Gasification furnace fault diagnosis method, electronic equipment and storage medium

Abstract

The invention discloses a gasifier fault diagnosis method, electronic equipment and a storage medium. The method comprises the following steps: acquiring and screening historical operation data of the gasifier; according to the historical operation data and a whale optimization algorithm based on an enhanced search strategy, establishing an auto-encoder, and performing variable optimization on the auto-encoder to obtain a key variable; establishing a fault early warning model based on the key variable; and collecting real-time operation data of the gasifier, inputting the real-time operation data into the fault early warning model to obtain an early warning signal, and performing fault preprocessing according to the early warning signal. According to the method, after variable optimization is carried out on the auto-encoder through the whale optimization algorithm of the enhanced search strategy, key variables of the auto-encoder are obtained, and a fault early warning model is established according to the key variables. Therefore, predictive slag blocking fault diagnosis is achieved through the fault early warning model, corresponding measures are taken in time, and stable operation of the gasifier system is guaranteed.

Description

Technical Field

[0001] The invention relates to the technical field related to chemical production automation, and in particular to a gasifier fault diagnosis method, electronic equipment and storage medium. Background Art

[0002] During the production and operation of the gasifier, due to the difference in coal quality and the uneven operating level of operators, ash and slag are easily accumulated at the slag mouth of the gasifier, forming slag blockage. The commonly used method for detecting slag blockage is to monitor the fluctuation of the pressure difference at the slag mouth of the gasifier. After slag blockage occurs, the pressure difference at the slag mouth of the gasifier fluctuates significantly. The traditional gasifier slag blockage detection method relies on the detection of the slag mouth pressure difference instrument, which has a time delay problem. The detection result is after the slag blockage occurs, and predictive slag blockage fault diagnosis cannot be performed. Summary of the invention

[0003] Based on this, it is necessary to provide a gasifier fault diagnosis method, electronic equipment and storage medium to address the technical problems of the prior art.

[0004] The present invention provides a gasifier fault diagnosis method, comprising:

[0005] Obtain and filter historical operation data of gasifier;

[0006] According to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, an autoencoder is established, and variable optimization is performed on the autoencoder to obtain key variables;

[0007] Based on the key variables, a fault early warning model is established;

[0008] The real-time operation data of the gasifier is collected, the real-time operation data is input into the fault warning model to obtain a warning signal, and the fault preprocessing is performed according to the warning signal.

[0009] Furthermore, the obtaining and screening of the historical operation data of the gasifier includes:

[0010] Obtain historical operation data of the gasifier from the database;

[0011] The historical operation data is screened for abnormal data using a Fourier transform method.

[0012] Furthermore, the method of using Fourier transform to screen abnormal data from the historical operation data includes:

[0013] Transforming the time domain data in the historical operation data into frequency domain data by Fourier transform;

[0014] A filter frequency range is set, and the historical operation data corresponding to the frequency domain data outside the filter frequency range is deleted as abnormal data.

[0015] Furthermore, the whale optimization algorithm of the enhanced search strategy includes:

[0016] The mathematical model of the whale optimization algorithm for building an enhanced search strategy is:

[0017]

[0018]

[0019]

[0020]

[0021]

[0022] Among them, j represents the dimension of the whale sample, i represents the whale number in the population, represents the whale with number i in the jth dimension, represents the optimal whale individual in the jth dimension, represents a whale individual randomly selected from the jth dimension, represents the jump factor of the j-th dimension, represents the global survey distance of the whale, represents the local development distance of the whale, and They represent random numbers that obey normal distribution and uniform distribution respectively, A represents the jump factor, RW j represents the jump factor step size of the jth dimension, t represents the current number of iterations, and t max is the maximum number of iterations.

[0023] Furthermore, the autoencoder is established according to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and the autoencoder is optimized for variables to obtain key variables, including:

[0024] Establishing an autoencoder including an encoding process and a decryption process, wherein the encoding process includes encoding input data of a first dimension and then outputting encoded data of a second dimension, and the decoding process includes decoding the encoded data to obtain decoded data of the first dimension, wherein the first dimension and the second dimension correspond to different variables, and the number of variables corresponding to the first dimension is greater than the number of variables corresponding to the second dimension;

[0025] Establishing a reconstruction loss model of the autoencoder according to the whale optimization algorithm;

[0026] The historical data is input into the autoencoder, and the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest is taken as the optimal dimension, and the variable corresponding to the optimal dimension is taken as the key variable.

