Hydroelectric generating set stability prediction method and system

Abstract

The invention discloses a hydroelectric generating set stability prediction method and system, and aims to solve the problem of insufficient generalization ability of stability prediction under extreme working conditions such as a flood period and a dry season in the prior art. The method comprises the following steps: preprocessing historical operation data containing extreme working conditions of a target hydroelectric generating set; acquiring extreme working condition data of the associated cross-domain unit; constructing a stability prediction model fusing physical mechanism embedding and cross-domain transfer learning; and inputting real-time data, and outputting prediction results of the vibration amplitude, the throw value and the winding temperature rise. The system correspondingly comprises a unit historical data processing module, a unit screening and data acquisition module, a physical mechanism-cross-domain migration prediction model construction module and a real-time data input and prediction output module. The method can supplement extreme working condition training data, improves the prediction precision under the extreme working condition, guarantees the stable operation and maintenance of the hydroelectric generating set under all working conditions, and is high in practicability.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically, to a method and system for predicting the stability of hydropower units. Background Technology

[0002] Hydropower units are core equipment for clean energy production, and their operational stability directly affects the safety and energy efficiency of the power system. Therefore, stability prediction technology has become a key focus of hydropower operation and maintenance. Existing prediction methods have formed a multi-technology integration system, mainly including two categories: traditional monitoring and intelligent models. Traditional methods collect data through sensors such as vibration, temperature, and flow, and combine this with FFT spectrum analysis and trend fitting to extract features, achieving early warning of faults under normal operating conditions. In intelligent models, time series models such as SVM, random forest, and LSTM can capture nonlinear data correlations. Some solutions also introduce digital twin technology to construct virtual models, combining multi-source data fusion simulation. These methods can already meet basic requirements under normal operating conditions.

[0003] However, existing technologies have significantly insufficient generalization ability under extreme operating conditions (flood season, dry season). Under extreme operating conditions, factors such as water flow impact and cavitation effects cause sudden changes in unit parameters, but due to the scarcity of historical data, data-driven models lack sufficient training samples; most intelligent models do not embed the coupled physical mechanism of "hydraulic-electromechanical-structure", and when the data distribution is off, they are prone to outputting prediction results that violate the law; moreover, existing models mostly rely on cloud computing, and under extreme operating conditions, parameters change rapidly, and data transmission and computing delays reduce the timeliness of predictions, making it difficult to support timely operation and maintenance decisions. Summary of the Invention

[0004] This invention proposes a method and system for predicting the stability of hydropower units, aiming to improve the generalization ability of stability prediction under extreme operating conditions (flood season and dry season), solve the problems of scarce data, prediction that easily violates physical laws and lacks timeliness in existing technologies, and ensure stable operation and maintenance of the units under all operating conditions.

[0005] The technical solution of the present invention is as follows: A method for predicting the stability of a hydropower unit includes the following steps: Historical operating data of the target hydropower unit is acquired and preprocessed. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season conditions and dry season conditions. Identify the cross-domain hydropower units that are related to the target hydropower unit, and then obtain the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data; Based on the historical operating data and cross-domain extreme operating condition data of the target hydropower unit, a stability prediction model integrating physical mechanism embedding and cross-domain transfer learning is constructed. The real-time operating data of the target hydropower unit is input into the stability prediction model, and the stability prediction results for a future preset time period are output. The stability prediction results include the predicted values ​​of vibration amplitude, sway value and winding temperature rise.

[0006] In one possible implementation, the preprocessing includes data cleaning, outlier removal, and feature extraction; the data cleaning uses sliding window filtering to remove noise interference from the original data; the outlier removal identifies and deletes abnormal data that deviates from the normal distribution based on the 3σ criterion; and the feature extraction includes extracting frequency features from vibration signals through FFT spectrum analysis and extracting temperature rise trend features from temperature signals through trend fitting. The feature extraction also includes performing multi-scale empirical mode decomposition on flow rate data and head data to extract time-frequency joint features of sudden changes in flow velocity and sudden rises and falls in head under extreme conditions; at the same time, Pearson correlation analysis is performed on rotational speed data and power data to explore the electromechanical coupling features between the two and enrich the feature dimensions of the model input.

[0007] In one possible implementation, the correlation between the cross-regional hydropower unit and the target hydropower unit is determined by comprehensively considering the following indicators: the matching degree of the rated capacity of the units, the similarity of the hydrological characteristics of the respective river basins, and the consistency of the hydraulic-electromechanical structure. The process of acquiring cross-domain extreme operating condition data employs federated learning technology to protect privacy. Specifically, this includes: the local nodes of the cross-domain hydropower units perform homomorphic encryption on their own extreme operating condition data, extracting only the feature vectors of the data and uploading them to the federated aggregation node; the aggregation node performs federated average aggregation on the encrypted feature vectors uploaded by each cross-domain node, generating a unified cross-domain extreme operating condition feature set, which is then fed back to the model training node of the target hydropower unit, without transmitting the original data of the cross-domain hydropower units throughout the entire process.

