A method and system for analyzing and predicting faults of a heating furnace operation data

Abstract

The application discloses a heating furnace operation data analysis and fault prediction method and system, in particular to the technical field of fault prediction; is through the nozzle state parameter and fuel flow data of heating furnace, analyzes the correlation characteristics of nozzle state fluctuation and fuel non-uniform mixing, extracts the fuel distribution non-uniformity index; based on the infrared thermal imaging data of the hearth temperature field, the drift trajectory of hot spots is identified, the hot spot drift rate and direction prediction result is obtained; combined with the fuel distribution non-uniformity index, the local overheating risk probability is calculated, the heating furnace risk grade atlas is constructed, and the high-risk area is identified, the abnormal working condition type is matched by combining the historical fault database, the fault early warning instruction containing the hot spot drift path prediction is generated; and according to the fault early warning instruction, the fuel injection pressure and nozzle opening of the fault nozzle are adjusted in real time, the effective identification and risk assessment of hot spot drift trend are realized, and the precision of heating furnace fault prediction and the timeliness of response are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction, and more particularly, to a heating furnace operation data analysis and fault prediction method and system. BACKGROUND

[0002] During the continuous operation of the heating furnace, the instability of the nozzle state can cause non-uniform mixing of fuel distribution, thereby causing frequent generation of local hot spots in the furnace interior and presenting a drifting distribution.

[0003] The existing operation data analysis method usually relies on static or linear models, which is difficult to dynamically track the hot spot change trend, resulting in difficulty in accurately depicting and early warning the hot spot behavior, and affecting the accuracy of fault prediction and the timeliness of response. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a heating furnace operation data analysis and fault prediction method and system to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A heating furnace operation data analysis and fault prediction method, comprising the following steps:

[0007] S1: obtaining real-time operation data of the heating furnace, performing noise filtering and feature standardization processing, and generating a preprocessed operation data set;

[0008] S2: based on the preprocessed operation data set, analyzing the correlation characteristics of nozzle state fluctuation and non-uniform mixing of fuel, and generating a fuel distribution non-uniformity index;

[0009] S3: based on the preprocessed operation data set, using a spatio-temporal convolutional neural network to identify the drift trajectory of the hot spot, and generating a hot spot drift rate and direction prediction result;

[0010] S4: based on the fuel distribution non-uniformity index and the hot spot drift rate and direction prediction result, calculating a local overheating risk probability, and generating a heating furnace risk level atlas;

[0011] S5: according to the heating furnace risk level atlas, dividing a high-risk area, matching an abnormal working condition type in combination with a historical fault database, and generating a fault warning instruction containing hot spot drift path prediction;

[0012] S6: based on the fault warning instruction, adjusting the injection pressure and nozzle opening degree of the fault nozzle in real time.

[0013] In a preferred embodiment, S1, in particular:

[0014] Real-time acquisition of the nozzle state parameters of the heating furnace, the infrared thermal imaging data of the furnace temperature field, and the fuel flow data;

[0015] The collected nozzle state parameters are subjected to missing value processing and abnormal value filtering, the infrared thermal imaging data of the furnace temperature field are subjected to noise filtering, data smoothing and pixel-level feature extraction, the fuel flow data are subjected to stationarity analysis and abnormal data interpolation processing, and then feature standardization processing is performed, thereby generating a preprocessed operation data set containing the nozzle state parameters, the infrared thermal imaging data of the furnace temperature field, and the fuel flow data.

[0016] In a preferred embodiment, S2, specifically:

[0017] Based on the nozzle state parameters and the fuel flow data, nozzle injection pressure fluctuation features, nozzle opening change features and fuel flow change features are extracted;

[0018] The correlation strength between the nozzle injection pressure fluctuation features and the fuel flow change features is analyzed through time series correlation calculation;

[0019] The frequency coupling degree between the nozzle opening change features and the fuel flow change features is obtained by using a frequency spectrum analysis method;

[0020] According to the correlation strength and the frequency coupling degree, a fuel distribution non-uniformity index is determined by using a principal component analysis method.

[0021] In a preferred embodiment, S3, specifically:

[0022] The infrared thermal imaging data of the furnace temperature field are subjected to grid-based regional segmentation by using a spatial sliding window division method, and the local hotspot image sequence obtained by the grid-based regional segmentation is input into a spatio-temporal convolutional neural network;

[0023] The spatial features of the local hotspot image sequence are extracted layer by layer by using a spatial convolutional layer, thereby obtaining the spatial position distribution features of the hotspots;

[0024] The spatial position distribution features are subjected to time series feature extraction by using a time convolutional layer, thereby identifying the drift change law of the hotspot positions;

[0025] According to the drift change law of the hotspot positions, the position change distance and angle of the hotspots within a unit time are calculated, thereby generating hotspot drift rate and direction prediction results.

[0026] In a preferred embodiment, S4, specifically:

[0027] The spatial correlation coefficient between the fuel distribution non-uniformity index and the hotspot drift rate is calculated;

[0028] Calculate the time coupling degree coefficient between the fuel distribution non-uniformity index and the hotspot drift direction;

[0029] According to the spatial correlation coefficient and the time coupling degree coefficient, the local overheating risk probability of the local area where the hotspot is located is calculated through a dynamic weight distribution algorithm;

[0030] The local overheating risk probability of the local area where the hotspot is located is classified into risk levels to generate a heating furnace risk level map.

[0031] In a preferred embodiment, S5, specifically:

[0032] According to the heating furnace risk level map, the local area where the hotspot is located in the hearth is divided into a high-risk area when the local overheating risk probability exceeds a preset risk probability threshold;

[0033] Based on the high-risk area, the abnormal working condition type in the historical fault database is searched to obtain an abnormal working condition type matching result corresponding to the local overheating risk probability;

[0034] According to the abnormal working condition type matching result, the hotspot drift rate and the hotspot drift direction prediction result are mapped into the hearth temperature field to generate a fault warning instruction containing hotspot position drift path prediction.

[0035] In a preferred embodiment, S6, specifically:

[0036] According to the hotspot position drift path prediction in the fault warning instruction, a fault nozzle associated with the hotspot position drift path is determined;

[0037] Based on the fault nozzle associated with the hotspot position drift path, a target adjustment value of the nozzle injection pressure and a target adjustment value of the nozzle opening degree are determined;

[0038] The target adjustment value of the nozzle injection pressure and the real-time collected nozzle injection pressure are calculated by difference to generate a nozzle injection pressure adjustment control amount;

[0039] The target adjustment value of the nozzle opening degree and the real-time collected nozzle opening degree are calculated by difference to generate a nozzle opening degree adjustment control amount;

[0040] According to the nozzle injection pressure adjustment control amount and the nozzle opening degree adjustment control amount, the injection pressure and the nozzle opening degree of the fault nozzle are adjusted in real time.

