Method, system, device and medium for analyzing tail gas data of waste incineration device

Abstract

The application discloses a waste incineration device tail gas data analysis method, system, equipment and medium, relates to the technical field of waste incineration, and the method comprises the following steps: acquiring incineration process parameter data and tail gas image data generated by a plurality of waste incineration devices when incinerating the same batch of waste; for each waste incineration device, taking the waste incineration device as a target device, based on non-target tail gas image data corresponding to a non-target device, first correlation mining is carried out on target tail gas image data corresponding to the target device, and corresponding image correlation characteristics are obtained; based on the incineration process parameter data, second correlation mining is carried out on the image correlation characteristics, and corresponding correlation parameter image characteristics are obtained; the correlation parameter image characteristics corresponding to each waste incineration device are analyzed respectively, and tail gas analysis results are obtained. The application has the effect of improving the monitoring accuracy of tail gas.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of waste incineration, and in particular to a tail gas data analysis method, system, device and medium for a waste incineration device. BACKGROUND

[0002] Waste incineration treatment is an important way of modern urban and industrial waste management. While achieving waste reduction and harmless treatment, it also faces environmental risks that may be caused by tail gas emissions. The tail gas of incineration contains various pollutants, and its emission status is directly related to environmental quality and public health safety. Therefore, accurate monitoring and effective analysis of incineration tail gas have become a key link to ensure that incineration facilities meet the emission standards.

[0003] However, the existing tail gas monitoring and analysis methods still have obvious deficiencies in accuracy and reliability. First, the commonly used judgment mechanism based on fixed thresholds (such as setting a single safety range for a specific process parameter or setting a static limit for the visual features of tail smoke) cannot adapt to the complex changes of actual working conditions. Since the composition and heat value of incinerated waste often fluctuate greatly, this rigid threshold judgment method often leads to false positives or false negatives, greatly limiting the accuracy of the monitoring results. Second, the existing analysis methods usually process process parameter data and visual feature data separately, lacking in-depth exploration of the internal correlation mechanism between multi-source data. The limitations of this analysis method make it difficult for the system to accurately distinguish whether the root cause of tail gas anomalies is from the state anomaly of the device itself or from the mismatch between process parameter configuration and waste characteristics, further affecting the accuracy and guidance value of the diagnostic conclusion.

[0004] In summary, the current waste incineration device tail gas monitoring and analysis method still has significant deficiencies in dealing with complex working condition changes, multi-source data fusion, and anomaly diagnosis accuracy. It is urgent to improve the accuracy and reliability of monitoring and analysis through technical improvement to meet the increasingly stringent environmental protection requirements and management needs. SUMMARY

[0005] In order to improve the monitoring accuracy of the tail gas of the waste incineration device, the present application provides a tail gas data analysis method, system, device and medium for a waste incineration device.

[0006] In a first aspect, the present application provides a tail gas data analysis method for a waste incineration device, which adopts the following technical solution:

[0007] The tail gas data analysis method for a waste incineration device comprises:

[0008] obtain incineration process parameter data and tail gas image data generated by multiple waste incineration devices incinerating the same batch of waste, wherein the incineration process parameter data is time series data of various incineration process parameters, and the tail gas image data is image data formed by image collection on a tail gas discharge port of the waste incineration device;

[0009] For each waste incineration device, the waste incineration device is taken as a target device, and first correlation mining is performed on target tail gas image data corresponding to the target device based on non-target tail gas image data corresponding to a non-target device, to obtain corresponding image correlation features;

[0010] Second correlation mining is performed on the image correlation features based on the incineration process parameter data, to obtain corresponding correlation parameter image features;

[0011] The correlation parameter image features corresponding to each of the waste incineration devices are analyzed respectively, to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent state abnormality of each of the waste incineration devices and configuration rationality of corresponding incineration process parameters.

