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

Abstract

The invention discloses a thermal power generating unit operation state monitoring method and system and a storage medium, and belongs to the technical field of thermal power generating unit state monitoring. The thermal power generating unit comprises a main device and an auxiliary device, and the operation state monitoring method of the main device comprises the steps that operation data of the main device are acquired, a target parameter true value and an associated parameter of a target parameter are obtained, and the associated parameter of the target parameter serves as an input parameter sequence; inputting the input parameter sequence into the main equipment target parameter prediction model, extracting features in the input parameter sequence, and capturing dynamic changes in the high-dimensional feature sequence to obtain a target parameter prediction value; and calculating an RMSE value according to the target parameter true value and the target parameter predicted value, and comparing the RMSE value with a first threshold value to judge whether the current operation state of the main equipment is normal or not. According to the invention, different methods are adopted to monitor the main equipment and the auxiliary equipment separately, and accurate monitoring of the operation state of the thermal power generating unit is realized according to different setting methods.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal power unit status monitoring, and in particular relates to a method, system and storage medium for monitoring the operating status of a thermal power unit. Background Art

[0002] With the advancement of the "dual carbon" goals, renewable energy sources (such as wind and solar power) are being connected to the power grid on a large scale. The randomness and volatility of these energy sources place higher demands on the stability and flexibility of the power grid. In this context, the operational quality requirements for thermal power units are increasing.

[0003] However, due to the complexity of the operation of thermal power units and the involvement of multiple parameters, it is difficult to accurately predict their parameters with existing technologies. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method, system and storage medium for monitoring the operating status of a thermal power unit, which adopt different methods to separately monitor the main equipment and auxiliary equipment, and realize accurate prediction of the parameters of the main equipment and auxiliary equipment.

[0005] The present invention provides the following technical solutions:

[0006] In a first aspect, a method for monitoring the operating status of a thermal power unit is provided. The thermal power unit includes a main device and an auxiliary device. The method for monitoring the operating status of the main device includes:

[0007] Obtaining the operating data of the main device, obtaining the true value of the target parameter and the associated parameters of the target parameter, and using the associated parameters of the target parameter as the input parameter sequence;

[0008] Input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value;

[0009] Calculating an RMSE value based on the true value of the target parameter and the predicted value of the target parameter, and comparing the RMSE value with a first threshold to determine whether the current operating state of the main device is normal;

[0010] The operating status monitoring method of the auxiliary equipment includes:

[0011] Obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and build a memory matrix based on the historical observation data;

[0012] An estimation vector is obtained by multivariate state estimation technology based on memory matrix calculation;

[0013] The similarity between the current observation data of the auxiliary device and the estimated vector is calculated, and compared with the second threshold to determine whether the current operating state of the auxiliary device is normal.

[0014] As an optional technical solution of the present invention, obtaining the operating data of the master device and obtaining the true value of the target parameter and the associated parameter of the target parameter includes: using a parameter in the operating data of the master device that represents the operating state of the master device as the target parameter;

[0015] Calculate the Pearson correlation coefficient between the target parameter and the remaining parameters, and according to the pre-set correlation coefficient threshold, take the parameter with a correlation coefficient greater than the correlation coefficient threshold as the associated parameter of the target parameter.

[0016] As an optional technical solution of the present invention, the main device target parameter prediction model includes a temporal convolutional network and a gated recurrent unit. The temporal convolutional network is used to extract features from the input parameter sequence to obtain a high-order feature sequence; the gated recurrent unit is used to capture dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value.

[0017] As an optional technical solution of the present invention, the temporal convolutional network includes several extended causal convolutional layers; each extended causal convolutional layer includes an extended causal convolution and a residual connection block;

[0018] The extended causal convolution introduces a dilation factor to extract features from the input parameter sequence, which is expressed as:

[0019] ;

[0020] in, represents the output feature of the extended causal convolution layer at the current time step t, k represents the convolution kernel size, represents the convolution kernel weight, Represents the first Value, i represents the position index of the convolution kernel in the time series;

[0021] Based on the output features of the dilated causal convolution at the current time step t , the final output sequence of the extended causal convolution is expressed as ,in, represents the final output sequence, Indicates the Output features at time steps;

[0022] The residual connection block expands the final output sequence of the causal convolution at each layer Added to the input parameter sequence, expressed as:

[0023] ;

[0024] Where X represents the input parameter sequence, Represents the residual connection result;

[0025] Residual connection result of the current extended causal convolution layer As input to the next dilated causal convolutional layer;

[0026] Several extended causal convolutional layers in the temporal convolutional network perform multi-layer convolution on the input parameter sequence to obtain a high-order feature sequence, which is expressed as:

[0027] ;

[0028] in, represents the high-order feature sequence, express The feature vector at each time step.