[0027] Furthermore, the reconstruction loss model is:

[0028]

[0029] Among them, R loss (f, g) is the reconstruction loss value, f is the encoding process of the autoencoder, g is the decoding process of the autoencoder, n is the number of samples, i is the value of the number of samples, RWOA means using the improved whale optimization algorithm to perform global search optimization on the autoencoder parameters, is the loss function, X i is the true value of the input to the loss function, The predicted value output by the loss function.

[0030] Furthermore, the step of inputting the historical data into the autoencoder and taking the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest as the optimal dimension includes:

[0031] Initializing the autoencoder parameters;

[0032] Perform the following iterative operations:

[0033] The historical data is input into the autoencoder to obtain the decoded data output by the autoencoder, the decoded data and the historical operation data are input into the reconstruction loss model to obtain the reconstruction loss value output by the reconstruction loss model, and when the reconstruction loss value is minimized, the iterative operation is stopped, and the second dimension when the reconstruction loss value output by the reconstruction loss model is minimized is taken as the optimal dimension.

[0034] Furthermore, the establishment of a fault warning model based on the key variables includes:

[0035] Using a random forest algorithm, the trend characteristics of the key variables are sorted from large to small based on importance metrics;

[0036] A fault early warning model is established by using the trend characteristics of the key variables whose importance metrics are greater than a set threshold.

[0037] The present invention provides an electronic device, comprising:

[0038] at least one processor; and,

[0039] a memory communicatively connected to at least one of the processors; wherein,

[0040] The memory stores instructions that can be executed by at least one of the processors. The instructions are executed by at least one of the processors to enable the at least one of the processors to perform the gasifier fault diagnosis method as described above.

[0041] The present invention provides a storage medium, wherein the storage medium stores computer instructions. When a computer executes the computer instructions, the storage medium is used to execute all steps of the gasifier fault diagnosis method as described above.

[0042] The present invention establishes an autoencoder based on the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and performs variable optimization on the autoencoder to obtain key variables; then establishes a fault warning model based on the key variables; finally, collects real-time operation data during the production and operation of the gasifier, inputs the fault warning model to obtain a warning signal, and performs fault preprocessing based on the warning signal. Thus, predictive slag blocking fault diagnosis is achieved, and corresponding measures are taken in time to ensure the stable operation of the gasifier system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a working flow chart of a gasifier fault diagnosis method according to an embodiment of the present invention;

[0044] Figure 2 This is a working flow chart of a gasifier fault diagnosis method according to another embodiment of the present invention;

[0045] Figure 3 A flowchart of a whale optimization algorithm for enhancing search strategy according to another embodiment of the present invention;

[0046] Figure 4 A fault diagnosis flow chart of a fault warning model of a whale optimization algorithm and an autoencoder based on an enhanced search strategy according to another embodiment of the present invention;

[0047] Figure 5 A warning signal diagram output by a fault warning model in another embodiment of the present invention;

[0048] Figure 6 The figure is a schematic diagram of the hardware structure of an electronic device of the present invention. DETAILED DESCRIPTION

[0049] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. The same components are represented by the same reference numerals. It should be noted that the words "front", "rear", "left", "right", "upper" and "lower" used in the following description refer to directions in the accompanying drawings, and the words "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0050] like Figure 1The figure shows a working flow chart of a gasifier fault diagnosis method according to an embodiment of the present invention, comprising:

[0051] Step S101, obtaining and filtering historical operation data of the gasifier;

[0052] Step S102, establishing an autoencoder according to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and performing variable optimization on the autoencoder to obtain key variables;

[0053] Step S103, establishing a fault warning model based on the key variables;

[0054] Step S104, collecting real-time operation data of the gasifier, inputting the real-time operation data into the fault warning model to obtain a warning signal, and performing fault preprocessing according to the warning signal.

[0055] Specifically, step S101 completes the collection of historical operating timing data of the gasifier from the database, and then eliminates abnormal data to implement data filtering to maintain the validity of the original data; step S102 establishes an autoencoder based on the historical operating data and the whale optimization algorithm based on the enhanced search strategy, and then uses the whale optimization algorithm with the enhanced search strategy to perform variable optimization on the autoencoder to obtain key variables that can respond to the slag blockage of the gasifier in advance; step S103 establishes a gasifier slag blockage fault warning model based on the key variables; step S104 collects real-time operating data of the gasifier, inputs the real-time operating data into the fault warning model to obtain a warning signal, and performs fault preprocessing according to the warning signal.

