Intelligent operation method based on automatic generation function of hydropower station unit operation report

Abstract

The intelligent operation method based on the automatic generation function of the hydropower station unit operation report includes determining the template for the report to be generated and its corresponding data feature items, extracting the feature data items from the monitoring database, cleaning the extracted feature database based on the neural network algorithm, mining and analyzing the feature sequence data based on the deep learning model to obtain its internal change rules and connection characteristics, structuring and analyzing the mining results, automatically generating the unit operation report, and finally manually reviewing, finalizing, and outputting the report. The present invention is more intelligent and precise, greatly improving work efficiency. By directly calling the database method, it overcomes the shortcomings of traditional manual data input and comprehensive analysis, liberates manpower, improves efficiency, provides an objective basis for the health status evaluation and fault diagnosis of power station equipment, and greatly improves the operation and maintenance efficiency of the power station.

Description

Technical Field

[0001] The present invention belongs to the field of large-scale hydropower station operation methods, and in particular relates to an intelligent operation method based on the automatic generation function of hydropower station unit operation reports. Background Art

[0002] In today's interconnected energy landscape, the safe and stable operation of hydropower plant equipment plays an increasingly crucial role in power system interconnection and energy security. Power plant equipment reports are a crucial component of a power plant's monitoring system. They clearly display the operating parameters and trends of key components such as the plant's 500kV transmission system, transformers, reactors, hydro-generator units and their auxiliary equipment, technical water supply system, speed control system, and excitation system. This allows for an assessment of the health of power plant equipment and the identification of potential faults.

[0003] Currently, most power plant reports require a significant amount of time and effort to collect, organize, and analyze data before reaching conclusions. With the rapid development of power systems, this approach is no longer sufficient. This increases labor and time costs. The complexity and diversity of power plant equipment make comprehensive analysis impossible. Furthermore, the resulting reports are easily influenced by objective human factors, resulting in inaccurate results and a delay in identifying potential equipment failures, which in turn affects the health of the equipment. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent operation method based on the automatic generation function of hydropower station unit operation reports, which is more intelligent and precise, greatly improving work efficiency. By directly calling the database method, it overcomes the shortcomings of traditional manual data input and comprehensive analysis, liberates manpower, improves efficiency, and provides an objective basis for the health status evaluation and fault diagnosis of power station equipment, greatly improving the operation and maintenance efficiency of the power station.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] The intelligent operation method based on the automatic generation function of the hydropower station unit operation report includes the following steps:

[0007] S1. Determine the template for the report to be generated and its corresponding data feature items;

[0008] S2, extracting feature data items from the monitoring database;

[0009] S3, cleaning the extracted feature database based on the neural network algorithm;

[0010] S4. Mining and analyzing feature sequence data based on deep learning models to obtain its internal change patterns and connection characteristics;

[0011] S5. Structural processing and analysis of the mining results;

[0012] S6. Automatically generate unit operation report;

[0013] S7. Perform manual review, revision and finalization, and then output the report.

[0014] Preferably, in step S1, the report template and its corresponding data feature items refer to: all relevant quantities of a certain functional module within a specific time period (feature quantities of the previous day, feature quantities of the previous week, feature quantities of the first half of the month, etc.), and all relevant quantities refer to all status data information.

[0015] Preferably, in step S2, the process of extracting characteristic data items from the monitoring database is as follows: uploading the status data information (video, audio, switch quantity, analog quantity, status quantity, etc.) collected by the hydropower station equipment monitoring system to a unified upper-end LCU to form a complete database of the equipment. The database is updated in real time, retaining all current and previous information, and automatically extracting relevant status data information based on the template of the report to be generated in step S1 and its corresponding data characteristic items.

[0016] Preferably, in step S3, the neural network algorithm is a gated recurrent unit network model algorithm based on the LSTM model, and the input gate and the forget gate in the LSTM model are combined into the update gate in the GRU model;

[0017] Update Gate z t Determines how much information from the previous moment is retained at the current moment. A larger value means more information from the previous moment is retained, and vice versa.

[0018] r t Controls ignoring the information of the previous moment. The larger the value, the less information of the previous moment is retained.

[0019] The input of the GRU model is x t , combined with z t and r t Get the output h t The formulas are shown in (1)-(4):

[0020] (1);

[0021] (2);

[0022] (3);

[0023] (4);

[0024] Where, Represents input x t With the previous output h t-1 The combination of W (z) 、 U (z) 、 W (r) 、 U (r) 、 U and W represents the training parameter matrix, Represents the element composite relationship, σ and tanh are the hyperbolic tangent functions of Sigmoid and tanh respectively.