[0012] By adopting the technical solution, the incineration process parameter data and the tail gas image data generated by multiple waste incineration devices incinerating the same batch of waste are obtained, wherein the incineration process parameter data is time series data of various incineration process parameters, and the tail gas image data is image data formed by image collection on a tail gas discharge port of the waste incineration device. Then, for each waste incineration device, the waste incineration device is taken as a target device, and first correlation mining is performed on target tail gas image data corresponding to the target device based on non-target tail gas image data corresponding to a non-target device, to obtain corresponding image correlation features. Then, second correlation mining is performed on the image correlation features based on the incineration process parameter data, to obtain corresponding correlation parameter image features. Then, the correlation parameter image features corresponding to each of the waste incineration devices are analyzed respectively, to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent state abnormality of each of the waste incineration devices and configuration rationality of corresponding incineration process parameters. In the above method, by transversely comparing tail gas image features and multi-modal feature fusion of each waste incineration device under the same batch of waste, effective identification of the running state of a single waste incineration device is realized, and equipment failure and process parameter configuration problems are accurately distinguished. The limitations of traditional single threshold judgment method are overcome, thereby enhancing the pertinence and reliability of the analysis results and improving the monitoring accuracy of the tail gas of the waste incineration device.

[0013] Optionally, the step of performing first correlation mining on target tail gas image data corresponding to the target device based on non-target tail gas image data corresponding to a non-target device, to obtain corresponding image correlation features, comprises:

[0014] performing semantic encoding on target tail gas image data corresponding to the target device to obtain target image semantic features, and performing semantic encoding on non-target tail gas image data corresponding to each non-target device to obtain non-target image semantic features;

[0015] performing association mining on the target image semantic features based on the non-target image semantic features by using a first association mining unit in the tail gas analysis network to obtain a corresponding association weight distribution;

[0016] performing weighted summation on the non-target image semantic features based on the association weight distribution to obtain a reference background feature;

[0017] splicing the reference background feature and the target image semantic features to obtain corresponding image association features.

[0018] By adopting the above technical solutions, in order to obtain the corresponding association weight distribution, the target image semantic features are obtained by performing semantic encoding on target tail gas image data corresponding to the target device, and the non-target image semantic features are obtained by performing semantic encoding on non-target tail gas image data corresponding to each non-target device. Then, the first association mining unit in the tail gas analysis network is used to perform association mining on the target image semantic features based on the non-target image semantic features to obtain the corresponding association weight distribution. Then, the reference background feature is obtained by performing weighted summation on the non-target image semantic features based on the association weight distribution. Then, the corresponding image association features are obtained by splicing the reference background feature and the target image semantic features.

[0019] Optionally, the step of performing association mining on the target image semantic features based on the non-target image semantic features by using the first association mining unit in the tail gas analysis network to obtain the corresponding association weight distribution comprises:

[0020] loading the non-target image semantic features and the target image semantic features into the first association mining unit, wherein the first association mining unit comprises a first linear mapping layer, a second linear mapping layer, and a third linear mapping layer;

[0021] performing linear mapping on the target image semantic features by using the first linear mapping layer to generate query features;

[0022] performing linear mapping on the non-target image semantic features by using the second linear mapping layer to generate key features;

[0023] performing linear mapping on the non-target image semantic features based on the third linear mapping layer to generate value features;

[0024] determine a similarity distribution between the query feature and the key feature, and scale and normalize the similarity distribution to generate a corresponding association weight distribution.

[0025] By adopting the technical scheme, in order to obtain the corresponding association weight distribution, the non-target image semantic feature and the target image semantic feature are loaded into the first association mining unit, wherein the first association mining unit comprises a first linear mapping layer, a second linear mapping layer and a third linear mapping layer, then the target image semantic feature is linearly mapped through the first linear mapping layer to generate the query feature, then the non-target image semantic feature is linearly mapped through the second linear mapping layer to generate the key feature, then the non-target image semantic feature is linearly mapped based on the third linear mapping layer to generate the value feature, then the similarity distribution between the query feature and the key feature is determined, and the similarity distribution is scaled and normalized to generate the corresponding association weight distribution.

[0026] Optionally, the step of performing second association mining on the image association feature based on the incineration process parameter data to obtain a corresponding association parameter image feature comprises:

[0027] performing semantic embedding on the incineration process parameter data to obtain corresponding incineration process parameter embedding features;

[0028] performing semantic mining on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data;

[0029] performing second association mining on the image association feature based on the incineration process parameter semantic features to obtain a corresponding association parameter image feature.