[0056] The present invention establishes a fault warning model of a gasifier based on a whale optimization algorithm and an autoencoder based on an enhanced search strategy, optimizes the fault warning model by using the whale optimization algorithm based on an enhanced search strategy, collects operation data in real time and inputs it into the fault warning model after variable optimization to obtain a real-time fault warning signal. This enables timely acquisition of fault warning signals of the gasifier operation, timely intervention of corresponding production conditions to avoid slag blockage in the gasifier, and ensures stable operation of the gasifier.

[0057] like Figure 2 The figure shows a working flow chart of a gasifier fault diagnosis method in another embodiment of the present invention, comprising:

[0058] Step S201, obtaining and filtering historical operation data of the gasifier.

[0059] In one of the embodiments, the obtaining and screening of the historical operation data of the gasifier includes: obtaining the historical operation data of the gasifier from a database; and screening the historical operation data for abnormal data using a Fourier transform method.

[0060] Specifically, the historical operation time series data of the gasifier are collected from the database, but the missing values ​​and outliers in the data will affect the accuracy of the model. The Fourier transform method is used to screen the abnormal data. First, the non-stationary time series data (that is, the time series data with no obvious regularity in dynamic changes) is Fourier transformed, and the mathematical expectation and covariance function of the whole data are calculated. The spectrum analysis error of the analytical signal and the original signal is established, and the abnormal data with spectrum fluctuation greater than the set value is eliminated to achieve data filtering and maintain the validity of the original data. For the stationary time series data, there is no need to perform Fourier transform and it can be directly used as valid data.

[0061] This embodiment achieves the screening of effective historical operation data of the gasifier.

[0062] In one of the embodiments, the method of using Fourier transform to filter abnormal data from the historical operation data includes: transforming the time domain data in the historical operation data into frequency domain data using Fourier transform; setting a filter frequency range, and deleting the historical operation data corresponding to the frequency domain data outside the filter frequency range as abnormal data.

[0063] Specifically, the historical operation data is first subjected to Fourier transformation to be transformed into time domain data, satisfying the following conditions:

[0064] E(|X(t) 2 |)<∞

[0065] E(|X(t)|)=m

[0066] C(X(t1),X(t2))=C(t1-t2)

[0067] Among them, X() represents time domain data, E() represents the mathematical expectation of the entire data, C() represents the covariance function, m represents a constant, t represents time, and t1 and t2 represent different time points;

[0068] The time domain data is transformed into frequency domain data using Fourier transform, satisfying the following conditions:

[0069]

[0070]

[0071] Where N represents the number of data, x(n) is a non-periodic function, X(ω) is the representation of the function in the frequency domain, and e -jωn and They are all complex exponential functions, ω is the angular frequency;

[0072] The collected time domain data is synchronously converted into frequency domain data, that is, the frequency domain distribution of the original data is obtained using Fourier transform, the frequency domain analytical signal distribution is established, the frequency error is analyzed, the filter frequency range is set, and data filtering is completed; data that exceeds the set filter frequency range is filtered out, and the data within the set filter frequency range is left as valid data.

[0073] This embodiment uses Fourier transform to filter the historical operation data.

[0074] Step S202, establishing an autoencoder according to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and performing variable optimization on the autoencoder to obtain key variables.

[0075] In one embodiment, the whale optimization algorithm for strengthening the search strategy includes: constructing a mathematical model of the whale optimization algorithm for strengthening the search strategy as follows:

[0076]

[0077]

[0078]

[0079]

[0080]

[0081] Among them, j represents the dimension of the whale sample, i represents the whale number in the population, represents the whale with number i in the jth dimension, represents the optimal whale individual in the jth dimension, represents a whale individual randomly selected from the jth dimension, represents the jump factor of the j-th dimension, represents the global survey distance of the whale, represents the local development distance of the whale, and They represent random numbers that obey normal distribution and uniform distribution respectively, A represents the jump factor, RW j represents the jump factor step size of the jth dimension, t represents the current number of iterations, and t max is the maximum number of iterations.