[0008] In one possible implementation, when constructing the stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning, the cross-domain extreme operating condition data is first subjected to generative data augmentation with physical mechanism constraints to generate synthetic extreme operating condition data that conforms to the physical characteristics of the target hydropower unit. The generative data augmentation is achieved through a physical constraint generative adversarial network. The generator of this network embeds the differential equation of the hydropower unit shaft system vibration and the turbine flow-head-efficiency characteristic curve as physical constraints during the data generation process. The discriminator verifies the generated data based on the physical laws of the historical extreme operating condition data of the target hydropower unit and outputs effective synthetic data through iterative adversarial training.

[0009] In one possible implementation, the stability prediction model includes: The physical mechanism prior layer is used to provide the model with objective physical constraints of the hydropower unit. Specifically, the physical constraints are implemented in the following ways: the coupling relationship between the bearing temperature rise and speed and load of the hydropower unit, and the relationship between the hydraulic excitation frequency and the water flow velocity are transformed into mathematical constraint formulas; during the model training process, if the prediction result violates the mathematical constraint formulas, a physical consistency penalty term is added to the loss function to forcibly correct the model parameters and ensure that the prediction result conforms to the objective physical laws of the hydropower unit. The cross-domain transfer learning layer is used to solve the problem of cross-domain data distribution offset and realize cross-domain extreme working condition knowledge transfer. It adopts a domain adaptive network to realize cross-domain knowledge transfer, which includes: extracting features from cross-domain extreme working condition data and target hydropower unit data respectively to obtain source domain features and target domain features; learning the distribution difference between the two types of features through a domain discriminator; and adjusting the feature extraction network parameters based on the maximum mean difference minimization criterion to make the distribution of source domain features and target domain features tend to be consistent, thus solving the problem of cross-domain data distribution offset. The time-series prediction layer is used to accurately capture the dynamic changes of stability indicators under extreme conditions. It adopts an improved bidirectional long short-term memory network with an added attention mechanism. The attention mechanism weights and enhances the key time-step features under extreme conditions, strengthens the model's ability to capture abrupt changes in stability indicators under extreme conditions, and improves prediction accuracy.

[0010] In one possible implementation, a dynamic correction process for the prediction results is also included: real-time acquisition of the current operating parameters of the target hydropower unit, calculation of the deviation between the measured value and the corresponding predicted value input by the model; if the deviation exceeds a preset threshold, the hydraulic characteristic constraint formula of the physical mechanism prior layer is called to compensate and correct the prediction results of vibration amplitude and swing value until the deviation is lower than the preset threshold, thereby eliminating prediction drift under extreme operating conditions.

[0011] In one possible implementation, the real-time prediction module of the stability prediction model is deployed on the local edge node of the target hydropower unit; the edge node communicates directly with the unit's vibration sensor, temperature sensor, and flow sensor to realize local acquisition and predictive inference of real-time operating data, reduce the delay of data transmission to the cloud, and ensure the real-time performance of prediction results under extreme operating conditions.

[0012] A hydropower unit stability prediction system, comprising: The historical data processing module of the unit acquires and preprocesses the historical operating data of the target hydropower unit. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season conditions and dry season conditions. The unit screening and data acquisition module identifies cross-domain hydropower units that are related to the target hydropower unit, and then acquires the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data. The physical mechanism-cross-domain migration prediction model construction module constructs a stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning based on the historical operation data and cross-domain extreme working condition data of the target hydropower unit. The real-time data input and prediction output module inputs the real-time operating data of the target hydropower unit into the stability prediction model and outputs the stability prediction results for a future preset time period. The stability prediction results include the predicted values ​​of vibration amplitude, swing value and winding temperature rise.

[0013] A computer device, including A memory that stores computer-readable instructions; A processor, which, when executing the computer-readable instructions, implements the steps of the hydropower unit stability prediction method as described above.

[0014] A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the hydropower unit stability prediction method described above.