[0041] On the other hand, the present application provides a heating furnace operation data analysis and fault prediction system, comprising:

[0042] The data acquisition module: acquires the real-time operation data of the heating furnace, removes noise and performs feature standardization processing to generate a preprocessed operation data set;

[0043] Feature extraction module: based on the pre-processed operation data set, analyze the correlation characteristics of nozzle state fluctuation and fuel non-uniform mixing, and generate fuel distribution non-uniformity index;

[0044] Trajectory recognition module: based on the pre-processed operation data set, identify the drift trajectory of hot spots by using a spatio-temporal convolutional neural network, and generate hot spot drift rate and direction prediction results;

[0045] Risk assessment module: based on the fuel distribution non-uniformity index and the hot spot drift rate and direction prediction results, calculate the local overheating risk probability, and generate the heating furnace risk level atlas;

[0046] Early warning generation module: according to the heating furnace risk level atlas, divide the high-risk area, match the abnormal working condition type combined with the historical fault database, and generate the fault early warning instruction containing the hot spot drift path prediction;

[0047] Nozzle control module: based on the fault early warning instruction, real-time adjust the injection pressure and nozzle opening of the fault nozzle.

[0048] The technical effects and advantages of the heating furnace operation data analysis and fault prediction method and system of the present application are:

[0049] By obtaining the heating furnace nozzle state parameters, the furnace temperature field infrared thermal imaging data and the fuel flow data, the pre-processed operation data set is constructed, the original data quality and the spatio-temporal consistency are improved; by extracting the correlation characteristics of nozzle state fluctuation and fuel non-uniform mixing, the fuel distribution non-uniformity index is quantified, the expression of nozzle disturbance effect is realized; the spatio-temporal convolutional neural network is introduced to identify the hot spot drift trajectory, the identification of hot spot dynamic evolution trend is improved; the local overheating risk probability is calculated combined with the fuel distribution characteristics and the hot spot drift rate and direction prediction results, the heating furnace risk level atlas is established, the quantitative expression and regional positioning of abnormal risk are realized; by matching the corresponding abnormal working condition type in the historical fault database, the fault early warning instruction containing the hot spot drift path is generated, the pertinence of fault identification is improved; by adjusting the injection pressure and opening parameters of the fault nozzle, the intervention control of hot spot risk path is completed, the identification and precise intervention of heating furnace hot spot migration risk are realized, the precision and response efficiency of fault prediction are improved. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The figure is a schematic diagram of the heating furnace operation data analysis and fault prediction method of the present application;

[0051] Figure 2 The figure is a structural schematic diagram of the heating furnace operation data analysis and fault prediction system of the present application. DETAILED DESCRIPTION

[0052] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0053] Embodiment 1

[0054] Figure 1 A heating furnace operation data analysis and fault prediction method is given, which comprises the following steps:

[0055] S1: acquiring real-time operation data of the heating furnace, performing noise filtering and feature standardization processing, and generating a preprocessed operation data set;

[0056] S2: based on the preprocessed operation data set, analyzing the correlation characteristics of nozzle state fluctuation and fuel non-uniform mixing, and generating a fuel distribution non-uniformity index;

[0057] S3: based on the preprocessed operation data set, using a spatio-temporal convolutional neural network to identify the drift trajectory of the hot spot, and generating a hot spot drift rate and direction prediction result;

[0058] S4: based on the fuel distribution non-uniformity index and the hot spot drift rate and direction prediction result, calculating a local overheating risk probability, and generating a heating furnace risk level atlas;

[0059] S5: according to the heating furnace risk level atlas, dividing a high-risk area, matching an abnormal working condition type in combination with a historical fault database, and generating a fault warning instruction containing a hot spot drift path prediction;

[0060] S6: based on the fault warning instruction, adjusting the injection pressure and nozzle opening degree of the fault nozzle in real time.

[0061] S1: acquiring real-time operation data of the heating furnace, performing noise filtering and feature standardization processing, and generating a preprocessed operation data set, comprising:

[0062] real-time acquisition of nozzle state parameters, furnace temperature field infrared thermal imaging data, and fuel flow data of the heating furnace;

[0063] The nozzle state parameters are acquired in real time by a pressure sensor and an opening degree sensor installed at the nozzle valve. The nozzle state parameters include fuel injection pressure of the nozzle and actual opening degree of the nozzle. The acquisition frequency of the nozzle state parameters is set to one per second, forming time series data of the nozzle state parameters.

[0064] The infrared thermal imaging sensor array installed on the top of the furnace chamber and the wall around the furnace chamber is used to collect the infrared thermal imaging data of the furnace temperature field in real time at a fixed frequency. The collection frequency of the infrared thermal imaging sensor array is set to collect once every second, and each collection of data is an infrared thermal imaging image with spatial resolution, including real-time temperature information of each area inside the furnace. Each pixel point of the infrared thermal imaging data of the furnace temperature field is marked with spatial coordinates, forming the spatial temperature distribution data inside the furnace.

[0065] The fuel flow sensor installed on the fuel delivery pipeline is used to collect fuel flow data in real time. The fuel flow data is the volume of fuel entering the nozzle through the delivery pipeline per unit time, and the collection frequency is set to once every second. Each collection of fuel flow data is accompanied by a time stamp, forming a real-time record of fuel flow data.

[0066] The collected nozzle state parameters are subjected to missing value processing and abnormal value filtering. The furnace temperature field infrared thermal imaging data is subjected to noise filtering, data smoothing and pixel-level feature extraction. The fuel flow data is subjected to stationarity analysis and abnormal data interpolation processing, and then subjected to feature standardization processing to generate a preprocessed operation data set containing nozzle state parameters, furnace temperature field infrared thermal imaging data and fuel flow data.

[0067] Missing value processing: If the nozzle injection pressure or nozzle opening degree cannot be successfully measured at a certain time point, the missing value is filled in by interpolation based on the measured values of the nozzle injection pressure or nozzle opening degree at adjacent time points. The interpolation calculation method is to use the arithmetic mean of the measured values at several consecutive time points before and after the missing value as the supplement of the missing value. Abnormal value filtering: Calculate the difference between the measured value of the nozzle injection pressure or nozzle opening degree at each time point and the continuous multiple historical measured values, and compare the above difference with the pre-set allowable fluctuation threshold. If the difference exceeds the allowable fluctuation threshold, the corresponding measured value is determined to be an abnormal value, and the above interpolation calculation method is used to correct the abnormal value.

