Water and electricity intelligent early warning method and system driven by data knowledge fusion

Abstract

The invention discloses a water and electricity intelligent early warning method and system driven by data knowledge fusion. The method comprises the following steps: S1, acquiring data; s2, preprocessing the data; s3, artificial intelligence is used for modeling; s4, performing equipment real-time early warning; s5, collecting data and expert knowledge; s6, constructing an FTA fault tree logic model; s7, carrying out early warning analysis; and S8, verifying early warning accuracy. According to the invention, intelligent early warning can be carried out by using an artificial intelligence technology, and accuracy verification can be carried out on an early warning result through expert knowledge.

Description

Technical Field

[0001] The present invention belongs to the field of hydropower intelligent technology, and in particular relates to a hydropower intelligent early warning method and system driven by data knowledge fusion. Background Art

[0002] As hydropower equipment becomes increasingly complex and intelligent, how to effectively carry out equipment early warning and fault diagnosis has become an urgent problem to be solved. Traditional hydropower equipment early warning methods mainly rely on expert knowledge and experience, lacking accuracy and real-time performance. In recent years, the development of artificial intelligence technology has provided new ideas for hydropower equipment early warning. However, single artificial intelligence technology also has problems such as insufficient generalization ability and sensitivity to data quality. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a hydropower intelligent early warning method and system driven by data knowledge fusion, which combines the two parts of a priori and a posteriori, and can not only use artificial intelligence technology for intelligent early warning, but also verify the accuracy of the early warning results through expert knowledge.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: A hydropower intelligent early warning method and system driven by data knowledge fusion, comprising the following steps: S1. Get data; S2, preprocessing the data; S3, using artificial intelligence modeling; S4. Real-time warning of equipment; S5. Collect data and expert knowledge; S6. Construct FTA fault tree logic model; S7. Conduct early warning analysis; S8. Verify the accuracy of the warning.

[0005] Preferably, in step S1, the acquired data includes sensor information and operating condition data; and time series data within a period of time is read; The sensor information includes measurement point location information and measurement attribute information; the operating condition data includes one or more data including active power, water head, guide vane opening, unit speed and tailwater pipe outlet temperature.

[0006] Preferably, in step S2, the data preprocessing process includes the following steps: S2.1. Pivot data: convert the time series data format into a wide table mode. The columns in the wide table mode include time, sensor information, and working condition data. S2.2. Perform null value processing: If the null value is a numeric value, fill in the missing attribute value based on the average value of the attribute in all other objects; if the null value is a non-numeric value, fill in the missing attribute value with the value with the largest number of times of the attribute in all other objects according to the principle of majority in statistics; S2.3, perform frequency reduction processing: reduce the frequency of all data from second-level data to 1-minute-level data, and the frequency reduction method is average processing, that is, take the average of the data in the time window of each minute; S2.4, steady-state processing: obtain the active power fluctuation range threshold of the input unit, and judge when the unit is in stable operation based on this range; S2.5, Out-range data processing: Out-range refers to the operating range of the device under test. The operating range information of the input measuring point is obtained and data is screened based on the information.

[0007] Preferably, S3, the step of adopting artificial intelligence modeling is: S3.1. Automatically extract features and target points: The extraction methods used include target point generation and feature point dimensionality reduction; S3.2, algorithm selection and training: Automatically select and train the FTA fault tree logic model based on the characteristics and target measurement points provided by S3.1.

[0008] Preferably, S4, real-time warning of equipment: deploy the FTA fault tree logic model obtained in step S3 in the industrial Internet platform, set a scheduling cycle, the FTA fault tree logic model runs periodically once, obtains data of one cycle each time, and determines whether the data is a warning.

[0009] Preferably, S5, the step of collecting data and expert knowledge is: S5.1. Collect historical data: Collect historical operation data of the hydropower station, including the temperature range of the upper guide shoe, abnormal temperature records and corresponding fault events; S5.2. Obtain expert knowledge: Obtain professional knowledge on the possible causes of abnormal upper conductor temperature, fault types and their solutions by interviewing hydropower station operation and maintenance experts and consulting technical documents and manuals.