[0015] Compared with the prior art, the present invention has the following main advantages: The method of this invention acquires and preprocesses historical operating data of the target hydropower unit covering extreme conditions (flood season and dry season), and simultaneously identifies cross-domain hydropower units associated with the target unit and acquires their extreme condition data, thereby supplementing model training samples under extreme conditions and alleviating the problem of scarce extreme condition data in existing technologies. Based on the historical data of the target unit and cross-domain extreme data, a stability prediction model integrating physical mechanism embedding and cross-domain transfer learning is constructed, enabling the model to both learn the laws of cross-domain extreme conditions and be constrained by physical mechanisms, avoiding prediction results that violate the coupled physical laws of "hydraulic-electromechanical-structural" relationships, and solving the problem of insufficient generalization ability in existing technologies. By inputting the real-time operating data of the target hydropower unit into the above-mentioned fusion model, the predicted results of vibration amplitude, sway value, and winding temperature rise within a preset future time period are output, achieving accurate acquisition of stability index prediction values ​​under extreme conditions, providing data support for operation and maintenance decisions, and improving the problem of poor prediction performance of existing technologies under extreme conditions. Attached Figure Description

[0016] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the method in Example 1; Figure 2 This is a system structure block diagram of Example 2; Figure 3 This is a block diagram of the computer device structure in Example 3. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Example 1

[0020] like Figures 1-2 As shown in Example 1, a method for predicting the stability of a hydropower unit is proposed, including the following steps: S100. Acquire historical operating data of the target hydropower unit and perform preprocessing. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season operating conditions and dry season operating conditions. Step S100 involves acquiring historical operating data of the target hydropower unit covering both normal and extreme operating conditions (flood season and dry season) and performing preprocessing. This provides a local data foundation containing extreme operating condition features for model training, making up for the initial shortcoming of scarce extreme operating condition data for a single unit. Furthermore, preprocessing improves data quality (such as reducing noise and unifying feature formats), providing more reliable input for subsequent model learning and avoiding model training bias caused by messy raw data.

[0021] S200. Identify the cross-domain hydropower units that are related to the target hydropower unit, and then obtain the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data. Step S200 overcomes the limitation of insufficient sample size for extreme operating conditions of a single unit by identifying cross-domain hydropower units that are correlated with the target hydropower unit and acquiring their extreme operating condition data. Correlation screening (e.g., based on unit parameters, watershed characteristics, etc.) ensures that cross-domain data and the target unit have similar physical or operational characteristics, avoiding the introduction of invalid data. This provides the model with richer and more applicable extreme operating condition learning materials, solving the problem in existing technologies where the scarcity of extreme data makes it difficult for models to learn the patterns of extreme operating conditions.

[0022] S300. Based on the historical operating data of the target hydropower unit and cross-domain extreme working condition data, a stability prediction model integrating physical mechanism embedding and cross-domain transfer learning is constructed. Step S300 constructs a stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning. On the one hand, physical mechanism embedding (such as the coupling law of "hydraulic-electromechanical-structural") can constrain the model output to conform to the objective operating law of the unit, avoiding prediction results that violate physical common sense when data distribution shifts under extreme operating conditions. On the other hand, cross-domain transfer learning can eliminate the distribution difference between cross-domain data and target unit data, enabling the model to effectively absorb knowledge from cross-domain extreme operating conditions and improve its adaptability to local extreme operating conditions. The combination of the two makes the model both data-driven flexibility and the reliability of physical laws, significantly improving the prediction accuracy under extreme operating conditions.

[0023] S400. Input the real-time operating data of the target hydropower unit into the stability prediction model and output the stability prediction results for a future preset time period. The stability prediction results include the predicted values ​​of vibration amplitude, swing value and winding temperature rise.

[0024] Step S400 inputs real-time operating data into the aforementioned fusion model and outputs predicted results such as vibration amplitude, sway value, and winding temperature rise within a preset time period. This allows for the early detection of changes in unit stability indicators under extreme operating conditions. Compared to existing technologies that suffer from prediction lag or insufficient accuracy under extreme conditions, this step provides timely and accurate prediction results, allowing maintenance personnel sufficient time to respond, reducing unplanned downtime or equipment damage caused by sudden failures under extreme conditions, and improving the safety and economy of hydropower unit operation under all conditions.

[0025] This embodiment also includes a dynamic correction process for the prediction results: real-time acquisition of the current operating parameters of the target hydropower unit (measured values ​​of head, flow rate, and rotational speed), calculation of the deviation between the measured value and the corresponding predicted value input by the model; if the deviation exceeds a preset threshold, the hydraulic characteristic constraint formula of the physical mechanism prior layer is called to compensate and correct the prediction results of vibration amplitude and swing value until the deviation is lower than the preset threshold, thereby eliminating prediction drift under extreme operating conditions.

[0026] The preset threshold is set based on the design parameters of the target hydropower unit (such as the rated head fluctuation range and allowable speed deviation) and the normal deviation range under historical extreme operating conditions. Specifically, it can be determined by statistically analyzing the deviation distribution between the measured values ​​and theoretical calculation values ​​of the target unit under extreme operating conditions in the past 3-5 years.

[0027] The hydraulic characteristic constraint formulas include, but are not limited to: vibration amplitude-head-flow correlation formulas derived from the turbine similarity law (e.g., vibration amplitude is positively correlated with the product of the square of the head and the flow rate), and swing value-speed-load correction formulas based on the shaft dynamics model.