[0082] Specifically, according to the logic of the whale optimization algorithm, local encirclement and global search for prey position update are enhanced, and a jump step factor is created to avoid the algorithm from completing local convergence too early; the whale optimization algorithm can be abstractly summarized into three strategies, namely local encirclement, bubble net attack and global search. The mathematical models of local encirclement and bubble net attack are introduced in detail as follows:

[0083] The local enclosure model is: X(t+1)=X*(t)-A·|C·X*(t)-X(t)|;

[0084] The attack model of Bubble Network is: X(t+1)=D′·e bl ·cos(2·π·l)+X*(t);

[0085] When a whale approaches its prey, it conducts local encirclement and bubble net attacks simultaneously. Assuming that the whale has a 50% probability of conducting local encirclement and bubble net attacks, the mathematical model is described as follows:

[0086]

[0087] When performing a global search, when |A|≥1, the whale population no longer updates its position based on the current optimal solution, but updates its position based on a randomly selected whale, in order to increase the search range and find the optimal solution to maintain the diversity of the population. Therefore, the global search model is:

[0088] X(t+1)=X rand (t)-A·|C·X rand (t)-X(t)|

[0089] Where X represents the search position, X* represents the current optimal solution position, t represents the current iteration number, A and C are constant vectors used to control the whale's swimming mode, D′=|X*(t)-X(t)| represents the distance between the current whale individual and the current optimal individual, b is a constant that determines the spiral shape, l is a random number distributed in [-1,1], X rand Represents a randomly selected search position, where p is a random number between [-1,1].

[0090] In order to make the whale algorithm converge faster and more accurately, different evolutionary strategies are integrated into different dimensions of the algorithm to improve the convergence accuracy of the whale algorithm. The whale algorithm is subjected to enhanced search for local encirclement and global search of prey position updates. The mathematical model of the whale optimization algorithm with enhanced search strategy is constructed as follows:

[0091]

[0092]

[0093]

[0094]

[0095]

[0096] Among them, j represents the dimension of the whale sample, i represents the whale number in the population, represents the whale with number i in the jth dimension, represents the optimal whale individual in the jth dimension, represents a whale individual randomly selected from the jth dimension, represents the jump factor of the j-th dimension, represents the global survey distance of the whale, represents the local development distance of the whale, and They represent random numbers that obey normal distribution and uniform distribution respectively, A represents the jump factor, RW j represents the jump factor step size of the jth dimension, t represents the current number of iterations, and t max is the maximum number of iterations, (2-t / t max ) indicates that the data structure decreases linearly from 2 to 1.

[0097] This embodiment creates a jump step factor to prevent the algorithm from completing local convergence too early, improves the stability and reliability of the algorithm, further enhances the computing power of the algorithm, uses an autoencoder to reduce the dimension of data, and uses an improved whale optimization algorithm to perform global optimization of network parameters, extracts hidden variables from input variables into a latent space, and the hidden variables are the key variables after dimensionality reduction.

[0098] like Figure 3 The flowchart of the whale optimization algorithm for strengthening the search strategy of another embodiment of the present invention is shown. Based on the above introduction, the implementation steps of the whale optimization algorithm are as follows: step S301, initialization, whale population size N, maximum number of iterations, position of each individual in the whale population; step S302, calculate the fitness value of each individual, record the optimal fitness value and the corresponding position vector; step S303, update the algorithm control parameters; step S304, when p≥0.5, update the individual position of the whale through the bubble network attack model; when p<0.5 and |A|≤1, update the individual position of the whale through formula (1); when p<0.5 and |A|>1, update the individual position of the whale through formula (2); step S305, check whether the newly generated individual exceeds the search limit, if so, initialize the new position; calculate the fitness of all individuals, and update the current optimal solution; step S306, if the termination condition is met, output the optimal individual, that is, the optimal solution found by the algorithm, otherwise, return to step S303.

[0099] Preferably, through correlation analysis of parameter variables, key variables characterizing the slag blocking phenomenon are found to include gasification furnace slag port pressure difference, CO content, methane content and water-coal slurry concentration.