[0010] Preferably, the method of step S6 is: S6.1. Determine the top event: According to the results of accident investigation and statistical analysis, refer to the frequency of accidents and the severity of accident losses, confirm the top event and logic gate; S6.2. Determine intermediate events: According to the analyzed event causes, create intermediate process events as intermediate event fault tree nodes in sequence; S6.3. Determine basic events: locate the basic event nodes of the fault tree by the basic events that cause the intermediate accidents; S6.4. Check the accident tree: Check whether the prepared accident tree conforms to the principles of logical analysis, check the connection status of the logic gates, check whether the upper-level events are the inevitable results of the lower-level events, whether the lower-level events are the sufficient cause events of the upper-level events, and check whether all direct cause events are found; S6.5, Binding node devices and fault diagnosis algorithms: Based on the collected data and expert knowledge, associate the logical devices where the fault mode is located with the fault tree nodes, and associate the hydropower-specific algorithms according to the node logical devices, so as to call the algorithm engine to diagnose the node occurrence probability and fault occurrence in the subsequent FTA fault diagnosis; S6.6. Save the FTA fault tree logic model: Confirm that it is correct and save the entire static tree. The fault tree drawing is now completed; S6.7. Release the FTA fault tree logic model and instantiate the FTA fault tree logic model: First, publish the FTA fault tree logic model to make the newly created FTA fault tree logic model effective. Only the effective FTA fault tree logic model can be instantiated. Then, perform the instantiation operation, select the plant station corresponding to the FTA fault tree logic model to be instantiated and the instantiated equipment tree, and enter the FTA fault tree logic model instantiation process; During the instantiation process, the logical devices bound to the nodes of the FTA fault tree logic model are converted into physical devices of the plant instantiation device tree, and the logical fault diagnosis tools bound to the nodes are converted into instantiation tools; S6.8, configure the scheduling task: S6.9, fault alarm: perform periodic fault diagnosis according to the scheduling task configured in step S6.7, and generate a fault diagnosis result instance diagram, i.e., a complete FTA fault tree; the FTA fault tree identifies the fault condition and occurrence probability of each event node.

[0011] Preferably, the method of step S6.8 is: a) Create a scheduled task to periodically call the instantiated FTA fault tree logic model for fault diagnosis; b) Diagnosis process: When the event node is configured with a diagnostic tool, the diagnostic tool configured on the node is called in sequence to obtain the node's alarm status and occurrence probability, and the occurrence probability is substituted into the FTA fault tree logic model.

[0012] 9. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized in that: the steps of step S8 are as follows: S8.1. Expert evaluation: When the prior module issues an early warning, the FTA fault tree logic model is used to perform fault diagnosis and generate possible fault types and solutions; at the same time, experts are asked to evaluate the early warning results based on their knowledge and experience; S8.2. Comparison of results: Compare the diagnostic results of the FTA fault tree logic model with the expert's evaluation results; if the two are consistent or highly similar, the accuracy of the early warning results is confirmed; otherwise, the early warning results need to be further reviewed and adjusted; S8.3 Feedback Mechanism: Establish a feedback mechanism to feed the verification results back to the a priori early warning system so that it can adjust and optimize itself according to the actual situation; S8.4, Intelligent Fault Diagnosis and Expert Evaluation: After the a priori module issues a warning signal, an advanced fault tree analysis model is used for intelligent fault diagnosis. The FTA fault tree logic model can automatically identify potential fault modes and generate a series of possible fault types and their solutions. At the same time, experts are used to conduct in-depth evaluation of the warning results to ensure the accuracy and comprehensiveness of the diagnosis. S8.5. Comparison of intelligent diagnosis and expert evaluation results: Compare and analyze the intelligent diagnosis results of the FTA fault tree logic model with the expert evaluation results; if the two are highly consistent, it means that the accuracy of the early warning results has been verified; if there is a difference, the early warning results need to be carefully reviewed and necessary adjustments are made to ensure the reliability of the final results; S8.6, Closed-loop feedback and optimization mechanism: Establish a closed-loop feedback mechanism to feed back the verification results to the prior warning system; the system will self-learn and optimize based on the feedback information to improve the accuracy and efficiency of the warning; at the same time, this mechanism also helps to build a continuously improving intelligent warning system so that it can better adapt to the changing environment and needs; Once the accuracy of the early warning results is verified, the FTA fault tree logic model should be able to output specific fault types and corresponding solutions.

[0013] A hydropower intelligent early warning system driven by data knowledge fusion adopts the hydropower intelligent early warning method driven by data knowledge fusion.

[0014] The present invention can achieve the following beneficial effects: 1. Data knowledge fusion: The present invention proposes an intelligent early warning framework driven by data knowledge fusion, which combines the dual advantages of data-driven and knowledge-driven, and improves the accuracy and reliability of early warning.

