Cascade hydropower remote centralized control plan generation method and system

Abstract

The invention relates to the technical field of hydropower system control, in particular to a cascade hydropower remote centralized control plan generation method and system, and the method comprises the steps: obtaining multi-source heterogeneous dynamic data of a target cascade hydropower station; performing statistical reconstruction on the multi-source heterogeneous dynamic data to generate a dynamic feature tensor; performing risk identification on the target cascade hydropower station according to the dynamic feature tensor to obtain a risk index sequence of the target cascade hydropower station; performing dynamic weighted stacking on a preset plan template set based on the risk index sequence to generate a self-adaptive plan set; and performing dynamic matching in the adaptive plan set based on the real-time operation state of the cascade hydropower remote centralized control system, and determining an optimal plan so as to control the cascade hydropower remote centralized control system based on the optimal plan. Through dynamic data fusion and risk quantitative analysis, the cascade hydropower station risk early warning and emergency plan real-time adaptive matching is realized, and the emergency response precision and timeliness are significantly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydropower system control, and in particular to a method and system for generating a cascade hydropower remote centralized control plan. Background Art

[0002] Remote centralized control systems for cascade hydropower stations are crucial for achieving efficient and safe utilization of hydropower resources across a river basin. To address the potential chain reactions and security risks associated with unexpected events such as equipment failures, natural disasters, and cyberattacks, the industry has generally established emergency response plan matching mechanisms to ensure rapid response and mitigate losses when incidents occur.

[0003] However, the relevant technical solutions have significant technical limitations in actual application, which restricts the timeliness and effectiveness of emergency response. This is mainly reflected in the difficulty in effectively integrating multi-dimensional dynamic information that reflects the real risk situation of the system, resulting in incomplete and in-time risk perception. At the same time, its core plan generation and matching mechanism is relatively rigid, mainly relying on preset fixed rules or limited static historical data models, and lacking the ability to dynamically adjust according to real-time risk status and system operating environment. As a result, when faced with complex, changing or newly emerging risk scenarios, the recommended emergency plans are often not adaptable enough and have low accuracy, and cannot meet the needs of rapid and accurate emergency response of cascade hydropower remote control systems.

[0004] Therefore, how to break through the bottleneck of existing technology and build a technical solution that can deeply integrate multi-source heterogeneous dynamic data, accurately portray dynamic risk situations, and intelligently generate and dynamically match optimal plans based on this, so as to significantly improve the emergency response speed and disposal effect of the remote centralized control system of cascade hydropower stations in complex and changing environments, has become a core technical problem that needs to be solved urgently. Summary of the Invention

[0005] In order to solve the above-mentioned problems in the prior art, the present invention provides a method and system for generating a remote centralized control plan for cascade hydropower.

[0006] A first aspect of the present invention provides a method for generating a cascade hydropower remote centralized control plan, comprising: Acquire multi-source heterogeneous dynamic data of target cascade hydropower stations; Statistically reconstructing the multi-source heterogeneous dynamic data to generate a dynamic feature tensor, wherein the dynamic feature tensor is used to characterize the temporal evolution and statistical distribution characteristics of the multi-source heterogeneous dynamic data in a structured multidimensional space; performing risk identification on the target cascade hydropower station according to the dynamic characteristic tensor to obtain a risk index sequence of the target cascade hydropower station; Based on the risk index sequence, a set of preset emergency plan templates is dynamically weighted and superimposed to generate an adaptive emergency plan set; Based on the real-time operating status of the cascade hydropower remote centralized control system, dynamic matching is performed in the adaptive plan set to determine the optimal plan, so as to control the cascade hydropower remote centralized control system based on the optimal plan.

[0007] A second aspect of the present invention provides a system for generating a remote centralized control plan for cascade hydropower, comprising: Data acquisition module, used to obtain multi-source heterogeneous dynamic data of the target cascade hydropower station; a statistical reconstruction module, configured to statistically reconstruct the multi-source heterogeneous dynamic data to generate a dynamic feature tensor, wherein the dynamic feature tensor is used to characterize the temporal evolution and statistical distribution characteristics of the multi-source heterogeneous dynamic data in a structured multidimensional space; a risk identification module, configured to perform risk identification on the target cascade hydropower station according to the dynamic characteristic tensor, and obtain a risk index sequence of the target cascade hydropower station; A plan generation module is used to dynamically weight and superimpose a set of preset plan templates based on the risk index sequence to generate an adaptive plan set; The plan matching module is used to dynamically match the adaptive plan set based on the real-time operating status of the cascade hydropower remote centralized control system, determine the optimal plan, and control the cascade hydropower remote centralized control system based on the optimal plan.

[0008] The beneficial effects of the present invention are embodied in: First, by integrating multi-source heterogeneous dynamic data and constructing dynamic feature tensors, a unified and structured representation of complex operating states is achieved, significantly improving the comprehensiveness and timeliness of risk perception. Second, based on a dynamic risk index sequence, the plan templates are dynamically weighted and superimposed to generate a highly adaptive plan set, overcoming the rigidity and low matching defects of traditional plan libraries. Finally, by dynamically matching and executing the optimal plan based on real-time operating status, the accuracy, agility, and pertinence of the emergency response are ensured, greatly improving the safety assurance capabilities and emergency control efficiency of the cascade hydropower remote centralized control system in complex and changing environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A flow chart of a method for generating a remote centralized control plan for cascade hydropower provided by the present invention; Figure 2 A schematic flow chart of a method for generating an adaptive emergency plan provided by the present invention; Figure 3 This is a structural diagram of a cascade hydropower remote centralized control plan generation system provided by the present invention. DETAILED DESCRIPTION

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0011] The existing technologies and their main problems can be summarized as follows: 1. Static plan calling mechanism based on fixed rules: Existing systems generally use pre-set, fixed rules (such as specific fault types or threshold violations) to trigger corresponding static emergency plans. While this mechanism enables automated invocation, it lacks dynamic adjustment capabilities. It cannot adjust the specific content or execution intensity of the plan based on real-time changes in the overall risk level (for example, the simultaneous occurrence of multiple faults or the combined impact of environmental factors) or the current overall system operating status, resulting in insufficient adaptability and flexibility.

