Hydroelectric unit state prediction system based on deep learning

Abstract

The application discloses a water and electricity unit state prediction system based on deep learning, and relates to the technical field of water and electricity unit monitoring.The system comprises the following steps: collecting a water and electricity unit through a data sensor network to obtain multiple operation sensing data; performing deep learning on multiple operation time periods to construct a state prediction channel; synchronizing the multiple operation sensing data to the state prediction channel for analysis to generate multiple operation state prediction results, perform operation monitoring on the water and electricity unit, and generate operation monitoring results; verifying the multiple operation state prediction results to generate prediction feedback information and update and optimize the state prediction channel; and intelligently predicting the state of the water and electricity unit.The application solves the technical problem that the prior art is inaccurate in predicting the operation state of a water and electricity unit, thereby reducing the reliability and operation efficiency of the water and electricity unit, and achieves the technical effect of accurately and intelligently predicting the operation state of the water and electricity unit and improving the reliability and operation efficiency of the water and electricity unit.

Description

Technical Field

[0001] This invention relates to the field of hydropower monitoring technology, and more specifically to a hydropower unit state prediction system based on deep learning. Background Technology

[0002] In today's power generation sector, hydropower units play a crucial role as an important clean energy power generation device. However, traditional methods for monitoring and predicting the condition of hydropower units have certain limitations. On the one hand, traditional methods typically rely on data acquisition from a single sensor and simple data analysis, making it difficult to comprehensively and accurately reflect the complex operating conditions of hydropower units. Moreover, for changes in the operating conditions of hydropower units across multiple operating periods and different operating cycles, traditional methods often lack effective processing tools and struggle to capture dynamic changes over time. On the other hand, traditional condition prediction mainly relies on empirical models and fixed rules, lacking the ability to deeply mine and learn from large amounts of historical operating data, resulting in limited prediction accuracy and adaptability. Furthermore, when faced with abnormal operating conditions, traditional methods lack sufficient diagnostic and location capabilities, making it difficult to quickly identify the source of the anomaly, leading to low efficiency in troubleshooting and maintenance.

[0003] Existing technologies suffer from inaccurate predictions of hydropower generator operating conditions, leading to low reliability and operating efficiency of hydropower units. Summary of the Invention

[0004] This application provides a deep learning-based hydropower unit state prediction system to address the technical problem of inaccurate prediction of hydropower unit operating status in existing technologies, which leads to low reliability and operating efficiency of hydropower units.

[0005] In view of the above problems, this application provides a hydropower unit state prediction system based on deep learning.

[0006] This application provides a deep learning-based hydropower unit state prediction system, the system comprising:

[0007] The system comprises the following modules: a sensor data acquisition module, which acquires multiple operational sensor data in real time from the hydropower unit through a data sensor network based on multiple operational segments; a state prediction channel construction module, which constructs a state prediction channel by performing deep learning on the multiple operational sensor datasets according to the multiple operational segments; an operational monitoring result generation module, which synchronizes the multiple operational sensor data to the state prediction channel for analysis, generates multiple operational state prediction results, and monitors the hydropower unit's operation according to the multiple operational state prediction results and the operational environment dataset, generating operational monitoring results containing multiple state labels; a state prediction optimization channel generation module, which iterates through the multiple operational state prediction results according to the multiple state labels for verification, generates prediction feedback information, and updates and optimizes the state prediction channel based on the prediction feedback information, generating a state prediction optimization channel; and an intelligent prediction module, which intelligently predicts the state of the hydropower unit through the state prediction optimization channel.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] The operation sensor data acquisition module acquires multiple operation data points from the hydropower unit in real time through a data sensor network, based on multiple operating segments. The state prediction channel construction module constructs a state prediction channel by performing deep learning on the historical operation sensor dataset according to the multiple operating segments. The operation monitoring result generation module synchronizes the multiple operation sensor data to the state prediction channel for analysis, generating multiple operation state prediction results. Based on these prediction results and the operating environment dataset, the module monitors the hydropower unit's operation and generates operation monitoring results. The state prediction optimization channel generation module iterates through the multiple operation state prediction results according to multiple state labels, verifies them, generates prediction feedback information, and updates and optimizes the state prediction channel to generate an optimized state prediction channel. The intelligent prediction module uses the optimized state prediction channel to intelligently predict the state of the hydropower unit. This achieves accurate and intelligent prediction of the hydropower unit's operating state, improving its reliability and operating efficiency. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic diagram of the structure of a deep learning-based hydropower unit state prediction system is provided for embodiments of this application;

[0012] Figure 2 This application provides a schematic diagram of the output state prediction channel of a deep learning-based hydropower unit state prediction system for embodiments of the present application.

[0013] Figure labeling: 10 for operation sensor data acquisition, 20 for state prediction channel construction, 30 for operation monitoring result generation, 40 for state prediction optimization channel generation, and 50 for intelligent prediction. Detailed Implementation

[0014] This application provides a deep learning-based hydropower unit state prediction system to address the technical problem of inaccurate prediction of hydropower unit operating status in existing technologies, which leads to low reliability and operating efficiency of hydropower units.

[0015] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0016] Example 1, as Figure 1 As shown, this application provides a deep learning-based hydropower unit state prediction system, the system comprising:

[0017] The operation sensor data acquisition module 10 is based on multiple operating segments and collects multiple operating sensor data from the hydropower unit in real time through a data sensor network.

[0018] Specifically, the operational sensor data acquisition module 10 focuses on acquiring operational sensor data of the hydropower unit, and its operation unfolds across multiple operating phases. Within each operating phase, a data sensor network comprehensively and in real-time monitors the hydropower unit. This data sensor network is a complex network composed of various types of sensors, carefully deployed at key locations within the hydropower unit. For example, temperature sensors are installed on the turbine runner, bearings, and volute to monitor temperature changes; pressure sensors are placed at pipe connections and pressure chambers to acquire pressure data; and vibration sensors are located near the unit's support structure and key transmission components to detect vibration frequency and amplitude. Through these widely distributed and functionally diverse sensors, a large amount of raw data on the hydropower unit's operating status is acquired in each operating phase. This raw data covers various aspects of unit operation, including but not limited to parameters such as temperature, pressure, vibration, and flow rate. Over time, continuous acquisition across multiple operating phases yields numerous operational sensor data points. This data forms a rich dataset, providing a foundation for subsequent analysis and processing, and is a crucial prerequisite for the accurate operation of the entire hydropower unit condition prediction system.

[0019] The state prediction channel construction module 20 is used to construct a state prediction channel by performing deep learning on the multiple runtime segments based on the historical running sensor dataset.