[0015] 2. Combination of a priori module and a posteriori module: The a priori module uses artificial intelligence technology to perform intelligent early warning, and the a posteriori module uses expert knowledge to verify accuracy, forming a closed-loop verification mechanism for the early warning results.

[0016] 3. Combining intelligent early warning with expert systems: Combining artificial intelligence technologies (such as machine learning and deep learning) with expert systems not only leverages the capabilities of artificial intelligence in data processing and analysis, but also utilizes expert knowledge and experience in fault diagnosis.

[0017] 4. Verification and correction of warning results: The a posteriori module verifies the warning results of the a priori module and can correct or ignore the warning results in real time, thus improving the flexibility and adaptability of the warning system.

[0018] 5. A hydropower intelligent early warning system driven by data knowledge fusion, by combining the priori and a posteriori parts, fully utilizes the advantages of artificial intelligence technology and expert knowledge, and improves the accuracy and reliability of hydropower equipment early warning. The priori module uses artificial intelligence technology to process and analyze data, which can automatically mine the potential laws and patterns in the data, realize intelligent early warning of hydropower equipment, and improve the real-time and automation of early warning. The a posteriori module uses expert knowledge to verify the accuracy of the early warning results, which can timely discover and correct the false alarms and missed alarms of the priori module, and improve the accuracy and credibility of the early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 This is a data knowledge fusion-driven hydropower intelligent early warning method and system structure diagram of the present invention. DETAILED DESCRIPTION

[0020] The preferred solution is Figure 1 As shown, a hydropower intelligent early warning method and system driven by data knowledge fusion, S1. Get data; S2, preprocessing the data; S3, using artificial intelligence modeling; S4. Real-time warning of equipment; S5. Collect data and expert knowledge; S6. Construct FTA fault tree logic model; S7. Conduct early warning analysis; S8. Verify the accuracy of the warning.

[0021] Furthermore, in step S1, the acquired data includes sensor information and working condition data; and the time series data within a period of time (nearly one year) is read; The sensor information includes measurement point location information and measurement attribute information; the operating condition data includes one or more data including active power, water head, guide vane opening, unit speed and tailwater pipe outlet temperature.

[0022] For example: sensor information includes 1# upper guide tile temperature, 2# upper guide tile temperature..., 12# upper guide tile temperature.

[0023] Furthermore, in step S2, the data preprocessing process includes the following steps: S2.1. Pivot data: convert the time series data format into a wide table mode. The columns in the wide table mode include time, sensor information, and working condition data. Because the time series format is not easy to analyze, it is converted to wide table mode. The converted columns are time, 1# upper guide bearing temperature, 2# upper guide bearing temperature..., 12# upper guide bearing temperature, active power, head, guide vane opening, unit speed, and tailwater pipe outlet temperature.

[0024] S2.2. Perform null value processing: If the null value is a numeric value, fill in the missing attribute value based on the average value of the attribute in all other objects; if the null value is a non-numeric value, fill in the missing attribute value with the value with the largest number of times of the attribute in all other objects according to the principle of majority in statistics; After data pivoting, there will be timestamp misalignment, that is, some columns have data at the same time, while others do not. This is because the data was not uploaded at that time during the data collection process, such as equipment failure, sensor failure, communication problems, etc. These null values ​​will affect subsequent data analysis and modeling. If the null value is numeric, the missing attribute value is filled in based on the average value of the attribute in all other objects; if the null value is non-numeric, the missing attribute value can be filled in based on the majority principle in statistics, using the value with the highest number of values ​​of the attribute in all other objects.

[0025] S2.3, perform frequency reduction processing: reduce the frequency of all data from second-level data to 1-minute-level data, and the frequency reduction method is average processing, that is, take the average of the data in the time window of each minute; Frequency reduction is a common operation in the analysis of time series data. The principle is to change the sampling frequency of the data or the length of the time period. Frequency reduction is the process of converting high-frequency data into low-frequency data. It can reduce the amount of data, improve the efficiency of early warning, and can also be used for purposes such as smoothing data and removing noise.

[0026] S2.4, steady-state processing: obtain the active power fluctuation range threshold of the input unit, and judge when the unit is in stable operation based on this range; The purpose of steady-state screening is to distinguish between fluctuating data and steady-state data. Under non-steady-state conditions, the characteristics of the equipment change dramatically, and the operating data cannot truly reflect the relationship between the input and output of the system. Therefore, through steady-state screening, these noises can be removed to better reflect the true operating status of the system.