[0012] 2. Risk perception that relies on a single data source: The data foundation of existing emergency plan matching systems has significant limitations. Their automated data collection primarily focuses on electrical equipment operating parameters (such as voltage, current, and frequency), but fails to effectively integrate heterogeneous data such as environmental parameters (hydrology and meteorology), human factors, and network security status, which are closely related to power plant safety operations. This single data source and fragmented information leads to incomplete and inaccurate assessments of overall system risk, directly impacting the effectiveness of subsequent emergency plan selection.

[0013] 3. Static classification model trained based on historical data: Some improvement plans attempt to apply machine learning algorithms (such as decision trees and support vector machines) to recommend emergency plans by training historical data patterns. However, these methods have inherent flaws: (1) Strong dependence on training samples and delayed updates: Model performance is highly dependent on limited and potentially outdated historical sample data.

[0014] (2) Lack of real-time dynamic response: It is difficult to effectively integrate and respond to real-time dynamic data flows that reflect the current risk evolution.

[0015] (3) Weak generalization ability: When faced with new risk types not included in the training data or sudden scenarios with complex coupling of multiple factors, the model matching accuracy drops significantly.

[0016] (4) Poor interpretability: The model decision-making process is usually not transparent, which makes it difficult for operators to understand and trust the recommended results.

[0017] 4. Static and rigid plan library management: The existing system's plan library is generally simple to build but difficult to maintain: (1) Content rigidity: Plan items are usually static, pre-defined texts or processes, lacking the ability to dynamically generate or adjust content based on real-time risks.

[0018] (2) Inefficient management: The emergency plan database may have redundancy, incomplete coverage, and unclear classification, which may make it difficult to find emergency plans or cause content conflicts.

[0019] (3) Update lag: The update and optimization of the plan library mainly relies on manual intervention, lacking automated and intelligent evaluation, optimization and synchronization mechanisms, making it difficult to keep up with the rapid changes in risk situations and the accumulation of system operation experience.

[0020] In view of this, the present invention proposes a method and system for generating a remote centralized control plan for cascade hydropower to solve the above technical problems.

[0021] Example 1: Reference Manual Figure 1 , which shows a flow chart of a method for generating a remote centralized control plan for cascade hydropower provided by the present invention.

[0022] An embodiment of the present invention provides a method for generating a cascade hydropower remote centralized control plan, comprising: S1: Acquire multi-source heterogeneous dynamic data of the target cascade hydropower station.

[0023] This step aims to build a comprehensive, real-time and reliable data foundation for subsequent risk identification and plan generation.

[0024] In specific implementation, dynamic data collection is achieved through a distributed sensing network deployed throughout the cascade hydropower station area. The network covers: Equipment operation parameter acquisition unit: including generator winding temperature sensor, bearing vibration accelerometer, turbine flow meter, gate opening sensor, bus voltage / current transformer, etc.; Environmental status monitoring unit: including reservoir water level radar, dam body piezometer, basin rainfall station, meteorological temperature and humidity sensors, geological disaster monitoring instrument, etc.; Human operation and network security unit: integrates the operation instruction log of the dispatching main control system, access control security records, network traffic probes and intrusion detection system alarm information.

[0025] All data are aggregated to the remote control center in real time through a multi-channel concurrent transmission mechanism, forming a four-dimensional heterogeneous data stream covering equipment operation, environmental status, human operation and network security, namely the above-mentioned multi-source heterogeneous dynamic data.

[0026] S2: Statistically reconstruct multi-source heterogeneous dynamic data to generate dynamic feature tensors. The dynamic feature tensors are used to characterize the temporal evolution and statistical distribution characteristics of multi-source heterogeneous dynamic data in structured multidimensional space.

[0027] In one possible implementation, performing statistical reconstruction on multi-source heterogeneous dynamic data in S2 to generate a dynamic feature tensor specifically includes: Performing data preprocessing on the initial multi-source heterogeneous data to eliminate dynamic anomalies in the initial multi-source heterogeneous data and obtain preprocessed multi-source heterogeneous data; A sliding feature reconstruction mechanism is introduced to perform sliding feature reconstruction on each type of variable in the preprocessed multi-source heterogeneous data to generate a time-aligned structured feature set. Based on the time axis, each type of variable in the structured feature set is encapsulated as a structural feature tensor to obtain a dynamic feature tensor.

[0028] This step uses time series feature reconstruction technology to transform the original dynamic dataset into a structured data representation with time series consistency and statistical expression capabilities. Specifically, it includes the following operations: After receiving the multi-source heterogeneous dynamic data of the target cascade hydropower station, the state-aware data cleaning algorithm can be used to solve typical problems such as packet loss, duplication, and delay in the sensor link. During the execution of the algorithm, the acquisition time window is set to , for each sensor variable sequence of multi-source heterogeneous dynamic data x ( t ), firstly, continuity detection and exception elimination are performed. If , then it is considered that there is a mutation abnormality at this point. If its continuous occurrence probability is less than , the value is marked as an isolated anomaly and removed, and the missing value is filled by linear interpolation to generate preprocessed multi-source heterogeneous data.

[0029] Among them, the parameters is the transition criterion threshold, is the continuous abnormality tolerance ratio threshold, both of which can be calculated based on the historical statistics of device operation.

[0030] In one possible implementation, a sliding feature reconstruction mechanism is introduced to perform sliding feature reconstruction on each type of variable in the preprocessed multi-source heterogeneous data to generate a time-aligned structured feature set, specifically including: Within the preset sliding window, calculate the statistical set corresponding to each type of variable in the preprocessed multi-source heterogeneous data, including the mean and weighted variance; Based on the statistical quantity set, the statistical quantity of each type of variable in the multi-source heterogeneous data at each moment is encapsulated as a binary structural feature to obtain a binary structural feature set; The binary structural features in the binary structural feature set are temporally assembled to obtain a temporally aligned structured feature set.