[0068] The infrared thermal imaging data of the furnace temperature field is filtered by median filtering. Specifically, the infrared thermal imaging data is divided into multiple local image regions of fixed size, and the temperature measurement values of all pixels in each local image region are sequentially sorted point by point, and the median of the sorted sequence is taken to replace the original measurement value, so as to eliminate the influence of high-frequency random noise. The data smoothing processing is to take the temperature values of all pixels in the adjacent region at each pixel point of the infrared thermal imaging image and calculate the weighted average value, and the weight value is determined according to the distance of the adjacent pixels to the pixel, that is, the closer the distance of the adjacent pixels, the greater the weight value, and the farther the distance of the adjacent pixels, the smaller the weight value. The above method eliminates abnormal mutation points and maintains the smoothness of the infrared thermal imaging data of the furnace temperature field. The pixel-level feature extraction is to extract the local temperature change rate of each pixel from the filtered and smoothed infrared thermal imaging data of the furnace temperature field, and the calculation method is to divide the temperature difference between the current time and the previous time of a pixel by the temperature value at the previous time, so as to represent the temperature change trend and local temperature fluctuation feature of each pixel point at consecutive time.

[0069] The time series stationarity analysis is performed on the real-time measured fuel flow data, that is, the variance of the fuel flow data in a continuous time window is calculated, and the fluctuation degree of the variance of the fuel flow data in a plurality of continuous time windows is determined. If the fluctuation amplitude of the variance of the fuel flow data exceeds the preset allowable threshold, it is determined that the data segment is non-stationary and is marked. For the abnormal data in the non-stationary data segment, the linear interpolation method is used to correct the abnormal data, that is, the effective measurement values before and after the abnormal data are taken, and the correction value of the abnormal data is calculated according to the time distance of the abnormal data relative to the effective measurement values before and after the abnormal data, so as to ensure the continuity and stationarity of the fuel flow data.

[0070] For each parameter data sequence, the arithmetic mean and variance of the data sequence are calculated, each data value in the data sequence is subtracted from the arithmetic mean and then divided by the square root of the variance, that is, the data values in the data sequence are standardized by dividing the difference between the data values and the sequence mean by the standard deviation. After the feature standardization processing, the nozzle state parameter, the infrared thermal imaging data of the furnace temperature field, and the fuel flow data are represented by dimensionless standardized values, and a pre-processing operation data set containing the nozzle state parameter, the infrared thermal imaging data of the furnace temperature field, and the fuel flow data is generated.

[0071] S2: Based on the pre-processing operation data set, the correlation characteristics of the nozzle state fluctuation and the fuel non-uniform mixing are analyzed, and a fuel distribution non-uniformity index is generated, including:

[0072] Based on the nozzle state parameter and the fuel flow data, the nozzle injection pressure fluctuation feature, the nozzle opening change feature, and the fuel flow change feature are extracted;

[0073] The extraction method of the nozzle injection pressure fluctuation feature is that the difference between the measured value of the nozzle injection pressure and the arithmetic mean value of the nozzle injection pressure measured values in the continuous time window is calculated in a plurality of continuous time windows, and the arithmetic mean value of the absolute values of the differences is calculated to represent the nozzle injection pressure fluctuation feature, which is used to represent the fluctuation degree of the nozzle injection pressure in the continuous time.

[0074] The extraction method of the nozzle opening degree change feature is that the difference between the measured value of the nozzle actual opening degree and the measured value of the nozzle actual opening degree at the previous moment is calculated in a plurality of continuous time windows, and the sum of the absolute values of the differences is calculated and then divided by the number of calculation differences to obtain the ratio as the nozzle opening degree change feature, which is used to represent the change rate and change amplitude of the nozzle actual opening degree.

[0075] The extraction method of the fuel flow change feature is that the difference between the measured value of the fuel flow and the arithmetic mean value of the fuel flow measured values in the time window is calculated in a plurality of continuous time windows, and the sum of the absolute values of the differences is calculated and then divided by the number of difference values in the time window to obtain the ratio as the fuel flow change feature, which is used to reflect the fluctuation condition of the fuel flow.

[0076] The correlation strength between the nozzle injection pressure fluctuation feature and the fuel flow change feature is calculated by time series correlation analysis;

[0077] The calculation process of the correlation strength is that the nozzle injection pressure fluctuation feature sequence and the fuel flow change feature sequence are aligned in the time dimension; then the nozzle injection pressure fluctuation feature sequence and the fuel flow change feature sequence are standardized, and in a plurality of continuous same time windows, the value at each time point of the nozzle injection pressure fluctuation feature sequence is multiplied by the value at the same time point of the fuel flow change feature sequence to obtain a series of product values. The product values represent the synchronous change of the nozzle injection pressure fluctuation feature and the fuel flow change feature at each time point. The product values are arithmetically accumulated, and the data points contained in the selected time window are arithmetically averaged to obtain the average product value.

[0078] The arithmetic mean values of the data points in the selected time window of the nozzle injection pressure fluctuation feature sequence and the fuel flow change feature sequence are calculated respectively. After obtaining the respective arithmetic mean values, the two arithmetic mean values are multiplied to form the product value of the arithmetic mean values.

[0079] The product value of the average product value minus the arithmetic average value is obtained, and a correlation coefficient between the nozzle injection pressure fluctuation feature and the fuel flow change feature is obtained; the absolute value of the correlation coefficient is the correlation strength between the nozzle injection pressure fluctuation feature and the fuel flow change feature. The correlation strength is used to represent the influence degree of the nozzle injection pressure fluctuation on the fuel flow change, and reflects the dynamic correlation between the nozzle injection pressure fluctuation feature and the fuel flow change feature. If the correlation coefficient is a positive value, it indicates that there is a positive correlation between the nozzle injection pressure fluctuation feature and the fuel flow change feature, and the synchronous change trend of the fuel flow will be enhanced with the increase of the nozzle injection pressure fluctuation; if the correlation coefficient is a negative value, it indicates that there is a negative correlation between the nozzle injection pressure fluctuation feature and the fuel flow change feature, and the synchronous change trend of the fuel flow will be weakened with the increase of the nozzle injection pressure fluctuation; the greater the absolute value of the correlation coefficient, the greater the correlation strength between the nozzle injection pressure fluctuation feature and the fuel flow change feature, and vice versa.

