Thermal power generating unit intelligent monitoring method and system based on data driving and expert experience cooperation

Abstract

The invention discloses a thermal power generating unit intelligent monitoring method and system based on data driving and expert experience collaboration, and the method comprises the steps: selecting a feature parameter group, extracting historical data in the feature parameter group, and carrying out the data management; based on historical data of the characteristic parameters, an early warning model based on probability statistics, a dynamic threshold early warning model based on similarity modeling and an early warning model based on artificial intelligence are constructed respectively, and a multi-algorithm fusion early warning model is jointly formed; based on expert experience, constructing a thermal power generating unit operation abnormity rule base, and based on the thermal power generating unit operation abnormity rule base, constructing an operation abnormity diagnosis model; the data acquisition module is used for acquiring production process data and monitoring data quality; inputting the collected real-time operation data into different early warning models, calculating the health degree of each subsystem in real time by using the different early warning models, and selecting the minimum value of the results of the different early warning models as the health degree of the subsystem for each subsystem; and judging whether the operation health degree of the monitored object meets requirements or not.

Description

Technical Field

[0001] The present invention relates to a method and system for intelligent monitoring of thermal power units, and in particular to a method and system for intelligent monitoring of thermal power units based on data drive and expert experience collaboration. Background Art

[0002] As thermal power units develop towards high parameters, large capacity, and high automation, the possibility of unit failure and the degree of harm caused by failure will further increase. However, the existing DCS fixed value alarm system of thermal power plants is biased towards equipment protection, and the alarm setting does not change with the working conditions, resulting in high false alarm and missed alarm rates. In addition, when an abnormality or failure occurs, there are hundreds of DCS alarm parameters, and the existing alarm system fails to filter alarms, covering a large number of invalid alarm parameters. Therefore, it is urgent to strengthen the monitoring and diagnosis of unit abnormalities, increase the construction of intelligent thermal power, and realize the intelligent monitoring and fault diagnosis of thermal power units to support the wide range and high-speed regulation of thermal power units, improve the quality, flexibility, and operational reliability of thermal power auxiliary peak regulation response under the new power system, and ensure the power grid security of the new power system. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method and system for intelligent monitoring of thermal power units based on data-driven and expert experience collaboration, so as to improve the accuracy of alarms and reduce missed alarms and false alarms.

[0004] Technical solution: The present invention comprises the following steps:

[0005] Select a feature parameter group and extract historical data of the feature parameter group for data governance;

[0006] Construct a multi-algorithm fusion warning model: Based on the historical data of feature parameters, construct a warning model based on probability statistics, a dynamic threshold warning model based on similarity modeling, and a warning model based on artificial intelligence, which together constitute a multi-algorithm fusion warning model;

[0007] Constructing an abnormal operation diagnosis model: Based on expert experience, construct an abnormal operation rule base for thermal power units, and based on the abnormal operation rule base for thermal power units, construct an abnormal operation diagnosis model;

[0008] Data collection: collect production process data and monitor data quality;

[0009] Input the collected data into different early warning models and calculate the health of each subsystem: Input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem in real time, and select the minimum value among the results of different early warning models for each subsystem as the health of the subsystem;

[0010] Determine whether the operating health of the monitored object meets the requirements.

[0011] The multi-algorithm fusion early warning model is used to calculate the health of each subsystem of the thermal power unit.

[0012] The health of each subsystem automatically sets the alarm threshold according to the health result of the test set, the first-level alarm threshold L1 = μ-3σ, the second-level alarm threshold L2 = μ-2σ, and the third-level alarm threshold L3 = μ-σ, where μ is the mean health of the test set and σ is the standard deviation of the health of the test set.

[0013] The test set uses mean absolute error, mean square error, and root mean square error evaluation indicators to evaluate the accuracy of the multi-algorithm fusion warning model.

[0014] The abnormal operation rule base of the thermal power unit is constructed by adopting the knowledge representation method of the production system.

[0015] The operation abnormality diagnosis model includes abnormality type, severity level, diagnosis rules, fault cause, and operation guidance.

[0016] The determining whether the operational health of the monitored objects meets the requirements is specifically: determining whether the operational health of all monitored objects meets the normal operating requirements; if not, issuing an abnormal alarm and providing an alarm level according to the health result.

[0017] After the abnormal alarm is issued, the operation abnormality diagnosis model is called to automatically diagnose the cause of the unit abnormality and provide operating instructions.

[0018] The data governance includes removing default values, data during outages, data during abnormal operation periods, and over-limit data to generate a training data set.