[0031] In order to maintain the continuity of data features and the sensitivity of system state changes, the present invention avoids the trend information loss caused by the traditional extreme value normalization method by introducing a sliding feature reconstruction mechanism based on local structure statistics.

[0032] Specifically, let the sliding window length be , the current time is , in the interval Defined in t Exponential decay plus minus mean at time With weighted variance They are: in, It is an exponential decay factor, which is used to control the influence of historical data with a long time distance on the current statistics, reflecting the feature of "high weight in the near future", that is, giving more weight to recent data. It is a measure of the degree of variable fluctuation. It can capture short-term drastic changes or differences in stable states in time-varying systems and is often used in anomaly detection and fluctuation modeling.

[0033] Calculate μ(t) and , forming a time-aligned structured feature set, and then stacking all feature vectors along the time axis to generate a three-dimensional structure tensor ,in is the total number of sampling moments, is the number of sensor variables.

[0034] This step significantly improves the accuracy of risk identification and the adaptability of emergency plan matching for cascade hydropower stations through statistical reconstruction and dynamic feature tensor generation of multi-source heterogeneous dynamic data. Specifically, the dynamic anomaly elimination mechanism effectively eliminates sensor noise and isolated anomalies to ensure data quality; sliding feature reconstruction uses exponentially decaying weighted mean and variance to quantify short-term fluctuations while retaining time series trends, avoiding information loss caused by traditional normalization; the three-dimensional dynamic feature tensor uniformly encapsulates multivariate statistical features to form a structured expression of spatiotemporal coupling, providing a high-dimensional data foundation for risk factor quantification. Ultimately, the problems of data noise interference and insufficient feature expression are fundamentally resolved, making subsequent risk identification results more reliable and emergency plan generation more accurately adapted to dynamic risk situations.

[0035] S3: Based on the dynamic characteristic tensor, risk identification is performed on the target cascade hydropower station to obtain a risk index sequence of the target cascade hydropower station.

[0036] In a possible implementation, risk identification of the target cascade hydropower station is performed based on the dynamic characteristic tensor in S3 to obtain a risk index sequence of the target cascade hydropower station, specifically including: According to the preset risk factor set, risk factors are extracted from the dynamic feature tensor to obtain the target risk factor set; Extract the sensor variable feature subset associated with each type of risk factor in the target risk factor set to obtain a risk feature vector set; Based on the risk feature vector set, a weighted nonlinear transformation is performed on each type of risk factor in the target risk factor set to obtain a set of risk activity scalar values; Based on the risk activity scalar value set and the preset core indicator data of each factor in the target risk factor set, a standardized calculation is performed to obtain a comprehensive risk index set; the preset core indicator data includes historical frequency, current activity and potential impact intensity; The comprehensive risk index vectors in the comprehensive risk index set are subjected to time series assembly processing to obtain a risk index sequence.

[0037] This step uses the dynamic feature tensor as input and uses a unified feature extraction framework to identify and quantify four pre-defined risk factors for cascade hydropower stations: Environmental risk factors: Based on the statistical characteristics of hydrology (water level, flow) and meteorology (rainfall, wind speed), a flood probability model and extreme weather impact index are constructed; Equipment risk factors: Analyze the time-varying patterns of unit vibration, temperature, and electrical parameters to quantify the risks of mechanical fatigue and electrical failure; Operational risk factors: Analyze the timing anomalies in the dispatch instruction log and locate the probability of misoperation based on access control system data; Information security risk factors: Correlate the fluctuation variance of network traffic probes with the frequency of intrusion alarms to build an attack threat level model.

[0038] Of course, the preset risk factor set in the present invention may also include other types of risk factors, which is not specifically limited in the present invention.

[0039] Specifically, there are The potential risk factors are denoted as , each type of risk factor is associated with a set of characteristic variables, let The original variable vector associated with the class risk factor is: , in, Indicates the i The dimension of the variable corresponding to the class risk factor.

[0040] In order to accurately identify the activity level of risk factors in the current state, a risk activity function is constructed , defined as follows: in, Indicates the i The first risk j The weight coefficient of a variable is used to indicate the contribution of the variable to the overall risk; Indicates the The first risk The risk-sensitive function of each variable is selected based on the physical meaning and fluctuation characteristics of the variable, and nonlinear enhancement or threshold activation processing is performed; Indicates time Moment Risk Category j The collected values ​​of variables. This function constitutes a functional expression of the activity level of the risk factor at the current moment and is the basis for subsequent quantification.

[0041] In order to achieve a unified measurement of risks across categories, a risk index is introduced S i As a quantified result, The risk index of a risk category is calculated by comprehensively considering three core indicators: the historical frequency, current activity, and potential impact strength of risk factors. That is, the preset core indicator data includes the historical frequency, current activity, and potential impact strength of each factor category. The definition is as follows: in, Indicates the i Normalization coefficient of risk factor class to ensure that all risk indices are in the same dimension; Indicates the current activity of the current risk factor, specifically by calculating the risk activity function f i ( t ) in the time window [ t - w +1, t ], reflecting the average value of i The activity trend of the risk factor within the current period (rather than the instantaneous status at a single moment); F i Indicates the i The frequency of occurrence of risk factors of this type in historical records, namely the historical frequency mentioned above, can be obtained by statistically analyzing the frequency of the first i The ratio of the number of times a risk factor of this type is triggered to the total number of observations is calculated to reflect the frequency of occurrence of this type of risk in history;L i To indicate the i The potential loss value of the risk factor class, that is, the potential impact intensity mentioned above, L i Indicates the i The expected impact of this type of risk on system operation once it occurs, with the value determined based on the historical accident database, expert experience, or operation simulation evaluation; 、 、 are the weighted coefficients of the above three indicators, respectively, satisfying , used to express the relative importance of each risk quantification factor in the comprehensive assessment.

