediate propagation to the final manifestation.

[0062] In some embodiments, the step of recursively calculating the propagation contribution at the apparent node along the accident causality graph to locate the root cause node includes: The representation node that characterizes the final external result of the accident is determined, and the causal propagation strength from the upstream node to the representation node is recursively accumulated along the directed causal edge starting from the representation node to obtain the propagation contribution of each candidate node. The candidate node with the highest propagation contribution and no preceding cause node is selected as the root cause node. The abnormal parameters of the root cause node are corrected to a healthy reference state, and the accident risk potential is reassessed. If the decrease in risk potential meets the verification threshold, the root cause localization is confirmed to be correct.

[0063] In the apparent node determination phase, the system extracts apparent event nodes that directly trigger accident alarms, protection actions, or unit shutdowns based on the accident level determination results. These nodes correspond to the events at the downstream end of the accident causal graph, representing the final manifest result of the accident. For example, in an accident scenario where guide vane feedback chain drift leads to increased vibration, the event of continuous vibration exceeding limits is determined as the apparent node es, which has attributes such as trigger time, abnormal amplitude, and duration. All determined apparent event nodes constitute the apparent node set Es = {es1, es2, ..., esm}, serving as the starting point for root cause inversion. Secondly, in the propagation contribution recursive calculation phase, the system uses each apparent node as the inversion endpoint and accumulates the causal influence of each upstream node on the apparent node layer by layer along the reverse direction of the directed causal edge.

[0064] For each representation node *es*, its propagation contribution to itself is initialized to *fss* = 1. For nodes in the causal graph that are not directly or indirectly connected to the representation node, their propagation contribution is defined as *fis* = 0. For any non-representation node *ei*, its propagation contribution to the representation node *es* is calculated according to the recursive formula *fis* = Σ{j:(ei,ej)∈C} *cij*·*fjs*, where *cij* is the causal propagation strength (i.e., the causal edge weight) from node *ei* to downstream node *ej*, and *fjs* is the propagation contribution of downstream node *ej* to the representation node *es*. This recursive process proceeds upstream layer by layer from the representation node in the reverse direction along the causal edges. Each time a causal edge is passed, the propagation strength *cij* of that edge is used to attenuate and accumulate the propagation contribution *fjs* already obtained by the downstream node, until all upstream nodes in the causal graph that have causal paths with the representation node are traversed.

[0065] To ensure the stability of the recursive solution, the system uses an iterative approach for computation, i.e., fis(r+1)=Σ{j:(ei,ej)∈C} cij·fjs(r), and sets the convergence condition as the difference between two consecutive iterations |fis(r+1)-fis(r)|<10. - ³, the iteration stops when the propagation contribution of all nodes meets the convergence condition. Through the above recursive calculation, each candidate node obtains a propagation contribution value that comprehensively reflects its cumulative influence on the apparent nodes through each causal path. The larger the value, the more significant the node's contribution to the final manifest result of the accident. Next, in the root cause node determination stage, the system sorts the propagation contribution of all candidate nodes, selects the candidate node with the largest propagation contribution, and further verifies whether the node satisfies the condition of no preceding cause node, that is, there is no other node ek in the causal graph such that the directed causal edge (ek, ei) ∈ C. If the node satisfies both the conditions of the largest propagation contribution and no preceding cause node, it is determined to be the root cause node of this accident.

[0066] When the difference in the propagation contribution of multiple candidate nodes is less than a preset threshold of 10 - When the value is 2, the system determines these nodes as joint root cause nodes. For example, in a specific accident scenario, the propagation contribution of the initial fault node e0 of the guide vane feedback chain drift to the vibration manifestation node es, after recursive calculation and superposition of contributions from all paths, reaches f0s = 0.92. Furthermore, this node has no preceding cause nodes in the causal graph, therefore it is determined to be a root cause node. Finally, in the root cause verification stage, the system corrects the abnormal parameters corresponding to the determined root cause nodes to the healthy reference state, that is, restores the abnormal parameters corresponding to the guide vane feedback chain drift to normal values. Then, the corrected parameters are substituted back into the accident state evolution equation X(k+1) = AX(k) + BU(k) + Dw(k) to recalculate the accident risk potential R(k).