[0025] Preferably, the feature database cleaning process using the GRU model is as follows:

[0026] 1) Select feature sequence data as input to the GRU cleaning model;

[0027] 2) Perform maximum and minimum normalization on the collected data. The implementation process is shown in formula (5), and the data is divided into training set data and test set data according to the ratio of 2:1; X p 、 X 、 X max 、 X min Respectively represent the normalized data, arbitrary value of the feature sequence, maximum value, and minimum value; formula (5) is as follows:

[0028] (5);

[0029] 3) Train and learn the training set data, understand its internal change rules, and calculate the test set data:

[0030] |Current data value - previous data set| <Q?;

[0031] Q represents the threshold value set specifically to determine whether the data is a normal value, a missing value, or an abnormal value. This model selects this threshold as 0.1;

[0032] If the calculated value is less than Q, it indicates that the value is normal, otherwise it is a missing value or an abnormal value;

[0033] 4) Set the parameters of the GRU cleaning model, reconstruct and repair the missing values ​​and outlier data in the test set, perform denormalization, and finally analyze the model based on the evaluation indicators; the evaluation indicators are as follows:

[0034] (6);

[0035] Preferably, in step S4, the process of mining and analyzing the feature sequence data based on the deep learning model to obtain its internal change rules and connection characteristics is: using a recurrent neural network to learn and mine the feature sequence cleaned in step S3, and fully grasp its internal change rules and related connections.

[0036] Preferably, the recurrent neural network consists of an input layer, a hidden layer and an output layer. t The input sequence of the RNN model at this moment is x =[ x 1, x 2, … x n ], according to formula (7), we can get the hidden layer sequence at this time v =( v 1, v 2, … v n ) and the output layer sequence y =( y 1, y 2, … y n ), by analyzing the potential connection between the hidden layer sequence and the output layer sequence, the change law between the input feature vectors can be obtained, formula (7) is as follows:

[0037] (7);

[0038] Where, v t-1 for t The state of the hidden layer at time -1; D 、 E are the weight coefficient matrices between the input layer to the hidden layer, and between the hidden layer and the output layer respectively; F is the self-feedback weight coefficient matrix; b Indicates the offset; f is the activation function.

[0039] Preferably, in step S5, the process of structuring and analyzing the mining results is as follows: running the natural language processing method to realize natural language communication between humans and computers, structuring the results obtained in step S4, completing error correction, classification, understanding and grammatical analysis of natural language, and expressing them in a standardized formal language in a computer.

[0040] Preferably, in step S6, the automatic generation process of the unit operation report is: the text, table, picture and status data processed by natural language in step S5 are automatically generated according to the template format requirements of the report in step S1.

[0041] Preferably, in step S7, the process of manual review, correction and finalization, and then outputting the report is: manually reviewing and proofreading the report content generated in step S7. If errors are found in the manually reviewed report content, steps S1 to S7 are repeated until the report content and conclusions are manually confirmed to be correct, and finally a complete structured report content that meets the requirements is output.

[0042] The present invention can achieve the following beneficial effects:

[0043] The present invention uses currently advanced neural network algorithms, deep learning models, and natural language processing methods as tools to solve the problems of feature data cleaning and rapid grasp of internal change rules, and realizes the automatic generation of hydropower station equipment operation report content, which is more intelligent and precise, greatly improving work efficiency. By directly calling the database method, it overcomes the shortcomings of traditional manual data input and comprehensive analysis, liberates manpower, improves efficiency, provides an objective basis for power station equipment health status evaluation and fault diagnosis, and greatly improves power station operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below with reference to the accompanying drawings and examples:

[0045] Figure 1 This is the LSTM network model diagram of the present invention;

[0046] Figure 2 This is the GRU network model diagram of the present invention;

[0047] Figure 3 This is the RNN topology diagram of the present invention;

[0048] Figure 4 This is a structural diagram of the convolutional neural network of the present invention;

[0049] Figure 5 Flowchart of the present invention;

[0050] Figure 6 This is a connection diagram of the diagnostic and analysis function modules of the device of the present invention;

[0051] Figure 7 This is a monthly diagnostic analysis report of the temperature and oil level of the three bearings of the unit of the present invention;

[0052] Figure 8 This is a monthly diagnostic analysis report chart of the temperature and oil level of the three bearings of the unit of the present invention. DETAILED DESCRIPTION

[0053] The preferred solution is Figures 1 to 8 As shown in the figure, the intelligent operation method based on the automatic generation function of the hydropower station unit operation report is

[0054] S1. Determine the template for the report to be generated and its corresponding data feature items; the specific process is as follows:

[0055] All relevant quantities of a functional module within a specific time period (feature quantity of the previous day, feature quantity of the previous week, feature quantity of the first half of the month, etc.) are fully reflected. All relevant quantities refer to all status data information.