[0042] Specifically, with regard to the value and method of obtaining the potential impact intensity, the actual loss data (such as economic losses, downtime duration, etc.) when similar risks occur in historical records can be statistically collected, and the average impact degree can be obtained through fitting or weighted calculation as a basic reference; or experts in the hydropower field can classify or score the risk consequences based on the characteristics of the power station (such as scale and importance), and quantify them into the calculation; or a system simulation platform (such as a digital twin model) can be used to simulate the system response after the risk occurs, evaluate the impact on core functions (power generation, flood control, etc.), and output the simulation calculation value. The present invention does not make specific limitations on this.

[0043] Furthermore, in order to integrate the originally scattered single-moment risk indices into time-correlated sequence data, so as to intuitively present the changes in the activity levels of various risk factors over time (such as the continuous increase in equipment risk indexes and the sudden fluctuations in environmental risk indexes), and thus provide a structured time-series basis for identifying risk evolution laws and dynamically matching emergency plans, the invention performs time-series assembly processing on the comprehensive risk index vectors in the comprehensive risk index set, and orderly integrates the comprehensive risk index vectors of different time nodes according to the time dimension to form a continuous and traceable risk evolution sequence, that is, to obtain a risk index sequence. The specific processing method is as follows: 1. Time dimension alignment: Based on the preset unified timestamp (such as milliseconds or seconds), the comprehensive risk index vector (such as t 1 moment S 1( t 1), S 2( t 1),..., S m ( t m )], t 2 moments S 1( t 2), S 2( t2),..., S m ( t m )] etc.) are sorted in chronological order to ensure that each vector corresponds to a unique time point and avoid time confusion.

[0044] 2. Sequence structure construction: Arrange the sorted comprehensive risk index vectors vertically along the time axis to form a two-dimensional time series matrix. For example, if in the time window [ t 0, t N ] N +1 moment vector, the risk index sequence can be expressed as: [ S 1( t 0), S 2( t 0),..., S m ( t 0)], / / t Risk index vector at time 0] [ S 1( t 1), S 2( t 1),...,S m ( t 1)], / / t Risk index vector at time 1] ... [ S 1( t N ), S 2( t N ),..., S m ( t N )] / / t N Risk index vector at the moment] Among them, each row corresponds to the comprehensive index of all risk factors at a time point, and each column corresponds to the index change trend of a certain type of risk factor in the full time window.

[0045] 3. Time continuity processing: If there are missing time nodes (such as data collection failure at a certain moment), linear interpolation or filling the mean of adjacent moments is used to supplement the missing vector to ensure the continuity of the sequence in the time dimension and avoid the impact of data gaps on the subsequent analysis of risk evolution trends.

[0046] 4. Dynamic Window Association: Bind the assembled time series to a preset time sliding window (such as a 5-minute or 10-minute window) so that the sequence can reflect both the real-time risk status (the vector at the current moment) and the short-term evolution trend of the risk (the change in the vector within the window), providing dynamic input for subsequent plan matching based on time series features.

[0047] Thus, through the above method m The comprehensive risk index set of each risk factor is output according to the time dimension to form a risk index sequence S ( t )=[ S 1( t ) ,S 2( t ) , ... ,S m ( t )], as the trigger basis for subsequent intelligent plan generation and the calculation basis for plan matching. S ( t )express t The comprehensive risk index collection at each moment, S 1( t ) represents the comprehensive risk index of the first category risk factor, S 2( t ) represents the comprehensive risk index of the second category risk factors, S 3( t ) indicates the m The comprehensive risk index of similar risk factors, and so on. This series has time dependence and variable interpretation, and can dynamically reflect the main risk types and weight structures currently faced by the system.

[0048] In summary, this module achieves dynamic assembly, parameter reconstruction, and execution evolution of response strategies by constructing a risk-driven structural activation model, combining historical emergency plan templates with real-time risk characteristics. Its key features are the use of risk indices to construct an index mapping for emergency plan structures and the introduction of a feedback-driven structural weight fine-tuning mechanism. This enables continuous self-adaptation and online optimization capabilities for emergency plan generation, significantly improving adaptability and response efficiency in the complex and ever-changing cascade hydropower remote centralized control system.

[0049] S4: Based on the risk index sequence, the preset plan template set is dynamically weighted and superimposed to generate an adaptive plan set.

[0050] Intelligent plan generation is the core link in realizing the functional transformation of "risk index drive-response strategy construction" in this invention. Its role is to dynamically construct a set of cascade hydropower remote centralized control plans that accurately correspond to the current system risk situation based on the risk factor identification results and quantitative indicators output by the previous module.

[0051] Reference Manual Figure 2 , which shows a flow chart of an adaptive plan generation method provided by the present invention.

[0052] In a possible implementation, the step S4 dynamically weights and superimposes a preset set of emergency plan templates based on the risk index sequence to generate an adaptive emergency plan set, specifically including sub-steps S401 to S403: S401: Generate a risk response impact vector based on a preset response weight set and risk index sequence; S402: Mapping the risk response impact vector to a plan structure factor vector through a structural activation mapping function; S403: Dynamically weighting and superimposing the preset plan template set based on the plan structure factor vector to generate an adaptive plan set.

[0053] After obtaining the risk factor set R 1 ,R 2 , ... ,R m and its corresponding risk index series S 1( t ) ,S 2( t ) , ... ,S m ( t ), static emergency plan templates can be transformed into multi-response path plans with variable structures and adaptive parameters through further modeling and calculation, thus realizing the transition from data-driven to strategy-generated.

[0054] Specifically, for any time t , the total number of identified risk factors is m , risk index series ,in S i ( t ) indicates the i Risks of this type t To capture the nonlinear weight relationship of each risk’s contribution to the plan structure, we first construct the risk response impact vector , which is defined as: in For the i The response weight corresponding to the risk type satisfies the normalization condition The weight represents the system's attention level or control preference for various risk factors, which can be trained based on historical execution priorities. This vector is passed as input to the plan structure to activate the mapping function. , generate the plan structure index vector z ( t ): in is the strategy structure weight matrix, is the bias term, is a nonlinear activation function (Sigmoid or Softmax can be used), is the contingency plan structural factor vector, representing the configuration weight of the stress response under the current system state. Each dimension in this vector represents the activation strength of a contingency plan component in the overall response scheme, which is used to guide subsequent strategy assembly and parameter customization.

