Thermal power generating unit operation state monitoring method, system and device and storage medium

Abstract

The invention discloses a thermal power generating unit operation state monitoring method, system and device and a storage medium, and belongs to the technical field of thermal power generating unit state monitoring. Comprising the steps of obtaining operation state parameter data of a thermal power generating unit, and determining a target state parameter and a key state parameter of the thermal power generating unit in the operation state parameter data; inputting the key state parameter into a pre-trained thermal power prediction model to obtain a target state parameter prediction value; obtaining a target state parameter actual value of the thermal power generating unit, and calculating a DTW similarity score between the target state parameter actual value and the target state parameter predicted value; when the DTW similarity score is lower than a preset early warning threshold value, it is judged that the state of the thermal power generating unit is abnormal, and an abnormal alarm signal is output. According to the method, the target state parameter prediction value is obtained through the thermal power prediction model, the abnormal state judgment index is constructed by calculating the DTW similarity score, the abnormal operation state of the thermal power generating unit can be accurately detected, and the method is suitable for complex and changeable working conditions of the thermal power generating unit.

Description

Technical Field

[0001] This invention belongs to the field of thermal power unit condition monitoring technology, specifically relating to a method, system, device, and storage medium for monitoring the operating status of thermal power units. Background Technology

[0002] Thermal power is a key pillar of global electricity supply, accounting for approximately 60% to 65% of total global electricity generation. Due to its continuous, large-scale power generation capacity and the ability to adjust output according to demand, thermal power is crucial for providing baseload power and maintaining grid stability. Furthermore, with its mature technology and stable costs, thermal power remains an important component of the current energy sector. As the core energy conversion equipment in thermal power plants, thermal power units directly impact economic performance and environmental efficiency. Therefore, monitoring the condition of thermal power units is essential.

[0003] Chinese patent application CN118730499A discloses a method for monitoring the status of a thermal power unit's regulating device based on a recurrent neural network. This method, based on real-time operating data of the thermal power unit, uses a recurrent neural network model to model and monitor the dynamic response characteristics of the regulating device after a series of preprocessing steps. By combining the recurrent neural network model with feedback response analysis technology, this method can assess the status of the thermal power unit's regulating device and provide early warnings of abnormal conditions. However, using a simple recurrent neural network cannot fully extract the correlation characteristics between data parameters, and for the complex and variable operating conditions of thermal power units, it cannot effectively and timely predict faults in advance, necessitating improvement. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system, device and storage medium for monitoring the operating status of thermal power units, which can accurately detect abnormal operating status of thermal power units and is applicable to the complex and ever-changing operating conditions of thermal power units.

[0005] This invention provides the following technical solution:

[0006] In the first aspect, a method for monitoring the operating status of a thermal power unit is provided, including: acquiring operating status parameter data of the thermal power unit, and determining the target status parameters and key status parameters of the thermal power unit from the operating status parameter data;

[0007] The key state parameters are input into a pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters;

[0008] Obtain the actual values ​​of the target state parameters of the thermal power unit, and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters;

[0009] When the DTW similarity score is lower than the preset warning threshold, the thermal power unit is determined to be in an abnormal state, and an abnormal alarm signal is output.

[0010] As an optional technical solution of the present invention, the step of acquiring the operating status parameter data of the thermal power unit and determining the target status parameters and key status parameters of the thermal power unit from the operating status parameter data includes:

[0011] The state parameters in the operating state parameter data that cause unit failure are taken as target state parameters;

[0012] Calculate the correlation between all state parameters and the target state parameters, and select several state parameters with the highest correlation as key state parameters, where the correlation is expressed as:

[0013] ;

[0014] in, Indicates correlation. Indicates the number of observations. Indicates the first The difference in rank between the observations in the state parameters and the target state parameters.

[0015] As an optional technical solution of the present invention, before the key state parameters are input into the thermal power prediction model, the key state parameters are standardized and converted into a distribution with a mean of 0 and a standard deviation of 1.

[0016] As an optional technical solution of the present invention, the thermal power prediction model includes an improved Transformer model and a NAMs model. Key state parameters are input into the improved Transformer model and the NAMs model respectively to obtain the base value portion of the predicted target state parameter values. and nonlinear correction part The predicted values ​​of the target state parameters are obtained by adding them together. , is represented as:

[0017] .