[0030] By adopting the technical scheme, in order to obtain the corresponding association parameter image feature, the incineration process parameter data is subjected to semantic embedding to obtain corresponding incineration process parameter embedding features, then the incineration process parameter embedding features are subjected to semantic mining to obtain incineration process parameter semantic features corresponding to the incineration process parameter data, and then the image association feature is subjected to second association mining based on the incineration process parameter semantic features to obtain a corresponding association parameter image feature.

[0031] Optionally, the step of performing semantic mining on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data comprises:

[0032] The incineration process parameter embedding feature is loaded into a deep mining unit of the tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping, and the second mapping branch includes linear mapping and a Sigmoid activation function;

[0033] The incineration process parameter embedding feature is subjected to self-attention processing to obtain incineration process parameter attention features corresponding to the incineration process parameter embedding feature;

[0034] The incineration process parameter attention features are subjected to mapping operation through the first mapping branch to obtain incineration process parameter candidate features corresponding to the incineration process parameter attention features;

[0035] The incineration process parameter attention features are subjected to mapping operation through the second mapping branch to obtain incineration process parameter gating features corresponding to the incineration process parameter attention features;

[0036] The incineration process parameter candidate features are subjected to element-by-element multiplication to obtain incineration process parameter semantic features corresponding to the incineration process parameter data.

[0037] By adopting the above technical solution, in order to obtain incineration process parameter semantic features, the incineration process parameter embedding feature is loaded into a deep mining unit of the tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping, and the second mapping branch includes linear mapping and a Sigmoid activation function, then the incineration process parameter embedding feature is subjected to self-attention processing to obtain incineration process parameter attention features corresponding to the incineration process parameter embedding feature, then the incineration process parameter attention features are subjected to mapping operation through the first mapping branch to obtain incineration process parameter candidate features corresponding to the incineration process parameter attention features, then the incineration process parameter attention features are subjected to mapping operation through the second mapping branch to obtain incineration process parameter gating features corresponding to the incineration process parameter attention features, and then the incineration process parameter candidate features are subjected to element-by-element multiplication to obtain incineration process parameter semantic features corresponding to the incineration process parameter data.

[0038] Optionally, the step of performing second association mining on the image association features based on the incineration process parameter data to obtain corresponding association parameter image features, comprises:

[0039] The incineration process parameter data and the image association features are loaded into a second association mining unit of the tail gas analysis network, wherein the second association mining unit includes a first association branch and a second association branch;

[0040] The first association branch is used for associatively mining the image association feature based on the incineration process parameter semantic feature, and outputting a first association feature, wherein the first association branch is internally provided with a first semantic space conversion matrix, and the first semantic space conversion matrix is used for converting the incineration process parameter semantic feature from a current semantic space to a semantic space where the image association feature is located.

[0041] The second association branch is used for associatively mining the incineration process parameter semantic feature based on the image association feature, and outputting a second association feature, wherein the second association branch is internally provided with a second semantic space conversion matrix, and the second semantic space conversion matrix is used for converting the image association feature from a current semantic space to a semantic space where the incineration process parameter semantic feature is located.

[0042] The first association feature, the second association feature and the target feature are weighted or spliced to synthesize a corresponding association parameter image feature, wherein the target feature is the image association feature or the incineration process parameter semantic feature.

[0043] In order to obtain a corresponding association parameter image feature, the incineration process parameter data and the image association feature are loaded into a second association mining unit of the tail gas analysis network, wherein the second association mining unit includes a first association branch and a second association branch. Then, the first association branch is used for associatively mining the image association feature based on the incineration process parameter semantic feature, and outputting a first association feature, wherein the first association branch is internally provided with a first semantic space conversion matrix, and the first semantic space conversion matrix is used for converting the incineration process parameter semantic feature from a current semantic space to a semantic space where the image association feature is located. Then, the second association branch is used for associatively mining the incineration process parameter semantic feature based on the image association feature, and outputting a second association feature, wherein the second association branch is internally provided with a second semantic space conversion matrix, and the second semantic space conversion matrix is used for converting the image association feature from a current semantic space to a semantic space where the incineration process parameter semantic feature is located. Then, the first association feature, the second association feature and the target feature are weighted or spliced to synthesize a corresponding association parameter image feature, wherein the target feature is the image association feature or the incineration process parameter semantic feature.