[0055] exist z ( t ) is determined, the system enters the plan generation stage, and further dynamically weights and superimposes the preset plan template set based on the plan structure factor vector to generate an adaptive plan set.

[0056] In a possible implementation, the step S403 dynamically weights and superimposes the preset plan template set based on the plan structure factor vector to generate an adaptive plan set, specifically including: Performing structural normalization processing on the plan structure factor vector to obtain the plan structure factor vector after structural normalization processing; Obtain the structural activation weight matrix corresponding to the plan structure factor vector after structural normalization; Based on the structure activation weight matrix, the combination mode of the preset plan templates is dynamically scheduled to obtain the dynamic combination mode of the preset plan template set; Based on the dynamic combination method, the preset plan template set is dynamically weighted and superimposed to obtain an adaptive plan set.

[0057] Assume that the set of plan templates is ,in T j Indicates the j A strategy template, including response logic, execution module, control parameter interface and scheduling control process. The generated plan structure is composed of multiple templates, and its combination is based on the structure activation value. z ( t ) for scheduling. To ensure the rationality and stability of structure generation, the system introduces structure normalization processing so that the template call strength meets the following requirements: On this basis, each final plan P ( t ) can be expressed as the weighted superposition result of the template set: in P ( t ) is the final generated centralized control plan, For the A basic template can include voltage regulation schemes, load switching strategies, node control logic, personnel response paths, etc. Although the representation is a linear combination in symbolic form, the actual execution is manifested as module-level execution assembly, that is, selecting a template based on activation strength and loading its parameter configuration to build the control process.

[0058] In a possible implementation, after dynamically weighting and superimposing the preset plan template set based on the plan structure factor vector to generate an adaptive plan set, the method further includes: Get the execution effect evaluation function; Evaluate the effect of the adaptive plan set according to the execution effect evaluation function to obtain the effect evaluation result; The effect evaluation results include the execution efficiency indicators of each plan in the adaptive plan set during the execution process. The execution efficiency indicators include response time, control stability and goal achievement degree; Obtain historical execution performance data; Based on historical execution effect data and effect evaluation results, the structure activation weight matrix corresponding to the plan structure factor vector is gradient updated to adaptively adjust the adaptive plan set.

[0059] In order to improve the adaptability and execution quality of plan generation, the system introduces an evolutionary mechanism based on historical feedback after the structure is generated. Define the historical execution effect evaluation function , measure the response time, control stability, target achievement and other indicators of the current plan in actual execution. k The effectiveness of the plan execution within the round E ( P ( t - k +1)) , ... ,E ( P ( t )), and the activation matrix W Perform gradient-based fine-tuning: in is the learning rate, Represents the gradient information of the evaluation function with respect to the structure matrix. This mechanism enables continuous adaptive adjustment of the plan structure weights, allowing the system to continuously enhance effective strategies and automatically weaken ineffective structures during long-term operation, thereby achieving self-evolution and optimization of the adaptive plan set.

[0060] S5: Based on the real-time operating status of the cascade hydropower remote centralized control system, dynamic matching is performed in the adaptive plan set to determine the optimal plan, so as to control the cascade hydropower remote centralized control system based on the optimal plan.

[0061] As the terminal response scheduling component of the system of the present invention, the automatic matching and optimization of emergency plans is responsible for accurately matching the constructed emergency plans with the current operating status of the cascade hydropower remote control system and dynamically executing them. Its core value lies in achieving high adaptability matching and effect optimization between the emergency plans and the real-time control environment.

[0062] This step is based on the plan set P ( t ), and combined with the state vector obtained by the system operation monitoring module By constructing a dynamic matching scoring function and a response path selection mechanism, the optimal strategy path is screened out and its execution parameters are fine-tuned to improve the immediacy and control accuracy of the regulation effect.

[0063] In one possible implementation, based on the real-time operating status of the cascade hydropower remote centralized control system, dynamic matching is performed in the adaptive plan set to determine the optimal plan, specifically including: Build a dynamic matching scoring function; The matching degree between each plan in the adaptive plan set and the real-time operation status is calculated through a dynamic matching scoring function to obtain a matching degree set; Determine candidate emergency plans based on preset matching rules and matching degree sets; Gradient fine-tuning is performed on the key control parameter sets of the candidate emergency plans to optimize the preset objective function values ​​and obtain the optimal plan.

[0064] In terms of specific design, The state vector set corresponding to the system operation state at the moment is The state vector set includes the target state vectors corresponding to the system operation parameters such as the current water level of each level of reservoirs, unit load status, communication delay, and master control strategy indicators. The adaptive plan set is , each plan They all contain response strategies with adjustable structures and weighted parameters, which can specifically include execution structures (such as control logic topology, scheduling steps), parameter configurations (such as adjustment thresholds, execution timing) and historical execution records. P ( t ) and further fine-tune the parameters, introducing a dynamic matching scoring function , the calculation of dynamic matching scoring function should be based on and The correlation characteristics of the three indicators are quantified item by item: Dynamic matching scoring function Defined as: in, For the contingency plan The topological deviation index of the execution structure and the current system control logic, is the error prediction of the main indicators of the system (such as water level and load) under the current state of the emergency plan strategy. is the average response effect loss of the plan under similar historical conditions, that is, the historical response loss, is the importance weight of each indicator, satisfying This function is used to evaluate the compatibility between each candidate plan and the current system state. The smaller the matching score, the better the plan.

[0065] Specifically, the calculation methods of the above indicators include: 1. Topological deviation index : Comparison Plan The execution structure of the system is consistent with the topology of the current system control logic. For example, if the current control logic of the system is "cascade joint scheduling priority", and the plan If the execution structure is "single plant independent regulation", the deviation is quantified by the topological difference algorithm (such as node matching calculation based on graph theory), and the topological deviation index The larger the value, the worse the structural adaptability.