[0027] S2.5, Out-range data processing: Out-range refers to the operating range of the device under test. The operating range information of the input measuring point is obtained and data is screened based on the information.

[0028] If the operating range of the upper guide shoe is 0-80 degrees, the data less than 0 or greater than 80 in the 12 upper guide shoe columns will be screened; Furthermore, S3, the steps of using artificial intelligence modeling are: S3.1. Automatic extraction of features and target measurement points: including target measurement point generation and feature measurement point dimensionality reduction; in: Generation of target measuring points: In this example, a total of 12 measuring points [1# upper guide tile temperature, 2# upper guide tile temperature..., 12# upper guide tile temperature] will be converted into the "maximum upper guide tile temperature" measuring point.

[0029] Feature point dimension reduction: Use machine learning algorithms to process data and extract useful features from it. It transforms or projects the original data into a lower-dimensional space to better capture the structural information of the data. This can reduce the data dimension and improve the accuracy of the algorithm. In this example, 5-dimensional data [active power, head, guide vane opening, unit speed, tailwater pipe outlet temperature] is input, and the PCA algorithm is used to reduce the data to 3 dimensions [active power, tailwater pipe outlet temperature, head + guide vane opening + unit speed].

[0030] S3.2, algorithm selection and training: The model is automatically selected and trained based on the features and target measurement points provided by S3.1.

[0031] This step tries a variety of different machine learning algorithms and uses automated techniques to optimize the model's hyperparameters. The target is manually selected as the maximum upper guide tile temperature. The data is divided into a training set and an evaluation set in a ratio of 7 to 3. Based on the current scenario and the existing information in the knowledge base (such as shown in Table 1), XGBOOST and LSTM are selected for training in the training set. The Rsquare of the three algorithms for the target measurement point is calculated on the evaluation set, and the model with the highest accuracy (that is, the model with the highest R-square value) is selected. Table 1 Knowledge base existing information table

[0032] Scenario 1: Tile and oil temperature warnings The tile temperature and oil temperature are important indicators that reflect the friction, wear and lubrication of the turbine. The data of these measuring points usually have periodic changes and are affected by factors such as workload and ambient temperature, showing certain seasonality and periodicity.

[0033] When warning this type of measuring point, due to insufficient fault samples, regression algorithms are usually used for modeling. The reasons for selecting Xgboost and LSTM are: Xgboost is an integrated learning algorithm suitable for processing structured data, which can effectively handle complex relationships between features and has strong data prediction capabilities for tile temperature and oil temperature measuring points. LSTM is a recurrent neural network suitable for processing time series data, which can capture long-term dependencies of data and has better prediction effects for data with periodicity and trend.

[0034] Scenario 2: Bearing runout warning Data characteristics of bearing runout measurement points: Frequency domain characteristics: The data of vibration measurement points usually include frequency domain characteristics, such as the frequency distribution and spectrum shape of the vibration signal.

[0035] Time domain characteristics: Vibration data also has time domain characteristics, including the amplitude, periodic changes, trends, etc. of the vibration signal.

[0036] Randomness: The vibration data is random to a certain extent, that is, the vibration signal may be affected by factors such as the external environment and working conditions and present uncertainty.

[0037] Periodicity: Some vibration data may have periodic variations, such as periodic oscillations or vibrations.

[0038] Advantages of algorithm modeling: Advantages of CNN: Since vibration data contains frequency domain and time domain characteristics, CNN can effectively extract these features from the data, use convolution operations to capture frequency domain features, and reduce data dimensions through pooling operations, making the model more efficient.

[0039] Advantages of LSTM: Vibration data usually has the characteristics of time series. LSTM can process this kind of sequence data well and can capture long-term dependencies in the data, such as periodic changes and trends of vibration signals.

[0040] Furthermore, S4 provides real-time warning for equipment: The model obtained in step S3 is deployed in the industrial Internet platform, and a scheduling cycle is set. The model runs periodically, obtains data of one cycle each time, and determines whether the data is a warning.

[0041] The cycle is 10 minutes.

[0042] Further, S5, collect data and expert knowledge: S5.1. Collect historical data: Collect historical operation data of the hydropower station, including the temperature range of the upper guide shoe, abnormal temperature records and corresponding fault events; In particular, data related to the upper guide temperature include the temperature range during normal operation, abnormal temperature records, corresponding fault events, etc.