[0044] Optionally, the step of associatively mining the image association feature based on the incineration process parameter semantic feature through the first association branch and outputting a first association feature includes:

[0045] The first semantic space conversion matrix of the first association branch is used for performing a semantic space conversion operation on the incineration process parameter semantic feature to obtain a corresponding incineration process parameter conversion feature.

[0046] determine a mapping parameter relationship between the incineration process parameter conversion feature and the image correlation feature, and normalize the mapping parameter relationship to obtain a normalized relationship;

[0047] based on the normalized relationship, the image correlation feature is weighted and processed, and a first correlation feature is output.

[0048] By adopting the above technical solution, the semantic space conversion matrix of the first correlation branch is used to perform semantic space conversion operation on the incineration process parameter semantic feature to obtain the corresponding incineration process parameter conversion feature, then the mapping parameter relationship between the incineration process parameter conversion feature and the image correlation feature is determined, and the mapping parameter relationship is normalized to obtain the normalized relationship, then based on the normalized relationship, the image correlation feature is weighted and processed, and the first correlation feature is output.

[0049] In a second aspect, the present application also provides a tail gas data analysis system of a waste incineration device, which adopts the following technical solution:

[0050] The tail gas data analysis system of the waste incineration device comprises:

[0051] A data acquisition module is configured to acquire incineration process parameter data and tail gas image data generated by a plurality of waste incineration devices when incinerating a same batch of waste, wherein the incineration process parameter data is time series data of each incineration process parameter, and the tail gas image data is image data formed by image acquisition of a tail gas discharge port of the waste incineration device.

[0052] A first correlation mining module is configured to, for each waste incineration device, take the waste incineration device as a target device, based on non-target tail gas image data corresponding to a non-target device, perform first correlation mining on target tail gas image data corresponding to the target device to obtain corresponding image correlation features.

[0053] A second correlation mining module is configured to, based on the incineration process parameter data, perform second correlation mining on the image correlation features to obtain corresponding correlation parameter image features.

[0054] A tail gas analysis result generation module is configured to analyze the correlation parameter image features corresponding to each waste incineration device respectively to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent state abnormality of each waste incineration device and configuration rationality of corresponding incineration process parameters.

[0055] In a third aspect, the present application also provides a computer device, which adopts the following technical solution:

[0056] A computer device comprises a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor implements the method in the first aspect when executing the computer program.

[0057] In a fourth aspect, the present application further provides a computer readable storage medium, which adopts the following technical solution:

[0058] A computer readable storage medium stores a computer program capable of being loaded and executed by a processor to implement the method in the first aspect.

[0059] To sum up, the present application at least has the following beneficial technical effects: obtaining the incineration process parameter data and the tail gas image data generated by the plurality of waste incineration devices for incinerating the same batch of waste, wherein the incineration process parameter data is the time series data of each incineration process parameter, and the tail gas image data is the image data formed by image collection on the tail gas discharge port of the waste incineration device; then for each waste incineration device, taking the waste incineration device as a target device, based on the non-target tail gas image data corresponding to a non-target device, the target tail gas image data corresponding to the target device is subjected to first correlation mining to obtain corresponding image correlation features; then based on the incineration process parameter data, the image correlation features are subjected to second correlation mining to obtain corresponding correlation parameter image features; then the correlation parameter image features corresponding to each waste incineration device are analyzed respectively to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent the state abnormality of each waste incineration device and the configuration rationality of the corresponding incineration process parameters; in the above method, by transversely comparing the tail gas image features and the multi-modal feature fusion of each waste incineration device under the same batch of waste, the effective identification of the running state of a single waste incineration device is realized, and the equipment failure and the process parameter configuration problem are accurately distinguished, the limitations of the traditional single threshold judgment method are overcome, the pertinence and reliability of the analysis results are enhanced, and the monitoring accuracy of the tail gas of the waste incineration device is improved. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 is a whole process schematic diagram of the embodiment of the present application.

[0061] Figure 2 is a structure schematic diagram of the system of the present application.

[0062] Figure 3 is a structure block diagram of the computer device of the present application. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0064] The embodiment of the present application discloses a waste incineration device tail gas data analysis method.

[0065] With reference to Figure 1 , the waste incineration device tail gas data analysis method comprises:

[0066] Step S11, obtaining incineration process parameter data and tail gas image data generated by multiple waste incineration devices for incinerating the same batch of waste.