[0020] Specifically, the state prediction channel construction module 20 is a key component of the entire hydropower unit state prediction system, responsible for constructing channels for predicting the state of the hydropower unit. First, it acquires a historical operation sensor dataset, which contains records of various operating parameters of the hydropower unit over multiple operating periods. These records provide a comprehensive reflection of the unit's operating state. For example, it covers information such as temperature changes, pressure fluctuations, vibration conditions, and flow data at different times. Then, deep learning is performed on the historical operation sensor dataset according to multiple operating periods. Before deep learning, the historical operation sensor dataset needs to be preprocessed according to multiple operating periods. This includes data cleaning, removing outliers and erroneous data, and data standardization to ensure that data with different characteristics have similar scales for better learning. The preprocessed historical operation sensor data is sequentially input into the LSTM network in time series. For each time step, the data passes through the input gate into the memory unit. Inside the memory unit, the forget gate determines which information needs to be forgotten based on the current input and the state at the previous time step. For example, if a parameter change in a certain operating period has a minor impact on the subsequent state, the forget gate may weaken or forget the relevant information. Simultaneously, the input gate determines which new information needs to be stored in the memory unit. For example, if a new operational feature emerges in the current time period and has a significant impact on subsequent predictions, the input gate will include it in the memory unit. After processing by the forget gate and the input gate, the information in the memory unit is output through the output gate. The output information serves as part of the input for the next time step and also as part of the network's output for the current time step. This output gradually reflects the patterns and regularities in historical operational sensor data over time through accumulation and continuous learning. Through long-term learning and processing of a large amount of historical operational sensor data, the LSTM network can learn the complex relationships and dynamic change patterns of hydropower units in different operating periods. Based on these learned patterns, a state prediction channel capable of predicting the future state of hydropower units is constructed. When new operational sensor data is input, this channel can use the previously learned knowledge to predict the future state of the hydropower units.

[0021] The operation monitoring result generation module 30 is used to synchronize the multiple operation sensor data to the state prediction channel for analysis, generate multiple operation state prediction results, and perform operation monitoring on the hydropower unit according to the multiple operation state prediction results and the operation environment dataset to generate operation monitoring results, which include multiple state labels.

[0022] Specifically, the operation monitoring result generation module 30 synchronizes multiple operation sensor data acquired from the operation sensor data acquisition module to the state prediction channel. This synchronization process ensures that the data can be analyzed within the pre-constructed state prediction channel. The state prediction channel is built based on an LSTM network algorithm that learns from historical operation sensor datasets, enabling it to predict the operating status of hydropower units. When operation sensor data enters the channel, it is processed by algorithms and models within the channel to generate multiple operation status prediction results. These results reflect the possible future operating states of the hydropower unit, such as predicting a temperature increase in a component or pressure fluctuations in a certain area. Next, based on these operation status prediction results and combined with an operating environment dataset, the hydropower unit's operation is monitored. This dataset includes information on environmental factors related to hydropower unit operation, such as water temperature, water level, and water quality. By comprehensively considering the prediction results and environmental factors, a more comprehensive understanding of the actual operating conditions of the hydropower unit can be obtained. For example, if a component's temperature is predicted to rise, combining this with current water temperature and flow rate, it is possible to more accurately determine whether the component is in an abnormal state. Finally, based on the above comprehensive analysis, operational monitoring results are generated, which include multiple status labels. These labels describe whether the hydropower unit's operating status is normal, whether there are any abnormalities, and the type of abnormality. For example, there will be different status labels such as "normal operation," "abnormal temperature," and "abnormal pressure fluctuation." These labels can provide important information for subsequent maintenance and decision-making.

[0023] The state prediction optimization channel generation module 40 is used to verify the multiple running state prediction results by traversing the multiple state labels, generate prediction feedback information, update and optimize the state prediction channel according to the prediction feedback information, and generate a state prediction optimization channel.

[0024] Specifically, the state prediction optimization channel generation module 40 is responsible for optimizing the state prediction channel to improve the accuracy and reliability of predictions. First, it iterates through and verifies multiple state prediction results generated by the operation monitoring result generation module, checking the accuracy of the corresponding state prediction result for each state label. For example, if the state label shows "temperature abnormality" for a component, it checks whether the prediction result correctly predicted the abnormal temperature of that component. During this verification process, the module analyzes in detail the deviation between the prediction result and the actual operating state. Based on the deviations found during the verification process, the module generates prediction feedback information. This feedback information includes various details about the prediction accuracy, such as which prediction results are inaccurate, the specific value of the deviation, and the cause of the deviation. For example, if a large deviation is found between the predicted and actual temperature values ​​of a component, the feedback information will clearly indicate the component, the deviation value, and the factors that may have caused the deviation, such as sensor failure or insufficient consideration of environmental factors. Finally, the state prediction channel is updated and optimized based on the generated prediction feedback information. The module adjusts various parameters and algorithm structures in the state prediction channel. For example, if a certain type of data is found to have a significant impact on the prediction results, but its weights are not set appropriately in the channel, then its weights will be readjusted; if a certain algorithm is found to be ineffective in handling specific situations, the algorithm will be replaced or improved. Through these update and optimization measures, an optimized state prediction channel is ultimately generated, enabling it to predict the state of hydropower units more accurately.

[0025] The intelligent prediction module 50 is used to intelligently predict the state of the hydropower unit through the state prediction optimization channel.

[0026] Specifically, the intelligent prediction module 50 utilizes a state prediction optimization channel to intelligently predict the state of the hydropower unit. Upon receiving current operating data from the hydropower unit, the intelligent prediction module 50 inputs this data into the state prediction optimization channel. This channel is a prediction model optimized by the preceding modules, incorporating patterns and rules learned from historical operating sensor data, as well as algorithms and parameter settings to correct for inaccurate predictions. Within the state prediction optimization channel, the input operating data is processed according to the algorithms and rules set by the channel. Through these processes, the intelligent prediction module 50 ultimately outputs predictions of the hydropower unit's future operating state. These predictions include forecasts of temperature changes, pressure changes, vibration conditions, and so on. These predictions help to understand the unit's operating trends in advance, enabling the implementation of appropriate maintenance measures and ensuring the safe and stable operation of the hydropower unit.