[0043] S5.2. Obtain expert knowledge: Obtain professional knowledge on the possible causes of abnormal upper conductor temperature, fault types and their solutions by interviewing hydropower station operation and maintenance experts and consulting technical documents and manuals.

[0044] Furthermore, the method of step S6 is: S6.1. Determine the top event: According to the results of accident investigation and statistical analysis, refer to the frequency of accidents and the severity of accident losses, confirm the top event and logic gate; Drag a top event from the toolbar to the canvas, and set the top event number (note that the top event number is T) and name. Here, the upper conductor temperature anomaly is used as the top event of the FTA fault tree logic model; S6.2. Determine intermediate events: According to the analyzed event causes, create intermediate process events (such as cooling system failure, insufficient lubrication, etc.) as intermediate event fault tree nodes in sequence; Click the top event to add an intermediate event and logic gate, and set the intermediate event number (note that the intermediate event number starts with M + a number) and name.

[0045] S6.3. Determine basic events: locate the basic event nodes of the fault tree by the basic events that cause the intermediate accidents (such as insufficient cooling water flow, lubricating oil pump failure, etc.); Click the intermediate event node, add a basic event, and set the basic event number (note that the basic event number starts with X + a number), name, and occurrence probability (the occurrence probability is between 0 and 1, including 0 and 1).

[0046] S6.4. Check the accident tree: Check whether the prepared accident tree conforms to the principles of logical analysis, check the connection status of the logic gates, check whether the upper-level events are the inevitable results of the lower-level events, whether the lower-level events are the sufficient cause events of the upper-level events, and check whether all direct cause events are found; S6.5, Binding node devices and fault diagnosis algorithms: Based on the collected data and expert knowledge, associate the logical devices where the fault mode is located with the fault tree nodes, and associate the hydropower-specific algorithms according to the node logical devices, so as to call the algorithm engine to diagnose the node occurrence probability and fault occurrence in the subsequent FTA fault diagnosis; S6.6. Save the fault tree model: Confirm that everything is correct and save the entire static tree. The fault tree drawing is now completed.

[0047] S6.7. Publish the logical model and instantiate the logical model; First, publish the logical model to make the newly created FTA fault tree logical model effective. Only the effective model can be instantiated. Then perform the instantiation operation, select the plant station corresponding to the model to be instantiated and the instantiation device tree, and enter the model instantiation process; During the instantiation process, the logical devices bound to the nodes of the FTA fault tree logic model are converted into physical devices of the plant instantiation device tree, and the logical fault diagnosis tools bound to the nodes are converted into instantiation tools; S6.8, configure the scheduling task: S6.9, Fault alarm: According to the scheduling task configured in step S6.7, periodic fault diagnosis is performed to generate a fault diagnosis result instance diagram, i.e., a complete FTA fault tree. The tree identifies the fault conditions and occurrence probabilities of each event node.

[0048] Further, the method of step S6.8 is: a) Create a scheduled task and periodically call the instantiated model for fault diagnosis; b) Diagnosis process: When the event node is configured with a diagnostic tool, the diagnostic tool configured on the node is called in sequence to obtain the node's alarm status and occurrence probability, and the occurrence probability is substituted into the fault tree model.

[0049] If the intermediate nodes of the diagnostic tool are not configured, the node occurrence probability is calculated according to the fault diagnosis method (minimum cut set, minimum path set), etc. until the top event alarm status and occurrence probability are finally calculated.

[0050] Furthermore, the steps of step S8 are as follows: S8.1. Expert evaluation: When the prior module issues an early warning, the FTA fault tree logic model is used to perform fault diagnosis and generate possible fault types and solutions. At the same time, experts are asked to evaluate the early warning results based on their knowledge and experience.

[0051] After the early warning system issues an early warning signal, an expert evaluation is first required. Expert evaluation refers to inviting professionals with rich knowledge and experience in the field to review and analyze the early warning results. In order to make an accurate diagnosis, the FTA (Fault Tree Analysis) model is used to analyze potential failure modes and their potential causes. The FTA model allows for gradual analysis from the top system level to the bottom until the specific root cause of the failure is determined.

[0052] In this step, experts will use their experience to compare the possible fault types and solutions generated by the FTA model. This process helps to verify whether the warning is accurate and also provides a reference for subsequent processing. Experts may propose additional test plans, troubleshooting methods or preventive measures to ensure the effectiveness of the warning results.