[0028] This dynamic correction employs a closed-loop mechanism of real-time monitoring, deviation calculation, and dynamic correction, enabling the prediction results to dynamically adapt to rapidly changing operating environments under extreme conditions. This avoids the accuracy degradation of static prediction models during sudden changes in operating conditions. Simultaneously, it relies on physical mechanism constraint formulas for correction (rather than simply depending on statistical data patterns), ensuring that the correction process always conforms to the objective laws of the unit's hydraulic-electromechanical coupling. This further enhances the reliability of prediction results under extreme conditions and provides a more accurate basis for operation and maintenance decisions.

[0029] In this embodiment, the preprocessing includes data cleaning, outlier removal, and feature extraction. Data cleaning employs a sliding window filter to remove noise interference from the original data. Outlier removal identifies and deletes abnormal data that deviates from the normal distribution based on the 3σ criterion. Feature extraction includes extracting frequency features from vibration signals through FFT spectral analysis and extracting temperature rise trend features from temperature signals through trend fitting. The window size of the sliding window filter is dynamically adjusted according to the fluctuation frequency of extreme operating conditions (e.g., reducing the window to retain abrupt change features when water flow impacts are frequent during flood season). The scale parameters of the multi-scale empirical mode decomposition are set based on the abrupt change period of water flow and head under extreme operating conditions to ensure effective decomposition of transient abrupt change components in non-stationary signals.

[0030] The sliding window filtering in the preprocessing is used to remove high-frequency noise (such as sensor jitter interference) from the original data, and the 3σ criterion accurately identifies and removes outliers that deviate from the normal distribution (such as instantaneous false alarm data under extreme conditions). Together, they ensure the reliability of the data and avoid noise or outliers from misleading the model learning. This solves the bottleneck of model accuracy caused by the uneven quality of the original monitoring data. The feature extraction also includes performing multi-scale empirical mode decomposition on flow rate data and head data to extract time-frequency joint features of sudden changes in flow velocity and sudden rises and falls in head under extreme conditions; at the same time, Pearson correlation analysis is performed on rotational speed data and power data to explore the electromechanical coupling features between the two and enrich the feature dimensions of the model input.

[0031] The feature extraction stage uses Pearson correlation analysis to uncover the electromechanical coupling characteristics of speed and power, which significantly enriches the feature dimensions of the model input. In particular, it strengthens the key discriminative features under extreme conditions, enabling the model to more accurately identify the stability change patterns unique to extreme conditions. This breaks through the limitations of traditional feature extraction, which only focuses on normal conditions and fails to capture the features of extreme conditions.

[0032] After feature extraction, all features are standardized (e.g., using Z-score standardization) to map feature values ​​to a uniform numerical range, eliminating the interference of dimensional differences on model training and improving the model's learning efficiency for extreme working condition features.

[0033] In this embodiment, the correlation between the cross-domain hydropower unit and the target hydropower unit is determined by comprehensively considering the following indicators: the matching degree of the rated capacity of the units, the similarity of the hydrological characteristics of the respective river basins, and the consistency of the hydraulic-electromechanical structure. This determination method avoids blindly selecting cross-domain data (such as selecting unit data with excessively large capacity differences or completely different hydrological environments), ensuring that the cross-domain extreme operating condition data and the target unit have physical compatibility. This solves the problem of low effectiveness of cross-domain data leading to model learning bias in the prior art, and provides a high-quality data foundation for the subsequent model to absorb cross-domain extreme operating condition knowledge.

[0034] The rated capacity matching degree of the units requires that the deviation of the rated capacity between the cross-regional units and the target units be ≤ a preset ratio (e.g., ±20%); the similarity of the hydrological characteristics of the watershed is calculated based on the cosine similarity of the annual average runoff, the duration of the flood and dry seasons, and the sediment content of the water flow, and the similarity is ≥ a preset threshold (e.g., 0.7); the hydraulic-electromechanical structure consistency requires that the turbine model, shaft support structure, and generator type of the cross-regional units and the target units be consistent in at least two aspects.

[0035] The acquisition of cross-domain extreme operating condition data employs federated learning technology to protect privacy. Specifically, this involves: local nodes of the cross-domain hydropower units performing homomorphic encryption on their own extreme operating condition data, extracting only the feature vectors of the data and uploading them to the federated aggregation node; the aggregation node performing federated average aggregation on the encrypted feature vectors uploaded by each cross-domain node, generating a unified cross-domain extreme operating condition feature set, and then feeding it back to the model training node of the target hydropower unit, without transmitting the original data of the cross-domain hydropower units throughout the process. Federated learning technology, by homomorphically encrypting cross-domain data, uploading only feature vectors, and performing federated average aggregation, achieves cross-domain extreme operating condition data sharing while preventing the transmission of original data. This technology breaks down data silos (addressing the core pain point of scarce extreme data for a single unit), completely avoids the risk of cross-domain power plant data privacy leakage (such as sensitive information like power plant operating parameters and equipment status), balances data sharing needs with privacy protection requirements, and overcomes the shortcoming of existing technologies where cross-domain data acquisition and privacy security are mutually exclusive.