[0067] The system compares the accident risk potential before and after the correction. If the decrease in risk potential meets the verification threshold of (Rbefore-Rafter) / Rbefore≥80%, and the apparent node no longer triggers abnormal events, then the root cause is determined to be correctly located. For example, in a specific embodiment, the accident risk potential Rbefore before correction = 0.58, and the accident risk potential Rafter after correction = 0.087. The decrease in risk potential is (0.58-0.087) / 0.58≈85%, which is greater than the preset verification threshold of 80%. Furthermore, the vibration apparent node no longer triggers over-limit events, thus confirming that the guide vane feedback chain drift is the correct root cause of this accident.

[0068] Corresponding to the control method for hydropower stations described above, this invention also proposes a control device for hydropower stations. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0069] Figure 2 This is a schematic diagram of the structure of a control device for a hydropower station provided in an embodiment of this disclosure, as shown below. Figure 2 As shown, it includes: The state determination unit 21 is used to map the multi-source heterogeneous monitoring data of the hydropower station to a unified state space, obtain the current comprehensive operating status, and determine the state offset based on the real-time operating condition health reference status. Risk rating unit 22 is used to calculate accident risk and determine accident level based on the state offset and multidimensional coupling correction factor; The optimization and handling unit 23 is used to calculate and execute the optimal handling action when the accident level reaches the warning condition, under the constraints of the unit operation safety boundary and handling action, with the optimization objectives of suppressing accident risk, reducing action cost and ensuring smooth action.

[0070] Furthermore, in one possible implementation of this disclosure embodiment, the state determination unit 21 is further configured to: The collected multi-source heterogeneous monitoring data is processed by protocol parsing, time synchronization and normalization to construct the current comprehensive operating status vector; Based on historical healthy operation samples and real-time operating parameters, a health reference state vector matching the current operating conditions is determined; The difference between the integrated operating state vector and the health reference state vector is calculated to obtain the state offset vector.

[0071] Furthermore, in one possible implementation of this disclosure embodiment, the risk rating unit 22 is further used for: The state offset is weighted and accumulated within a sliding time window to obtain the accident energy accumulation value. The alarm mutation factor, which reflects the characteristics of alarm timing aggregation, the topological coupling factor, which reflects the intensity of abnormality propagation along the process chain, and the logic abnormality factor, which reflects the abnormal state of protection logic, are integrated to correct the accident energy accumulation value. The modified accident energy accumulation value is transformed into a standardized accident risk through a mapping function.

[0072] Furthermore, in one possible implementation of this disclosure, the optimization processing unit 23 is further configured to: Establish a comprehensive cost function with multiple objectives: minimizing the potential risk of accidents in the future prediction time domain, minimizing the cost of response actions, and maximizing the smoothness of action changes. Based on the unit state evolution model, and under the constraints of equipment action limits and state safety thresholds, the sequence of actions that minimizes the comprehensive cost function is solved, and the optimal actions are executed.

[0073] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 3 As shown, it also includes a root cause localization unit 24; The root cause localization unit is used for: Extract discrete accident event nodes and calculate the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph. The propagation contribution is recursively calculated in reverse along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident.

[0074] Furthermore, in one possible implementation of this disclosure, the root cause localization unit 24 is further configured to: Based on at least one of the correlation features among temporal proximity, parameter change synergy, and process topology connectivity, the causal propagation strength between event node pairs is quantified. When the causal propagation strength meets the preset association threshold, a directed causal edge is constructed between the corresponding event nodes to obtain a directed weighted accident causal graph that characterizes the accident evolution path.

[0075] Furthermore, in one possible implementation of this disclosure, the root cause localization unit 24 is further configured to: The representation node that characterizes the final external result of the accident is determined, and the causal propagation strength from the upstream node to the representation node is recursively accumulated along the directed causal edge starting from the representation node to obtain the propagation contribution of each candidate node. The candidate node with the highest propagation contribution and no preceding cause node is selected as the root cause node. The abnormal parameters of the root cause node are corrected to a healthy reference state, and the accident risk potential is reassessed. If the decrease in risk potential meets the verification threshold, the root cause localization is confirmed to be correct.

[0076] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0077] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0078] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0079] like Figure 4As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 402 or a computer program loaded from storage unit 408 into RAM (Random Access Memory) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. I / O (Input / Output) interface 405 is also connected to bus 404.

[0080] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0081] The computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as the control methods for a hydroelectric power station. For example, in some embodiments, the control methods for a hydroelectric power station may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the aforementioned control method for the hydropower station by any other suitable means (e.g., by means of firmware).