[0018] As an optional technical solution of the present invention, the improved Transformer model includes an encoder and a fully connected layer. The encoder includes several sequentially stacked coding layers. Each coding layer includes a multi-head nonlinear self-attention mechanism layer, a feedforward network layer, a residual connection and normalization layer, and a Rollout module.

[0019] By inputting the key state parameters into the improved Transformer model, the base value part of the predicted target state parameters is obtained. ,include:

[0020] The key state parameters are normalized and split into a query matrix Q, a key matrix K, and a value matrix V.

[0021] Non-stationary information is extracted from key state parameters using a multilayer perceptron, and represented as follows:

[0022] ;

[0023] ;

[0024] in, The standard deviation of the key state parameters, This represents the mean of the key state parameters. and All of these represent non-stationary factors. Indicates key state parameters. Represents a multilayer perceptron;

[0025] Adding non-stationary information to the query matrix Q, key matrix K, and value matrix V, respectively, is represented as follows:

[0026] ;

[0027] ;

[0028] ;

[0029] in, This indicates a query matrix that incorporates non-stationary information. This represents the key matrix incorporating non-stationary information. This represents the value matrix incorporating non-stationary information, and T represents the matrix transpose. and Representing the query matrix Bond matrix The mean, Represents a matrix where all values ​​are 1;

[0030] The query matrix will incorporate non-stationary information. Key matrix Sum matrix In the input multi-head nonlinear self-attention mechanism layer, the attention score is calculated and represented as:

[0031] ;

[0032] in, This represents the softmax function. Representing the query matrix Key matrix The number of columns;

[0033] The output of each attention head in the multi-head nonlinear self-attention mechanism layer is represented as follows:

[0034] ;

[0035] in, This represents the output of the g-th attention head. , , These represent the query matrices of the g-th attention head. Key matrix Sum matrix The transformation matrix;

[0036] The outputs of all attention heads are concatenated to obtain the concatenated result of the multi-head nonlinear self-attention mechanism layer, which is represented as follows:

[0037] ;

[0038] in, Let h represent the linear transformation matrix, and h represent the number of attention heads. Indicates a connection operation. This represents the splicing result of the multi-head nonlinear self-attention mechanism layer;

[0039] The concatenation results of the multi-head nonlinear self-attention mechanism are sequentially input into the feedforward network layer, the residual connection and normalization layer, and the Rollout module. The Rollout module performs a weighted summation of the outputs of the residual connection and normalization layer, as shown below:

[0040] ;

[0041] in, This represents the output of the Rollout module. Indicates the first Key state parameters in the output of the residual connection and normalization layer express The weight matrix;

[0042] The outputs of the Rollout modules in each coding layer are summed through a fully connected layer to obtain the base value part of the predicted target state parameters. , is represented as:

[0043] ;

[0044] in, This indicates a fully connected layer.

[0045] As an optional technical solution of the present invention, the NAMs model constructs a sub-network for each key state parameter and calculates the function value corresponding to each key state parameter, expressed as follows:

[0046] ;

[0047] ;

[0048] in, Indicates the first The key state parameters are obtained from the hidden layer output after passing through the first fully connected layer. Indicates the first There are several key state parameters, where ReLU represents the activation function. and These represent the weights and biases of the first fully connected layer, respectively. and These represent the weights and biases of the second fully connected layer, respectively. Indicates the first The function values ​​corresponding to the key state parameters;

[0049] Summing the function values ​​corresponding to all key state parameters yields the nonlinear correction component. , is represented as:

[0050] ;

[0051] in, This indicates the number of key state parameters.

[0052] As an optional technical solution of the present invention, the step of obtaining the actual value of the target state parameter of the thermal power unit and calculating the DTW similarity score between the actual value and the predicted value of the target state parameter includes:

[0053] The Euclidean distance between the actual and predicted values ​​of the target state parameters is calculated and expressed as:

[0054] ;

[0055] in, This represents the actual value of the l-th target state parameter. This represents the predicted value of the j-th target state parameter. This represents the Euclidean distance between the actual value of the l-th target state parameter and the predicted value of the j-th target state parameter;

[0056] The DTW similarity score is calculated based on the Euclidean distance between the actual and predicted values ​​of the target state parameters, and is expressed as follows:

[0057] ;

[0058] ;

[0059] in, Indicates DTW distance, express Three points to point The minimum distance, Indicates the total amount accumulated. The final distance obtained by the minimum distance, This represents the DTW similarity score between the actual value sequence X of the target state parameters and the predicted value sequence Y of the target state parameters.