[0053] S8.2. Comparison of results: Compare the diagnostic results of the FTA fault tree logic model with the expert's evaluation results. If the two are consistent or highly similar, the accuracy of the early warning results is confirmed; otherwise, the early warning results need to be further reviewed and adjusted.

[0054] When the experts have completed their evaluation of the warning results, the next step is to compare the evaluation results with the diagnostic results of the FTA model. The purpose of this step is to ensure coordination and consistency between the two.

[0055] If the results of the two are consistent or highly similar, the accuracy of the warning can be confirmed and appropriate action can be taken. However, if the results of the two are inconsistent or there are significant differences, it may indicate that there is a problem with the warning system or that the expert assessment is wrong. In this case, further in-depth analysis is required to find out the cause of the difference. Possible solutions include re-examining the FTA model, adjusting the parameters of the warning system, or re-conducting the expert assessment.

[0056] S8.3 Feedback Mechanism: Establish a feedback mechanism to feed back the verification results to the prior warning system so that it can adjust and optimize itself according to the actual situation.

[0057] In order to continuously improve the performance of the early warning system, an effective feedback mechanism needs to be established. The feedback mechanism refers to the feedback of insights gained from the result comparison to the early warning system so that it can self-adjust and optimize. Through the feedback mechanism, the early warning system can learn and adapt to new patterns and changes, thereby improving its accuracy and reliability.

[0058] In this mechanism, the early warning system will receive the results of expert evaluation and comparison, and adjust its algorithms and models based on these feedbacks. Through continuous iteration, the early warning system will be able to perform more accurately and stably in future warnings. This feedback mechanism not only helps to improve the performance of the early warning system, but also provides a guarantee for the reliability of the overall system.

[0059] Through the synergy of expert evaluation, result comparison and feedback mechanism, this process of verifying the accuracy of early warning ensures the reliability and accuracy of the early warning system, thus providing a solid foundation for the safety and stability of the overall system. Once the accuracy of the early warning results is verified, the FTA model should be able to output specific fault types and corresponding solutions.

[0060] S8.4, Intelligent Fault Diagnosis and Expert Evaluation: After the prior module issues a warning signal, an advanced fault tree analysis (FTA) model is used for intelligent fault diagnosis. The model can automatically identify potential failure modes and generate a series of possible failure types and their solutions. At the same time, domain experts are invited to use their rich knowledge and experience to conduct in-depth evaluation of the warning results to ensure the accuracy and comprehensiveness of the diagnosis.

[0061] S8.5. Comparison of intelligent diagnosis and expert evaluation results Compare and analyze the intelligent diagnosis results of the FTA fault tree logic model with the expert evaluation results. If the two are highly consistent, it means that the accuracy of the early warning results has been verified; if there is a difference, the early warning results need to be carefully reviewed and adjusted as necessary to ensure the reliability of the final results.

[0062] S8.6. Closed-loop feedback and optimization mechanism A closed-loop feedback mechanism is established to feed back the verification results to the prior warning system. The system will self-learn and optimize based on the feedback information to improve the accuracy and efficiency of the warning. At the same time, this mechanism also helps to build a continuously improving intelligent warning system so that it can better adapt to the changing environment and needs.

[0063] Once the accuracy of the early warning results is verified, the FTA fault tree logic model should be able to output specific fault types and corresponding solutions.

[0064] A hydropower intelligent early warning system driven by data knowledge fusion adopts the hydropower intelligent early warning method driven by data knowledge fusion. It includes: The a priori module is responsible for using artificial intelligence technology to provide intelligent early warning for hydropower equipment. Specifically, the a priori module collects the operating data of hydropower equipment, uses artificial intelligence technologies such as machine learning and deep learning to process and analyze the data, and mines the potential laws and patterns in the data, thereby achieving intelligent early warning for hydropower equipment.

[0065] The a posteriori module is responsible for verifying the accuracy of the warning results of the a priori module using expert knowledge. Specifically, the a posteriori module builds a fault diagnosis model for hydropower equipment by collecting historical data and expert knowledge of hydropower equipment. When the a priori module issues a warning, the a posteriori module inputs the warning result into the fault diagnosis model and verifies the accuracy of the warning result using expert knowledge. If the verification passes, the warning result is confirmed; otherwise, the warning result is corrected or ignored.