[0056] For example, if you need to obtain the monthly analysis data table of the unit's generator temperature and pure water system, the corresponding data feature items are: the maximum, minimum, and average values ​​of the unit's operating head; the maximum, minimum, and average values ​​of the unit's active power; the maximum, minimum, and average values ​​of the unit's dead power; the maximum, minimum, and average values ​​of the rotor temperature; the maximum, minimum, and average values ​​of the excitation voltage; the maximum, minimum, and average values ​​of the excitation current; the maximum and minimum temperatures of the stator slots and their corresponding slot numbers, the maximum and minimum slot temperature differences; the maximum and minimum temperatures of the stator core and their corresponding The corresponding slot number, the maximum and minimum slot temperature difference; the maximum and average temperatures of the upper and lower tooth pressure plates of the stator core tooth pressure plate; the maximum, minimum and average temperatures of the water outlet of the stator winding; the average temperature of the cold air and hot air of the air cooler, the average water temperature of the inlet and outlet, the inlet and outlet water temperature of the cooling water; the maximum, minimum and average values ​​of the water level of the pure water expansion tank, the maximum, minimum and average values ​​of the pure water inlet temperature, the maximum, minimum and average values ​​of the pure water outlet temperature, the average values ​​of the pure water circuit pressure, pure water flow, secondary cooling water flow, main circuit conductivity, and ion exchanger outlet conductivity.

[0057] S2. Extract characteristic data items from the monitoring database. The specific process is as follows:

[0058] The status data (video, audio, switching, analog, and status information) collected by the hydropower station's equipment monitoring system is uploaded to a unified upper-level LCU, forming a complete equipment database. This database is updated in real time, retaining all current and previous information. Based on the report template and its corresponding data feature items generated in step S1, relevant status data is automatically extracted. The feature database is based on information collected by all hydropower station equipment monitoring systems and is rationally categorized and organized according to time tags, allowing for feature extraction as needed.

[0059] S3. Clean the extracted feature database based on the neural network algorithm. The specific process is as follows:

[0060] Since the data acquisition system, transmission system, signal processing system and other systems are affected by various factors, the monitored and collected data may contain "missing values" or "outliers". It is unreliable to perform equipment status analysis based on these abnormal data. Therefore, the neural network algorithm (gated recurrent unit network model, GRU) is used to train and learn the feature sequence. This can grasp the potential change rules of the sequence itself, realize the reconstruction and repair of missing values ​​and outliers, and restore the information of the original feature data sequence to the greatest extent, providing reliable data quality for judging the operating status of the equipment. The gated recurrent unit network model is an upgrade based on the long short-term memory network (LSTM) model, which solves the problems of complex LSTM parameter training and long training time. Its internal structure is as follows. Figure 1-2 As shown, the input gate and forget gate in the LSTM model are merged into the update gate in the GRU model. z t It determines how much information from the previous moment is retained at the current moment. The larger the value, the more information from the previous moment is retained, and vice versa. r t Controls ignoring the information of the previous moment. The larger the value, the less information of the previous moment is retained.

[0061] The input of the GRU model is x t , combined with z t and r t Get the output h t The formulas are shown in (1)-(4):

[0062] (1);

[0063] (2);

[0064] (3);

[0065] (4);

[0066] Where, Represents input x t With the previous output h t-1 The combination of W (z) 、 U (z) 、 W (r) 、 U (r) 、U and W represents the training parameter matrix, Represents the element composite relationship, σ and tanh are the hyperbolic tangent functions of Sigmoid and tanh respectively.