[0100] In one embodiment, the autoencoder is established according to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and the autoencoder is optimized for variables to obtain key variables, including:

[0101] Establishing an autoencoder including an encoding process and a decryption process, wherein the encoding process includes encoding input data of a first dimension and then outputting encoded data of a second dimension, and the decoding process includes decoding the encoded data to obtain decoded data of the first dimension, wherein the first dimension and the second dimension correspond to different variables, and the number of variables corresponding to the first dimension is greater than the number of variables corresponding to the second dimension;

[0102] Establishing a reconstruction loss model of the autoencoder according to the whale optimization algorithm;

[0103] The historical data is input into the autoencoder, and the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest is taken as the optimal dimension, and the variable corresponding to the optimal dimension is taken as the key variable.

[0104] Specifically, the autoencoder (AE) is an unsupervised feature dimension reduction and feature expression method, which consists of an encoder and a decoder. It is a neural network with the same input and training target. The parameters of the AE are trained through reconstruction loss. The whale optimization algorithm (RWOA) with enhanced search strategy is used to perform global search optimization on the variables of the AE neural network, thereby reducing the dimension of the input variables into key variables, saving the key variables in the latent space, and realizing data reconstruction, so that the reconstruction loss reaches the minimum value. By reducing the dimension of the input vector, the complexity of the clustering algorithm is reduced; data visualization is achieved to facilitate the observation of the distribution characteristics of the data; and the pressure of data transmission is reduced. Data reconstruction can be achieved by simply transferring the trained network parameters and the reduced-dimensional data.

[0105] This embodiment obtains key variables by performing variable optimization on the autoencoder.

[0106] In one embodiment, the reconstruction loss model is:

[0107]

[0108] Among them, R loss(f, g) is the reconstruction loss value, f is the encoding process of the autoencoder, g is the decoding process of the autoencoder, n is the number of samples, i is the value of the number of samples, RWOA means using the improved whale optimization algorithm to perform global search optimization on the autoencoder parameters, is the loss function, X i is the true value of the input to the loss function, The predicted value output by the loss function.

[0109] Specifically, the autoencoder includes an encoder and a decoder. The encoder encodes the input variables of the autoencoder and then reduces the dimension to output to the latent space. Then, the decoder decodes the variables in the latent space and outputs the predicted variables. The reconstruction loss model calculates the mean square error between the output predicted variables and the input original variables. For each encoding method, the mean square error of the encoding method is calculated through n samples. That is, for each encoding method, n original variables are input to obtain n predicted variables, thereby obtaining the mean square error of the encoding method.

[0110] Among them, a group of input variables can be represented by an input data. After the autoencoder encodes the input data, the encoded data corresponding to the encoding method is obtained to represent a group of variables. The encoded data is stored in the latent space, that is, the input variables are encoded and then output to the latent space in a reduced dimension. Then the decoder decrypts the encoded data to obtain the predicted data used to represent the predicted variables. The mean square error is calculated for the input data and the predicted data.

[0111] For example, binary data can be used to represent a set of input variables. As an example, for the input variables of the gasification furnace slag port pressure difference, CO content, methane content and water-coal slurry concentration, a four-bit binary value can be used to represent it, and each bit of the binary data represents whether the corresponding variable is selected. The input data used to represent the input variable is 1111, indicating that the gasification furnace slag port pressure difference, CO content, methane content and water-coal slurry concentration are all selected. After being encoded by the encoder, the selected variable of a coding method is represented by the coded data 0101, indicating that the CO content and water-coal slurry concentration are selected, but the gasification furnace slag port pressure difference and water-coal slurry concentration are not selected.

[0112] For each encoding method and the corresponding mean square error, the whale optimization algorithm is input. The encoded data of each encoding method is used as a whale sample. The fitness value of each whale sample is the mean square error of the encoding method. The objective function is to minimize the mean square error, that is, the fitness value. The distance of the whale sample is to calculate the distance between the encoded data corresponding to the two encoding methods, for example, to calculate the Euclidean distance between the two encoded data. Through the whale optimization algorithm, the optimal whale sample is obtained, that is, the whale sample corresponding to the minimum mean square error. When the mean square error is the smallest, the encoding method of the corresponding autoencoder is the optimal encoding method, and the variables stored in the hidden space of the autoencoder corresponding to the optimal encoding method are the optimal variables. Through the iterative calculation of the whale optimization algorithm, the minimum mean square error can be quickly found, thereby obtaining the optimal variable of the autoencoder, and the optimal variable is the key variable to be found.

[0113] This embodiment implements modeling for optimizing autoencoder parameters.