[0080] The frequency coupling degree between the nozzle opening change feature and the fuel flow change feature is obtained by using a spectrum analysis method.

[0081] The nozzle opening change feature sequence and the fuel flow change feature sequence are respectively subjected to fast Fourier transform, and respective frequency spectrums are obtained. The frequency spectrum is composed of a frequency value and an amplitude corresponding to the frequency value. The amplitudes of the respective frequency spectrums of the nozzle opening change feature sequence and the fuel flow change feature sequence are multiplied by frequency points, and a cross spectrum is obtained; then the amplitude of the cross spectrum is subjected to arithmetic average value calculation, and an average cross spectrum amplitude is obtained; finally, the average cross spectrum amplitude is divided by the product of the arithmetic average value of the frequency spectrum amplitude of the nozzle opening change feature sequence and the arithmetic average value of the frequency spectrum amplitude of the fuel flow change feature sequence, and then square root is taken, and the frequency coupling degree between the nozzle opening change feature and the fuel flow change feature is obtained. The frequency coupling degree reflects the dynamic interaction relationship between the nozzle opening change and the fuel flow change in the frequency domain.

[0082] According to the correlation strength and the frequency coupling degree, a principal component analysis method is used to determine a fuel distribution non-uniformity index.

[0083] The principal component analysis method is as follows: the coupling degree of the correlation strength and the frequency is taken as the input data of the principal component analysis method, and a data matrix of the principal component analysis is constructed; each row of the data matrix corresponds to the correlation strength and the frequency coupling degree in a time window, and each column represents a specific index, that is, the first column is the value of the correlation strength, and the second column is the value of the frequency coupling degree. The covariance matrix calculation is performed on the data matrix, and the eigenvalue and eigenvector of the covariance matrix are obtained; the covariance matrix is a symmetric matrix, the diagonal elements are the variances of the correlation strength and the frequency coupling degree, and the non-diagonal elements are the covariance values between the correlation strength and the frequency coupling degree. The eigenvectors are arranged in descending order of eigenvalue, and the proportion of each eigenvalue in the total sum of all eigenvalues is calculated. From the largest eigenvalue, the cumulative proportion is added in turn, and the process is stopped when the cumulative proportion first exceeds the pre-set cumulative contribution rate threshold (for example, 95%). At this time, the eigenvectors corresponding to all the eigenvalues that have been added are selected as the principal components. The correlation strength value and the frequency coupling degree value are multiplied by the weight coefficients in the corresponding principal components, and the weight coefficients are the elements corresponding to the correlation strength and the frequency coupling degree in the eigenvectors. Then, the two feature products are arithmetically summed, and the result is the fuel distribution non-uniformity index in the corresponding time window. The fuel distribution non-uniformity index can reflect the coupling relationship between the nozzle injection pressure fluctuation, the nozzle opening change and the fuel flow change, and reflect the uniformity of the fuel mixing and distribution in the furnace.

[0084] S3: based on the pre-processed operation data set, a spatio-temporal convolutional neural network is used to identify the drift trajectory of the hot spot, and a hot spot drift rate and direction prediction result is generated, including:

[0085] The spatial sliding window division method is used to divide the furnace temperature field infrared thermal imaging data into grid regions, and the local hot spot image sequence obtained by dividing the grid regions is input into the spatio-temporal convolutional neural network;

[0086] The spatial sliding window division method is to define a rectangular window with a fixed size in the furnace temperature field infrared thermal imaging image. The rectangular window is slid point by point in the infrared thermal imaging image according to a certain step size, and each time the rectangular window contains a plurality of pixel points to form an independent grid region. Through the sliding window sliding, the entire furnace internal temperature field infrared thermal imaging data is divided into a plurality of grid regions connected to each other and having a fixed size. Each grid region contains temperature information corresponding to a local spatial range in the furnace temperature field, and the local spatial range has a fixed spatial position coordinate in the furnace.

[0087] The grid regions obtained by the above division are regarded as independent local hotspot image regions respectively, and each local hotspot image region is sorted in chronological order to form a plurality of local hotspot image sequences, each of which corresponds to the temperature change process of a fixed spatial position in the hearth over time, and each frame of image in the local hotspot image sequence contains temperature information of all pixels in the corresponding spatial region. For example, in the infrared thermal imaging data of the hearth temperature field, if the position near the center of the top of the hearth is divided into an independent grid region, then the continuous multiple frames of images of the grid region changing over time together constitute the local hotspot image sequence of the grid region, so as to reflect the temperature change characteristics of the hotspot at the center of the top of the hearth at different times.

[0088] The local hotspot image sequence is input into a space-time convolutional neural network to extract the spatial features and time features of the local hotspot image sequence and identify the drift change rule of the hotspot region position. The structure of the space-time convolutional neural network includes a spatial convolutional layer and a time convolutional layer. The spatial convolutional layer is used to extract the spatial features in the hotspot region, and the time convolutional layer is used to identify the dynamic change process of the spatial features in the time dimension.

[0089] The spatial features of the local hotspot image sequence are extracted layer by layer by the spatial convolutional layer to obtain the spatial position distribution features of the hotspot.

[0090] The spatial features of the input local hotspot image sequence are extracted by the spatial convolutional layer. The spatial convolutional layer uses a plurality of convolution kernels, and each convolution kernel performs pixel-by-pixel spatial convolution operation on each frame of image of the local hotspot image sequence. The process of spatial convolution operation is as follows: the convolution kernel slides in each frame of image of the local hotspot image sequence, and the arithmetic cumulative value of the multiplication of all pixels in the convolution kernel and the corresponding weight coefficients of the convolution kernel is obtained at each sliding position, and the convolution result at the corresponding position is stored in the spatial feature map as a spatial feature value. A plurality of spatial feature maps are generated by repeating the above process using a plurality of convolution kernels. The spatial feature maps jointly constitute the spatial position distribution features of the local hotspot region, which reflect the temperature change pattern of the local hotspot region in the spatial dimension and the temperature spatial distribution inside the hotspot.

[0091] The time sequence features of the spatial position distribution features are extracted by the time convolutional layer to identify the drift change rule of the hotspot position.

[0092] The processing procedure of the time convolution layer is: performing convolution operation in the time dimension between continuous multiple spatial feature maps, that is, by defining a time window convolution kernel with a certain length, the corresponding spatial positions of the continuous multiple spatial feature maps are convolved one by one in the time step. The convolution calculation process is: at each spatial position, the weight coefficients of the time window convolution kernel are multiplied with the spatial feature values of the continuous multiple time points corresponding to the spatial position, and the above multiplication result is calculated as an arithmetic cumulative value, as the time sequence feature value of the spatial position. By repeating the above convolution calculation process for all spatial positions, a time sequence feature sequence corresponding to all spatial positions of the hot spot area is generated. The time sequence feature sequence reflects the drift characteristic law of the spatial position of the hot spot area with time, that is, the spatial position change trajectory feature of the hot spot area with time.