[0067] Preferably, the GRU model has the advantage of better processing time series feature data, which can fully explore the potential relationships between data sequences, automatically repair bad data in the feature data set, and restore the information of the original data sequence to the greatest extent, providing reliable data quality for judging the health status of the equipment. The specific steps are as follows:

[0068] The process of using the GRU model to clean the feature database is as follows:

[0069] 1) Select feature sequence data as input to the GRU cleaning model;

[0070] 2) Perform maximum and minimum normalization on the collected data. The implementation process is shown in formula (5), and the data is divided into training set data and test set data according to the ratio of 2:1; X p 、 X 、 X max 、 X min Respectively represent the normalized data, arbitrary value of the feature sequence, maximum value, and minimum value; formula (5) is as follows:

[0071] (5);

[0072] 3) Train and learn the training set data, understand its internal change rules, and calculate the test set data:

[0073] |Current data value - previous data set| <Q?;

[0074] Q represents the threshold value set specifically to determine whether the data is a normal value, a missing value, or an abnormal value. This model selects this threshold as 0.1;

[0075] If the calculated value is less than Q, it indicates that the value is normal, otherwise it is a missing value or an abnormal value;

[0076] 4) Set the parameters of the GRU cleaning model, reconstruct and repair the missing values ​​and outlier data in the test set, perform denormalization, and finally analyze the model based on the evaluation indicators; the evaluation indicators are as follows:

[0077] (6);

[0078] S4. Mining and analyzing feature sequence data based on a deep learning model to obtain its internal variation patterns and connection characteristics. The specific process is as follows:

[0079] The recurrent neural network is used to learn and mine the feature sequence after cleaning in step S3 to fully grasp its internal change rules and related connections.

[0080] Preferably, the recurrent neural network RNN ​​is a special neural network that can automatically analyze the potential relationship between data and grasp the internal change rules. The recurrent neural network consists of an input layer, a hidden layer and an output layer. t The input sequence of the RNN model at this moment is x =[ x 1, x 2, … x n ], according to formula (7), we can get the hidden layer sequence at this time v =( v 1, v 2, … v n ) and the output layer sequence y =( y 1, y 2, … y n ), by analyzing the potential connection between the hidden layer sequence and the output layer sequence, the change law between the input feature vectors can be obtained, formula (7) is as follows:

[0081] (7);

[0082] Where, v t-1 for t The state of the hidden layer at time -1; D 、 E are the weight coefficient matrices between the input layer to the hidden layer, and between the hidden layer and the output layer respectively; F is the self-feedback weight coefficient matrix; b Indicates the offset; f is the activation function.

[0083] S5. Structural processing and analysis of the mining results; the specific process is as follows:

[0084] Run the natural language processing method to realize natural language communication between humans and machines, perform structured processing on the results obtained in step S4, complete the error correction, classification, understanding and grammatical analysis of natural language, and express it in a standardized formal language in the computer. Convolutional neural network CNN is a feedforward neural network that uses local connections and shared weights to train and learn historical language features. It can master higher-level and more abstract language features without manually defining training sets, and automatically completes this learning process. Its structure consists of input layer, convolution layer, pooling layer, fully connected layer and output layer, as shown in the attached figure. Figure 4 shown.

[0085] S6. Automatically generate the unit operation report; the specific process is as follows:

[0086] The text, table, picture and status data processed by the natural language in step S5 are automatically generated according to the template format requirements reported in step S1.

[0087] S7, conduct manual review, revision and finalization, and then output the report. The specific process is as follows:

[0088] The report content generated in step S7 is manually reviewed and proofread. If errors are found in the manually reviewed report content, steps S1 to S7 are repeated until the report content and conclusions are manually confirmed to be correct. Finally, a complete and structured report content that meets the requirements is output.

[0089] The specific process of the present invention is as shown in the attached Figure 5 As shown, in this embodiment, the operation data of large hydropower station equipment is analyzed. The specific equipment diagnosis and analysis function module is shown in the attached Figure 6 As shown, limited to the length of this invention, the monthly diagnosis and analysis function module of the temperature and oil level of the three bearings of the unit is used as an example to introduce the final monthly diagnosis and analysis report of the temperature and oil level of the three bearings of the unit. Figure 7-8 shown.