[0067] Among them, the incineration process parameter data is the time sequence data of each incineration process parameter, and the tail gas image data is the image data formed by image collection on the tail gas discharge port of the waste incineration device.

[0068] It should be noted that in step S11, the operation data of multiple devices under the same input condition (i.e. processing the same batch of waste) is obtained, the incineration process parameter data reflects the internal control state of the incineration process, such as the sequence of changes of furnace temperature, air volume, pressure, etc. with time, which is a quantitative description of the process; and the tail gas image data records the external visual performance of the plume of the discharge port, such as color, concentration, texture, shape, etc., which is a visualization of the process result.

[0069] Step S12, for each waste incineration device, taking the waste incineration device as a target device, based on the non-target tail gas image data corresponding to the non-target device, performing first correlation mining on the target tail gas image data corresponding to the target device, to obtain corresponding image correlation features.

[0070] It should be noted that step S12 performs horizontal correlation analysis of image data across devices, by taking one of the devices as a target device and comparing its tail gas image (target data) with the reference benchmark formed by the tail gas images (non-target data) of all other devices (non-target devices). This first correlation mining is essentially a group comparison analysis, aiming to identify the individuality or abnormality of the target device from the commonality of multiple devices. For example, when most devices process the same batch of waste, the color of the tail smoke is light gray, while the color of the tail smoke of the target device is significantly black, which can be effectively captured and quantified through this correlation analysis, thereby obtaining image correlation features representing the difference between the target device and the group.

[0071] Step S13, based on the incineration process parameter data, performing second correlation mining on the image correlation features to obtain corresponding correlation parameter image features.

[0072] It should be noted that step S12 realizes the deep fusion and secondary mining of process parameter data and image correlation features, and the purpose is to establish a quantitative correlation model between the internal process state and the external tail gas visual performance. The second correlation mining aims to explore how the setting, fluctuation or combination of process parameters affect and lead to the differences in image features found in step S12; through this mining, simple visual abnormalities can be linked to specific process operations, thereby converting image correlation features representing external phenomena into more explanatory correlation parameter image features that also contain internal process causes.

[0073] Step S14, respectively analyzing the correlation parameter image features corresponding to each waste incineration device to obtain tail gas analysis results.

[0074] Among them, the tail gas analysis results are used to represent the state abnormality of each waste incineration device and the configuration rationality of the corresponding incineration process parameters.

[0075] It should be noted that in step S14, by analyzing the correlation parameter image features unique to each device (such as classification, regression or anomaly detection judgment through pre-set models, rules or algorithms), two levels of diagnostic conclusions are finally output: one is "state abnormality", that is, to judge whether the incineration device itself (such as burners, furnaces, nozzles, etc. hardware) has faults or performance degradation; the second is configuration rationality, that is, to evaluate whether the current process parameters (such as temperature, air volume, etc.) set for the device match the characteristics of the waste being processed and whether they are in the optimal interval, thereby indicating the specific direction of operation and maintenance adjustment.

[0076] In the above embodiment, the waste incineration device obtains the incineration process parameter data and the tail gas image data generated by incinerating the same batch of waste, wherein the incineration process parameter data is time series data of each incineration process parameter, and the tail gas image data is image data obtained by image acquisition of the tail gas discharge port of the waste incineration device. Then, for each waste incineration device, the waste incineration device is taken as a target device, the target tail gas image data corresponding to the target device is first associated and mined based on the non-target tail gas image data corresponding to the non-target device, the corresponding image association features are obtained, then the image association features are second associated and mined based on the incineration process parameter data, the corresponding association parameter image features are obtained, and then the association parameter image features corresponding to each waste incineration device are analyzed respectively to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent the state abnormality of each waste incineration device and the configuration rationality of the corresponding incineration process parameters. In the above method, by comparing the tail gas image features of each waste incineration device under the same batch of waste and the multi-modal feature fusion, the running state of a single waste incineration device is effectively identified, the equipment failure and the process parameter configuration problem are accurately distinguished, the limitations of the traditional single threshold judgment method are overcome, the pertinence and reliability of the analysis results are enhanced, and the monitoring accuracy of the tail gas of the waste incineration device is improved.