[0060] Secondly, a thermal power unit operation status monitoring system is provided, including: a data acquisition module, used to acquire the operation status parameter data of the thermal power unit, and determine the target status parameters and key status parameters of the thermal power unit in the operation status parameter data;

[0061] The prediction module is used to input the key state parameters into a pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters.

[0062] The similarity calculation module is used to obtain the actual values ​​of the target state parameters of the thermal power unit and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters.

[0063] The early warning module is used to determine that the thermal power unit is in an abnormal state and output an abnormal alarm signal when the DTW similarity score is lower than a preset early warning threshold.

[0064] Thirdly, a thermal power unit operation status monitoring device is provided, including a processor and a storage medium;

[0065] The storage medium is used to store instructions;

[0066] The processor is configured to operate according to the instructions to execute the steps of the thermal power unit operating status monitoring method described in the first aspect.

[0067] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the thermal power unit operating status monitoring method described in the first aspect.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] This invention provides a method for monitoring the operating status of thermal power units. By using a thermal power prediction model, high-dimensional features and hidden relationships in key status parameters can be extracted more effectively, thereby improving the accuracy of abnormal status detection. The thermal power prediction model can adapt to the complex and variable operating conditions of thermal power units and maintain high performance under different operating conditions, thus having stronger universality. The use of the DTW method can detect abnormal states of thermal power units earlier, which provides staff with more time to intervene and handle the situation, thereby reducing unnecessary losses. Attached Figure Description

[0070] Figure 1 This is a flowchart of the main equipment operation status monitoring method in an embodiment of the present invention;

[0071] Figure 2 This is a time-domain diagram of the parameters of the thermal power unit condition assessment model in an embodiment of the present invention;

[0072] Figure 3 This is a status monitoring curve of a thermal power unit in an embodiment of the present invention. Detailed Implementation

[0073] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0074] Example 1

[0075] This embodiment provides a method for monitoring the operating status of thermal power units. This method improves upon the original Transformer model and combines it with the NAMs model to form a thermal power prediction model. Based on a specific threshold setting method, it can more accurately provide early warning of abnormal states.

[0076] like Figure 1 As shown, the specific steps include the following:

[0077] Step 1: Obtain the operating status parameter data of the thermal power unit, and determine the target status parameters and key status parameters of the thermal power unit from the operating status parameter data.

[0078] After acquiring the operating status parameter data of the thermal power unit, firstly identify and delete invalid data caused by sensor failure or abnormal acquisition; then, based on the characteristics of thermal power process and operating requirements, select characteristic parameters that are directly related to unit performance, efficiency or fault prediction, and remove irrelevant or low-value redundant information.

[0079] After filtering, the status parameters that cause unit failure in the operating status parameter data are taken as target status parameters.

[0080] Calculate the correlation between all state parameters and the target state parameters, and select several state parameters with the highest correlation as key state parameters, where the correlation is expressed as:

[0081] ;

[0082] in, Indicates correlation. Indicates the number of observations. Indicates the first The difference in rank between the observations in the state parameters and the target state parameters.

[0083] Before the key state parameters are input into the thermal power prediction model, the key state parameters are standardized and converted into a distribution with a mean of 0 and a standard deviation of 1, so that they are not too affected by extreme values.

[0084] Step 2: Input the key state parameters into the pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters.

[0085] The thermal power prediction model includes an improved Transformer model and a NAMs model. Key state parameters are input into the improved Transformer model and the NAMs model respectively to obtain the base value part of the predicted target state parameter values. and nonlinear correction part The predicted values ​​of the target state parameters are obtained by adding them together. , is represented as: .