[0066] 2. Error prediction amount : Based on real-time operating status , forecast plan The degree of deviation of key system indicators (water level, load, etc.) after execution. For example, based on the current reservoir water level and contingency plans The adjustment target value is calculated by using an error prediction model (such as linear regression or LSTM), The smaller the value, the higher the control accuracy of the plan on the target.

[0067] 3. Historical response loss : Search the history database with the current status Similar scenarios (matched by cosine similarity or Euclidean distance) to extract plans The average loss is calculated by weighted average of the execution effect data (such as response delay and adjustment overshoot) in similar scenarios. Reflects the historical reliability of the plan.

[0068] Further adopt the weighted summation formula (in =1, the weight is dynamically adjusted according to the current control priority of the system), for each plan Calculate the score.

[0069] Then traverse the adaptive plan set , all scores are aggregated to form a matching degree set [ , ,..., ], each element corresponds to the matching degree between a plan and the real-time status, providing a quantitative basis for the subsequent screening of candidate plans.

[0070] After obtaining the matching degree set (i.e. completing the matching score), the candidate emergency plans are determined based on the preset matching rules and matching degree set. Specifically, the preset matching rule can be to select the plan with the smallest matching score. As the optimal execution strategy at the current moment. Therefore, the optimization goal can be defined as: That is, select the plan with the smallest matching score among all candidate plans As a candidate emergency plan at the current moment.

[0071] It is worth noting that the gradient fine-tuning of the key control parameter set of the candidate emergency plan in the above-mentioned implementation method is consistent with the method of gradient fine-tuning of the key control parameter set of the optimal plan after controlling the cascade hydropower remote control system based on the optimal plan in the following text. The specific method will be elaborated below and will not be repeated here.

[0072] In a possible implementation, after controlling the cascade hydropower remote centralized control system based on the optimal plan, the method further includes: Obtain the real control indicators corresponding to the optimal plan; Based on the expected indicators and the actual control indicators, calculate the performance loss corresponding to the optimal plan; Calculate the gradient of the performance loss with respect to the weight vector in the dynamic matching scoring function; The weight vector is dynamically updated according to the gradient.

[0073] In order to further improve the execution quality of the plan, the module also introduces a local parameter optimization process to optimize the optimal plan. A set of control parameters Perform gradient fine-tuning to optimize the objective function value as much as possible while keeping the original response logic unchanged. The optimization iterative update form is as follows: in To optimize the step size, Represents matching function to parameters This process performs a finite number of iterations while meeting the real-time message processing and execution control time limits to ensure that parameter adjustments are completed within an acceptable time window, thereby improving execution efficiency and accuracy.

[0074] In addition, in order to improve the long-term stability of the system, the module also introduces a dynamic weight update mechanism, that is, according to the plan Dynamically adjust weights based on performance during actual execution value, so that it is more in line with the control objectives of the current operation stage of the system.

[0075] Assume that the actual control index in a certain operation cycle is y ( t ), then the performance loss function is introduced ,in, yref Represents the expected control index. The weight update adopts the following rules: in represents the update step size, It represents the gradient of the loss function with respect to the weight, and is used to guide the matching mechanism to adjust towards the direction of minimum performance loss to adapt to the long-term evolution of the system.

[0076] In summary, this implementation method, by constructing a risk-driven response matching evaluation mechanism and parameter fine-tuning process, not only achieves precise adaptation of the plan to the system status, but also endows the strategy execution with dynamic self-optimization capabilities, breaking through the bottleneck of static selection and parameter solidification of traditional response plans, and has significant dynamic adaptability and execution efficiency improvement value in the cascade hydropower remote control system.

[0077] Example 2 Reference Manual Figure 3 , showing a structural schematic diagram of a cascade hydropower remote centralized control plan generation system provided by the present invention.

[0078] An embodiment of the present invention provides a cascade hydropower remote centralized control plan generation system 20, comprising: The data acquisition module 201 is used to obtain multi-source heterogeneous dynamic data of the target cascade hydropower station; A statistical reconstruction module 202 is used to statistically reconstruct the multi-source heterogeneous dynamic data to generate a dynamic feature tensor, which is used to characterize the temporal evolution and statistical distribution characteristics of the multi-source heterogeneous dynamic data in a structured multidimensional space; a risk identification module 203, configured to perform risk identification on the target cascade hydropower station according to the dynamic characteristic tensor, and obtain a risk index sequence of the target cascade hydropower station; The emergency plan generation module 204 is configured to dynamically weight and superimpose a preset emergency plan template set based on the risk index sequence to generate an adaptive emergency plan set; The plan matching module 205 is used to dynamically match the adaptive plan set based on the real-time operating status of the cascade hydropower remote centralized control system and determine the optimal plan so as to control the cascade hydropower remote centralized control system based on the optimal plan.

[0079] In a possible implementation, the plan generation module 204 is specifically configured to: Generating a risk response impact vector according to a preset response weight set and the risk index sequence; Mapping the risk response impact vector into a plan structure factor vector through a structural activation mapping function; Based on the plan structure factor vector, the preset plan template set is dynamically weighted and superimposed to generate an adaptive plan set.

[0080] In a possible implementation, the plan generation module 204 is specifically configured to: Performing structural normalization processing on the plan structure factor vector to obtain a structurally normalized plan structure factor vector; Obtaining a structural activation weight matrix corresponding to the plan structure factor vector after the structural normalization processing; Based on the structure activation weight matrix, dynamically scheduling the combination mode of the preset plan templates to obtain the dynamic combination mode of the preset plan template set; Based on the dynamic combination method, the preset plan template set is dynamically weighted and superimposed to obtain an adaptive plan set.