[0114] In one embodiment, inputting the historical data into the autoencoder and taking the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest as the optimal dimension includes:

[0115] Initializing the autoencoder parameters;

[0116] Perform the following iterative operations:

[0117] The historical data is input into the autoencoder to obtain the decoded data output by the autoencoder, the decoded data and the historical operation data are input into the reconstruction loss model to obtain the reconstruction loss value output by the reconstruction loss model, and when the reconstruction loss value is minimized, the iterative operation is stopped, and the second dimension when the reconstruction loss value output by the reconstruction loss model is minimized is taken as the optimal dimension.

[0118] Specifically, initialize the autoencoder parameters, set the maximum number of iterations of the whale optimization algorithm, the population size, and other control parameters; use the autoencoder to calculate the model output, and calculate the mean square error (MSE) between the model output and the actual data. The RWOA algorithm uses the mean square error as the objective function to search for the best fault warning model parameters, so as to obtain the current best fault warning model parameters. The input parameter of the mean square error is the optimal whale individual of RWOA. The process of RWOA algorithm using mean square error as the objective function for optimization is as follows: use the initial value to obtain the first mean square error, and then perform the first optimization to obtain the second mean square error. The two mean square errors are compared and the smallest one is used to replace the current optimal value. Continue to iterate and optimize until the minimum mean square error is found. At this time, the optimal whale individual of RWOA is the optimal variable of the autoencoder. Figure 4The figure shows a fault diagnosis flow chart of another embodiment of the present invention based on the whale optimization algorithm of the enhanced search strategy and the fault warning model of the autoencoder. The autoencoder 401 reduces the dimension of the input variables of the gasifier into key variables, reduces the dimension of the input vector, and thus reduces the complexity of the clustering algorithm. The input variables have a certain structure, so the encoding work converts the complete input space into a meaningful lower dimension from this structure to achieve dimensionality reduction; realizes data visualization to facilitate observation of the distribution characteristics of the data; reduces the pressure of data transmission, and only needs to transfer the trained network parameters and the reduced-dimensional data to achieve data reconstruction. At the same time, the whale optimization algorithm 402 of the enhanced search strategy optimizes the encoded data of the autoencoder 401, and finally outputs the fault warning signal after processing and judgment by the fault warning model.

[0119] This embodiment reconstructs the loss model to reduce the dimensionality of the input variables and then uses the whale optimization algorithm to find the optimal encoding method and optimal dimension.

[0120] Step S203, using a random forest algorithm to sort the trend characteristics of the key variables from large to small based on the importance metric; using the trend characteristics of the key variables whose importance metric is greater than a set threshold, to establish a fault warning model.

[0121] Specifically, the random forest algorithm is used to measure the importance of trend characteristics of key variables, and the trend characteristics ranked top in importance are selected to establish a slag blockage fault early warning model.

[0122] Preferably, 14 numerical trend features of key variables characterizing the slag blocking phenomenon are extracted, including: mean, variance, standard deviation, extreme value, dispersion coefficient, root mean square value, peak index, peak factor, kurtosis factor, kurtosis factor, root amplitude, margin coefficient, pulse factor, and waveform index. The 14 trend features of key variables are selected by importance measurement using the random forest algorithm, and the top 7 trend features are selected to establish a slag blocking fault warning model. Table 1 shows the random forest importance ranking of the 14 trend features of key variables of the gasifier.

[0123] Table 1

[0124]

[0125]

[0126] Preferably, 10 slag blocking faults and 20 normal operating conditions occurring in the historical data are selected, and feature annotation of the fault and normal operating conditions is performed, wherein the fault condition is labeled "1" and the normal operating condition is labeled "0", and the length of the data acquisition time period of the key variables is set to 30 minutes, and the trend feature extraction and selection of step S3 is completed. From the selected 30 operating conditions (10 common faults and 20 normal operating conditions), 24 operating conditions are randomly selected as training set data and 6 operating conditions are used as test set data. The training set data is sent to the random forest model for model training, wherein the input variable is the trend features ranked top 7 in importance of the key variables, and the output variable is the labeled value of the operating condition. The test set data is sent to the trained model, and the test results are output to verify the effectiveness of the slag blocking early warning model.

[0127] Step S204, collecting real-time operating data of the gasifier, inputting the real-time operating data into the fault warning model to obtain a warning signal, and performing fault preprocessing according to the warning signal.