[0029] As an optional technical solution of the present invention, the gated cycle unit includes an update gate and a reset gate;

[0030] The update gate is used to fuse the state information of the current time step and the previous time step in the high-order feature sequence, which is expressed as:

[0031] ;

[0032] in, represents the output of the update gate, represents the sigmoid activation function, represents the update gate bias term, represents the update gate weight coefficient, Indicates the first feature vectors, Represents the previous time step The hidden state of

[0033] The reset gate is used to determine the forgotten state information, which is expressed as:

[0034] ;

[0035] in, represents the output of the reset gate, Represents the reset gate bias term, Indicates resetting the gate weight coefficient;

[0036] Calculate the candidate hidden state at the current moment, expressed as:

[0037] ;

[0038] in, represents the candidate hidden state at the current moment, represents the activation function, represents the candidate hidden state bias item at the current moment, Indicates the first The weight coefficient of the eigenvector, express The weight coefficient of represents the product of matrices;

[0039] The final hidden state at the current moment is expressed as:

[0040] ;

[0041] in, Represents the final hidden state at the current moment;

[0042] Based on the final hidden state at the current moment, the target parameter prediction value is obtained through nonlinear transformation, which is expressed as:

[0043] ;

[0044] in, represents the predicted value of the target parameter, Represents the final hidden state bias at the current moment, Represents the weight coefficient of the final hidden state at the current moment.

[0045] As an optional technical solution of the present invention, current observation data of the auxiliary device and historical observation data under normal operating conditions are obtained, and a memory matrix is ​​constructed based on the historical observation data, including:

[0046] The historical observation data of the auxiliary equipment under normal operating conditions is expressed as:

[0047] ;

[0048] in, Indicates that auxiliary equipment is The observation sequence at the moment, n represents the n observation points on the auxiliary equipment, Indicates that the nth observation point is The observation value at time, T represents transposition;

[0049] Select m observation sequences from the historical observation data to construct a memory matrix, which can be expressed as:

[0050] ;

[0051] Where D represents the memory matrix, Represents the historical observation data in the nth row and mth column of the memory matrix.

[0052] As an optional technical solution of the present invention, an estimated vector is obtained by calculating based on a memory matrix using a multivariate state estimation technique, including:

[0053] By calculating the similarity between the current observation data and the historical observation data in the memory matrix, a weight vector W is generated, which is expressed as:

[0054] ;

[0055] in, Represents the current observation data, Represents nonlinear operations;

[0056] Based on the weight vector W, the estimated vector is expressed as:

[0057] ;

[0058] in, represents the estimated vector.

[0059] As an optional technical solution of the present invention, calculating the similarity between the current observation data of the auxiliary device and the estimated vector includes:

[0060] The similarity between the current observation data and the estimated vector of the calculation auxiliary device is expressed as:

[0061] ;

[0062] Among them, s represents the similarity between the current observation data and the estimated vector, represents the Euclidean distance between the observation vector and the estimated vector, Represents the current observation data, represents the estimated vector.

[0063] In a second aspect, a thermal power unit operation status monitoring system is provided, comprising: the thermal power unit includes a main device and an auxiliary device, and the operation status monitoring system of the main device includes:

[0064] A first data acquisition module is configured to acquire operating data of a master device, obtain a true value of a target parameter and associated parameters of the target parameter, and use the associated parameters of the target parameter as an input parameter sequence;

[0065] The prediction module is used to input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value;

[0066] A first judgment module is used to calculate an RMSE value according to the true value of the target parameter and the predicted value of the target parameter, and compare the RMSE value with a first threshold to determine whether the current operating state of the main device is normal;

[0067] The operating status monitoring system of the auxiliary equipment includes:

[0068] The second data acquisition module is used to obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and to construct a memory matrix based on the historical observation data;

[0069] An estimation module, configured to obtain an estimation vector by using a multivariate state estimation technique based on a memory matrix;

[0070] The second judgment module is used to calculate the similarity between the current observation data of the auxiliary device and the estimated vector, and compare it with the second threshold to determine whether the current operating state of the auxiliary device is normal.