[0086] (1) The improved Transformer model includes an encoder and a fully connected layer. The encoder includes six sequentially stacked encoding layers. Each encoding layer includes a multi-head nonlinear self-attention mechanism layer, a feedforward network layer (FFN), a residual connection and normalization layer, and a Rollout module. The improved Transformer model replaces the linear multi-head self-attention mechanism of the traditional Transformer model with a multi-head nonlinear self-attention mechanism, and replaces the pooling layer of the traditional Transformer model with a Rollout module.

[0087] By inputting the key state parameters into the improved Transformer model, the base value part of the predicted target state parameters is obtained. ,include:

[0088] The key state parameters are normalized and split into a query matrix Q, a key matrix K, and a value matrix V.

[0089] Non-stationary information is extracted from key state parameters using a multilayer perceptron, and represented as follows:

[0090] ;

[0091] ;

[0092] in, The standard deviation of the key state parameters, This represents the mean of the key state parameters. and All of these represent non-stationary factors. Indicates key state parameters. This represents a multilayer perceptron.

[0093] Adding non-stationary information to the query matrix Q, key matrix K, and value matrix V, respectively, is represented as follows:

[0094] ;

[0095] ;

[0096] ;

[0097] in, This indicates a query matrix that incorporates non-stationary information. This represents the key matrix incorporating non-stationary information. This represents the value matrix incorporating non-stationary information, and T represents the matrix transpose. and Representing the query matrix Bond matrix The mean, This represents a matrix where all values ​​are 1.

[0098] The query matrix will incorporate non-stationary information. Key matrix Sum matrix In the input multi-head nonlinear self-attention mechanism layer, the attention score is calculated and represented as:

[0099] ;

[0100] in, This represents the softmax function. Representing the query matrix Key matrix The number of columns.

[0101] The output of each attention head in the multi-head nonlinear self-attention mechanism layer is represented as follows:

[0102] ;

[0103] in, This represents the output of the g-th attention head. , , These represent the query matrices of the g-th attention head. Key matrix Sum matrix The transformation matrix.

[0104] The outputs of all attention heads are concatenated to obtain the concatenated result of the multi-head nonlinear self-attention mechanism layer, which is represented as follows:

[0105] ;

[0106] in, Let h represent the linear transformation matrix, and h represent the number of attention heads. Indicates a connection operation. This represents the splicing result of the multi-head nonlinear self-attention mechanism layer. In this embodiment, the multi-head nonlinear self-attention mechanism layer includes eight attention heads.

[0107] The concatenation results of the multi-head nonlinear self-attention mechanism are sequentially input into the feedforward network layer, the residual connection and normalization layer, and the Rollout module. The Rollout module performs a weighted summation of the outputs of the residual connection and normalization layer, as shown below:

[0108] ;

[0109] in, This represents the output of the Rollout module. Indicates the first Key state parameters in the output of the residual connection and normalization layer express The weight matrix.

[0110] The outputs of the Rollout modules in each coding layer are summed through a fully connected layer to obtain the base value part of the predicted target state parameters. , is represented as:

[0111] ;

[0112] in, This indicates a fully connected layer.

[0113] (2) The NAMs model constructs a sub-network for each key state parameter and calculates the function value corresponding to each key state parameter, expressed as:

[0114] ;

[0115] ;

[0116] in, Indicates the first The key state parameters are obtained from the hidden layer output after passing through the first fully connected layer. Indicates the first There are several key state parameters, where ReLU represents the activation function. and These represent the weights and biases of the first fully connected layer, respectively. and These represent the weights and biases of the second fully connected layer, respectively. Indicates the first The function values ​​corresponding to the key state parameters.

[0117] Summing the function values ​​corresponding to all key state parameters yields the nonlinear correction component. , is represented as:

[0118] ;

[0119] in, This indicates the number of key state parameters.

[0120] Step 3: Obtain the actual values ​​of the target state parameters of the thermal power unit, and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters.

[0121] The Euclidean distance between the actual and predicted values ​​of the target state parameters is calculated and expressed as:

[0122] ;

[0123] in, This represents the actual value of the l-th target state parameter. This represents the predicted value of the j-th target state parameter. This represents the Euclidean distance between the actual value of the l-th target state parameter and the predicted value of the j-th target state parameter.