[0027] In one possible implementation, such as Figure 2 As shown, the state prediction channel construction module also includes:

[0028] The operation of the hydropower unit is analyzed based on the multiple operating periods, and multiple operating cycles are generated according to the analysis results. Historical operating data archives of the hydropower unit are retrieved, and historical operating sensor datasets are extracted from these archives. Features are extracted from the historical operating sensor datasets according to the multiple operating cycles to generate multiple historical operating features. Time series analysis is performed according to the multiple operating cycles and the multiple historical operating features to determine the operating time step of the hydropower unit. A Long Short-Term Memory (LSTM) network is used to perform deep learning on the historical operating sensor datasets according to the operating time step to generate an operating time series. The operating time series contains multiple operating time nodes, and each of these operating time nodes has at least one historical operating sensor data point. The multiple historical operating sensor data points are verified and evaluated based on the multiple operating time nodes to generate verification tokens. Evaluation is then performed based on the verification tokens, and a state prediction channel is output.

[0029] Specifically, firstly, detailed data on the hydropower unit's operation at each stage needs to be collected, including but not limited to parameters such as power output, water flow rate, unit speed, temperature changes, and pressure fluctuations. This data comprehensively reflects the unit's operating status at different times. Then, this data is comprehensively analyzed. For example, observing the power output curve over time may reveal that the unit's power output is relatively stable in some periods, while showing a clear upward or downward trend in others. Water flow rate may also show a correlation with power output; higher power output is accompanied by a correspondingly higher water flow rate. Simultaneously, temperature changes and pressure fluctuations also affect the unit's operation; excessively high temperatures or unstable pressures may indicate changes in the unit's operating status. Through detailed analysis and comparison of this data, based on its changing patterns and characteristics, the hydropower unit's operation process is divided into multiple operating cycles. Each operating cycle has unique operating characteristics. For instance, in one operating cycle, the unit is in a stable power generation state, with relatively stable parameters such as power output, water flow rate, temperature, and pressure; while in another operating cycle, the unit is in the process of starting up or shutting down, and these parameters will change significantly. This classification method helps to gain a deeper understanding of the operating mechanism of hydropower units and provides a more reasonable framework for subsequent data analysis and processing.

[0030] Retrieving historical operating data archives of the hydropower units is a crucial step in data acquisition. These archives store a vast amount of data records from the units' long-term operation, covering various aspects of operational information. From these archives, a historical operating sensor dataset is extracted. This dataset, obtained after filtering and organizing the original archive data, primarily contains unit operating parameter data collected by various sensors, such as temperature values ​​recorded by temperature sensors, pressure values ​​recorded by pressure sensors, flow values ​​recorded by flow sensors, and vibration parameters recorded by vibration sensors. This data directly reflects the unit's operating status at different times. Next, feature extraction is performed on the historical operating sensor dataset according to pre-defined operating cycles. Within each operating cycle, each parameter in the dataset is analyzed to extract features representative of the cycle's operating characteristics. For temperature data, features such as the average temperature, maximum temperature, minimum temperature, and temperature fluctuation range within the cycle are calculated. These features reflect the unit's temperature changes within the cycle, such as whether the unit is in a stable temperature environment or whether there are abnormal conditions of excessively high or low temperatures. For pressure data, features such as average pressure, maximum pressure, minimum pressure, and pressure change rate are calculated. These pressure features help determine whether the unit experienced stable pressure during the cycle and whether pressure changes were within the normal range. For flow data, features such as average flow rate, maximum flow rate, minimum flow rate, and flow rate trend are calculated. Flow characteristics reflect the water supply situation of the unit during the cycle, whether it meets power generation needs, and whether the flow rate is stable. By performing the above feature extraction operations on different parameters in each operating cycle, multiple historical operating features are finally generated. These historical operating features can more concisely and accurately describe the operating status and characteristics of the hydropower unit in each operating cycle, providing a more targeted data foundation for subsequent time series analysis and deep learning.

[0031] Considering multiple operating cycles, each representing a different stage of the hydropower unit's operation, these cycles differ in power output, temperature variations, pressure fluctuations, and flow rate adjustments. For example, one operating cycle corresponds to the unit's startup phase, during which power gradually increases, and temperature and pressure change accordingly; another operating cycle represents the stable operation phase, where parameters remain relatively stable. Simultaneously, combining multiple historical operating characteristics extracted from the historical sensor datasets of each operating cycle provides a more detailed description of the characteristics of each cycle. For instance, within a given operating cycle, features such as average temperature, temperature fluctuation range, and pressure change rate reflect the unit's operating status during that cycle. Time-series analysis requires studying the changes in these operating cycles and their corresponding historical operating characteristics over time, observing the transitions between different operating cycles, and the trends of characteristics within each cycle over time. For example, analyzing how power gradually stabilizes from the startup phase to the stable operation phase, how temperature and pressure reach stable values, and the timeframe over which these changes occur. By analyzing the time series of these operating cycles and historical operating characteristics, the operating time step of the hydropower unit is determined. This time step is an appropriate time interval that accurately captures key changes during unit operation, avoiding both overly large intervals that might miss important information and overly small intervals that would lead to complex and difficult-to-process data. For example, if the time step is set to 1 minute, the unit's operating status can be observed every minute during data analysis, providing a basis for subsequent deep learning and other operations.

[0032] The historical sensor dataset is processed according to a set runtime step, which determines the time interval at which data enters the network, ensuring that the data is received and processed by the LSTM network in an orderly manner. For example, if the runtime step is 5 minutes, then data segments of every 5 minutes are input into the network as a unit. Then, when a data unit enters the LSTM network, the network begins its learning process. The unique structure of the LSTM network includes input gates, forget gates, and output gates. The forget gate determines which information to forget based on the current input and the state of the previous time step. For example, if a sensor data point from a previous time period is not very important for the current prediction, the forget gate will reduce its weight in memory. Simultaneously, the input gate determines which new information to store in the memory unit based on the current input. For example, if there are new pressure fluctuation characteristics in the current time period, the input gate will include the relevant data in the memory unit. Inside the network, the information processed by the forget gate and input gate is updated in the memory unit and output through the output gate. The output information not only affects the input of the next time step but also contributes to the output of the current time step. In this way, the LSTM network can capture long-term dependencies and dynamic change patterns in the data. As time progresses and data is continuously input, the network eventually generates an operational time series. This operational time series contains multiple operational time nodes, each corresponding to a specific point in time (determined by the operational time step). For example, each 5 minutes corresponds to a node, and each operational time node has at least one historical operational sensor data point. These historical operational sensor data are processed and integrated by the network to reflect the operational status of the hydropower unit at the corresponding time point and its relationship with other time points, providing an important basis for subsequent analysis and prediction.