[0019] An intelligent monitoring system for thermal power units based on data-driven and expert experience collaboration, including:

[0020] Feature parameter group selection and data management module: extract historical data of feature parameter groups and perform data management;

[0021] Multi-algorithm fusion early warning model construction module: Based on the historical data of characteristic parameters, a multi-algorithm fusion early warning model is constructed to calculate the health of each subsystem of the thermal power unit;

[0022] Operation abnormality diagnosis model construction module: Based on expert experience, build an operation abnormality rule base for thermal power units, and based on the operation abnormality rule base for thermal power units, build an operation abnormality diagnosis model;

[0023] Data acquisition module: collects production process data and monitors data quality;

[0024] Health calculation module: input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem in real time, and select the minimum value of the results of different early warning models for each subsystem as the health of the subsystem;

[0025] Health judgment module: judge whether the operating health of all monitored objects meets the normal operating requirements. If not, an abnormal alarm is issued and the alarm level is given according to the health result. After the abnormal alarm is issued, the operation abnormality diagnosis model is called to automatically diagnose the cause of the unit abnormality and give operational guidance.

[0026] Beneficial effects: The intelligent monitoring method and system for thermal power units based on data drive and expert experience collaboration of the present invention can improve the monitoring efficiency. Compared with the traditional DCS alarm, it improves the alarm accuracy and reduces missed alarms and false alarms. Based on massive operating data, artificial intelligence technology and expert experience are used to realize unit operating status early warning and abnormal diagnosis and analysis, thereby reducing the workload of operating personnel while improving the reliability of power plant operation and reducing the probability of forced load reduction and non-stoppage due to faults. It can not only ensure the stability of power supply of the power grid, but also reduce the economic losses of the plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a diagnostic model for abnormal increase in reheat steam temperature according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described below in conjunction with the accompanying drawings.

[0029] Example 1

[0030] The intelligent monitoring method for thermal power units based on data drive and expert experience collaboration in this embodiment provides intelligent early warning, fault diagnosis and operation guidance based on real-time monitoring of the operating parameters of the thermal power units, helps operating personnel to quickly respond to and handle faults, and improves the operating efficiency and stability of the thermal power units. Specifically, it includes the following steps:

[0031] Step 1: Select characteristic parameter groups: For a specific subsystem or device, select characteristic parameters and determine the characteristic parameter group that describes the system operation characteristics based on correlation analysis, PCA, etc.

[0032] Step 2: Data governance: Extract historical data of the feature parameter group and perform data governance, including removing default values, data during outages, data during abnormal operation periods, and over-limit data, to generate a training data set.

[0033] Step 3: Construct a multi-algorithm fusion warning model: Based on the historical data of characteristic parameters, probability statistics, similarity modeling, artificial intelligence and other methods can be used to construct a warning model based on probability statistics, a dynamic threshold warning model based on similarity modeling, and a warning model based on artificial intelligence, which together constitute a multi-algorithm fusion warning model. The model can be used to calculate the health of each subsystem (equipment) of the thermal power unit. According to the health results of the test set, the alarm threshold is automatically set, the first-level alarm threshold L1 = μ-3σ, the second-level alarm threshold L2 = μ-2σ, and the third-level alarm threshold L3 = μ-σ, where μ is the mean health of the test set and σ is the standard deviation of the health of the test set.

[0034] Step 4: Construct an operation abnormality diagnosis model: Based on expert experience, the knowledge representation method of the production system is used to construct an operation abnormality rule base for thermal power units. Based on the operation abnormality rule base for thermal power units, an operation abnormality diagnosis model is constructed. The model includes abnormality type, severity level, diagnosis rules, fault cause, operation guidance and other contents.

[0035] Step 5: Data collection: Collect production process data from power plant SIS or DCS systems, monitor data quality, and monitor and alarm for dead numbers, jump numbers, etc.

[0036] Step 6. Input the collected data into different early warning models and calculate the health of each subsystem: Input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem (equipment) in real time, and select the minimum value of the results of different early warning models for each subsystem (equipment) as the health of the subsystem (equipment), that is, P = min (P1, P2, ..., Pn). Among them, P is the health of the subsystem (equipment), and Pi (i = 1, 2, ..., n) is the health of the subsystem (equipment) calculated using different models.

[0037] Step 7. Determine whether the operating health of the monitored objects meets the requirements: Determine whether the operating health of all monitored objects meets the normal operating requirements. If not, issue an abnormal alarm and give an alarm level based on the health results. After issuing an abnormal alarm, call the operation abnormality diagnosis model to automatically diagnose the cause of the unit abnormality and give operational guidance.