[0124] The DTW similarity score is calculated based on the Euclidean distance between the actual and predicted values ​​of the target state parameters, and is expressed as follows:

[0125] ;

[0126] ;

[0127] in, Indicates DTW distance, express Three points to point The minimum distance, Indicates the total amount accumulated. The final distance obtained by the minimum distance, This represents the DTW similarity score between the actual value sequence X of the target state parameters and the predicted value sequence Y of the target state parameters.

[0128] Step 4: When the DTW similarity score is lower than the pre-set warning threshold, the thermal power unit is determined to be in an abnormal state, and an abnormal alarm signal is output.

[0129] The warning threshold is determined based on historical health data learned by the thermal power prediction model during the training phase. Specifically, actual values ​​of target state parameters for multiple time periods are selected from the historical health data, and the DTW similarity score between the actual values ​​of the target state parameters for each time period and the corresponding predicted values ​​output by the thermal power prediction model is calculated. The distribution of all DTW similarity scores is statistically analyzed, and based on 3... The rules determine the warning threshold.

[0130] Example 2

[0131] This embodiment provides an application example of the method based on Embodiment 1.

[0132] Obtain the operating status parameter data of the thermal power unit, and determine the target status parameters and key status parameters of the thermal power unit from the operating status parameter data.

[0133] The data comes from a thermal power plant in Southwest China. It includes data for the entire day of May 17, 2024, when the unit was operating normally, and data for June 12, 2024, including data from when a fault occurred. The sampling frequency is set to once per second. The dataset contains 23 different parameters related to the steam turbine generator unit, with 45,367 sample points for each parameter.

[0134] Preprocessing of thermal power unit operating data: First, identify and delete invalid data caused by sensor failure or abnormal acquisition; then, based on the characteristics of thermal power process and operating requirements, select characteristic parameters that are directly related to unit performance, efficiency or fault prediction, and remove irrelevant or low-value redundant information.

[0135] The parameters that directly cause unit failures are set as target state parameters, and the Spearman rank correlation coefficient is used ( The correlation between all state parameters and the target state parameters was calculated. Parameters with a correlation greater than or equal to 0.75 were selected as key state parameters, and finally, six parameters were chosen. The time-domain plots of each parameter are shown below. Figure 2 As shown in Table 1, the names of each parameter are as follows.

[0136] Table 1 Key State Parameter Information Table

[0137] Serial Number parameter unit Serial Number parameter unit 1 Main steam valve #1 inlet steam temperature ℃ 4 High-pressure cylinder exhaust steam pressure MPa 2 Active power KW 5 PI output samples / 3 FSMTAVG output samples ℃ 6 TP output samples MPa

[0138] The acquired key state parameters cannot be directly used for state detection. Therefore, they are standardized to convert the data into a distribution with a mean of 0 and a standard deviation of 1, thus minimizing the impact of extreme values. Data from normal operation on May 17th is used as the training set (from sample point 0 to 20562), the last 10% of the training data is used as the validation set, and data from the time of the fault on June 12th is used as the test set (from sample point 20563 to 45,367). The training set is then input into the thermal power prediction model for model training.

[0139] In this embodiment, in the trained thermal power prediction model, the improved Transformer model sets the MLP dimension to 512, the hidden layer dimension to 128, and the number of heads in the multi-head nonlinear self-attention mechanism to 8. The fully connected layer has 6 input features and 1 output feature. In the NAMs model, the number of input features is 6, the number of output features is 1, the number of hidden layers is 2, and the hidden layer dimension is 128.

[0140] The method for setting the warning threshold is as follows: calculate the score every 50 data points with a step size of 1, traverse the entire dataset, and calculate the mean (μ) and standard deviation (σ) of the DTW similarity score based on the training set data. Then, use the warning threshold defined as μ-3σ for anomaly detection.

[0141] The test set is input into the trained thermal power prediction model to obtain predicted values. The actual values ​​of the target state parameters of the thermal power unit are input into the thermal power prediction model. The predicted values ​​and actual values ​​generated by the thermal power prediction model are compared using the DTW similarity score calculation. The score is calculated every 50 data points with a step size of 1, traversing the entire dataset. The resulting curve, with the horizontal axis representing the sample points and the vertical axis representing the similarity score, is the state monitoring curve. When the curve falls below a set warning threshold, the thermal power unit is determined to have an anomaly. Specifically... Figure 3 As shown, the horizontal axis represents the sample points of key state parameters. Each sample point has a corresponding actual sampling time, so the order of the sample points on the horizontal axis represents the time axis.