[0033] For each operational time point and its corresponding historical sensor data, the first step is to check the reasonableness of the time period. Considering the operational process and logic of the hydropower unit, different operational segments have specific physical processes and characteristics. For example, during the unit startup phase, some data related to stable operation may not yet be present or may be in an unstable state. At this time, it is necessary to determine whether there is any data that does not conform to the characteristics of the current time period. For instance, if precise temperature control data, which is only present under stable operating conditions, appears in the initial startup phase, this is unreasonable because the unit is still in the process of heating up and has not yet reached a stable temperature control stage. Similarly, if high power output data, typical of normal operation, appears during the shutdown phase, it is clearly inconsistent with reality. It is also necessary to check whether there is any data that is physically impossible in the current time period. For example, when the unit is stationary and not started, the flow sensor should not detect water flow data; if such data appears, it is likely due to sensor malfunction or data recording errors. This check of the reasonableness of the time period is an important step in ensuring data quality, helping to eliminate abnormal data caused by incorrect acquisition or non-consistent operational logic. In addition to checking the time period, it's also necessary to check the data range of each sensor. Different sensors measure physical quantities within reasonable ranges, depending on the design and operating characteristics of the hydropower unit. For example, temperature sensors should typically measure temperatures within the unit's tolerable temperature range. Data outside this range indicates a sensor malfunction or an abnormal condition in the unit, such as overheating or overcooling. Pressure sensors should also measure pressure data within a reasonable range. Abnormal high or low pressure data requires further investigation. By checking the data range, problematic data can be further identified.

[0034] Based on the results of the data verification and evaluation described above, evaluation metrics were determined for generating verification tokens. These metrics quantify the quality of the data, providing a basis for generating verification tokens. By comprehensively considering these metrics, a more holistic understanding of the overall data quality is achieved. Verification tokens are generated based on the determined evaluation metrics. A verification token is a numerical value, such as a score between 0 and 1. A higher score indicates better data quality, reflecting the overall state of the data after verification and evaluation.

[0035] The generated verification tokens are evaluated. If the verification tokens meet pre-set standards, such as numerical verification tokens exceeding a certain threshold (e.g., 0.8), or data object verification tokens meeting all requirements (e.g., the percentage of data not conforming to the reasonableness of the time period is below a certain value, or the percentage of data exceeding the data range is below a certain value), then the data quality meets the requirements, and the state prediction channel is output. If the verification tokens do not meet the standards, the data acquisition process needs to be re-examined to check for sensor malfunctions, data recording errors, etc. After correcting the problems found, the verification evaluation is repeated until the verification tokens meet the requirements. When the verification tokens meet the requirements, the output state prediction channel is a validated and optimized model structure that can predict the future state of the hydropower unit based on the relevant input data.

[0036] In one possible implementation, the state prediction channel building module further includes:

[0037] The historical operating sensor dataset is subjected to time-series standardization processing according to the stated operating time step to generate a historical sensor standard dataset. The historical sensor standard dataset is then segmented according to the stated operating time step to generate multiple standard data sample segments. Long-term analysis is performed based on these multiple standard data sample segments to generate a first analysis result. Multiple long-term standard data sample segments are extracted based on the first analysis result, and long-term dependency coefficients are calculated based on these multiple long-term standard data sample segments. Short-term analysis is performed based on these multiple standard data sample segments to generate a second analysis result. Multiple short-term standard data sample segments are extracted based on the second analysis result, and short-term dependency coefficients are calculated based on these multiple short-term standard data sample segments. The historical operating sensor dataset is captured according to the long-term dependency coefficients to generate a first information stream. The historical operating sensor dataset is captured according to the short-term dependency coefficients to generate a second information stream. The first and second information streams are stacked to determine multiple operating time nodes, and the operating time series is extracted by integrating these multiple operating time nodes.

[0038] Specifically, when performing time-series standardization on historical operational sensor datasets, the Z-score standardization algorithm is used. The dataset is divided into different time segments based on the runtime time step, ensuring temporal continuity within each segment. For each time segment, the mean and standard deviation are calculated. The mean reflects the central trend of the data within that time segment, while the standard deviation reflects the dispersion of the data. For each data point in the dataset, standardization is performed using the following formula: Standardized value = (Original value - Mean for that time segment) / Standard deviation for that time segment. For example, suppose a temperature sensor collects data at different time points. Within a specific runtime time step, the mean temperature data is 25℃, and the standard deviation is 2℃. If the temperature at a certain moment is 28℃, after standardization, the value becomes (28-25) / 2 = 1.5. Through this process, the historical operational sensor dataset is transformed into a historical sensor standard dataset.

[0039] The historical sensor standard dataset is divided according to the runtime time step, generating multiple standard data sample segments. Each standard data sample segment corresponds to a specific time interval, and its length is determined by the runtime time step. This segmentation method facilitates more detailed data analysis, decomposing a large dataset into multiple smaller data units with similar time scales, providing a foundation for subsequent long-term and short-term analyses.

[0040] A long-term analysis of multiple standard data sample segments is conducted. This analysis employs statistical and trend analysis methods to observe the patterns and characteristics of these data changes over a longer time span. The analysis generates a first analysis result, which includes information such as the long-term trends of certain parameters and the long-term correlations between different parameters. Next, based on the first analysis result, several representative and significant long-term standard data sample segments are extracted. These segments exhibit clear trends or special patterns in the long-term analysis. Then, by calculating the covariance and correlation coefficients between different parameters in these multiple long-term standard data sample segments, a long-term dependency coefficient is generated. This long-term dependency coefficient quantifies the dependence between different parameters of the hydropower unit during long-term operation, providing an important basis for assessing the long-term stability of the hydropower unit and predicting its future operating status.

[0041] Short-term analysis is performed on multiple standard data sample segments. This process focuses on data changes within a short timeframe, analyzing fluctuations in the operating parameters of hydropower units over a day, several hours, or even shorter periods. Through detailed study of this data, specific analytical methods are used to calculate statistics such as the mean, variance, and rate of change of the short-term data. This short-term analysis generates a second analysis result. This result includes some prominent short-term characteristics, such as rapid changes in certain parameters within a specific time period and strong short-term correlations between parameters. Based on the second analysis result, multiple short-term standard data sample segments are extracted. These sample segments are data parts considered to have important characteristics or representativeness in the short-term analysis. For example, if abnormal fluctuations in the vibration parameters of hydropower units are found within a certain hour, then the relevant data sample segments for that time period may be extracted for further analysis. Next, short-term dependency coefficients are calculated based on these multiple short-term standard data sample segments. The calculation method for this coefficient is similar to that of the long-term dependency coefficient, but with a greater emphasis on the short-term time scale. The covariance of different parameters in the short term is calculated to measure their dependence on each other over a short period. The short-term dependence coefficient reflects the characteristics of changes in the operating status of hydropower units and the degree of interaction between parameters in the short term. It can help to quickly understand the short-term operating stability and potential problems of hydropower units, providing a basis for timely adjustment measures or fault early warning.