[0128] Specifically, the key variable data during the operation of the gasifier are transmitted to the slag blocking fault warning model in real time. If there are signs of slag blocking, the model will output a warning signal in time to ensure the stable operation of the gasifier.

[0129] This embodiment realizes fault warning of the real-time operating condition of the gasifier.

[0130] The present invention trains and tests the autoencoder by using the filtered historical operation data, and then optimizes the parameters of the autoencoder according to the whale optimization algorithm with enhanced search strategy to obtain the key variables of the autoencoder; a fault warning model is established based on these key variables, and then the key variables in the production and operation process of the gasifier are collected in real time, and input into the fault warning model in real time, and the fault warning model outputs a warning signal in real time. Thus, predictive slag blocking fault diagnosis can be achieved, and corresponding measures can be taken in time to ensure the stable operation of the gasifier system.

[0131] Another embodiment

[0132] The operation data of October 25, 2021 was collected from the production database. The collection time interval was 1 minute and the time window was 10 minutes. The maximum number of iterations of all algorithms was set to 300, and the whale population size was set to 30. The encoder and decoder in the autoencoder model were built using long short-term memory network (LSTM) neurons. The key variables were selected as the gasification furnace slag mouth pressure difference, CO content, methane content, water-coal slurry concentration, mean, variance, standard deviation, extreme value, dispersion coefficient, root mean square value, and peak index as the model fault warning input variables, and the data was delivered to the trained fault warning model. Figure 5The figure shows the output warning signal of the fault warning model of this embodiment. The fault warning model outputs the slag blocking alarm signal at 22:00 on the same day. The gasifier device actually found the slag blocking and performed the oxygen extraction operation at 23:09. It can be seen that the fault warning model issued a warning signal one hour before the slag blocking occurred.

[0133] like Figure 6 The figure shows a hardware structure diagram of an electronic device of the present invention, including:

[0134] at least one processor 601; and,

[0135] A memory 602 is communicatively connected to at least one of the processors 601; wherein,

[0136] The memory 602 stores instructions that can be executed by at least one of the processors. The instructions are executed by at least one of the processors so that the at least one of the processors can perform the gasifier fault diagnosis method as described above.

[0137] Figure 6 A processor 601 is taken as an example.

[0138] The electronic device may further include: an input device 603 and a display device 604 .

[0139] The processor 601, the memory 602, the input device 603 and the display device 604 may be connected via a bus or other means, and the figure takes the connection via a bus as an example.

[0140] The memory 602 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the gasifier fault diagnosis method in the embodiment of the present application, for example, Figure 1 , Figure 2 , Figure 3 and Figure 4 The processor 601 executes various functional applications and data processing by running the non-volatile software programs, instructions and modules stored in the memory 602, that is, the gasifier fault diagnosis method in the above embodiment is implemented.

[0141] The memory 602 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the gasifier fault diagnosis method, etc. In addition, the memory 602 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 602 may optionally include a memory remotely arranged relative to the processor 601, and these remote memories may be connected to the device for executing the gasifier fault diagnosis method via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0142] The input device 603 can receive user clicks and generate signal inputs related to user settings and function controls of the gasifier fault diagnosis method. The display device 604 can include display devices such as a display screen.

[0143] The one or more modules are stored in the memory 602 , and when executed by the one or more processors 601 , the gasifier fault diagnosis method in any of the above method embodiments is executed.

[0144] The present invention trains and tests the autoencoder by using the screened historical operation data, and then optimizes the parameters of the autoencoder according to the whale optimization algorithm of the enhanced search strategy to obtain the key variables of the autoencoder; a fault warning model is established based on these key variables, and then the key variables in the production and operation process of the gasifier are collected in real time, and input into the fault warning model in real time, and the fault warning model outputs a warning signal in real time. In this way, the real-time working conditions of the gasifier operation process can be discovered in time, and corresponding warnings and interventions can be made in time, especially for the slag blocking fault of the gasifier, timely warning and processing can be carried out to ensure the smooth operation of the gasifier.

[0145] An embodiment of the present invention provides a storage medium, wherein the storage medium stores computer instructions. When a computer executes the computer instructions, the storage medium is used to execute all steps of the gasifier fault diagnosis method as described above.