[0142] like Figure 3 As shown, the horizontal red dashed line represents the warning threshold, the blue curve represents the detection curve, and the warning status is the time period corresponding to the sample point where the detection curve exceeds the warning threshold. The warning sample point is the 2773rd sample point, the warning time is 05:56 on June 12, 2024, and the actual time of the fault is 09:15 on June 12, 2024. This method can provide a warning of the fault 3.31 hours in advance.

[0143] Example 3

[0144] This embodiment provides a thermal power unit operation status monitoring system, including:

[0145] The data acquisition module is used to acquire the operating status parameter data of the thermal power unit and determine the target status parameters and key status parameters of the thermal power unit from the operating status parameter data.

[0146] The prediction module is used to input the key state parameters into a pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters.

[0147] The similarity calculation module is used to obtain the actual values ​​of the target state parameters of the thermal power unit and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters.

[0148] The early warning module is used to determine that the thermal power unit is in an abnormal state and output an abnormal alarm signal when the DTW similarity score is lower than a preset early warning threshold.

[0149] Example 4

[0150] This embodiment provides a thermal power unit operation status monitoring device, including a processor and a storage medium;

[0151] The storage medium is used to store instructions;

[0152] The processor is used to operate according to the instructions to execute the steps of the thermal power unit operation status monitoring method described in Embodiment 1.

[0153] Example 5

[0154] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the thermal power unit operation status monitoring method described in Embodiment 1.

[0155] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0157] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0158] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0159] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring the operating status of a thermal power unit, characterized in that, include: Acquire the operating status parameter data of the thermal power unit, and determine the target status parameters and key status parameters of the thermal power unit from the operating status parameter data; The key state parameters are input into a pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters; Obtain the actual values ​​of the target state parameters of the thermal power unit, and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters; When the DTW similarity score is lower than the preset warning threshold, the thermal power unit is determined to be in an abnormal state, and an abnormal alarm signal is output.

2. The method for monitoring the operating status of thermal power units according to claim 1, characterized in that, The process of acquiring operating status parameter data of thermal power units and determining target status parameters and key status parameters of thermal power units from the operating status parameter data includes: The state parameters in the operating state parameter data that cause unit failure are taken as target state parameters; Calculate the correlation between all state parameters and the target state parameters, and select several state parameters with the highest correlation as key state parameters, where the correlation is expressed as: ； in, Indicates correlation. Indicates the number of observations. Indicates the first The difference in rank between the observations in the state parameters and the target state parameters.

3. The method for monitoring the operating status of thermal power units according to claim 1, characterized in that, Before the key state parameters are input into the thermal power prediction model, the key state parameters are standardized and converted into a distribution with a mean of 0 and a standard deviation of 1.