[0071] In a third aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of the method for monitoring the operating status of a thermal power unit described in the first aspect are implemented.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] The present invention provides a method for monitoring the operating status of a thermal power unit. The method, through a TCN-GRU neural network model, can effectively capture the long-term and short-term dependencies in an input parameter sequence, significantly improving the prediction accuracy of target parameters. The method also monitors the main equipment and auxiliary equipment in the thermal power unit separately, enabling parameter prediction using different methods for different equipment, making the prediction results more accurate. Early fault monitoring can promptly detect potential equipment problems, reduce downtime, and lower maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 is a flow chart of a method for monitoring the operating status of a master device according to an embodiment of the present invention;

[0075] Figure 2 is a flow chart of a method for monitoring the operating status of an auxiliary device in an embodiment of the present invention;

[0076] Figure 3 It is a common convolution graph with a convolution kernel size of 3 in an embodiment of the present invention;

[0077] Figure 4 is a dilated convolution graph with a convolution kernel size of 3 in an embodiment of the present invention;

[0078] Figure 5 3 is a diagram showing the calculation results of the RMSE value of the bearing vibration in an embodiment of the present invention. DETAILED DESCRIPTION

[0079] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0080] Example 1

[0081] This embodiment provides a method for monitoring the operating status of a thermal power unit. The thermal power unit includes a main device and an auxiliary device. In this embodiment, the main device is a steam turbine and the auxiliary device is a fan, and different monitoring methods are used for the two.

[0082] like Figure 1 As shown, the operating status monitoring method of the master device specifically includes the following steps:

[0083] Step 1: Obtain the operating data of the main device, obtain the true value of the target parameter and the associated parameters of the target parameter, and use the associated parameters of the target parameter as the input parameter sequence.

[0084] 1.1. A parameter representing the operating status of the master device in the operating data of the master device is used as a target parameter.

[0085] This embodiment selects #1 bearing pad vibration as the target parameter.

[0086] 1.2. Calculate the Pearson correlation coefficient between the target parameter and the remaining parameters, and according to the pre-set correlation coefficient threshold, take the parameter with a correlation coefficient greater than the correlation coefficient threshold as the associated parameter of the target parameter.

[0087] In this embodiment, parameters with correlation coefficients greater than 0.5 are used as associated parameters of bearing shoe vibration, including #3 bearing shoe vibration, 1# bearing rotor vibration (X phase relative), 2# bearing rotor vibration (Y phase relative), 1# bearing metal temperature 1# (left side), 1# bearing metal temperature 2# (right side), 2# bearing metal temperature 1# (left side), 1# bearing return oil temperature, and main steam pressure.

[0088] Step 2: Input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value.

[0089] Among them, the main device target parameter prediction model includes a temporal convolutional network and a gated recurrent unit. The temporal convolutional network is used to extract features from the input parameter sequence to obtain a high-order feature sequence; the gated recurrent unit is used to capture dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value.

[0090] (1) The temporal convolutional network (TCN) includes several extended causal convolution layers; each extended causal convolution layer includes an extended causal convolution and a residual connection block, and the residual connection block is used to keep the network output and the output length the same. The extended causal convolution is different from the ordinary one-dimensional convolution, mainly in terms of causality and expansion: Unlike ordinary convolution, causal convolution has strict time constraints, and the output of the neuron is obtained by convolution of data with causal relationships. In the time series, causality can only be expressed as time Previous information to predict the moment The value of , thus satisfying the principle of temporal dependence. In order to capture the long-term temporal features between data, it is necessary to stack multiple layers of ordinary convolution when building the model, but this will make the model too deep and difficult to train. Dilated convolution can alleviate this problem because it can obtain a longer receptive field with fewer layers. Dilated convolution introduces a dilation factor, which is equivalent to a fixed step size between each two adjacent convolution kernels. Figure 3 and Figure 4 As shown in the figure, if the input receptive field is 15, 7 layers of ordinary convolution with a kernel size of 3 are required, while 3 layers of dilated convolution with the same size kernel are required.