[0146] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A gasifier fault diagnosis method, characterized in that: include: Obtain and filter historical operation data of gasifier; According to the historical operation data and the whale optimization algorithm based on the enhanced search strategy, an autoencoder is established, and variable optimization is performed on the autoencoder to obtain key variables; Based on the key variables, a fault early warning model is established; The real-time operation data of the gasifier is collected, the real-time operation data is input into the fault warning model to obtain a warning signal, and the fault preprocessing is performed according to the warning signal.

2. The gasifier fault diagnosis method according to claim 1, characterized in that: The obtaining and screening of the historical operation data of the gasifier includes: Obtain historical operation data of the gasifier from the database; The historical operation data is screened for abnormal data using a Fourier transform method.

3. The gasifier fault diagnosis method according to claim 2, characterized in that: The method of using Fourier transform to screen abnormal data from the historical operation data includes: Transforming the time domain data in the historical operation data into frequency domain data by Fourier transform; A filter frequency range is set, and the historical operation data corresponding to the frequency domain data outside the filter frequency range is deleted as abnormal data.

4. The gasifier fault diagnosis method according to claim 1, characterized in that: The whale optimization algorithm of the enhanced search strategy includes: The mathematical model of the whale optimization algorithm for building an enhanced search strategy is: Among them, j represents the dimension of the whale sample, i represents the whale number in the population, represents the whale with number i in the jth dimension, represents the optimal whale individual in the jth dimension, represents a whale individual randomly selected from the jth dimension, represents the jump factor of the j-th dimension, represents the global survey distance of the whale, represents the local development distance of the whale, and They represent random numbers that obey normal distribution and uniform distribution respectively, A represents the jump factor, RW j represents the jump factor step size of the jth dimension, t represents the current number of iterations, and t max is the maximum number of iterations.

5. The gasifier fault diagnosis method according to claim 1, characterized in that: The method of establishing an autoencoder based on the historical operation data and the whale optimization algorithm based on the enhanced search strategy, and optimizing the autoencoder to obtain key variables includes: Establishing an autoencoder including an encoding process and a decryption process, wherein the encoding process includes encoding input data of a first dimension and then outputting encoded data of a second dimension, and the decoding process includes decoding the encoded data to obtain decoded data of the first dimension, wherein the first dimension and the second dimension correspond to different variables, and the number of variables corresponding to the first dimension is greater than the number of variables corresponding to the second dimension; Establishing a reconstruction loss model of the autoencoder according to the whale optimization algorithm; The historical data is input into the autoencoder, and the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest is taken as the optimal dimension, and the variable corresponding to the optimal dimension is taken as the key variable.

6. The gasifier fault diagnosis method according to claim 5, characterized in that: The reconstruction loss model is: Among them, R loss (f, g) is the reconstruction loss value, f is the encoding process of the autoencoder, g is the decoding process of the autoencoder, n is the number of samples, i is the value of the number of samples, RWOA means using the improved whale optimization algorithm to perform global search optimization on the autoencoder parameters, is the loss function, X i is the true value of the input to the loss function, The predicted value output by the loss function.

7. The gasifier fault diagnosis method according to claim 5, characterized in that: The step of inputting the historical data into the autoencoder and taking the second dimension when the reconstruction loss value output by the reconstruction loss model is the smallest as the optimal dimension includes: Initializing the autoencoder parameters; Perform the following iterative operations: The historical data is input into the autoencoder to obtain the decoded data output by the autoencoder, the decoded data and the historical operation data are input into the reconstruction loss model to obtain the reconstruction loss value output by the reconstruction loss model, and when the reconstruction loss value is minimized, the iterative operation is stopped, and the second dimension when the reconstruction loss value output by the reconstruction loss model is minimized is taken as the optimal dimension.

8. The gasifier fault diagnosis method according to claim 1, characterized in that: The establishing of a fault warning model based on the key variables includes: Using a random forest algorithm, the trend characteristics of the key variables are sorted from large to small based on importance metrics; A fault early warning model is established by using the trend characteristics of the key variables whose importance metrics are greater than a set threshold.

9. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors to enable at least one of the processors to execute the gasifier fault diagnosis method according to any one of claims 1 to 8.

10. A storage medium, characterized in that: The storage medium stores computer instructions, and when a computer executes the computer instructions, it is used to execute all steps of the gasifier fault diagnosis method according to any one of claims 1 to 8.

Images