[0036] The homomorphic encryption adopts partial homomorphic encryption (such as the Paillier encryption algorithm) to ensure that the feature vector can directly participate in the aggregation operation in the encrypted state. After the federated aggregation node generates the cross-domain extreme working condition feature set, it also needs to perform 'feature alignment processing' with the local features of the target unit (such as unifying feature dimensions and standardizing feature value range) to ensure that the cross-domain features are consistent with the local feature format and can be directly used for model training.

[0037] In this embodiment, when constructing the stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning, the cross-domain extreme operating condition data is first subjected to generative data augmentation with physical mechanism constraints to generate synthetic extreme operating condition data that conforms to the physical characteristics of the target hydropower unit. The generative data augmentation is achieved through a physical constraint generative adversarial network. The generator of this network embeds the differential equation of the hydropower unit shaft vibration and the turbine flow-head-efficiency characteristic curve as physical constraints during the data generation process. The discriminator verifies the generated data based on the physical laws of the historical extreme operating condition data of the target hydropower unit, and outputs effective synthetic data through iterative adversarial training. The generated synthetic extreme operating condition data and the original historical extreme operating condition data of the target unit are mixed in a preset ratio (e.g., synthetic data accounts for 60%-80%, and the original data accounts for 20%-40%) and used together as the extreme operating condition training dataset. This avoids the model from becoming overly dependent on artificial data due to an excessive proportion of synthetic data, while ensuring that the original real data can guide the model to conform to the actual operating characteristics of the target unit. Physically constrained generative adversarial networks (GANs) enhance cross-domain extreme operating condition data, generating a large amount of synthetic extreme operating condition data that conforms to the physical characteristics of the target hydropower unit. This significantly expands the training sample size for extreme operating conditions, compensates for the lack of historical data on extreme operating conditions for a single unit, provides more sufficient learning material for the model, and avoids the problem in existing technologies where the model cannot fully learn the patterns of extreme operating conditions due to insufficient sample size.

[0038] The physical constraints are embedded in the generator by incorporating the constraint terms into the loss function. That is, the theoretical vibration range calculated by the differential equation of shaft vibration and the theoretical parameter range corresponding to the turbine flow-head-efficiency characteristic curve are transformed into a penalty term of the generator's loss function. When the generated data exceeds the theoretical range, the loss value is increased by the penalty term, forcing the generator to adjust its output to ensure that the generated data conforms to the physical constraints.

[0039] The discriminator verifies the physical laws of the generated data, specifically including two core indicators: first, whether the relationship between vibration amplitude, rotational speed, and load in the generated data conforms to the theoretical correlation derived from the differential equation of shaft vibration; second, whether the numerical relationship between flow rate, head, and efficiency in the generated data falls within a reasonable range of the turbine characteristic curve of the target unit. When both indicators are satisfied, the data is determined to be valid synthetic data.

[0040] In this embodiment, the stability prediction model includes: The physical mechanism prior layer provides the model with constraints based on the objective physical laws of hydropower units. Specifically, these physical constraints are achieved by converting the coupling relationship between bearing temperature rise and speed / load (derived from the heat conduction equation) and the correlation between hydraulic excitation frequency and water flow velocity (derived from the fluid dynamics momentum equation) into mathematical constraint formulas. During model training, if the prediction result violates these mathematical constraint formulas, a physical consistency penalty term is added to the loss function to forcibly correct the model parameters, ensuring that the prediction result conforms to the objective physical laws of hydropower units. These mathematical constraint formulas take effect in real-time during the forward propagation phase of model training. Each time the model outputs a prediction result, it is first compared with the theoretical range of the mathematical constraint formulas. The weight of the physical consistency penalty term is dynamically adjusted based on the importance of the constraint. For example, the weight of the constraint relating hydraulic excitation frequency and water flow velocity is higher than that of the bearing temperature rise coupling constraint (because hydraulic excitation has a more direct impact on stability under extreme operating conditions). The initial weight value can be set with reference to the frequency of failures caused by constraint violations in the target unit's historical extreme operating conditions.