[0077] As a further embodiment of the method, the step of first associated mining of the target tail gas image data corresponding to the target device based on the non-target tail gas image data corresponding to the non-target device to obtain the corresponding image association features comprises:

[0078] In step S21, the target tail gas image data corresponding to the target device is semantically encoded to obtain target image semantic features, and the non-target tail gas image data corresponding to each non-target device is semantically encoded to obtain non-target image semantic features.

[0079] It should be noted that by step S21, the original tail gas image data is converted into higher-level and more abstract numerical semantic features. Generally, a pre-trained convolutional neural network or a self-defined feature extraction model is used to compress and extract the visual information (such as color, texture, shape, smoke plume dynamic trend, etc.) contained in each image into a low-dimensional and dense feature vector. The target image semantic features and the non-target image semantic features obtained in this way can more effectively represent the core visual patterns of the tail gas of each device, avoiding the high dimensionality and redundancy of directly processing the original pixel data.

[0080] In step S22, the first associated mining unit in the tail gas analysis network is used to associate and mine the target image semantic features based on the non-target image semantic features to obtain the corresponding association weight distribution.

[0081] It should be noted that by step S21, the similarity or correlation of the target device with each non-target device in the exhaust visual mode is quantified, the first association mining unit (the implementation can be analogous to the attention mechanism) receives the target feature and the non-target feature set as input, generates an association weight distribution by calculating the similarity or correlation score between the target image semantic feature and each non-target image semantic feature, the distribution is a vector, each weight value in the vector corresponds to a non-target device, the higher the weight value, the more similar the exhaust visual mode of the non-target device to the target device, and the greater the contribution of the non-target device to the reference background in subsequent construction.

[0082] Step S23, based on the association weight distribution, the non-target image semantic features are weighted and summed to obtain the reference background feature.

[0083] It should be noted that by step S22, a dynamic and weighted "reference background" is constructed. Based on the association weight distribution calculated in step S22, all non-target image semantic features are weighted and summed, the features with high weights occupy a larger proportion in the summation result, and the features with low weights occupy a smaller proportion. The "reference background feature" obtained in this way is not a simple average state, but a "group normal state" or "expected state" that can better represent the similar running mode of the target device, thereby providing a highly contextualized benchmark for evaluating the abnormality of the target device.

[0084] Step S24, the reference background feature and the target image semantic feature are spliced to obtain the corresponding image association feature.

[0085] In the above embodiment, in order to obtain the corresponding association weight distribution, the target exhaust image data corresponding to the target device is semantically encoded to obtain the target image semantic feature, and the non-target exhaust image data corresponding to each non-target device is semantically encoded to obtain the non-target image semantic feature. Then, the first association mining unit in the exhaust analysis network is used to mine the association of the target image semantic feature based on the non-target image semantic feature to obtain the corresponding association weight distribution, then the non-target image semantic feature is weighted and summed based on the association weight distribution to obtain the reference background feature, and then the reference background feature and the target image semantic feature are spliced to obtain the corresponding image association feature.

[0086] As a further embodiment of the method, the step of using the first association mining unit in the exhaust analysis network to mine the association of the target image semantic feature based on the non-target image semantic feature to obtain the corresponding association weight distribution includes:

[0087] Step S31, loading the non-target image semantic features and the target image semantic features into a first association mining unit, wherein the first association mining unit comprises a first linear mapping layer, a second linear mapping layer and a third linear mapping layer.

[0088] Step S32, linearly mapping the target image semantic features through the first linear mapping layer to generate query features.

[0089] Step S33, linearly mapping the non-target image semantic features through the second linear mapping layer to generate key features.

[0090] Step S34, linearly mapping the non-target image semantic features based on the third linear mapping layer to generate value features.

[0091] Step S35, determining the similarity distribution between the query features and the key features, and performing scaling and normalization processing on the similarity distribution to generate a corresponding association weight distribution.

[0092] It should be noted that from step S31 to step S35, a calculation process of realizing first association mining based on an attention mechanism is described, which converts input features into different semantic spaces through multiple linear mapping layers, and then calculates the refined association weight between the target and the background. The core idea is to take the "target image semantic features" as the query (Query), take the "non-target image semantic features" set as the key (Key) and value (Value), calculate the similarity between the query and each key, and obtain a weight distribution. The distribution accurately quantifies the association degree of the target device and each non-target device in the exhaust visual mode.