[0081] In a possible implementation, the cascade hydropower remote centralized control plan generation system 20 further includes a first weight updating module 206, which is specifically configured to: After dynamically weighting and superimposing the preset plan template set based on the plan structure factor vector to generate an adaptive plan set, obtaining an execution effect evaluation function; Performing an effect evaluation on the adaptive plan set according to the execution effect evaluation function to obtain an effect evaluation result; The effect evaluation results include the execution performance indicators of each plan in the adaptive plan set during the execution process, and the execution performance indicators include response time, control stability and target achievement degree; Obtain historical execution performance data; Based on the historical execution effect data and the effect evaluation result, the structure activation weight matrix corresponding to the plan structure factor vector is gradient updated to adaptively adjust the adaptive plan set.

[0082] In a possible implementation, the plan matching module 205 is specifically configured to: Build a dynamic matching scoring function; Calculating the matching degree between each plan in the adaptive plan set and the real-time operating state by using the dynamic matching scoring function to obtain a matching degree set; Determining candidate emergency plans based on preset matching rules and the matching degree set; Gradient fine-tuning is performed on the key control parameter set of the candidate emergency plan to optimize the preset objective function value and obtain the optimal plan.

[0083] In a possible implementation, the cascade hydropower remote centralized control plan generation system 20 further includes a second weight updating module 207, and the second weight updating module 207 is specifically configured to: After controlling the cascade hydropower remote centralized control system based on the optimal plan, obtaining a real regulation index corresponding to the optimal plan; Calculating the performance loss corresponding to the optimal plan based on the expected indicator and the actual control indicator; Calculating the gradient of the performance loss with respect to a weight vector in the dynamic matching scoring function; The weight vector is dynamically updated according to the gradient.

[0084] In a possible implementation, the risk identification module 203 is specifically configured to: Extracting risk factors from the dynamic feature tensor according to a preset risk factor set to obtain a target risk factor set; Extracting a subset of sensor variable features associated with each type of risk factor in the target risk factor set to obtain a set of risk feature vectors; Based on the risk feature vector set, performing a weighted nonlinear transformation on each type of risk factor in the target risk factor set to obtain a risk activity scalar value set; Based on the risk activity scalar value set and the preset core indicator data of each type of factor in the target risk factor set, a standardized calculation is performed to obtain a comprehensive risk index set; the preset core indicator data includes historical frequency, current activity and potential impact intensity; The comprehensive risk index vectors in the comprehensive risk index set are subjected to time series assembly processing to obtain a risk index sequence.

[0085] In a possible implementation, the statistical reconstruction module 202 is specifically configured to: Performing data preprocessing on the multi-source heterogeneous dynamic data to eliminate dynamic anomalies in the multi-source heterogeneous dynamic data, thereby obtaining preprocessed multi-source heterogeneous data; A sliding feature reconstruction mechanism is introduced to perform sliding feature reconstruction on each type of variable in the preprocessed multi-source heterogeneous data to generate a time-series aligned structured feature set; Each type of variable in the structured feature set is encapsulated as a structured feature tensor based on a time axis to obtain a dynamic feature tensor.

[0086] In a possible implementation, the statistical reconstruction module 202 is specifically configured to: Calculating a set of statistics corresponding to each type of variable in the preprocessed multi-source heterogeneous data within a preset sliding window, wherein the set of statistics includes a mean and a weighted variance; Based on the statistical quantity set, encapsulating the statistical quantity of each type of variable in the multi-source heterogeneous data at each moment into a binary structural feature to obtain a binary structural feature set; The binary structural features in the binary structural feature set are temporally assembled to obtain a temporally aligned structured feature set.

[0087] The embodiment of the present invention provides a cascade hydropower remote centralized control plan generation system 20, which can implement the steps and effects of the cascade hydropower remote centralized control plan generation method in Example 1. To avoid repetition, the present invention will not elaborate on them.

[0088] The beneficial effects of the present invention are embodied in: First, by integrating multi-source heterogeneous dynamic data and constructing dynamic feature tensors, a unified and structured representation of complex operating states is achieved, significantly improving the comprehensiveness and timeliness of risk perception. Second, based on a dynamic risk index sequence, the plan templates are dynamically weighted and superimposed to generate a highly adaptive plan set, overcoming the rigidity and low matching defects of traditional plan libraries. Finally, by dynamically matching and executing the optimal plan based on real-time operating status, the accuracy, agility, and pertinence of the emergency response are ensured, greatly improving the safety assurance capabilities and emergency control efficiency of the cascade hydropower remote centralized control system in complex and changing environments.

[0089] In the description of the embodiments of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "center", "top", "bottom", "top", "bottom", "inside", "outside", "inner side", "outer side" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. Among them, "inside" refers to an internal or enclosed area or space. "Periphery" refers to the area surrounding a specific component or specific area.

[0090] In the description of the embodiments of the present invention, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly specifying the number of the technical features indicated. Therefore, a feature specified as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0091] In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," "connected," and "assembled" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to direct connections, indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention in specific circumstances.

[0092] In the description of the embodiments of the present invention, specific features, structures, materials or characteristics may be combined in an appropriate manner in any one or more embodiments or examples.

[0093] In describing the embodiments of the present invention, it should be understood that "-" and "~" represent a range between two values, and the range includes the endpoints. For example, "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0094] In describing the embodiments of the present invention, the term "and / or" is used herein to describe a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. Furthermore, the character " / " is generally used herein to indicate that the associated objects are in an "or" relationship.

[0095] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating a remote centralized control plan for cascade hydropower, characterized in that: include: Acquire multi-source heterogeneous dynamic data of target cascade hydropower stations; Statistically reconstructing the multi-source heterogeneous dynamic data to generate a dynamic feature tensor, wherein the dynamic feature tensor is used to characterize the temporal evolution and statistical distribution characteristics of the multi-source heterogeneous dynamic data in a structured multidimensional space; performing risk identification on the target cascade hydropower station according to the dynamic characteristic tensor to obtain a risk index sequence of the target cascade hydropower station; Based on the risk index sequence, a set of preset emergency plan templates is dynamically weighted and superimposed to generate an adaptive emergency plan set; Based on the real-time operating status of the cascade hydropower remote centralized control system, dynamic matching is performed in the adaptive plan set to determine the optimal plan, so as to control the cascade hydropower remote centralized control system based on the optimal plan.