4. The method for monitoring the operating status of thermal power units according to claim 1, characterized in that, The thermal power prediction model includes an improved Transformer model and a NAMs model. Key state parameters are input into the improved Transformer model and the NAMs model respectively to obtain the base value part of the predicted target state parameter values. and nonlinear correction part The predicted values ​​of the target state parameters are obtained by adding them together. , is represented as: 。 5. The method for monitoring the operating status of thermal power units according to claim 4, characterized in that, The improved Transformer model includes an encoder and a fully connected layer. The encoder includes several sequentially stacked coding layers. Each coding layer includes a multi-head nonlinear self-attention mechanism layer, a feedforward network layer, a residual connection and normalization layer, and a Rollout module. By inputting the key state parameters into the improved Transformer model, the base value part of the predicted target state parameters is obtained. ,include: The key state parameters are normalized and split into a query matrix Q, a key matrix K, and a value matrix V. Non-stationary information is extracted from key state parameters using a multilayer perceptron, and represented as follows: ； ； in, The standard deviation of the key state parameters, This represents the mean of the key state parameters. and All of these represent non-stationary factors. Indicates key state parameters. Represents a multilayer perceptron; Adding non-stationary information to the query matrix Q, key matrix K, and value matrix V, respectively, is represented as follows: ； ； ； in, This indicates a query matrix that incorporates non-stationary information. This represents the key matrix incorporating non-stationary information. This represents the value matrix incorporating non-stationary information, and T represents the matrix transpose. and Representing the query matrix Bond matrix The mean, Represents a matrix where all values ​​are 1; The query matrix will incorporate non-stationary information. Key matrix Sum matrix In the input multi-head nonlinear self-attention mechanism layer, the attention score is calculated and represented as: ； in, This represents the softmax function. Representing the query matrix Key matrix The number of columns; The output of each attention head in the multi-head nonlinear self-attention mechanism layer is represented as follows: ； in, This represents the output of the g-th attention head. , , These represent the query matrices of the g-th attention head. Key matrix Sum matrix The transformation matrix; The outputs of all attention heads are concatenated to obtain the concatenated result of the multi-head nonlinear self-attention mechanism layer, which is represented as follows: ； in, Let h represent the linear transformation matrix, and h represent the number of attention heads. Indicates a connection operation. This represents the splicing result of the multi-head nonlinear self-attention mechanism layer; The concatenation results of the multi-head nonlinear self-attention mechanism are sequentially input into the feedforward network layer, the residual connection and normalization layer, and the Rollout module. The Rollout module performs a weighted summation of the outputs of the residual connection and normalization layer, as shown below: ； in, This represents the output of the Rollout module. Indicates the first Key state parameters in the output of the residual connection and normalization layer express The weight matrix; The outputs of the Rollout modules in each coding layer are summed through a fully connected layer to obtain the base value part of the predicted target state parameters. , is represented as: ； in, This indicates a fully connected layer.

6. The method for monitoring the operating status of thermal power units according to claim 4, characterized in that, The NAMs model constructs a sub-network for each key state parameter and calculates the function value corresponding to each key state parameter, expressed as follows: ； ； in, Indicates the first The key state parameters are obtained from the hidden layer output after passing through the first fully connected layer. Indicates the first There are several key state parameters, where ReLU represents the activation function. and These represent the weights and biases of the first fully connected layer, respectively. and These represent the weights and biases of the second fully connected layer, respectively. Indicates the first The function values ​​corresponding to the key state parameters; Summing the function values ​​corresponding to all key state parameters yields the nonlinear correction component. , is represented as: ； in, This indicates the number of key state parameters.

7. The method for monitoring the operating status of thermal power units according to claim 1, characterized in that, The process of obtaining the actual values ​​of the target state parameters of the thermal power unit and calculating the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters includes: The Euclidean distance between the actual and predicted values ​​of the target state parameters is calculated and expressed as: ； in, This represents the actual value of the l-th target state parameter. This represents the predicted value of the j-th target state parameter. This represents the Euclidean distance between the actual value of the l-th target state parameter and the predicted value of the j-th target state parameter; The DTW similarity score is calculated based on the Euclidean distance between the actual and predicted values ​​of the target state parameters, and is expressed as follows: ； ； in, Indicates DTW distance, express Three points to point The minimum distance, Indicates the total amount accumulated. The final distance obtained by the minimum distance, This represents the DTW similarity score between the actual value sequence X of the target state parameters and the predicted value sequence Y of the target state parameters.

8. A thermal power unit operating status monitoring system, characterized in that, include: The data acquisition module is used to acquire the operating status parameter data of the thermal power unit and determine the target status parameters and key status parameters of the thermal power unit from the operating status parameter data. The prediction module is used to input the key state parameters into a pre-trained thermal power prediction model to obtain the predicted values ​​of the target state parameters. The similarity calculation module is used to obtain the actual values ​​of the target state parameters of the thermal power unit and calculate the DTW similarity score between the actual values ​​and the predicted values ​​of the target state parameters. The early warning module is used to determine that the thermal power unit is in an abnormal state and output an abnormal alarm signal when the DTW similarity score is lower than a preset early warning threshold.

9. A monitoring device for the operating status of a thermal power unit, characterized in that, Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the thermal power unit operation status monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps of the thermal power unit operation status monitoring method according to any one of claims 1 to 7.

Images