[0093] In the above embodiment, in order to obtain the corresponding association weight distribution, the non-target image semantic features and the target image semantic features are loaded into the first association mining unit, wherein the first association mining unit comprises a first linear mapping layer, a second linear mapping layer and a third linear mapping layer. Then, the target image semantic features are linearly mapped through the first linear mapping layer to generate query features, the non-target image semantic features are linearly mapped through the second linear mapping layer to generate key features, the non-target image semantic features are linearly mapped based on the third linear mapping layer to generate value features, and then the similarity distribution between the query features and the key features is determined. The similarity distribution is scaled and normalized to generate a corresponding association weight distribution.

[0094] As a further embodiment of the method, the step of performing second association mining on the image association features based on the incineration process parameter data to obtain corresponding association parameter image features comprises:

[0095] Step S41, semantic embedding is performed on the incineration process parameter data to obtain corresponding incineration process parameter embedding features.

[0096] Step S42, semantic mining is performed on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data.

[0097] Step S43, based on the incineration process parameter semantic features, second correlation mining is performed on the image correlation features to obtain corresponding correlation parameter image features.

[0098] It should be noted that, from step S41 to step S43, first, the time sequence process parameters are converted into dense vector representation through semantic embedding, the time sequence relationship and physical meaning between the parameters are retained, then high-level semantic information contained in the parameters is extracted through deep semantic mining to generate semantic features capable of comprehensively representing the process state, finally, a deep mapping relationship between the process parameter semantic features and the image correlation features is established through cross-modal correlation mining to form a fusion feature representation containing process state information and visual performance information. This hierarchical processing method effectively realizes the alignment and fusion of numerical process parameters and visual image features at the semantic level, providing high-quality feature input for subsequent analysis.

[0099] In the above embodiment, in order to obtain the corresponding correlation parameter image features, the incineration process parameter data is subjected to semantic embedding to obtain corresponding incineration process parameter embedding features, then the incineration process parameter embedding features are subjected to semantic mining to obtain incineration process parameter semantic features corresponding to the incineration process parameter data, and then based on the incineration process parameter semantic features, second correlation mining is performed on the image correlation features to obtain the corresponding correlation parameter image features.

[0100] As a further embodiment of the method, the step of performing semantic mining on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data comprises:

[0101] Step S51, the incineration process parameter embedding features are loaded into a deep mining unit of the tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping, and the second mapping branch includes linear mapping and Sigmoid activation function.

[0102] Step S52, self-attention processing is performed on the incineration process parameter embedding features to obtain incineration process parameter attention features corresponding to the incineration process parameter embedding features.

[0103] Step S53, mapping operation is performed on the incineration process parameter attention features through the first mapping branch to obtain incineration process parameter candidate features corresponding to the incineration process parameter attention features.

[0104] Step S54, through the second mapping branch, the incineration process parameter attention feature is mapped to obtain the incineration process parameter attention feature corresponding to the incineration process parameter gating feature.

[0105] Step S55, the incineration process parameter candidate feature and the incineration process parameter candidate feature are multiplied element by element to obtain the incineration process parameter data corresponding to the incineration process parameter semantic feature.

[0106] It should be noted that from step S51 to step S55, first, the long-range dependence relationship between each time point in the incineration process parameter sequence is captured through the self-attention module to generate attention features with more context awareness; then a double-branch structure is used for parallel processing: the first branch is responsible for generating candidate features, retaining the complete information of the original parameters; the second branch generates gating weights through the Sigmoid activation function to achieve selective filtering of the features; finally, the outputs of the two branches are multiplied element by element through the gating mechanism to form a refined semantic feature representation that retains important information and suppresses irrelevant noise, which effectively improves the model's ability to capture complex process parameter dynamic change patterns and the robustness of feature expression.

[0107] In the above embodiment, in order to obtain the incineration process parameter semantic feature, the incineration process parameter embedding feature is loaded into the deep mining unit of the tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping, the second mapping branch includes linear mapping and Sigmoid activation function, then the incineration process parameter embedding feature is processed by self-attention to obtain the incineration process parameter attention feature corresponding to the incineration process parameter embedding feature, then the incineration process parameter attention feature is mapped through the first mapping branch to obtain the incineration process parameter attention feature corresponding to the incineration process parameter gating feature, then the incineration process parameter attention feature is mapped through the second mapping branch to obtain the incineration process parameter attention feature corresponding to the incineration process parameter gating feature, then the incineration process parameter candidate feature and the incineration process parameter candidate feature are multiplied element by element to obtain the incineration process parameter data corresponding to the incineration process parameter semantic feature.