2. The method according to claim 1, characterized in that The step of dynamically weighting and superimposing a preset set of emergency plan templates based on the risk index sequence to generate an adaptive emergency plan set includes: Generating a risk response impact vector according to a preset response weight set and the risk index sequence; Mapping the risk response impact vector into a plan structure factor vector through a structural activation mapping function; Based on the plan structure factor vector, the preset plan template set is dynamically weighted and superimposed to generate an adaptive plan set.

3. The method according to claim 2, characterized in that The method of dynamically weighting and superimposing a preset plan template set based on the plan structure factor vector to generate an adaptive plan set includes: Performing structural normalization processing on the plan structure factor vector to obtain a structurally normalized plan structure factor vector; Obtaining a structural activation weight matrix corresponding to the plan structure factor vector after the structural normalization processing; Based on the structure activation weight matrix, dynamically scheduling the combination mode of the preset plan templates to obtain the dynamic combination mode of the preset plan template set; Based on the dynamic combination method, the preset plan template set is dynamically weighted and superimposed to obtain an adaptive plan set.

4. The method according to claim 2, characterized in that After dynamically weighting and superimposing the preset plan template set based on the plan structure factor vector to generate an adaptive plan set, the method further includes: Get the execution effect evaluation function; Performing an effect evaluation on the adaptive plan set according to the execution effect evaluation function to obtain an effect evaluation result; The effect evaluation results include the execution performance indicators of each plan in the adaptive plan set during the execution process, and the execution performance indicators include response time, control stability and target achievement degree; Obtain historical execution performance data; Based on the historical execution effect data and the effect evaluation result, the structure activation weight matrix corresponding to the plan structure factor vector is gradient updated to adaptively adjust the adaptive plan set.

5. The method according to claim 1, characterized in that The real-time operating status of the cascade hydropower remote centralized control system is dynamically matched in the adaptive plan set to determine the optimal plan, including: Build a dynamic matching scoring function; Calculating the matching degree between each plan in the adaptive plan set and the real-time operating state by using the dynamic matching scoring function to obtain a matching degree set; Determining candidate emergency plans based on preset matching rules and the matching degree set; Gradient fine-tuning is performed on the key control parameter set of the candidate emergency plan to optimize the preset objective function value and obtain the optimal plan.

6. The method according to claim 5, characterized in that After controlling the cascade hydropower remote centralized control system based on the optimal plan, the method further includes: Obtaining the actual control indicators corresponding to the optimal plan; Calculating the performance loss corresponding to the optimal plan based on the expected indicator and the actual control indicator; Calculating the gradient of the performance loss with respect to a weight vector in the dynamic matching scoring function; The weight vector is dynamically updated according to the gradient.

7. The method according to claim 1, characterized in that The step of performing risk identification on the target cascade hydropower station according to the dynamic characteristic tensor to obtain a risk index sequence of the target cascade hydropower station includes: Extracting risk factors from the dynamic feature tensor according to a preset risk factor set to obtain a target risk factor set; Extracting a subset of sensor variable features associated with each type of risk factor in the target risk factor set to obtain a set of risk feature vectors; Based on the risk feature vector set, performing a weighted nonlinear transformation on each type of risk factor in the target risk factor set to obtain a risk activity scalar value set; Based on the risk activity scalar value set and the preset core indicator data of each type of factor in the target risk factor set, a standardized calculation is performed to obtain a comprehensive risk index set; the preset core indicator data includes historical frequency, current activity and potential impact intensity; The comprehensive risk index vectors in the comprehensive risk index set are subjected to time series assembly processing to obtain a risk index sequence.

8. The method according to claim 1, characterized in that The statistical reconstruction of the multi-source heterogeneous dynamic data to generate a dynamic feature tensor includes: Performing data preprocessing on the multi-source heterogeneous dynamic data to eliminate dynamic anomalies in the multi-source heterogeneous dynamic data and obtain preprocessed multi-source heterogeneous data; A sliding feature reconstruction mechanism is introduced to perform sliding feature reconstruction on each type of variable in the preprocessed multi-source heterogeneous data to generate a time-series aligned structured feature set; Each type of variable in the structured feature set is encapsulated as a structured feature tensor based on a time axis to obtain a dynamic feature tensor.

9. The method according to claim 8, characterized in that The sliding feature reconstruction mechanism is introduced to perform sliding feature reconstruction on each type of variable in the preprocessed multi-source heterogeneous data to generate a time-aligned structured feature set, including: Calculating a set of statistics corresponding to each type of variable in the preprocessed multi-source heterogeneous data within a preset sliding window, wherein the set of statistics includes a mean and a weighted variance; Based on the statistical quantity set, encapsulating the statistical quantity of each type of variable in the multi-source heterogeneous data at each moment into a binary structural feature to obtain a binary structural feature set; The binary structural features in the binary structural feature set are temporally assembled to obtain a temporally aligned structured feature set.

10. A cascade hydropower remote centralized control plan generation system, characterized in that: include: Data acquisition module, used to obtain multi-source heterogeneous dynamic data of the target cascade hydropower station; a statistical reconstruction module, configured to statistically reconstruct the multi-source heterogeneous dynamic data to generate a dynamic feature tensor, wherein the dynamic feature tensor is used to characterize the temporal evolution and statistical distribution characteristics of the multi-source heterogeneous dynamic data in a structured multidimensional space; a risk identification module, configured to perform risk identification on the target cascade hydropower station according to the dynamic characteristic tensor, and obtain a risk index sequence of the target cascade hydropower station; A plan generation module is used to dynamically weight and superimpose a set of preset plan templates based on the risk index sequence to generate an adaptive plan set; The plan matching module is used to dynamically match the adaptive plan set based on the real-time operating status of the cascade hydropower remote centralized control system, determine the optimal plan, and control the cascade hydropower remote centralized control system based on the optimal plan.

Images