[0090] The above embodiments are merely preferred technical solutions of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the technical solutions set forth in the claims, including equivalent alternatives to the technical features of the technical solutions set forth in the claims. In other words, equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. An intelligent operation method based on the automatic generation function of hydropower station unit operation reports is characterized by: S1. Determine the template for the report to be generated and its corresponding data feature items; S2, extracting feature data items from the monitoring database; S3, cleaning the extracted feature database based on the neural network algorithm; S4. Mining and analyzing feature sequence data based on deep learning models to obtain its internal change patterns and connection characteristics; S5. Structural processing and analysis of the mining results; S6. Automatically generate unit operation report; S7. Manually review, revise and finalize the report before outputting it; The process of using the GRU model to clean the feature database is as follows: 1) Select feature sequence data as input to the GRU cleaning model; 2) Perform maximum and minimum normalization on the collected data. The implementation process is shown in formula (5), and the data is divided into training set data and test set data according to the ratio of 2:1; X p 、 X 、 X max 、 X min Respectively represent the normalized data, arbitrary value of the feature sequence, maximum value, and minimum value; formula (5) is as follows: （5）； 3) Train and learn the training set data, understand its internal change rules, and calculate the test set data: |Current data value - previous data set| <Q?; Q represents the threshold value set specifically to determine whether the data is a normal value, a missing value, or an abnormal value; If the calculated value is less than Q, it indicates that the value is normal, otherwise it is a missing value or an abnormal value; 4) Set the parameters of the GRU cleaning model, reconstruct and repair the missing values ​​and outlier data in the test set, perform denormalization, and finally analyze the model based on the evaluation indicators; the evaluation indicators are as follows: （6）； In step S4, the feature sequence data is mined and analyzed based on the deep learning model to obtain its internal change patterns and connection characteristics. The process is as follows: the feature sequence cleaned in step S3 is learned and mined using a recurrent neural network to fully grasp its internal change patterns and related connections; The recurrent neural network consists of an input layer, a hidden layer and an output layer. t The input sequence of the RNN model at this moment is x =[ x 1, x 2, … x n ], according to formula (7), we can get the hidden layer sequence at this time v =( v 1, v 2, … v n ) and the output layer sequence y =( y 1, y 2, … y n ), by analyzing the potential connection between the hidden layer sequence and the output layer sequence, the change law between the input feature vectors can be obtained, formula (7) is as follows: （7）； Where, v t-1 for t The state of the hidden layer at time -1; D 、 E are the weight coefficient matrices between the input layer to the hidden layer, and between the hidden layer and the output layer respectively; F is the self-feedback weight coefficient matrix; b Indicates the offset; f is the activation function.

2. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 1 is characterized in that: In step S1 , the report template and its corresponding data feature items refer to: fully reflecting all relevant quantities of a certain functional module within a specific time period, and all relevant quantities refer to including all status data information.

3. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 2 is characterized in that: In step S2, the process of extracting characteristic data items from the monitoring database is as follows: uploading the status data information collected by the hydropower station equipment monitoring system to a unified upper-end LCU to form a complete database of the equipment. The database is updated in real time, retaining all current and previous information, and automatically extracting relevant status data information based on the template of the report to be generated in step S1 and its corresponding data characteristic items.

4. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 3 is characterized by: In step S3, the neural network algorithm is a gated recurrent unit network model algorithm based on the LSTM model, and the input gate and forget gate in the LSTM model are combined into the update gate in the GRU model; Update Gate z t Determines how much information from the previous moment is retained at the current moment. A larger value means more information from the previous moment is retained, and vice versa. r t Controls ignoring the information of the previous moment. The larger the value, the less information of the previous moment is retained. The input of the GRU model is x t , combined with z t and r t Get the output h t The formulas are shown in (1)-(4): （1）； （2）； （3）； （4）； Where, Represents input x t With the previous output h t-1 The combination of W (z) 、 U (z) 、 W (r) 、 U (r) 、 U and W represents the training parameter matrix, Represents the element composite relationship, σ and tanh are the hyperbolic tangent functions of Sigmoid and tanh respectively.

5. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 1 is characterized in that: In step S5, the process of structuring and analyzing the mining results is as follows: running the natural language processing method to realize natural language communication between humans and computers, structuring the results obtained in step S4, completing the error correction, classification, understanding and grammatical analysis of natural language, and expressing them in a standardized formal language in the computer.

6. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 1 is characterized in that: In step S6, the automatic generation process of the unit operation report is as follows: the text, table, picture and status data processed by natural language in step S5 are automatically generated according to the template format requirements of the report in step S1.

7. The intelligent operation method based on the automatic generation function of the hydropower station unit operation report according to claim 1 is characterized in that: In step S7, the process of manual review, correction and finalization, and then outputting the report is as follows: the report content generated in step S7 is manually reviewed and proofread. If errors are found in the manually reviewed report content, steps S1 to S7 are repeated until the report content and conclusions are manually confirmed to be correct, and finally a complete and structured report content that meets the requirements is output.

Images