[0091] The extended causal convolution introduces a dilation factor to extract features from the input parameter sequence, which is expressed as:

[0092] ;

[0093] in, represents the output feature of the extended causal convolution layer at the current time step t, k represents the convolution kernel size, represents the convolution kernel weight, Represents the first A value, i represents the position index of the convolution kernel in the time series.

[0094] Based on the output features of the dilated causal convolution at the current time step t , the final output sequence of the extended causal convolution is expressed as ,in, represents the final output sequence, Indicates the In this embodiment, a Dropout operation is performed after the extended causal convolution of each layer to prevent overfitting, and the output is normalized to ensure numerical stability, and finally the final output sequence of the current extended causal convolution is obtained. .

[0095] The residual connection block expands the final output sequence of the causal convolution at each layer Added to the input parameter sequence, expressed as:

[0096] ;

[0097] Where X represents the input parameter sequence, Represents the residual connection result.

[0098] Residual connection result of the current extended causal convolution layer As the input to the next dilated causal convolutional layer.

[0099] Several extended causal convolutional layers in the temporal convolutional network perform multi-layer convolution on the input parameter sequence to obtain a high-order feature sequence, which is expressed as:

[0100] ;

[0101] in, represents the high-order feature sequence, express The feature vector at each time step.

[0102] (2) The Gated Recurrent Unit (GRU) is a variant of LSTM. Its network structure is simpler than LSTM. It simplifies the forget gate, input gate, and output gate in LSTM into an update gate and a reset gate. Because there is one less gate function, the number of parameters is also reduced. Therefore, the GRU includes an update gate and a reset gate.

[0103] The update gate is used to fuse the state information of the current time step and the previous time step in the high-order feature sequence. The activation function outputs an activation value between 0 and 1, which determines the amount of information that needs to be transmitted between the previous time step and the current time step, and fuses these two parts of information. It can be expressed as:

[0104] ;

[0105] in, represents the output of the update gate, represents the sigmoid activation function, represents the update gate bias term, represents the update gate weight coefficient, Indicates the first feature vectors, Represents the previous time step The hidden state of .

[0106] The reset gate is used to determine the forgotten state information, which is expressed as:

[0107] ;

[0108] in, represents the output of the reset gate, Represents the reset gate bias term, Indicates resetting the gate weight coefficient.

[0109] Calculate the candidate hidden state at the current moment, expressed as:

[0110] ;

[0111] in, represents the candidate hidden state at the current moment, represents the activation function, represents the candidate hidden state bias item at the current moment, Indicates the first The weight coefficient of the eigenvector, express The weight coefficient of Represents the product of matrices.

[0112] The final memory at the current moment is stored in the current unit and passed to the next unit, which determines the current memory content and the information that needs to be retained in the previous time step. The final hidden state at the current moment is expressed as:

[0113] ;

[0114] in, Represents the final hidden state at the current moment;

[0115] Based on the final hidden state at the current moment, the target parameter prediction value is obtained through nonlinear transformation, which is expressed as:

[0116] ;

[0117] in, represents the predicted value of the target parameter, Represents the final hidden state bias at the current moment, Represents the weight coefficient of the final hidden state at the current moment.

[0118] Step 3: Calculate a root mean square error (RMSE) value using the true value of the target parameter and the predicted value of the target parameter, and compare the RMSE value with a first threshold to determine whether the current operating status of the master device is normal.

[0119] Calculate the RMSE value of the target parameter's true value and predicted value, and set the maximum RMSE value in the corresponding time period during the model training process as the first threshold, such as Figure 5 As shown, the first threshold is 0.04. It is stipulated that when the RMSE of 300 or more consecutive sample points continuously exceeds the first threshold, the master device is judged to be in an abnormal state. Figure 5The abnormal monitoring point was 3.30 5:27:33, and the machine trip time was 3.30 5:53:12. The potential fault was discovered 26 minutes in advance. Through early fault monitoring, potential problems of equipment can be discovered in time and downtime can be reduced.

[0120] During the operation of power plant fans, due to wear, aging and fatigue of components, the operating conditions of fans gradually deteriorate over time and faults will gradually appear. Traditional warning threshold setting methods cannot identify these early signs of faults in a timely manner because these signs are hidden in the dynamic changes of fan monitoring parameters. By utilizing the characteristics of multivariate state estimation technology (MSET), the initial degradation point of the equipment state can be detected and the dynamic development process of the fault can be captured, thereby achieving early warning of fan faults. Figure 2 As shown, the operating status monitoring method of the auxiliary equipment includes:

[0121] Step 1: Obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and build a memory matrix based on the historical observation data.