[0042] Historical operational sensor datasets are captured using long-term dependency coefficients, which reflect the interrelationships of hydropower unit operating parameters over long time scales. These coefficients are then used to filter data points from the historical operational sensor datasets that exhibit significant long-term correlations. For example, if the long-term dependency coefficients indicate a strong positive correlation between temperature and pressure over the long term, the capture process will focus on data points where both temperature and pressure show specific trends. This targeted capture of historical operational sensor datasets generates a first information stream containing key information about the hydropower units over long time scales, such as long-term stable operating modes, slowly changing trends, and parameter combinations related to long-term performance.

[0043] Similarly, historical operational sensor datasets are captured using short-term dependency coefficients, which reflect the interrelationships and dynamic changes of hydropower unit operating parameters within a short period. This coefficient is used to quickly identify data points in historical operational sensor data that are significantly correlated within a short timeframe. For example, if the short-term dependency coefficients show a close relationship between vibration frequency and power output over a short period, then during the capture process, data points where vibration frequency and power output change simultaneously will be prioritized. This capture process generates a second information stream, which encompasses key information about the hydropower unit on a short-term timescale, such as sudden changes, short-term fluctuations, and rapid response parameter combinations. This is crucial for timely detection of short-term anomalies in the hydropower unit and rapid adjustment of operating strategies.

[0044] The first and second information streams are stacked. This stacking process integrates long-term and short-term information to comprehensively understand the operational status of hydropower units at different time scales. Through stacking, the interrelationship between long-term trends and short-term fluctuations, and their combined impact on the operation of hydropower units, are discovered. Based on the stacked information streams, multiple operational time points are identified. These time points are determined based on key changes, turning points, or significant data points in the information streams. For example, moments when long-term trends change significantly, or when short-term fluctuations reach peaks or troughs. Finally, these multiple operational time points are integrated to extract the operational time series. This operational time series integrates long-term and short-term information, reflecting the changes in the operational status of hydropower units at different points in time. It provides a comprehensive and accurate time series data foundation for subsequent analysis, forecasting, and decision-making.

[0045] In one possible implementation, the state prediction channel building module further includes:

[0046] The first information stream and the second information stream are aligned according to the stated running time step to generate a timestamp alignment result. Based on the timestamp alignment result, the first information stream and the second information stream are stacked according to the time dimension to generate a two-dimensional data matrix. The two-dimensional data matrix is ​​traversed to perform information change analysis and generate an information change curve. The multiple running time nodes are extracted according to the information change curve. The multiple running time nodes are integrated in multiple dimensions to generate a multi-dimensional vector set. The multi-dimensional vector set is arranged according to the multiple running cycles to determine the running time sequence.

[0047] Specifically, the first and second information streams are aligned according to their runtime steps. The runtime step, as a key time scale, ensures accurate temporal correspondence between the two streams. During this process, each data point is compared and adjusted based on its timestamp, ensuring accurate matching of data points with the same or similar timestamps, generating a timestamp alignment result. This alignment operation allows subsequent analysis to be conducted within a unified timeframe, avoiding erroneous analysis due to time inconsistencies.

[0048] Based on the timestamp alignment, stacking the first and second information streams along the time dimension to generate a two-dimensional data matrix is ​​a crucial data processing step. First, the timestamp alignment ensures the temporal correspondence between the two information streams, providing an accurate foundation for subsequent stacking operations. Then, during stacking, time steps are used as rows in the two-dimensional data matrix, meaning each row represents a specific point in time, while columns represent different dimensions, including various data features from both the first and second information streams. For example, one column might represent a specific parameter from the first information stream, and another column might represent a different parameter from the second. In this way, the two information streams are integrated into a single two-dimensional data matrix, allowing simultaneous observation of data changes across different dimensions within the same timeframe. This two-dimensional data matrix provides an intuitive and structured dataset for further data analysis, facilitating various statistical analyses, pattern recognition, and other operations, and contributing to a deeper understanding of the operational status of hydropower units at different times and in different aspects.

[0049] When processing a two-dimensional data matrix, information change analysis is performed by traversing the matrix. During the traversal, the changes in data values ​​across different dimensions at each time step are carefully observed. For example, the differences in values ​​across various data dimensions in adjacent time steps are compared, and indicators such as the rate of change and fluctuation amplitude are calculated to understand the dynamic characteristics of the data over time. Through a comprehensive analysis of the entire two-dimensional data matrix, an information change curve is generated. This curve visually displays the trend and magnitude of data changes at different time points. Multiple operating time nodes are extracted according to the information change curve. In this process, attention is focused on time points with significant data fluctuations, as these time points often indicate significant changes in the operating status of the hydropower unit. For example, if the values ​​of multiple data dimensions change significantly at a certain time point, then this time point may be extracted as an operating time node. Simultaneously, nodes indicating specific state changes are also selected, such as the moment of transition from a stable operating state to a fault state, or the moment of switching from one operating mode to another. These extracted operating time nodes are of great significance for understanding the operating dynamics and key events of the hydropower unit, providing crucial time references for subsequent analysis and decision-making.

[0050] Multidimensional integration is performed based on multiple operating time points. For each operating time point, data from all relevant dimensions, such as temperature, pressure, vibration, and flow rate, are integrated. These different dimensions of data are combined into a unified vector, thus integrating multiple pieces of information from a single time point. For example, at a specific operating time point, the temperature, pressure, vibration amplitude, and flow rate values ​​at that moment are arranged into a vector in a certain order. This vector fully represents the operating status characteristics of the hydropower unit at that time point. By processing each operating time point in this way, a multidimensional vector set is generated. This multidimensional vector set is then arranged according to multiple operating cycles, where the operating cycle is a time interval determined based on the actual operating conditions of the hydropower unit. During the arrangement process, it is ensured that each multidimensional vector corresponds to the correct time position, and these vectors are arranged sequentially according to time. In this way, the operating time series can be determined by the ordered arrangement of the multidimensional vector set. This operating time series not only contains detailed status information of the hydropower units at different points in time, but also organically combines data from different dimensions in the form of multidimensional vectors, providing a comprehensive and accurate time series data foundation for further analysis of the operating trends of hydropower units, fault diagnosis and prediction.