[0093] According to the drift change law of the hot spot position, the position change distance and angle of the hot spot in a unit time are calculated to generate the prediction result of the hot spot drift rate and direction;

[0094] The calculation process of the position change distance is: for each spatial position of the hot spot area, the spatial coordinate position difference of the hot spot position in space between the continuous two time frame images is determined, and then the spatial coordinate position difference is squared and summed and square root is taken, that is, the spatial movement distance of the hot spot area between the continuous two time frame images is obtained. The spatial movement distance is divided by the time interval between the adjacent two frame images, and the hot spot position drift rate is obtained.

[0095] The calculation process of the hot spot drift direction is: after determining the spatial position coordinate difference between the continuous two time frame images of the hot spot area, the angle of the hot spot area position change is calculated according to the proportional relationship between the vertical and horizontal components of the spatial position coordinate difference in the spatial coordinate system, which represents the spatial direction of the hot spot area movement, that is, the prediction result of the hot spot drift direction.

[0096] Through the above calculation, the prediction results of the hot spot drift rate and the hot spot drift direction are determined for each local hot spot area in the furnace interior.

[0097] S4: based on the fuel distribution non-uniformity index and the prediction results of the hot spot drift rate and direction, calculating the local overheating risk probability, and generating a heating furnace risk level map, including:

[0098] calculating the spatial correlation coefficient between the fuel distribution non-uniformity index and the hot spot drift rate;

[0099] The calculation process of the space correlation coefficient is as follows: according to the hot spot drift rate, different space regions of the furnace where the hot spots are located are divided, the average value of the fuel distribution non-uniformity index in a plurality of continuous time periods and the average value of the corresponding hot spot drift rate in the same time period are calculated for each space region of the furnace, the average value sequence of the fuel distribution non-uniformity index and the average value sequence of the hot spot drift rate are respectively subjected to feature standardization processing, so that the data scales of the two sequences are unified and have no dimension. The data points of the two standardized sequences corresponding to the space positions are multiplied one by one to obtain the product value at each space position. Then, the product values of all space positions are arithmetically summed and divided by the total number of space positions to obtain the average product value between the fuel distribution non-uniformity index and the hot spot drift rate. The arithmetic average values of all space position data points of the standardized sequence of the fuel distribution non-uniformity index and the standardized sequence of the hot spot drift rate are respectively calculated, and the two arithmetic average values are multiplied to obtain the average value product of the two sequences. The average product value is subtracted from the average value product to obtain the space correlation coefficient. The closer the space correlation coefficient is to zero, the weaker the correlation between the fuel distribution non-uniformity index and the hot spot drift rate in the spatial dimension; on the contrary, the larger the space correlation coefficient value, the stronger the spatial correlation between the fuel distribution non-uniformity index and the hot spot drift rate.

[0100] The time coupling degree coefficient between the fuel distribution non-uniformity index and the hot spot drift direction is calculated.

[0101] Based on the prediction results of the fuel distribution non-uniformity index and the hot spot drift direction, the time coupling degree coefficient between the fuel distribution non-uniformity index and the hot spot drift direction is calculated. The hot spot drift direction prediction result is the angle of change of the hot spot region position, and the fuel distribution non-uniformity index reflects the fuel mixing condition. The time coupling degree coefficient is used to represent the dynamic correlation degree between the time change of the fuel distribution non-uniformity index and the change of the hot spot drift direction.

[0102] The calculation process of the time coupling degree coefficient is as follows: in a plurality of continuous same time windows, the time sequence data of the fuel distribution non-uniformity index and the hot spot drift direction prediction result are respectively determined. The fuel distribution non-uniformity index sequence and the hot spot drift direction sequence are respectively subjected to standardization processing. The corresponding values of the standardized fuel distribution non-uniformity index sequence and the standardized hot spot drift direction sequence at each time point are multiplied to obtain the product value at each time point. Then, the arithmetic average of all product values is calculated to obtain the average product value between the fuel distribution non-uniformity index and the hot spot drift direction. At the same time, the arithmetic average values of the standardized fuel distribution non-uniformity index sequence and the standardized hot spot drift direction sequence are respectively calculated and multiplied to obtain the average value product. Then, the average product value is subtracted from the average value product to obtain the time coupling degree coefficient.

[0103] According to the spatial correlation coefficient and the time coupling degree coefficient, the local overheating risk probability of the local area where the hotspot is located is calculated through a dynamic weight distribution algorithm;

[0104] The ratio of the absolute value of the spatial correlation coefficient to the absolute value of the time coupling degree coefficient is calculated to determine the weight distribution proportion of the spatial correlation coefficient and the time coupling degree coefficient in the calculation of the local overheating risk probability. When the absolute value of the spatial correlation coefficient is greater than the absolute value of the time coupling degree coefficient, the spatial correlation coefficient is given a higher weight; when the absolute value of the time coupling degree coefficient is greater, the time coupling degree coefficient is given a higher weight. According to the weight proportion determined by the above method, the spatial correlation coefficient and the time coupling degree coefficient are weighted and summed to obtain the local overheating risk probability of the local area where the hotspot is located. The higher the local overheating risk probability, the greater the possibility of abnormal temperature rise in the local area where the hotspot is located.

[0105] The local overheating risk probability of the local area where the hotspot is located is classified into risk levels to generate a heating furnace risk level map;

[0106] A plurality of risk probability thresholds (e.g. low risk probability threshold, medium risk probability threshold, high risk probability threshold) are preset, the risk probability thresholds are arranged from small to large to form a plurality of risk level intervals from low to high; the local overheating risk probability is compared with each risk probability threshold, and the local overheating risk probability of the local area where the hotspot is located is classified into the corresponding risk level interval according to the comparison result; the higher the risk level interval, the more serious the risk level of local overheating; according to the risk level of the local area where the hotspot is located, the risk level is mapped into the furnace space coordinate system diagram in the form of different colors or marks, and a heating furnace risk level map containing the risk level distribution of all hotspot areas is constructed.