[0122] The historical observation data of the auxiliary equipment under normal operating conditions is expressed as:

[0123] ;

[0124] in, Indicates that auxiliary equipment is The observation sequence at the moment, n represents the n observation points on the auxiliary equipment, Indicates that the nth observation point is The observation value at time t, where T represents transpose.

[0125] The memory matrix is ​​the foundation of multivariate state estimation technology and a direct factor influencing its accuracy. For multivariate state estimation, using all historical data from normal states as the memory matrix provides the best predictive alarm results. However, due to the large volume of historical observation data and the long computation time, this is not suitable for factory applications. Therefore, in this embodiment, m historical observation vectors are selected to form the memory matrix. This maximizes the representation of the dynamic process of normal equipment operation while balancing computation time.

[0126] Select m observation sequences from the historical observation data to construct a memory matrix, which can be expressed as:

[0127] ;

[0128] Where D represents the memory matrix, Represents the historical observation data in the nth row and mth column of the memory matrix.

[0129] Step 2: Obtain the estimated vector based on the memory matrix through multivariate state estimation technology.

[0130] The estimated vector is the product of the memory matrix and the weight vector, expressed as:

[0131] ;

[0132] in, Represents the weight vector.

[0133] The residual between the current observation data and the estimated vector is expressed as: Under the constraint that ||ε||2 is the smallest, the weight vector W is obtained using the least squares method. The residual sum of squares is:

[0134] .

[0135] Where i represents the i-th row of the memory matrix, j represents the j-th column of the memory matrix, represents the residual sum of squares, Represents the jth element in the weight vector.

[0136] Will respectively Take the partial derivative and set it equal to 0, which is expressed as:

[0137] ;

[0138] Among them, k represents the partial derivative of the kth element in the weight vector.

[0139] Arranged:

[0140] ;

[0141] Convert it into matrix form:

[0142] ;

[0143] Arrange the weight vector expression:

[0144] .

[0145] In actual calculations, it is found that the linear correlation between the memory matrix state vectors makes D T D is irreversible, which leads to no solution for W. To solve this problem, nonlinear operation is introduced. Instead of the original multiplication operator, the weight vector W is generated by calculating the similarity between the current observation data and the historical observation data in the memory matrix, which is expressed as:

[0146] ;

[0147] in, Represents the current observation data, Represents a nonlinear operation.

[0148] Based on the weight vector W, the estimated vector is expressed as:

[0149] ;

[0150] in, represents the estimated vector.

[0151] Step 3: Calculate the similarity between the current observation data of the auxiliary device and the estimated vector, and compare it with the second threshold to determine whether the current operating state of the auxiliary device is normal.

[0152] The estimated vector is obtained by calculating the current observation data through MSET, in which there are different degrees of deviation between the various features of the two. In order to express the degree of deviation, the residual between the features was used in the earliest applications. However, considering the large number of features in the actual application of power plants, it is not easy to obtain intuitive information. Therefore, a fused similarity function is used to represent the deviation between the estimated vector and the observation vector. Normal vectors are generally represented by Euclidean distance, cosine similarity or Pearson correlation coefficient. Taking into account the linear distance and angle of the vectors, the similarity function fused by Euclidean distance and cosine similarity is used to calculate the similarity between the current observation data of the auxiliary equipment and the estimated vector, which is expressed as:

[0153] ;

[0154] Among them, s represents the similarity between the current observation data and the estimated vector, represents the Euclidean distance between the observation vector and the estimated vector, Represents the current observation data, represents the estimated vector.

[0155] The range of s is (0,1]. The closer the s value is to 1, the higher the similarity between the two vectors is, that is, the more normal the auxiliary equipment is running.

[0156] In this embodiment, the second threshold value adopts a dynamic threshold value. Combining the interval estimation idea in statistics, it can be deduced from Chebyshev inequality that for any random variable with finite mean and finite variance, its value range is [ ],in and are the mean and variance of the random variable, It is related to the false positive rate a. From this, we can deduce the similarity at N moments The value range C is ,in, ; ; ;

[0157] That is, the similarity of N moments is required to solve the problem. As time goes by, the amount of data will also increase, and the amount of calculation will also increase. Therefore, the improvement is:

[0158] ;

[0159] .