[0051] In one possible implementation, the state prediction channel building module further includes:

[0052] The multiple operational sensor data are synchronized to the state prediction channel and stored in the memory module of the state prediction channel. The memory module contains the historical operational sensor dataset. Distance calculations are performed between the multiple operational sensor data and the historical operational sensor dataset using the memory module to determine multiple Euclidean distance data. It is determined whether the multiple Euclidean distance data are within the expected distance interval; Euclidean distance data not within the expected distance interval are extracted to determine multiple abnormal distances. These multiple abnormal distances are used as indexes to traverse the historical operational sensor dataset, generating multiple sensor data to be updated. According to the multiple operational cycles, the multiple operational sensor data are mapped and replaced with the multiple sensor data to be updated, and the memory module is updated to obtain a memory update module. Regression prediction is performed based on the memory update module and the multiple operational sensor datasets to generate multiple operational state prediction results.

[0053] Specifically, multiple operational sensor data are synchronously transmitted to the state prediction channel, and these data are stored in the memory module of the state prediction channel. This memory module already contains historical operational sensor datasets, and the addition of new operational sensor data further enriches the data set. In this way, the integration of old and new data in the same module is achieved, providing a more comprehensive data foundation for subsequent analysis and processing.

[0054] By utilizing the memory module, the distance between the newly stored multiple operational sensor data and the historical operational sensor dataset is calculated. The Euclidean distance calculation method is used here. By calculating for each corresponding data point, multiple Euclidean distance data can be determined. These distance data reflect the degree of numerical difference between the new operational sensor data and the historical data, and can help quantify the similarity or difference between the two.

[0055] These multiple Euclidean distance data points are analyzed to determine if they fall within the expected distance range. This expected distance range is a pre-defined range based on the normal operation of the hydropower unit and the statistical characteristics of the data. If a particular Euclidean distance data point does not fall within this expected range, it indicates a significant difference between the new operational sensor data and historical data, suggesting an anomaly. These Euclidean distance data points that do not fall within the expected distance range are extracted and identified as multiple abnormal distances.

[0056] Using these multiple outlier distances as indices, the historical sensor dataset is traversed. During the traversal, historical data points corresponding to the outlier distances are found; these data points are the parts that need to be updated, generating multiple sensor data sets to be updated.

[0057] Following the sequence of multiple operating cycles, new operating sensor data is mapped to and replaces these multiple sensor data to be updated. In this way, potentially problematic historical data is replaced with new, accurate data, thereby updating the memory module and obtaining a memory update module. This updated module contains more accurate hydropower unit operating data and can better reflect the current operating status.

[0058] The memory update module is a collection of the latest and most accurate operational sensor data obtained after a series of previous processing steps. This module integrates historical and new operational sensor data, and updates and corrects potentially abnormal data, resulting in higher reliability and representativeness. Multiple operational sensor datasets are collections of data collected in real time during the operation of the hydropower unit, including parameters measured by various sensors, such as temperature, pressure, and vibration.

[0059] Multiple operational sensor datasets are real-time data collected by various sensors of the hydropower unit at the current moment. These datasets cover key parameters such as temperature, pressure, and vibration, and can promptly reflect the current operating status of the hydropower unit. Long Short-Term Memory (LSTM) networks are a type of deep learning model well-suited for processing time-series data. During regression prediction, the data from the memory update module, along with the multiple operational sensor datasets, are input into the LSTM network. The network first preprocesses the input data, performing operations such as normalization to ensure the data is within an appropriate range. Then, the LSTM network progressively processes the input data, analyzing trends, periodicity, and other characteristics in historical data, and combining this with current operational sensor data to predict the future operating status of the hydropower unit. The memory unit plays a crucial role in this process, remembering long-term dependencies, enabling the network to better understand the operating patterns of the hydropower unit. As data flows through the network, the LSTM network continuously adjusts its internal parameters to minimize the error between predicted and actual values. Through repeated training and optimization, the network gradually improves the accuracy of its predictions. Ultimately, the LSTM network generates multiple operational status predictions, including specific numerical predictions of various parameters of the hydropower unit over a future period, such as temperature trends and pressure fluctuation ranges.

[0060] In one possible implementation, the operation monitoring result generation module further includes:

[0061] The hydropower unit's operation is evaluated based on the aforementioned operating environment dataset. An expected operating state threshold is set based on the evaluation results. It is then determined whether the multiple predicted operating state results are greater than or equal to the expected operating state threshold. If the multiple predicted operating state results are greater than or equal to the expected operating state threshold, these results are added to the normal operating state data group, and multiple normal operating state prediction labels are generated. The normal operating state data group and the multiple normal operating state prediction labels have a corresponding relationship. If the multiple predicted operating state results are less than the expected operating state threshold, these results are added to the abnormal operating state data group, and multiple abnormal operating state prediction labels are generated. The abnormal operating state data group and the multiple abnormal operating state prediction labels have a corresponding relationship. Finally, the multiple normal operating state prediction labels and the multiple abnormal operating state prediction labels are added to the operating monitoring results.

[0062] Specifically, a comprehensive operational assessment of the hydropower units is conducted using an operating environment dataset, which encompasses various factors such as water temperature, water flow velocity, and external climate conditions. Through in-depth analysis of this data, combined with the operating parameters of the hydropower units, a comprehensive judgment is made on their operational performance under specific environmental conditions. Based on the assessment results, a threshold for an expected operating state is set. This threshold is a crucial standard for measuring whether the hydropower units are in a good operating condition, and its determination requires comprehensive consideration of multiple factors to ensure its rationality and accuracy.

[0063] Multiple operational status predictions are evaluated by comparing each prediction with a desired operational status threshold. If a prediction meets the condition of being greater than or equal to the threshold, the corresponding hydropower unit is considered to be operating in a relatively ideal state. This prediction is then added to the normal operating status data group. Simultaneously, to facilitate subsequent identification and analysis, multiple normal operating status prediction tags are generated, ensuring a correspondence between the normal operating status data group and these tags. This allows for quick retrieval of the corresponding normal operating status data when needed.

[0064] If multiple operational status predictions are lower than the expected operational status threshold, it indicates a potential problem with the corresponding hydropower unit's operational status. Therefore, these predictions that fall below the expected operational status threshold are added to the abnormal operational status data group. Similarly, to facilitate the investigation and handling of abnormal operational statuses, multiple abnormal operational status prediction labels are generated, ensuring a correspondence between the abnormal operational status data group and these labels.

[0065] Finally, by adding multiple normal operation prediction labels and multiple abnormal operation prediction labels to the operation monitoring results, the operation monitoring results can clearly display the classification of the hydropower unit's operating status. By viewing the operation monitoring results, one can quickly understand the operating status of the hydropower unit. For normal operation status, continuous monitoring and optimization can be carried out, while for abnormal operation status, corresponding measures can be taken in a timely manner for diagnosis and repair, so as to ensure the safe, stable and efficient operation of the hydropower unit.