[0041] The physical mechanism prior layer transforms the bearing temperature rise coupling relationship derived from the heat conduction equation and the hydraulic excitation correlation relationship derived from the fluid dynamics momentum equation into mathematical constraints, and adds a physical consistency penalty term to the loss function, forcing the model prediction results to conform to the objective physical laws of the unit. This design fundamentally solves the problem that existing pure data-driven models are prone to outputting results that violate physical common sense (such as vibration amplitude exceeding the shaft system bearing limit, and temperature rise being unrelated to load) when data distribution shifts under extreme operating conditions, providing a hard constraint on the reliability of the prediction results.

[0042] The cross-domain transfer learning layer addresses the problem of cross-domain data distribution offset and enables knowledge transfer under extreme cross-domain operating conditions. It employs a domain-adaptive network to achieve this, specifically including: extracting features from both the cross-domain extreme operating condition data (source domain) and the target hydropower unit data (target domain) to obtain source domain features and target domain features; learning the distribution differences between the two types of features through a domain discriminator; and adjusting the feature extraction network parameters based on the criterion of minimizing the maximum mean difference to make the distributions of source domain features and target domain features more consistent, thus resolving the cross-domain data distribution offset problem. The feature extraction of the source and target domain data is implemented using a convolutional neural network (CNN) + fully connected layer. The CNN extracts local time-domain and frequency-domain features of the data, and the fully connected layer transforms them into fixed-dimensional feature vectors. The calculation of the maximum mean difference only targets core features strongly correlated with stability (such as vibration frequency, head change rate, and rotational speed fluctuation), rather than all features, to avoid misjudgments of distribution differences caused by irrelevant features.

[0043] The cross-domain transfer learning layer employs a domain-adaptive network. A domain discriminator identifies the differences in feature distribution between cross-domain data (source domain) and target unit data (target domain), and adjusts parameters based on the criterion of minimizing the maximum mean difference to bring the two feature distributions closer together. This approach effectively eliminates the distribution shift in cross-domain data caused by differences in unit characteristics and watershed environment, allowing effective knowledge from extreme cross-domain operating conditions to be accurately transferred to the target unit model. This solves the problems of low knowledge transfer efficiency and poor adaptability in existing cross-domain transfer technologies, further enhancing the model's learning ability for extreme operating conditions of the target unit.

[0044] The time-series prediction layer is used to accurately capture the dynamic changes of stability indicators under extreme operating conditions. It employs an improved bidirectional long short-term memory network with an added attention mechanism. This attention mechanism weights and enhances key time-step features under extreme operating conditions (features at the moment of sudden increase in vibration amplitude, moment of sudden change in head, and moment of peak fluctuation in rotational speed), strengthening the model's ability to capture abrupt change patterns of stability indicators under extreme conditions and improving prediction accuracy. The weighting method of the attention mechanism is soft attention, which calculates the importance score (0-1 range) of each time-step feature using the sigmoid activation function. The higher the score, the greater the weight. The importance score is calculated based on the cosine similarity between the time-step feature and historical fault features. The basic parameters of the bidirectional long short-term memory network are as follows: the number of hidden layers is set to 2-3, the number of hidden units is set to 128-256 (which can be adjusted according to the amount of data of the target unit; if the amount of data is small, take a smaller value), and the time-step window is set to 5-10 minutes (to match the abrupt change cycle of parameters under extreme operating conditions).

[0045] The time-series prediction layer incorporates an improved bidirectional long short-term memory network with an attention mechanism to weight and enhance key time-step features under extreme conditions (such as sudden increases in vibration and sudden changes in head), allowing the model to prioritize the instantaneous features that have the greatest impact on stability. This design specifically addresses the problems of existing time-series models being insensitive to parameter mutations and exhibiting prediction lag under extreme conditions. It can more accurately capture the dynamic trends of stability indicators, especially improving the prediction sensitivity to sudden anomalies under extreme conditions.

[0046] In this embodiment, the real-time prediction module of the stability prediction model is deployed on the local edge node of the target hydropower unit. This edge node directly communicates with the unit's vibration sensors, temperature sensors, and flow sensors, enabling local acquisition and predictive inference of real-time operating data. This reduces the latency of data transmission to the cloud and ensures the real-time performance of prediction results under extreme operating conditions. By deploying the real-time prediction module of the stability prediction model on the local edge node of the target hydropower unit, and having this edge node directly communicate with sensors such as vibration, temperature, and flow sensors, the core pain points of existing centralized cloud-based prediction solutions can be fundamentally addressed. Specifically, real-time operational data does not need to be transmitted to a remote cloud. Instead, collection and predictive inference are completed locally, significantly reducing the total time spent on data transmission and computation. This avoids prediction lags caused by rapid parameter changes under extreme conditions (such as a sudden increase in water flow impact during flood season or a sudden cavitation effect during dry season), ensuring that prediction results can support operational and maintenance decisions in a timely manner. At the same time, local edge node data processing reduces dependence on cloud bandwidth and avoids network congestion when multiple sensors transmit data concurrently, further ensuring the continuity of data processing under extreme conditions. Furthermore, local data processing reduces the cross-network transmission of raw operational data, which can indirectly reduce the risk of data leakage, balancing real-time performance and data security.