[0160] The similarity mean and variance obtained from the above formula are not only dynamic, changing with time T, but also each time the similarity at time T+1 is calculated, only the mean and variance at time T and the similarity at time T+1 are required, significantly reducing the amount of computation. A threshold is also set for the minimum similarity, using the minimum similarity between the current observation data of the test set under normal conditions and the estimated vector as the second threshold to improve warning accuracy.

[0161] Example 2

[0162] This embodiment provides a system for monitoring the operating status of a thermal power unit. The thermal power unit includes a main device and an auxiliary device. The system for monitoring the operating status of the main device includes:

[0163] A first data acquisition module is configured to acquire operating data of a master device, obtain a true value of a target parameter and associated parameters of the target parameter, and use the associated parameters of the target parameter as an input parameter sequence;

[0164] The prediction module is used to input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value;

[0165] A first judgment module is used to calculate an RMSE value according to the true value of the target parameter and the predicted value of the target parameter, and compare the RMSE value with a first threshold to determine whether the current operating state of the main device is normal;

[0166] The operating status monitoring system of the auxiliary equipment includes:

[0167] The second data acquisition module is used to obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and to construct a memory matrix based on the historical observation data;

[0168] An estimation module, configured to obtain an estimation vector by using a multivariate state estimation technique based on a memory matrix;

[0169] The second judgment module is used to calculate the similarity between the current observation data of the auxiliary device and the estimated vector, and compare it with the second threshold to determine whether the current operating state of the auxiliary device is normal.

[0170] Example 3

[0171] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method for monitoring the operating status of a thermal power unit described in Example 1 are implemented.

[0172] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0174] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0176] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as 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: The thermal power unit includes main equipment and auxiliary equipment, and the operating status monitoring method of the main equipment includes: Obtaining the operating data of the main device, obtaining the true value of the target parameter and the associated parameters of the target parameter, and using the associated parameters of the target parameter as the input parameter sequence; Input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value; Calculating an RMSE value based on the true value of the target parameter and the predicted value of the target parameter, and comparing the RMSE value with a first threshold to determine whether the current operating state of the master device is normal; The operating status monitoring method of the auxiliary equipment includes: Obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and build a memory matrix based on the historical observation data; An estimation vector is obtained by multivariate state estimation technology based on memory matrix calculation; The similarity between the current observation data of the auxiliary device and the estimated vector is calculated and compared with the second threshold to determine whether the current operating state of the auxiliary device is normal.

2. The method for monitoring the operating status of a thermal power unit according to claim 1, wherein: Obtain the operating data of the main device, obtain the actual value of the target parameter and the associated parameters of the target parameter, including: using a parameter representing an operating state of the master device in the operating data of the master device as a target parameter; Calculate the Pearson correlation coefficient between the target parameter and the remaining parameters, and according to the pre-set correlation coefficient threshold, take the parameter with a correlation coefficient greater than the correlation coefficient threshold as the associated parameter of the target parameter.

3. The method for monitoring the operating status of a thermal power unit according to claim 1, wherein: The master device target parameter prediction model includes a temporal convolutional network and a gated recurrent unit, wherein the temporal convolutional network is used to extract features from an input parameter sequence to obtain a high-order feature sequence; The gated recurrent unit is used to capture dynamic changes in high-dimensional feature sequences and obtain target parameter prediction values.

4. The method for monitoring the operating status of a thermal power unit according to claim 3, wherein: The temporal convolutional network includes several extended causal convolutional layers; each extended causal convolutional layer includes an extended causal convolution and a residual connection block; The extended causal convolution introduces a dilation factor to extract features from the input parameter sequence, which is expressed as: ； in, represents the output feature of the extended causal convolution layer at the current time step t, k represents the convolution kernel size, represents the convolution kernel weight, Represents the first Value, i represents the position index of the convolution kernel in the time series; Based on the output features of the dilated causal convolution at the current time step t , the final output sequence of the extended causal convolution is expressed as ,in, represents the final output sequence, Indicates the Output features at time steps; The residual connection block expands the final output sequence of the causal convolution at each layer Added to the input parameter sequence, expressed as: ； Where X represents the input parameter sequence, Represents the residual connection result; Residual connection result of the current extended causal convolution layer As input to the next dilated causal convolutional layer; Several extended causal convolutional layers in the temporal convolutional network perform multi-layer convolution on the input parameter sequence to obtain a high-order feature sequence, which is expressed as: ； in, represents the high-order feature sequence, express The feature vector at each time step.