[0066] In one possible implementation, the state prediction optimization channel generation module further includes:

[0067] Extract the multiple abnormal operation prediction labels and iterate through the multiple operation state prediction results to determine multiple abnormal operation state results; calculate the deviation based on the multiple abnormal operation state results to generate multiple abnormal deviation values; backtrack the multiple operation sensor data according to the multiple abnormal deviation values ​​to determine multiple abnormal data sources; compare and verify the multiple abnormal data sources with the multiple operation state prediction results to generate multiple abnormal prediction verification results; add the multiple abnormal prediction verification results to the prediction feedback information.

[0068] Specifically, multiple abnormal operation prediction labels are extracted, and these labels are used to iterate through multiple operation state prediction results. During the iteration process, it is possible to accurately determine which operation state prediction results belong to abnormal operation states, thus obtaining multiple abnormal operation state results. These results represent the potential operational states of the hydropower unit, providing specific targets for subsequent analysis.

[0069] After identifying the results of multiple abnormal operating states, deviation calculation is performed to more accurately quantify the severity of these abnormal states. A reference standard for normal operating conditions needs to be established, based on a stable operating range from historical data, theoretical model predictions, or a pre-defined specification range. Then, for each abnormal operating state result, it is compared with the reference standard for normal operating conditions. By calculating the difference, the degree of deviation between the abnormal result and the normal state is determined, thus generating multiple abnormal deviation values, each representing the magnitude of the difference between an abnormal operating state and the normal state.

[0070] Based on multiple abnormal deviation values, backtracking is performed on multiple operational sensor data. This process analyzes the relationship between each abnormal deviation value and the operational sensor data. By comparing data changes at different time points and the performance of each sensor, the source of these abnormal deviations is identified, thus finding multiple abnormal data sources. These abnormal data sources may include a specific sensor malfunction, interference during data transmission, or abnormal operation of a component within the hydropower unit. The identified multiple abnormal data sources are compared and verified with multiple operational state prediction results. Features and information extracted from the abnormal data sources are compared with previously predicted operational state results to check for consistency. If they match, the prediction results accurately reflect the source of the abnormality; if they do not match, further analysis is needed, which may indicate inaccurate identification of the abnormal data sources or limitations in the prediction model. After comparison and verification, multiple abnormal prediction verification results are generated to help accurately locate problems and take effective measures to resolve abnormal operational states of the hydropower unit.

[0071] Adding multiple anomaly prediction verification results to the prediction feedback information enriches and improves the assessment and monitoring of the hydropower unit's operating status. These anomaly prediction verification results include detailed analyses of abnormal operating states of the hydropower unit, such as the specific manifestations of the anomaly, possible data sources of the anomaly, and comparisons with the predicted operating status results. By adding these results to the prediction feedback information, the system can more comprehensively reflect the actual operating status of the hydropower unit. By reviewing the prediction feedback information, users can quickly understand the problems existing in the hydropower unit and their severity, thereby enabling timely implementation of corresponding measures for fault diagnosis and repair to ensure the safe and stable operation of the hydropower unit. Simultaneously, the prediction feedback information with added anomaly prediction verification results also provides valuable references for subsequent data analysis and model optimization, helping to continuously improve the hydropower unit's operation monitoring and prediction system.

[0072] In one possible implementation, the state prediction optimization channel generation module further includes:

[0073] Based on the results of the multiple abnormal operating states, cluster analysis is performed according to the multiple abnormal deviation values ​​to determine multiple deviation types; according to the multiple deviation types and the multiple abnormal deviation values, the multiple operating sensor data are backtracked according to the multiple operating cycles to draw a trend chart of operating sensor data changes; the trend chart of operating sensor data changes is traversed to locate fluctuations and determine the multiple abnormal data sources.

[0074] Specifically, when faced with multiple abnormal operating conditions, cluster analysis is performed based on multiple abnormal deviation values. The abnormal deviation value corresponding to each abnormal operating condition is carefully examined; these deviation values ​​reflect the degree of difference between the abnormal and normal states. Then, a clustering algorithm is used to group these abnormal results according to the similarity of their deviation values. For example, abnormal operating conditions with similar deviation values ​​are grouped into the same category, thus identifying multiple deviation types. This classification helps to better understand different types of abnormal situations, revealing that some anomalies are caused by similar reasons, such as specific sensor fault types or performance degradation of specific components of the hydropower unit. By identifying multiple deviation types, more targeted directions are provided for subsequent fault diagnosis and handling, enabling more efficient resolution of abnormal problems in hydropower units.

[0075] Based on multiple deviation types and their corresponding abnormal deviation values, backtracking is performed on multiple operational sensor data over multiple operational cycles. In this process, the data collected by each sensor in each operational cycle is analyzed one by one. Combining different deviation types and abnormal deviation values, the relationship between the data change characteristics and the deviations within a specific operational cycle is determined. By integrating this information, a trend chart of operational sensor data changes is plotted. This trend chart visually demonstrates how the data from each sensor changes over time in different operational cycles, and its correspondence with different deviation types and abnormal deviation values. For example, for a specific deviation type, the trend chart clearly shows that the data from the relevant sensors exhibits an upward, downward, or fluctuating trend within a specific operational cycle, thus providing a better understanding of the changes in the hydropower unit's operating status and offering important visual evidence for further analysis of the root causes of anomalies.

[0076] When processing the trend chart of operational sensor data changes, traversing the chart to locate fluctuations is a crucial step. Carefully observe the changes at each data point in the trend chart, analyzing the data trend step by step along the time axis, paying particular attention to areas where significant fluctuations occur. These fluctuations manifest as sharp increases, decreases, or irregular patterns in the data. For each possible fluctuation point, compare the data performance of different sensors at the same time point or time period. If multiple sensors show fluctuations simultaneously, this area is more likely to be the source of the anomaly. Simultaneously, combine this with previously identified deviation types and abnormal deviation values ​​to further analyze the relationship between these fluctuations and the anomaly. For example, if the temperature sensor data suddenly rises within a certain time period, and the pressure sensor data also shows abnormal changes, and the deviation type and abnormal deviation value for this time period also show significant deviations, then this time period and the related sensors can be preliminarily identified as an abnormal data source. Through a comprehensive traversal and analysis of the entire operational sensor data trend chart, multiple abnormal data sources are identified. These abnormal data sources may be due to a malfunction in a specific component of the hydroelectric generator, such as bearing wear causing abnormal fluctuations in temperature and vibration sensor data; or they may be due to external environmental factors, such as changes in water flow causing changes in pressure sensor data. Identifying these abnormal data sources provides a clear direction for subsequent fault diagnosis and repair work, enabling more targeted measures to improve the operational stability and reliability of hydropower units.