[0047] The communication between the edge nodes and the sensors adopts an industrial-grade standard protocol. The analog sensors such as vibration and temperature adopt the Modbus-RTU protocol, while the digital sensors such as flow rate and speed adopt the MQTT protocol (which supports low-power and high-reliability data interaction between edge nodes and sensors). The protocol needs to enable data verification function (such as CRC check) to avoid data transmission errors caused by electromagnetic interference under extreme operating conditions.

[0048] Example 1 can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0049] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0050] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0051] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0052] Example 2

[0053] like Figure 2 As shown in Example 2, a hydropower unit stability prediction system is proposed, comprising: The historical data processing module of the unit acquires and preprocesses the historical operating data of the target hydropower unit. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season conditions and dry season conditions. The unit screening and data acquisition module identifies cross-domain hydropower units that are related to the target hydropower unit, and then acquires the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data. The physical mechanism-cross-domain migration prediction model construction module constructs a stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning based on the historical operation data and cross-domain extreme working condition data of the target hydropower unit. The real-time data input and prediction output module inputs the real-time operating data of the target hydropower unit into the stability prediction model and outputs the stability prediction results for a future preset time period. The stability prediction results include the predicted values ​​of vibration amplitude, swing value and winding temperature rise.

[0054] The hydropower unit stability prediction system provided in Example 2 can realize all the processes of the hydropower unit stability prediction method in Example 1. The functions and technical effects of each module in the hydropower unit stability prediction system are the same as those of the hydropower unit stability prediction method in Example 1, and will not be repeated here.

[0055] Example 3

[0056] To address the aforementioned technical problems, Embodiment 3 also provides a computer device. Please refer to [link / reference needed] for details. Figure 3 .

[0057] The computer device 3 includes a memory 31, a processor 32, and a network interface 33 that are interconnected via a system bus. It should be noted that only the computer device 3 with components 31-33 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0058] The computer device 3 can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 can interact with the user through a keyboard, mouse, remote control, touchpad, or voice control device.

[0059] The memory 31 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 31 may be an internal storage unit of the computer device 3, such as the hard disk or memory of the computer device 3. In other embodiments, the memory 31 may also be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Of course, the memory 31 may include both the internal storage unit and its external storage device of the computer device 3. In this embodiment, the memory 31 is typically used to store the operating system and various application software installed on the computer device 3, such as the computer-readable instructions in Subject 1. In addition, the memory 31 can also be used to temporarily store various types of data that have been output or will be output.

[0060] In some embodiments, the processor 32 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 32 is typically used to control the overall operation of the computer device 3. In this embodiment, the processor 32 is used to execute computer-readable instructions stored in the memory 31 or to process data, for example, to execute computer-readable instructions for a hydropower unit stability prediction method.

[0061] The network interface 33 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 3 and other electronic devices.

[0062] Example 4

[0063] The present invention also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the hydropower unit stability prediction method described above.

[0064] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0065] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting the stability of a hydropower unit, characterized in that, Includes the following steps: Historical operating data of the target hydropower unit is acquired and preprocessed. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season conditions and dry season conditions. Identify the cross-domain hydropower units that are related to the target hydropower unit, and then obtain the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data; Based on the historical operating data and cross-domain extreme operating condition data of the target hydropower unit, a stability prediction model integrating physical mechanism embedding and cross-domain transfer learning is constructed. The real-time operating data of the target hydropower unit is input into the stability prediction model, and the stability prediction results for a future preset time period are output. The stability prediction results include the predicted values ​​of vibration amplitude, sway value and winding temperature rise.

2. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, The preprocessing includes data cleaning, outlier removal, and feature extraction. The data cleaning uses sliding window filtering to remove noise interference from the original data. The outlier removal is based on the 3σ criterion to identify and delete abnormal data that deviates from the normal distribution. The feature extraction includes extracting frequency features from vibration signals through FFT spectrum analysis and extracting temperature rise trend features from temperature signals through trend fitting. The feature extraction also includes performing multi-scale empirical mode decomposition on flow rate data and head data to extract time-frequency joint features of sudden changes in flow velocity and sudden rises and falls in head under extreme conditions; at the same time, Pearson correlation analysis is performed on rotational speed data and power data to explore the electromechanical coupling features between the two and enrich the feature dimensions of the model input.

3. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, The correlation between the cross-regional hydropower unit and the target hydropower unit is determined by the following indicators: the matching degree of the rated capacity of the units, the similarity of the hydrological characteristics of the respective river basins, and the consistency of the hydraulic-electromechanical structure. The method of obtaining cross-domain extreme operating condition data adopts federated learning technology to protect privacy, specifically including: the local node of the cross-domain hydropower unit performs homomorphic encryption on its own extreme operating condition data, and only extracts the feature vector of the data and uploads it to the federated aggregation node. The aggregation node performs federated average aggregation on the encrypted feature vectors uploaded by each cross-domain node, generates a unified cross-domain extreme working condition feature set, and then feeds it back to the model training node of the target hydropower unit. The original data of the cross-domain hydropower unit is not transmitted throughout the process.

4. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, When constructing the stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning, the cross-domain extreme operating condition data is first subjected to generative data augmentation with physical mechanism constraints to generate synthetic extreme operating condition data that conforms to the physical characteristics of the target hydropower unit. The generative data augmentation is achieved through a physical constraint generative adversarial network. The generator of this network embeds the differential equation of the hydropower unit shaft system vibration and the turbine flow-head-efficiency characteristic curve as physical constraints during the data generation process. The discriminator verifies the generated data based on the physical laws of the historical extreme operating condition data of the target hydropower unit and outputs effective synthetic data through iterative adversarial training.

5. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, The stability prediction model includes: The physical mechanism prior layer is used to provide the model with objective physical constraints of the hydropower unit. Specifically, the physical constraints are implemented in the following ways: the coupling relationship between the bearing temperature rise and speed and load of the hydropower unit, and the relationship between the hydraulic excitation frequency and the water flow velocity are transformed into mathematical constraint formulas; during the model training process, if the prediction result violates the mathematical constraint formulas, a physical consistency penalty term is added to the loss function to forcibly correct the model parameters and ensure that the prediction result conforms to the objective physical laws of the hydropower unit. The cross-domain transfer learning layer is used to solve the problem of cross-domain data distribution offset and realize cross-domain extreme working condition knowledge transfer. It adopts a domain adaptive network to realize cross-domain knowledge transfer, which includes: extracting features from cross-domain extreme working condition data and target hydropower unit data respectively to obtain source domain features and target domain features; learning the distribution difference between the two types of features through a domain discriminator; and adjusting the feature extraction network parameters based on the maximum mean difference minimization criterion to make the distribution of source domain features and target domain features tend to be consistent, thus solving the problem of cross-domain data distribution offset. The time-series prediction layer is used to accurately capture the dynamic changes of stability indicators under extreme conditions. It adopts an improved bidirectional long short-term memory network with an added attention mechanism. The attention mechanism weights and enhances the key time-step features under extreme conditions, strengthens the model's ability to capture abrupt changes in stability indicators under extreme conditions, and improves prediction accuracy.

6. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, It also includes a dynamic correction process for the prediction results: real-time acquisition of the current operating parameters of the target hydropower unit, calculation of the deviation between the measured value and the corresponding predicted value input by the model; if the deviation exceeds the preset threshold, the hydraulic characteristic constraint formula of the physical mechanism prior layer is called to compensate and correct the prediction results of vibration amplitude and swing value until the deviation is lower than the preset threshold, thereby eliminating prediction drift under extreme operating conditions.

7. The method for predicting the stability of a hydropower unit according to claim 1, characterized in that, The real-time prediction module of the stability prediction model is deployed on the local edge node of the target hydropower unit. The edge node communicates directly with the unit's vibration sensor, temperature sensor, and flow sensor to realize local acquisition and prediction inference of real-time operating data, reduce the delay of data transmission to the cloud, and ensure the real-time performance of prediction results under extreme operating conditions.

8. A hydropower unit stability prediction system, characterized in that, include: The historical data processing module of the unit acquires and preprocesses the historical operating data of the target hydropower unit. The historical operating data includes vibration signals, temperature signals, flow data, head data, speed data and power data under normal operating conditions, as well as corresponding operating data under extreme operating conditions, including flood season conditions and dry season conditions. The unit screening and data acquisition module identifies cross-domain hydropower units that are related to the target hydropower unit, and then acquires the extreme operating condition data of the cross-domain hydropower units as cross-domain extreme operating condition data. The physical mechanism-cross-domain migration prediction model construction module constructs a stability prediction model that integrates physical mechanism embedding and cross-domain transfer learning based on the historical operation data and cross-domain extreme working condition data of the target hydropower unit. The real-time data input and prediction output module inputs the real-time operating data of the target hydropower unit into the stability prediction model and outputs the stability prediction results for a future preset time period. The stability prediction results include the predicted values ​​of vibration amplitude, swing value and winding temperature rise.

9. A computer device, characterized in that, include A memory that stores computer-readable instructions; A processor that, when executing the computer-readable instructions, implements the steps of a hydropower unit stability prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of a hydropower unit stability prediction method as described in any one of claims 1 to 7.

Images