5. The method for monitoring the operating status of a thermal power unit according to claim 3, wherein: The gated recurrent unit includes an update gate and a reset gate; The update gate is used to fuse the state information of the current time step and the previous time step in the high-order feature sequence, which is expressed as: ； in, represents the output of the update gate, represents the sigmoid activation function, represents the update gate bias term, represents the update gate weight coefficient, Indicates the first feature vectors, Represents the previous time step The hidden state of The reset gate is used to determine the forgotten state information, which is expressed as: ； in, represents the output of the reset gate, Represents the reset gate bias term, Indicates resetting the gate weight coefficient; Calculate the candidate hidden state at the current moment, expressed as: ； in, represents the candidate hidden state at the current moment, represents the activation function, represents the candidate hidden state bias item at the current moment, Indicates the first The weight coefficient of the eigenvector, express The weight coefficient of represents the product of matrices; The final hidden state at the current moment is expressed as: ； in, Represents the final hidden state at the current moment; Based on the final hidden state at the current moment, the target parameter prediction value is obtained through nonlinear transformation, which is expressed as: ； in, represents the predicted value of the target parameter, Represents the final hidden state bias at the current moment, Represents the weight coefficient of the final hidden state at the current moment.

6. The method for monitoring the operating status of a thermal power unit according to claim 1, wherein: Acquire current observation data of the auxiliary device and historical observation data under normal operating conditions, and construct a memory matrix based on the historical observation data, including: The historical observation data of the auxiliary equipment under normal operating conditions is expressed as: ； in, Indicates that auxiliary equipment is The observation sequence at the moment, n represents the n observation points on the auxiliary equipment, Indicates that the nth observation point is The observation value at time, T represents transposition; Select m observation sequences from the historical observation data to construct a memory matrix, which can be expressed as: ； Where D represents the memory matrix, Represents the historical observation data in the nth row and mth column of the memory matrix.

7. The method for monitoring the operating status of a thermal power unit according to claim 6, wherein: The estimated vector is obtained by multivariate state estimation technology based on the memory matrix, including: By calculating the similarity between the current observation data and the historical observation data in the memory matrix, a weight vector W is generated, which is expressed as: ； in, Represents the current observation data, Represents nonlinear operations; Based on the weight vector W, the estimated vector is expressed as: ； in, represents the estimated vector.

8. The method for monitoring the operating status of a thermal power unit according to claim 7, wherein: Calculate the similarity between the current observation data and the estimated vector of the auxiliary device, including: The similarity between the current observation data and the estimated vector of the calculation auxiliary device is expressed as: ； Among them, s represents the similarity between the current observation data and the estimated vector, represents the Euclidean distance between the observation vector and the estimated vector, Represents the current observation data, represents the estimated vector.

9. A thermal power unit operating status monitoring system, characterized in that: The thermal power unit includes main equipment and auxiliary equipment, and the operating status monitoring system of the main equipment includes: A first data acquisition module is configured to acquire operating data of a master device, obtain a true value of a target parameter and associated parameters of the target parameter, and use the associated parameters of the target parameter as an input parameter sequence; The prediction module is used to input the input parameter sequence into the pre-trained master device target parameter prediction model, extract the features in the input parameter sequence, and capture the dynamic changes in the high-dimensional feature sequence to obtain the target parameter prediction value; A first judgment module is used to calculate an RMSE value according to the true value of the target parameter and the predicted value of the target parameter, and compare the RMSE value with a first threshold to determine whether the current operating state of the main device is normal; The operating status monitoring system of the auxiliary equipment includes: The second data acquisition module is used to obtain the current observation data of the auxiliary equipment and the historical observation data under normal operating conditions, and to construct a memory matrix based on the historical observation data; An estimation module, configured to obtain an estimation vector by using a multivariate state estimation technique based on a memory matrix; The second judgment module is used to calculate the similarity between the current observation data of the auxiliary device and the estimated vector, and compare it with the second threshold to determine whether the current operating state of the auxiliary device is normal.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for monitoring the operating status of a thermal power unit according to any one of claims 1 to 8 are implemented.

Images