[0107] S5: According to the heating furnace risk level map, the high-risk area is divided, the abnormal working condition type is matched by combining the historical fault database, and the fault warning instruction containing the hotspot drift path prediction is generated, including:

[0108] According to the heating furnace risk level map, the local area where the hotspot is located in the furnace is divided into a high-risk area when the local overheating risk probability exceeds the preset risk probability threshold;

[0109] The local overheating risk probability of the local area where each hot spot is located in the heating furnace risk level map is compared with the high risk probability threshold value preset in step S4 in sequence. For each local area where a hot spot is located, if the value of the corresponding local overheating risk probability is greater than or equal to the high risk probability threshold value, the local area where the hot spot is located is marked as a high-risk area. If the value of the corresponding local overheating risk probability is less than the high risk probability threshold value, the local area where the hot spot is located is divided into a non-high-risk area. Through the above method, whether the local area where all hot spots in the furnace are located belongs to a high-risk area is determined. The high-risk area represents a risk area where there is a significant trend of abnormal temperature rise and which may cause equipment failure.

[0110] Based on the high-risk area, the abnormal working condition type corresponding to the local overheating risk probability is obtained by searching the historical failure database.

[0111] The historical failure database is searched and analyzed. The historical failure database stores the abnormal working condition types and corresponding failure occurrence data records recorded since the heating furnace is operated. The abnormal working condition types include, but are not limited to, local hot spot abnormal drift, serious non-uniform fuel distribution, local overheating causing furnace structure damage, and the like. For each high-risk area, the local overheating risk probability is used to match and search the historical failure database. The specific method is as follows: each historical failure record stored in the historical failure database is selected. The historical failure record includes the local overheating risk probability of the local area where the hot spot is located and the corresponding abnormal working condition type. The absolute value of the difference between the local overheating risk probability of the current high-risk area and the local overheating risk probability in the historical failure record is calculated to obtain the matching difference value of each historical failure record. The matching difference values are sorted, and the abnormal working condition type corresponding to one or more historical failure records with the smallest matching difference value is determined as the abnormal working condition type matching result of the current high-risk area. The determination process of the abnormal working condition type matching result can accurately reflect the possible failure type of the local area where the current hot spot is located.

[0112] According to the abnormal working condition type matching result, the hot spot drift rate and hot spot drift direction prediction result are mapped to the furnace temperature field to generate a failure warning instruction containing hot spot position drift path prediction.

[0113] The hotspot position drift path prediction method is: determining the future moving direction of the hotspot according to the hotspot drift direction prediction result; determining the spatial distance of the hotspot moving in a unit time according to the hotspot drift rate prediction result; the hotspot drift rate prediction result represents the ratio of the spatial position moving distance of the hotspot area between two continuous time frame images to the time interval between the two images; in the furnace temperature field, taking the current position of the hotspot as the spatial reference point, determining the position coordinates of the hotspot in the future multiple time intervals along the direction determined by the hotspot drift direction prediction result and with the spatial distance determined by the hotspot drift rate prediction result as the interval, connecting the multiple predicted position coordinates in time sequence to form the hotspot position drift path prediction; the hotspot position drift path prediction indicates the spatial position change trend of the hotspot in the future continuous multiple time periods.

[0114] The space position area passed through by the hotspot position drift path prediction is analyzed point by point, and when the space position area passed through by the hotspot position drift path prediction approaches the sensitive area of important equipment or structure in the furnace, the information of the nozzle corresponding to the space position area, such as the spatial installation position of the nozzle, the number and type of the nozzle, is recorded. For the nozzle approaching the sensitive area in the process of the hotspot position drift path prediction, a failure warning instruction is generated, which contains the drift path information of the hotspot, such as the future moving direction of the hotspot, the time point when the hotspot is expected to reach the sensitive area, the drift rate of the hotspot, and the spatial coordinates of the current and future positions of the hotspot; and the recommended operation of the related nozzle is specified in the failure warning instruction, such as adjusting the fuel injection pressure or nozzle opening degree of the nozzle to avoid more serious local overheating failure caused by the hotspot in the furnace. The failure warning instruction embodies the hotspot position drift path prediction result and proposes the nozzle control suggestion.

[0115] S6: based on the failure warning instruction, adjusting the injection pressure and nozzle opening degree of the failure nozzle in real time, including:

[0116] According to the hotspot position drift path prediction in the failure warning instruction, the failure nozzle associated with the hotspot position drift path is determined;

[0117] The hotspot position drift path prediction result is the current position of the hotspot region in the furnace temperature field spatial coordinate system and the position change trend in the future continuous time periods. The spatial coordinates of the spatial position region passed through by the hotspot region position drift path prediction are recorded. According to the spatial position layout information of the nozzles in the furnace, all the spatial position regions passed through by the hotspot region position drift path prediction are analyzed to determine the fault nozzle directly related to the temperature rise of the hotspot region in the hotspot region position drift path prediction. The fault nozzle is the nozzle device corresponding to the region passed through by the hotspot region position drift path prediction. The method for determining the fault nozzle is that if the distance between the installation position of a nozzle and the spatial coordinate of the hotspot region position drift path prediction is less than a preset critical distance threshold, the nozzle is determined as the fault nozzle. Through the above method, the spatial position and the nozzle number information of the fault nozzle are determined to complete the positioning of the fault nozzle.

[0118] Based on the fault nozzle associated with the hotspot position drift path, the target adjustment value of the nozzle injection pressure and the target adjustment value of the nozzle opening degree are determined.

[0119] The method for determining the target adjustment value of the nozzle injection pressure is that the historical operation data of the fault nozzle is analyzed, the historical operation data including the correlation data between the nozzle injection pressure and the temperature of the corresponding hotspot region of the furnace temperature field; the influence relationship between the change of the nozzle injection pressure and the change of the temperature of the hotspot region in the historical operation data is analyzed to determine the best value range of the nozzle injection pressure when the temperature of the hotspot region is in a reasonable state; the upper and lower limit values of the best value range are weighted and averaged, specifically, the upper and lower limit values of the best value range are multiplied by their respective weights and then arithmetically added, wherein the weights are determined according to the current running state of the furnace; the result of the addition is the target adjustment value of the nozzle injection pressure; the target adjustment value of the nozzle injection pressure reflects the reasonable control target of the injection pressure of the fault nozzle when the temperature of the hotspot region is within a safe range.