[0066] The workflow of the a priori module is as follows: 1. Collect the operating data of the target equipment, including the equipment's operating parameters, characteristic parameters and other information.

[0067] 2. Preprocess the collected data, including data cleaning, data transformation, feature extraction and other operations, to obtain a data format suitable for machine learning or deep learning model processing.

[0068] 3. Use artificial intelligence technologies such as machine learning and deep learning to train and learn the processed data to explore the potential rules and patterns in the data.

[0069] 4. Based on the learned rules and patterns, build an intelligent early warning model for hydropower equipment and provide real-time early warning for hydropower equipment.

[0070] The workflow of the posterior module is as follows: 5. Collect historical data and expert knowledge of hydropower equipment, including information such as equipment failure types, failure causes, and failure solutions.

[0071] 6. Construct a fault diagnosis model for hydropower equipment based on the collected historical data and expert knowledge.

[0072] 7. When the prior module issues a warning, the warning result is input into the fault diagnosis model.

[0073] 8. Use expert knowledge to verify the accuracy of the early warning results. If the verification passes, the early warning results are confirmed and the corresponding fault type and solution are output; otherwise, the early warning results are corrected or ignored.

[0074] In summary, this patent improves the intelligence level of hydropower equipment early warning by innovatively integrating the advantages of data-driven and knowledge-driven, and enhances the system's generalization ability and the reliability of early warning results.

[0075] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A hydropower intelligent early warning method driven by data knowledge fusion, characterized in that The following steps are involved: S1. Get data; S2, preprocessing the data; S3, using artificial intelligence modeling; S4. Real-time warning of equipment; S5. Collect data and expert knowledge; S6. Construct FTA fault tree logic model; S7. Conduct early warning analysis; S8. Verify the accuracy of the warning.

2. According to claim 1, a hydropower intelligent early warning method driven by data knowledge fusion is characterized by: In step S1, the acquired data includes sensor information and working condition data; and the time series data within a period of time is read; The sensor information includes measurement point location information and measurement attribute information; the operating condition data includes one or more data including active power, water head, guide vane opening, unit speed and tailwater pipe outlet temperature.

3. According to claim 1, a hydropower intelligent early warning method driven by data knowledge fusion is characterized by: In step S2, the data preprocessing process includes the following steps: S2.

1. Pivot data: convert the time series data format into a wide table mode. The columns in the wide table mode include time, sensor information, and working condition data. S2.

2. Perform null value processing: If the null value is a numeric value, fill in the missing attribute value based on the average value of the attribute in all other objects; if the null value is a non-numeric value, fill in the missing attribute value with the value with the largest number of times of the attribute in all other objects according to the principle of majority in statistics; S2.3, perform frequency reduction processing: reduce the frequency of all data from second-level data to 1-minute-level data, and the frequency reduction method is average processing, that is, take the average of the data in the time window of each minute; S2.4, steady-state processing: obtain the active power fluctuation range threshold of the input unit, and judge when the unit is in stable operation based on this range; S2.5, Out-range data processing: Out-range refers to the operating range of the device under test. The operating range information of the input measuring point is obtained and data is screened based on the information.

4. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized by: S3. The steps of using artificial intelligence modeling are: S3.

1. Automatically extract features and target points: The extraction methods used include target point generation and feature point dimensionality reduction; S3.2, algorithm selection and training: Automatically select and train the FTA fault tree logic model based on the characteristics and target measurement points provided by S3.

1.

5. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 4 is characterized by: S4. Perform real-time equipment warning: deploy the FTA fault tree logic model obtained in step S3 in the industrial Internet platform, set a scheduling cycle, and run the FTA fault tree logic model periodically. Each time, obtain data for a cycle and determine whether the data is a warning.

6. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized by: S5. The steps for collecting data and expert knowledge are: S5.

1. Collect historical data: Collect historical operation data of the hydropower station, including the temperature range of the upper guide shoe, abnormal temperature records and corresponding fault events; S5.

2. Obtain expert knowledge: Obtain professional knowledge on the possible causes of abnormal upper conductor temperature, fault types and their solutions by interviewing hydropower station operation and maintenance experts and consulting technical documents and manuals.

7. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized by: The method of step S6 is: S6.

1. Determine the top event: According to the results of accident investigation and statistical analysis, refer to the frequency of accidents and the severity of accident losses, confirm the top event and logic gate; S6.