[0038] The present invention will be explained in detail below by taking the reheat steam system of unit #1 as an example in combination with the accompanying drawings and embodiments.

[0039] The intelligent monitoring method for thermal power units based on data drive and expert experience collaboration in this embodiment includes the following steps:

[0040] S1. For the reheat steam system of unit #1, select the characteristic parameter group.

[0041] The parameters associated with the reheat steam system of unit #1 include the total power generation load of unit #1, the total coal feed of unit #1, the cold reheat steam pressure at the inlet of reheater, the cold reheat steam temperature at the inlet of reheater, the desuperheating water flow of reheater, the spray water temperature of reheater desuperheater, the opening of reheater desuperheating water regulating valve, the inlet temperature A-1 of the first reheat inlet header, the inlet temperature A-2 of the first reheat inlet header, the steam pressure of the first reheat inlet header, the inlet temperature B-1 of the first reheat inlet header, the inlet temperature B-2 of the first reheat inlet header, the outlet steam temperature A-1 of the first reheat outlet header, the outlet steam temperature A-2 of the first reheat outlet header, the outlet steam temperature B-1 of the first reheat outlet header, the outlet steam temperature B-2 of the first reheat outlet header, the flue gas temperature 1 of the reheater, the flue gas temperature 2 of the reheater, the low re-outlet flue gas temperature 1 of #1 furnace, the low re-outlet flue gas temperature 2 of #1 furnace.

[0042] The Pearson correlation coefficients between the above parameters are calculated, and the parameters with a Pearson correlation coefficient of more than 0.8 are selected to form a characteristic parameter group. The parameters of the reheat steam system characteristic parameter group obtained by calculation include the reheater inlet cold reheat steam temperature, the first-stage reheat inlet header inlet temperature A-1, the first-stage reheat inlet header inlet temperature B-1, the first-stage reheat outlet header outlet steam temperature A-1, the first-stage reheat outlet header outlet steam temperature B-1, the total power generation load of the #1 unit, and the total coal supply of the #1 unit.

[0043] S2. Import the historical data of the feature parameter group for the past year as the original data set, and perform data governance on the original data set, including removing default values, data during shutdown periods, data during abnormal operation periods, over-limit data, etc., to generate a training data set.

[0044] S3. Based on the historical data of characteristic parameters, similarity modeling, artificial intelligence and other methods are used to build a dynamic threshold warning model based on similarity modeling and a warning model based on artificial intelligence, which together constitute a multi-algorithm fusion warning model for calculating the health of each subsystem (equipment) of the thermal power unit. According to the health results of the test set, the alarm threshold is automatically set. Specifically, it includes:

[0045] S31. Divide the data set by using the holdout method and selecting a ratio of 70%:30%. Directly divide the data set into two mutually exclusive parts, of which 70% is used as the training set and 30% is used as the test set.

[0046] S32. Perform data preprocessing on the training set and the test set. Use normalization methods on the training set and the test set respectively to transform the data to the [0,1] interval to generate new training set S and test set T.

[0047] S33. Based on the data of training set S, train the early warning model:

[0048] A nonlinear multivariate state estimation algorithm is used to construct a static historical memory matrix of the reheat steam system of unit #1 based on the data of the training set S. The observation vector is selected, the Euclidean distance between the observation vector and the static historical memory matrix is ​​calculated, and n historical observation vectors with a smaller Euclidean distance to the observation vector are selected to form a dynamic matrix. A weight vector is constructed so that the n historical observation vectors have the shortest Euclidean distance to the observation vector under the linear combination of the weight vector. This completes the construction of the similarity model M1 of the reheat steam system of unit #1. The inverse of the Euclidean distance at this time is the health P1 of the reheat steam system of unit #1 calculated by the nonlinear multivariate state estimation algorithm.

[0049] The BP neural network algorithm is adopted, and the total coal feed of unit #1 is selected as the input parameters, and the cold reheat steam temperature at the reheater inlet, the first-stage reheat inlet header inlet temperature A-1, the first-stage reheat inlet header inlet temperature B-1, the first-stage reheat outlet header outlet steam temperature A-1, the first-stage reheat outlet header outlet steam temperature B-1, and the total power generation load of unit #1 are selected as the output parameters. Based on the data of the training set S, a neural network model M2 of the reheat steam system of unit #1 is constructed, and a set of input observation vectors I1 and the corresponding output observation vectors O1 are selected. The Euclidean distance between the output vector O2 calculated by the neural network model for the input observation vector I1 and the output observation vector O1 is calculated. The reciprocal of the Euclidean distance is the health P2 of the reheat steam system of unit #1 calculated by the BP neural network algorithm.