[0120] The method for determining the target adjustment value of the nozzle opening degree is that the historical operation data of the fault nozzle is analyzed to analyze the historical correlation between the change of the nozzle opening degree and the change of the temperature of the hotspot region; when the temperature of the hotspot region is in a reasonable state, the best control value range of the nozzle opening degree is determined; the specific analysis method is to calculate the numerical interval between the nozzle opening degree corresponding to the lowest temperature of the hotspot region and the nozzle opening degree corresponding to the highest temperature of the hotspot region in the historical operation data; the upper and lower limits of the numerical interval are multiplied by the weights determined by the current hotspot region temperature state of the furnace, and the arithmetic sum of the products is calculated to obtain the target adjustment value of the nozzle opening degree; the target adjustment value of the nozzle opening degree reflects the ideal running value of the opening degree of the fault nozzle under the normal state of the temperature of the hotspot region. Through the above method, the target adjustment values of the two running parameters of the injection pressure and the opening degree of the fault nozzle are determined.

[0121] The target adjustment value of the nozzle injection pressure is subtracted from the real-time collected nozzle injection pressure to generate a nozzle injection pressure adjustment control amount;

[0122] The nozzle injection pressure adjustment control amount is generated by subtracting the target adjustment value of the nozzle injection pressure from the real-time collected nozzle injection pressure actual measurement value corresponding to the current time of the fault nozzle from the pretreatment running data set; the nozzle injection pressure adjustment control amount represents the gap between the fault nozzle injection pressure and the target state, and the nozzle injection pressure adjustment control amount has directionality, a positive value indicating that the nozzle injection pressure needs to be increased, and a negative value indicating that the nozzle injection pressure needs to be decreased.

[0123] The target adjustment value of the nozzle opening degree is subtracted from the real-time collected nozzle opening degree to generate a nozzle opening degree adjustment control amount;

[0124] The nozzle opening degree adjustment control amount is generated by subtracting the target adjustment value of the nozzle opening degree from the real-time collected nozzle actual opening degree measurement value corresponding to the current time of the fault nozzle from the pretreatment running data set; the nozzle opening degree adjustment control amount represents the control deviation between the fault nozzle opening degree and the target opening degree; the nozzle opening degree adjustment control amount has clear directionality, a positive value indicating that the nozzle opening degree needs to be increased, and a negative value indicating that the nozzle opening degree needs to be decreased.

[0125] The nozzle injection pressure adjustment control amount and the nozzle opening degree adjustment control amount are used to real-time adjust the injection pressure and the nozzle opening degree of the fault nozzle;

[0126] The nozzle injection pressure adjustment control amount and the nozzle opening degree adjustment control amount are input into a nozzle injection pressure automatic adjustment execution device and a nozzle opening degree automatic adjustment execution device respectively; the nozzle injection pressure automatic adjustment execution device automatically adjusts the pressure control component at the nozzle valve according to the nozzle injection pressure adjustment control amount, and changes the nozzle injection pressure by adjusting the channel diameter of the fuel flowing into the nozzle; when the nozzle injection pressure adjustment control amount is positive, the pressure control component at the nozzle valve acts to moderately increase the fuel flow into the nozzle, thereby increasing the nozzle injection pressure; when the nozzle injection pressure adjustment control amount is negative, the pressure control component at the nozzle valve acts to reduce the fuel flow into the nozzle, thereby reducing the nozzle injection pressure.

[0127] The nozzle opening degree automatic adjusting execution device adjusts the opening degree of the nozzle valve in real time according to the nozzle opening degree adjusting control quantity; when the nozzle opening degree adjusting control quantity is positive, the valve core mechanical structure of the nozzle valve automatically acts to increase the opening degree of the nozzle valve; when the nozzle opening degree adjusting control quantity is negative, the valve core mechanical structure automatically acts to reduce the opening degree of the nozzle valve; the real-time change of the nozzle opening degree ensures that the hot spot region temperature of the furnace is effectively controlled, and the hot spot region temperature is kept within the allowable range.

[0128] Through the real-time linkage adjustment of the nozzle injection pressure and the nozzle opening degree, effective dynamic control of the hot spot region temperature and its drift path is realized.

[0129] Embodiment 2

[0130] The embodiment 2 of the present application is different from the embodiment 1 in that the embodiment 2 introduces a heating furnace operation data analysis and fault prediction system.

[0131] Figure 2 A structure diagram of a heating furnace operation data analysis and fault prediction system is given, and the heating furnace operation data analysis and fault prediction system comprises:

[0132] The data acquisition module acquires real-time operation data of the heating furnace, performs noise filtering and feature standardization processing, and generates a preprocessed operation data set;

[0133] The feature extraction module analyzes the correlation features of nozzle state fluctuation and fuel non-uniform mixing based on the preprocessed operation data set, and generates a fuel distribution non-uniformity index;

[0134] The trajectory recognition module identifies the drift trajectory of the hot spot based on the preprocessed operation data set using a spatio-temporal convolutional neural network, and generates hot spot drift rate and direction prediction results;

[0135] The risk assessment module calculates the local overheating risk probability based on the fuel distribution non-uniformity index and the hot spot drift rate and direction prediction results, and generates a heating furnace risk grade atlas;

[0136] The early warning generation module divides the high-risk area according to the heating furnace risk grade atlas, matches the abnormal working condition type in combination with the historical fault database, and generates a fault early warning instruction containing hot spot drift path prediction;

[0137] The nozzle regulation and control module adjusts the injection pressure and the nozzle opening degree of the fault nozzle in real time based on the fault early warning instruction.

[0138] The above formulas are all dimensionless numerical calculations, the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and the preset parameters and threshold values in the formula are set by a person skilled in the art according to the actual situation.

[0139] The above-described embodiments can be implemented in part or in whole through software, hardware, firmware or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When loaded and executed by a computer, the computer instructions or computer programs cause the computer to perform all or part of the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through a wired (for example, infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0140] Those skilled in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0142] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed ones can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0143] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0144] In addition, the functional modules in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0145] The functions, if realized in the form of software functional modules and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part of the prior art or the part of the technical solutions of the present application can be embodied in the form of software products, and the computer software product is stored in a storage medium, including a plurality of instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and various program codes that can be stored in the medium.

[0146] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0147] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the protection scope of the present application.