2. Determine intermediate events: According to the analyzed event causes, create intermediate process events as intermediate event fault tree nodes in sequence; S6.

3. Determine basic events: locate the basic event nodes of the fault tree by the basic events that cause the intermediate accidents; S6.

4. Check the accident tree: Check whether the prepared accident tree conforms to the principles of logical analysis, check the connection status of the logic gates, check whether the upper-level events are the inevitable results of the lower-level events, whether the lower-level events are the sufficient cause events of the upper-level events, and check whether all direct cause events are found; S6.5, Binding node devices and fault diagnosis algorithms: Based on the collected data and expert knowledge, associate the fault tree nodes with the logical devices where their fault modes are located, and associate the hydropower-specific algorithms according to the node logical devices, so as to call the algorithm engine to diagnose the node occurrence probability and fault occurrence during the subsequent FTA fault diagnosis; S6.

6. Save the FTA fault tree logic model: Confirm that it is correct and save the entire static tree. The fault tree drawing is now completed; S6.

7. Release the FTA fault tree logic model and instantiate the FTA fault tree logic model: First, publish the FTA fault tree logic model to make the newly created FTA fault tree logic model effective. Only the effective FTA fault tree logic model can be instantiated. Then, perform the instantiation operation, select the plant station corresponding to the FTA fault tree logic model to be instantiated and the instantiated equipment tree, and enter the FTA fault tree logic model instantiation process; During the instantiation process, the logical devices bound to the nodes of the FTA fault tree logic model are converted into physical devices of the plant instantiation device tree, and the logical fault diagnosis tools bound to the nodes are converted into instantiation tools; S6.8, configure the scheduling task: S6.9, fault alarm: perform periodic fault diagnosis according to the scheduling task configured in step S6.7, and generate a fault diagnosis result instance diagram, i.e., a complete FTA fault tree; the FTA fault tree identifies the fault condition and occurrence probability of each event node.

8. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized by: The method of step S6.8 is: a) Create a scheduled task to periodically call the instantiated FTA fault tree logic model for fault diagnosis; b) Diagnosis process: When the event node is configured with a diagnostic tool, the diagnostic tool configured on the node is called in sequence to obtain the node's alarm status and occurrence probability, and the occurrence probability is substituted into the FTA fault tree logic model.

9. The hydropower intelligent early warning method driven by data knowledge fusion according to claim 1 is characterized by: The steps of step S8 are as follows: S8.

1. Expert evaluation: When the prior module issues an early warning, the FTA fault tree logic model is used to perform fault diagnosis and generate possible fault types and solutions; at the same time, experts are asked to evaluate the early warning results based on their knowledge and experience; S8.

2. Comparison of results: Compare the diagnostic results of the FTA fault tree logic model with the expert's evaluation results; if the two are consistent or highly similar, the accuracy of the early warning results is confirmed; otherwise, the early warning results need to be further reviewed and adjusted; S8.3 Feedback Mechanism: Establish a feedback mechanism to feed the verification results back to the a priori early warning system so that it can adjust and optimize itself according to the actual situation; S8.4, Intelligent Fault Diagnosis and Expert Evaluation: After the a priori module issues a warning signal, an advanced fault tree analysis model is used for intelligent fault diagnosis. The FTA fault tree logic model can automatically identify potential fault modes and generate a series of possible fault types and their solutions. At the same time, experts are used to conduct in-depth evaluation of the warning results to ensure the accuracy and comprehensiveness of the diagnosis. S8.

5. Comparison of intelligent diagnosis and expert evaluation results: Compare and analyze the intelligent diagnosis results of the FTA fault tree logic model with the expert evaluation results; if the two are highly consistent, it means that the accuracy of the early warning results has been verified; if there is a difference, the early warning results need to be carefully reviewed and necessary adjustments are made to ensure the reliability of the final results; S8.6, Closed-loop feedback and optimization mechanism: Establish a closed-loop feedback mechanism to feed back the verification results to the prior warning system; the system will self-learn and optimize based on the feedback information to improve the accuracy and efficiency of the warning; at the same time, this mechanism also helps to build a continuously improving intelligent warning system so that it can better adapt to the changing environment and needs; Once the accuracy of the early warning results is verified, the FTA fault tree logic model should be able to output specific fault types and corresponding solutions.

10. A hydropower intelligent early warning system driven by data knowledge fusion, characterized by: A hydropower intelligent early warning method driven by data knowledge fusion according to any one of claims 1-9 is adopted.

Images