[0050] S34. Based on the test set T, use the mean absolute error, mean square error, root mean square error and other evaluation indicators to evaluate the accuracy of the similarity model M1 and the neural network model M2 respectively. If the model accuracy does not meet the requirements, retrain the model and iterate the original model until the model accuracy meets the requirements.

[0051] S35. According to the health results of the test set, the mean μ1 and μ2 of the health obtained by the similarity model and the health obtained by the neural network model, as well as the standard deviation σ1 and σ2 are calculated respectively. According to the first-level alarm threshold L1=μ-3σ, the second-level alarm threshold L2=μ-2σ, and the third-level alarm threshold L3=μ-σ, the alarm thresholds of the similarity model M1 and the neural network model M2 are automatically set respectively.

[0052] S4. Based on expert experience, a diagnostic model for all abnormal operation of the reheat steam system is constructed. Taking the abnormal increase of reheat steam temperature as an example, the phenomenon of abnormal increase of reheat steam temperature is first sorted out into logical rules, and the fault causes and operation instructions are sorted out, as shown in Table 1 below. Then, according to the diagnostic rules, a diagnostic model for abnormal increase of reheat steam temperature is constructed using graphical configuration, such as Figure 1 shown.

[0053] Table 1 Fault Causes and Operation Guidelines for Abnormal Increase in Reheat Steam Temperature

[0054]

[0055]

[0056] S5. Collect the production process data of the power plant SIS system, normalize the collected real-time operation data, and input it into the similarity model M1 and neural network model M2 of the reheat steam system of Unit #1. Calculate the health degrees P1 and P2 of the reheat steam system of Unit #1 using models M1 and M2 respectively, and take the minimum value as the health degree of the reheat steam system of Unit #1, that is, P = min(P1, P2). For example, if P1 < P2, then P = P1, and the alarm threshold is taken as the calculation result of model M1, that is, the mean value μ = μ1 and standard deviation σ = σ1 are used to calculate the alarm threshold. The first-level alarm threshold L1 = μ - 3σ, the second-level alarm threshold L2 = μ - 2σ, and the third-level alarm threshold L3 = μ - σ.

[0057] S6. Judge whether the health degree P of the reheat steam system of Unit #1 meets the normal operation requirements, and give the alarm level according to the health degree. For example: when P < L1, issue a first-level alarm, and call all the operation abnormality diagnosis models of the reheat steam system. According to the calculation results of the operation abnormality diagnosis models, give the abnormality type, severity level, fault cause, and operation guidelines. For example, if the main steam flow monitored in real time is greater than the set value, the deviation between the reheat steam temperature and the standard reheat steam temperature is greater than the threshold and lasts for 5 minutes, then an alarm for abnormal increase in reheat steam temperature is issued, and the corresponding fault cause and operation guidelines are given.

[0058] Embodiment 2

[0059] The intelligent monitoring system for thermal power units based on the collaboration of data-driven and expert experience in this embodiment includes:

[0060] Characteristic parameter group selection module: For a specific subsystem or device, select characteristic parameters, and determine the characteristic parameter group describing the system operation characteristics according to correlation analysis, PCA, etc.

[0061] Data governance module: Extract the historical data of the characteristic parameter group and perform data governance.

[0062] Multi-algorithm fusion early warning model construction module: Based on the historical data of characteristic parameters, probability statistics, similarity modeling, artificial intelligence and other methods can be used to respectively construct a warning model based on probability statistics, a dynamic threshold warning model based on similarity modeling, and a warning model based on artificial intelligence, which together constitute a multi-algorithm fusion early warning model. The model can be used to calculate the health of each subsystem (equipment) of the thermal power unit. According to the health results of the test set, the alarm threshold is automatically set, the first-level alarm threshold L1 = μ-3σ, the second-level alarm threshold L2 = μ-2σ, and the third-level alarm threshold L3 = μ-σ, where μ is the mean health of the test set and σ is the standard deviation of the health of the test set.

[0063] Operation abnormality diagnosis model construction module: Based on expert experience, the knowledge representation method of the production system is used to build the operation abnormality rule base of the thermal power unit. Based on the operation abnormality rule base of the thermal power unit, the operation abnormality diagnosis model is constructed. The model includes abnormality type, severity level, diagnosis rules, fault cause, operation guidance and other contents.