[0077] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0078] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0079] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A deep learning-based hydropower unit state prediction system, characterized in that, The system includes: The operation sensor data acquisition module is based on multiple operating segments and collects multiple operating sensor data from the hydropower unit in real time through a data sensor network. A state prediction channel construction module is used to construct a state prediction channel by performing deep learning on the multiple runtime segments based on the historical running sensor dataset. The operation monitoring result generation module is used to synchronize the multiple operation sensor data to the state prediction channel for analysis, generate multiple operation state prediction results, and perform operation monitoring of the hydropower unit according to the multiple operation state prediction results and the operation environment dataset to generate operation monitoring results, which include multiple state labels. A state prediction optimization channel generation module is used to verify the multiple running state prediction results by traversing the multiple state labels, generate prediction feedback information, update and optimize the state prediction channel according to the prediction feedback information, and generate a state prediction optimization channel. The intelligent prediction module is used to intelligently predict the state of the hydropower unit through the state prediction optimization channel. The state prediction channel construction module further includes: Based on the multiple operating periods, the operation of the hydropower unit is analyzed, and multiple operating cycles are generated according to the analysis results. Retrieve historical operating data archives of hydropower units, extract historical operating sensor datasets based on historical operating data archives, extract features from historical operating sensor datasets according to the multiple operating cycles, and generate multiple historical operating features; Time series analysis is performed based on the multiple operating cycles and the multiple historical operating characteristics to determine the operating time step of the hydropower unit. A long short-term memory network is used to perform deep learning on the historical running sensor dataset according to the running time step to generate a running time series. The running time series contains multiple running time nodes, and each of the multiple running time nodes has at least one historical running sensor data. The multiple historical operational sensor data are verified and evaluated based on the multiple operational time nodes to generate a verification token. The status prediction channel is then output based on the verification token for evaluation. The state prediction channel construction module also includes: Based on the running time step, the historical running sensor dataset is subjected to data time series standardization processing to generate a historical sensor standard dataset. The historical sensor standard dataset is divided according to the running time step to generate multiple standard data sample segments; Long-term analysis is performed based on the multiple standard data sample segments to generate a first analysis result. Multiple long-term standard data sample segments are extracted based on the first analysis result. Long-term dependency coefficients are calculated based on the multiple long-term standard data sample segments. A short-term analysis is performed based on the multiple standard data sample segments to generate a second analysis result. Multiple short-term standard data sample segments are extracted based on the second analysis result. Short-term dependency coefficients are calculated based on the multiple short-term standard data sample segments. The historical operational sensor dataset is captured according to the long-term dependency coefficient to generate a first information stream; The historical operational sensor dataset is captured according to the short-term dependency coefficient to generate a second information stream; Based on the stacking of the first information stream and the second information stream, multiple running time nodes are determined, and the running time sequence is extracted by integrating the multiple running time nodes. The state prediction channel construction module also includes: The first information stream and the second information stream are aligned according to the stated running time step to generate a timestamp alignment result; Based on the timestamp alignment result, the first information stream and the second information stream are stacked according to the time dimension to generate a two-dimensional data matrix; The information change analysis is performed by traversing the two-dimensional data matrix to generate an information change curve, and the multiple running time nodes are extracted according to the information change curve. The multiple running time nodes are integrated in multiple dimensions to generate a multidimensional vector set. The multidimensional vector set is then arranged according to the multiple running cycles to determine the running time sequence.

2. The deep learning-based hydropower unit state prediction system as described in claim 1, characterized in that, The state prediction channel construction module also includes: The multiple operational sensor data are synchronized to the state prediction channel and stored in the memory module of the state prediction channel. The memory module contains the historical operational sensor dataset. The memory module calculates the distance between the multiple operational sensor data and the historical operational sensor dataset to determine multiple Euclidean distance data. Determine whether the multiple Euclidean distance data are within the expected distance range, extract the Euclidean distance data that are not within the expected distance range, and identify multiple abnormal distances; Using the multiple abnormal distances as indexes, the historical operating sensor dataset is traversed to generate multiple sensor data to be updated. According to the multiple operating cycles, the multiple operating sensor data are mapped and replaced with the multiple sensor data to be updated, and the memory module is updated to obtain the memory update module; Based on the memory update module and the multiple operational sensor datasets, regression prediction is performed to generate the multiple operational state prediction results.

3. The deep learning-based hydropower unit state prediction system as described in claim 1, characterized in that, The operation monitoring result generation module also includes: Based on the aforementioned operating environment dataset, the hydropower unit is evaluated for operation, and the expected operating state threshold is set according to the evaluation results. Determine whether the multiple predicted operating states are greater than or equal to the expected operating state threshold; If the multiple operation status prediction results are greater than or equal to the expected operation status threshold, then the operation status prediction results that are greater than or equal to the expected operation status threshold are added to the normal operation status data group, and multiple normal operation prediction labels are generated. The normal operation status data group and the multiple normal operation prediction labels have a corresponding relationship. If the multiple operation status prediction results are less than the expected operation status threshold, the operation status prediction results that are less than the expected operation status threshold are added to the abnormal operation status data group, and multiple abnormal operation prediction tags are generated. The abnormal operation status data group and the multiple abnormal operation prediction tags have a corresponding relationship. Add the multiple normal operation prediction labels and the multiple abnormal operation prediction labels to the operation monitoring results.

4. The deep learning-based hydropower unit state prediction system as described in claim 3, characterized in that, The state prediction optimization channel generation module further includes: Extract the multiple abnormal operation prediction labels, traverse the multiple operation state prediction results, and determine the multiple abnormal operation state operation results; Based on the results of the operation in the multiple abnormal operating states, deviation calculation is performed to generate multiple abnormal deviation values; The multiple operational sensor data are backtracked according to the multiple abnormal deviation values ​​to identify multiple abnormal data sources. The multiple abnormal data sources are compared and verified with the multiple operational state prediction results to generate multiple abnormal prediction verification results. The multiple anomaly prediction verification results are added to the prediction feedback information.

5. The deep learning-based hydropower unit state prediction system as described in claim 4, characterized in that, The state prediction optimization channel generation module further includes: Based on the results of the multiple abnormal operating states, cluster analysis is performed according to the multiple abnormal deviation values ​​to determine multiple deviation types; Based on the multiple deviation types and the multiple abnormal deviation values, the multiple operational sensor data are backtracked according to the multiple operational cycles, and a trend chart of operational sensor data change is drawn. The fluctuations are located by traversing the trend chart of the operational sensor data, and the multiple abnormal data sources are identified.

Images