Claims

1. A method for analyzing operating data and predicting faults in a heating furnace, characterized in that, Includes the following steps: S1: Acquire real-time operating data of the heating furnace, perform noise filtering and feature standardization processing, and generate a pre-processed operating data set; S2: Based on the preprocessed operational data set, analyze the correlation characteristics between nozzle state fluctuations and non-uniform fuel mixing, and generate a fuel distribution non-uniformity index, specifically: Based on nozzle state parameters and fuel flow data, the characteristics of nozzle injection pressure fluctuation, nozzle opening change and fuel flow change are extracted. The correlation strength between nozzle injection pressure fluctuation characteristics and fuel flow rate variation characteristics was analyzed by time series correlation calculation. The frequency coupling between nozzle opening variation characteristics and fuel flow rate variation characteristics was obtained using spectral analysis. Based on the correlation strength and frequency coupling degree, principal component analysis is used to determine the fuel distribution nonuniformity index; S3: Based on the preprocessed runtime dataset, a spatiotemporal convolutional neural network is used to identify the drift trajectory of hotspots and generate prediction results for hotspot drift rate and direction. Specifically: The spatial sliding window method is used to segment the infrared thermal imaging data of the furnace temperature field into gridded regions, and the local hotspot image sequence obtained by the gridded region segmentation is input into the spatiotemporal convolutional neural network. Spatial features of local hotspot image sequences are extracted layer by layer using spatial convolutional layers to obtain the spatial location distribution features of hotspots; Temporal features of spatial location distribution are extracted using temporal convolutional layers to identify the drift and change patterns of hotspot locations; Based on the drift and change pattern of hotspot locations, the distance and angle of position change of hotspots per unit time are calculated, and the hotspot drift rate and direction prediction results are generated. S4: Based on the fuel distribution non-uniformity index and the prediction results of hot spot drift rate and direction, calculate the probability of local overheating risk and generate a risk level map of the heating furnace. S5: Based on the risk level map of the heating furnace, high-risk areas are divided, and abnormal operating conditions are matched with the historical fault database to generate fault warning instructions that include hot spot drift path prediction. S6: Based on fault warning commands, adjust the injection pressure and nozzle opening of the faulty nozzle in real time.

2. The method for analyzing operating data and predicting faults in a heating furnace according to claim 1, characterized in that, S1, specifically: Real-time acquisition of nozzle status parameters, furnace temperature field infrared thermal imaging data, and fuel flow data of the heating furnace; The collected nozzle state parameters are processed for missing values ​​and outliers are filtered out. The furnace temperature field infrared thermal imaging data are subjected to noise filtering, data smoothing and pixel-level feature extraction. The fuel flow data is subjected to stationarity analysis and outlier interpolation. After feature standardization, a preprocessed running dataset containing nozzle state parameters, furnace temperature field infrared thermal imaging data and fuel flow data is generated.

3. The method for analyzing operating data and predicting faults in a heating furnace according to claim 2, characterized in that, S4, specifically: Calculate the spatial correlation coefficient between the fuel distribution nonuniformity index and the hot spot drift rate; Calculate the time coupling coefficient between the fuel distribution nonuniformity index and the hot spot drift direction; Based on the spatial correlation coefficient and the temporal coupling coefficient, the probability of local overheating risk in the local area where the hotspot is located is calculated using a dynamic weight allocation algorithm. The risk level of local overheating risk in the local area where the hot spot is located is classified into risk levels to generate a risk level map of the heating furnace.

4. The method for analyzing operating data and predicting faults in a heating furnace according to claim 3, characterized in that, S5, specifically: According to the risk level map of the heating furnace, the local area where the probability of local overheating in the furnace exceeds the preset risk probability threshold is designated as a high-risk area; Based on high-risk areas, retrieve abnormal operating condition types from the historical fault database to obtain abnormal operating condition type matching results corresponding to the probability of local overheating risk. Based on the abnormal operating condition type matching results, the hot spot drift rate and hot spot drift direction prediction results are mapped to the furnace temperature field to generate a fault warning command that includes the hot spot location drift path prediction.

5. The method for analyzing operating data and predicting faults in a heating furnace according to claim 4, characterized in that, S6, specifically: Based on the hotspot location drift path prediction in the fault warning command, identify the faulty nozzle associated with the hotspot location drift path; Based on the faulty nozzle associated with the hot spot location drift path, determine the target adjustment values ​​for nozzle injection pressure and nozzle opening. The difference between the target adjustment value of the nozzle injection pressure and the real-time collected nozzle injection pressure is calculated to generate the nozzle injection pressure adjustment control quantity. The difference between the target adjustment value of the nozzle opening and the real-time collected nozzle opening is calculated to generate the nozzle opening adjustment control quantity. Based on the nozzle injection pressure adjustment control amount and the nozzle opening adjustment control amount, the injection pressure and nozzle opening of the faulty nozzle are adjusted in real time.

6. A heating furnace operation data analysis and fault prediction system, used to implement the heating furnace operation data analysis and fault prediction method according to any one of claims 1-5, characterized in that, include: Data acquisition module: acquires real-time operating data of the heating furnace, performs noise filtering and feature standardization processing, and generates a preprocessed operating data set; Feature extraction module: Based on the preprocessed operational data set, it analyzes the correlation characteristics between nozzle state fluctuations and non-uniform fuel mixing, and generates a fuel distribution non-uniformity index, specifically: Based on nozzle state parameters and fuel flow data, the characteristics of nozzle injection pressure fluctuation, nozzle opening change and fuel flow change are extracted. The correlation strength between nozzle injection pressure fluctuation characteristics and fuel flow rate variation characteristics was analyzed by time series correlation calculation. The frequency coupling between nozzle opening variation characteristics and fuel flow rate variation characteristics was obtained using spectral analysis. Based on the correlation strength and frequency coupling degree, principal component analysis is used to determine the fuel distribution nonuniformity index; The trajectory recognition module, based on a preprocessed runtime dataset, uses a spatiotemporal convolutional neural network to identify the drift trajectories of hotspots and generates predictions of hotspot drift rates and directions. Specifically: The spatial sliding window method is used to segment the infrared thermal imaging data of the furnace temperature field into gridded regions, and the local hotspot image sequence obtained by the gridded region segmentation is input into the spatiotemporal convolutional neural network. Spatial features of local hotspot image sequences are extracted layer by layer using spatial convolutional layers to obtain the spatial location distribution features of hotspots; Temporal features of spatial location distribution are extracted using temporal convolutional layers to identify the drift and change patterns of hotspot locations; Based on the drift and change pattern of hotspot locations, the distance and angle of position change of hotspots per unit time are calculated, and the hotspot drift rate and direction prediction results are generated. Risk assessment module: Based on the fuel distribution non-uniformity index and the prediction results of hot spot drift rate and direction, calculate the probability of local overheating risk and generate a risk level map of the heating furnace; Early warning generation module: Based on the risk level map of the heating furnace, high-risk areas are divided, and abnormal operating condition types are matched with the historical fault database to generate fault early warning instructions that include hot spot drift path prediction. Nozzle control module: Based on fault warning commands, adjust the injection pressure and nozzle opening of the faulty nozzle in real time.

Images