[0064] Data acquisition module: collects production process data of power plant SIS or DCS systems, monitors data quality, and monitors and alarms for dead counts, jump counts, etc.

[0065] Health calculation module: input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem (equipment) in real time, and select the minimum value of the results of different early warning models for each subsystem (equipment) as the health of the subsystem (equipment), that is, P = min (P1, P2, ..., Pn). Among them, P is the health of the subsystem (equipment), and Pi (i = 1, 2, ..., n) is the health of the subsystem (equipment) calculated using different models.

[0066] Health judgment module: judge whether the operating health of all monitored objects meets the normal operating requirements. If not, an abnormal alarm is issued and the alarm level is given according to the health result. After the abnormal alarm is issued, the operation abnormality diagnosis model is called to automatically diagnose the cause of the unit abnormality and give operational guidance.

Claims

1. A method for intelligent monitoring of thermal power units based on data-driven and expert experience collaboration, characterized in that: The following steps are involved: Construct a multi-algorithm fusion early warning model: select feature parameter groups, extract historical data of feature parameter groups, and perform data governance; Construct a multi-algorithm fusion early warning model based on historical data of characteristic parameters; Constructing an abnormal operation diagnosis model: Based on expert experience, construct an abnormal operation rule base for thermal power units, and based on the abnormal operation rule base for thermal power units, construct an abnormal operation diagnosis model; Data collection: collect production process data and monitor data quality; Input the collected data into different early warning models and calculate the health of each subsystem: Input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem in real time, and select the minimum value among the results of different early warning models for each subsystem as the health of the subsystem; Determine whether the operating health of the monitored object meets the requirements.

2. According to claim 1, a method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration is characterized in that: The multi-algorithm fusion early warning model is used to calculate the health of each subsystem of the thermal power unit.

3. According to claim 2, a method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration is characterized in that: The health of each subsystem automatically sets the alarm threshold according to the health result of the test set, the first-level alarm threshold L1 = μ-3σ, the second-level alarm threshold L2 = μ-2σ, and the third-level alarm threshold L3 = μ-σ, where μ is the mean health of the test set and σ is the standard deviation of the health of the test set.

4. According to claim 3, a method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration is characterized in that: The test set uses mean absolute error, mean square error, and root mean square error evaluation indicators to evaluate the accuracy of the multi-algorithm fusion warning model.

5. According to claim 1, a method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration is characterized in that: The abnormal operation rule base of the thermal power unit is constructed by adopting the knowledge representation method of the production system.

6. The method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration according to claim 1 or 5, characterized in that: The operation abnormality diagnosis model includes abnormality type, severity level, diagnosis rules, fault cause, and operation guidance.

7. The method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration according to claim 1 is characterized in that: The determining whether the operational health of the monitored objects meets the requirements is specifically: determining whether the operational health of all monitored objects meets the normal operating requirements; if not, issuing an abnormal alarm and providing an alarm level according to the health result.

8. The method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration according to claim 7 is characterized in that: After the abnormal alarm is issued, the operation abnormality diagnosis model is called to automatically diagnose the cause of the unit abnormality and provide operating instructions.

9. The method for intelligent monitoring of thermal power units based on data drive and expert experience collaboration according to claim 1 is characterized in that: The data governance includes removing default values, data during outages, data during abnormal operation periods, and over-limit data to generate a training data set.

10. An intelligent monitoring system for thermal power units based on data drive and expert experience collaboration, characterized in that: include: Feature parameter group selection and data management module: extract historical data of feature parameter groups and perform data management; Multi-algorithm fusion early warning model construction module: Based on the historical data of characteristic parameters, a multi-algorithm fusion early warning model is constructed to calculate the health of each subsystem of the thermal power unit; Operation abnormality diagnosis model construction module: Based on expert experience, build an operation abnormality rule base for thermal power units, and based on the operation abnormality rule base for thermal power units, build an operation abnormality diagnosis model; Data acquisition module: collects production process data and monitors data quality; Health calculation module: input the collected real-time operation data into different early warning models, use different early warning models to calculate the health of each subsystem in real time, and select the minimum value of the results of different early warning models for each subsystem as the health of the subsystem; Health judgment module: judge whether the operating health of all monitored objects meets the normal operating requirements. If not, an abnormal alarm is issued and the alarm level is given according to the health result. After the abnormal alarm is issued, the operation abnormality diagnosis model is called to automatically diagnose the cause of the unit abnormality and give operational guidance.

Images