[0082] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0083] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0084] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0085] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0086] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0087] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0088] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0089] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0090] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A control method for a hydropower station, characterized in that, include: The multi-source heterogeneous monitoring data of the hydropower station is mapped to a unified state space to obtain the current comprehensive operating status, and the state offset is determined based on the health reference status of the real-time operating conditions. Based on the state offset and multidimensional coupling correction factor, the accident risk is calculated and the accident level is determined. When the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and the handling action, the optimal handling action is calculated and executed with the optimization objectives of suppressing accident risk, reducing action cost, and ensuring smooth action.

2. The method according to claim 1, characterized in that, The process of mapping multi-source heterogeneous monitoring data of hydropower stations to a unified state space, obtaining the current comprehensive operating status, and determining the state offset based on the real-time operating condition health reference status includes: The collected multi-source heterogeneous monitoring data is processed by protocol parsing, time synchronization and normalization to construct the current comprehensive operating status vector; Based on historical healthy operation samples and real-time operating parameters, a health reference state vector matching the current operating conditions is determined; The difference between the integrated operating state vector and the health reference state vector is calculated to obtain the state offset vector.

3. The method according to claim 1, characterized in that, The process of calculating accident risk and determining accident level based on the state offset and multidimensional coupling correction factor includes: The state offset is weighted and accumulated within a sliding time window to obtain the accident energy accumulation value. The alarm mutation factor, which reflects the characteristics of alarm timing aggregation, the topological coupling factor, which reflects the intensity of abnormality propagation along the process chain, and the logic abnormality factor, which reflects the abnormal state of protection logic, are integrated to correct the accident energy accumulation value. The modified accident energy accumulation value is transformed into a standardized accident risk through a mapping function.

4. The method according to claim 1, characterized in that, When the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action, the optimal response actions are calculated and executed, including: Establish a comprehensive cost function with multiple objectives: minimizing the potential risk of accidents in the future prediction time domain, minimizing the cost of response actions, and maximizing the smoothness of action changes. Based on the unit state evolution model, and under the constraints of equipment action limits and state safety thresholds, the sequence of actions that minimizes the comprehensive cost function is solved, and the optimal actions are executed.

5. The method according to claim 1, characterized in that, When the accident level reaches the warning condition, under the constraints of the unit's operational safety boundary and response actions, and with the optimization objectives of suppressing accident risk, reducing action costs, and ensuring smooth action execution, the method further includes: Extract discrete accident event nodes and calculate the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph. The propagation contribution is recursively calculated in reverse along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident.

6. The method according to claim 5, characterized in that, The step of extracting discrete accident event nodes and calculating the causal propagation strength between event nodes based on multidimensional correlation features to construct an accident causal graph includes: Based on at least one of the correlation features among temporal proximity, parameter change synergy, and process topology connectivity, the causal propagation strength between event node pairs is quantified. When the causal propagation strength meets the preset association threshold, a directed causal edge is constructed between the corresponding event nodes to obtain a directed weighted accident causal graph that characterizes the accident evolution path.

7. The method according to claim 5, characterized in that, The step of recursively calculating the propagation contribution along the cause-effect graph of the accident at the apparent node to locate the root cause node of the accident includes: The representation node that characterizes the final external result of the accident is determined, and the causal propagation strength from the upstream node to the representation node is recursively accumulated along the directed causal edge starting from the representation node to obtain the propagation contribution of each candidate node. The candidate node with the highest propagation contribution and no preceding cause node is selected as the root cause node. The abnormal parameters of the root cause node are corrected to a healthy reference state, and the accident risk potential is reassessed. If the decrease in risk potential meets the verification threshold, the root cause localization is confirmed to be correct.

8. A control device for a hydropower station, characterized in that, include: The state determination unit is used to map multi-source heterogeneous monitoring data of hydropower stations to a unified state space, obtain the current comprehensive operating status, and determine the state offset based on the real-time operating condition health reference status. The risk rating unit is used to calculate the accident risk and determine the accident level based on the state offset and the multidimensional coupling correction factor. The optimization and handling unit is used to calculate and execute the optimal handling action when the accident level reaches the warning condition, under the constraints of the unit operation safety boundary and handling action, with the optimization objectives of suppressing accident risk, reducing action cost and ensuring smooth action.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images