[0108] As a further embodiment of the method, based on the incineration process parameter data, the step of performing second association mining on the image association feature to obtain the corresponding association parameter image feature comprises:

[0109] Step S61, the incineration process parameter data and the image association feature are loaded into the second association mining unit of the tail gas analysis network, wherein the second association mining unit includes a first association branch and a second association branch.

[0110] Step S62, through the first association branch, the image association features are associated and mined based on the incineration process parameter semantic features, and the first association features are output. The first association branch is built-in with a first semantic space conversion matrix, which is used to convert the incineration process parameter semantic features from the current semantic space to the semantic space where the image association features are located.

[0111] Step S63, through the second association branch, the incineration process parameter semantic features are associated and mined based on the image association features, and the second association features are output. The second association branch is built-in with a second semantic space conversion matrix, which is used to convert the image association features from the current semantic space to the semantic space where the incineration process parameter semantic features are located.

[0112] Step S64, the first association features, the second association features and the target features are weighted or spliced, and the corresponding association parameter image features are synthesized. The target features are the image association features or the incineration process parameter semantic features.

[0113] It should be noted that from step S61 to step S64, the bidirectional semantic alignment and information interaction between the process parameters and the image features are realized through a double-branch parallel architecture. Specifically, the first association branch maps the process parameter semantics to the image feature space, explores the influence mode of the process parameters on the visual performance, and the second association branch maps the image features to the parameter semantic space, and mines the process state information reflected by the visual features. Finally, through a multi-source feature fusion strategy, the bidirectional interaction features and the original features are integrated to form a joint feature representation that contains both the process parameter influence mechanism and the visual performance feedback. This bidirectional cross-mining mechanism effectively breaks through the limitations of traditional one-way feature fusion, and realizes the deep semantic cooperation between multi-modal data.

[0114] In the above embodiment, in order to obtain the corresponding associated parameter image feature, the incineration process parameter data and the image associated feature are first loaded into the second association mining unit of the tail gas analysis network, wherein the second association mining unit includes a first association branch and a second association branch, then through the first association branch, the image associated feature is associated and mined based on the incineration process parameter semantic feature, and a first associated feature is output, wherein the first association branch is internally provided with a first semantic space conversion matrix, which is used to convert the incineration process parameter semantic feature from the current semantic space to the semantic space where the image associated feature is located, then through the second association branch, the incineration process parameter semantic feature is associated and mined based on the image associated feature, and a second associated feature is output, wherein the second association branch is internally provided with a second semantic space conversion matrix, which is used to convert the image associated feature from the current semantic space to the semantic space where the incineration process parameter semantic feature is located, then the first associated feature, the second associated feature and the target feature are weighted or spliced to synthesize the corresponding associated parameter image feature, wherein the target feature is the image associated feature or the incineration process parameter semantic feature.

[0115] As a further embodiment of the method, the step of outputting the first associated feature by associating and mining the image associated feature based on the incineration process parameter semantic feature through the first association branch includes:

[0116] Step S71, performing a semantic space conversion operation on the incineration process parameter semantic feature through the first semantic space conversion matrix of the first association branch to obtain a corresponding incineration process parameter conversion feature.

[0117] Step S72, determining the mapping parameter relationship between the incineration process parameter conversion feature and the image associated feature, and performing normalization processing on the mapping parameter relationship to obtain a normalized relationship.

[0118] Step S73, performing weighting processing on the image associated feature based on the normalized relationship to output the first associated feature.

[0119] It should be noted that the technical principle of step S62 is basically the same as that of step S61, and the technical principle of step S62 can refer to steps S71 to S73.

[0120] In the above embodiment, in order to obtain the first associated feature, the incineration process parameter semantic feature is subjected to a semantic space conversion operation through the first semantic space conversion matrix of the first association branch to obtain a corresponding incineration process parameter conversion feature, then the mapping parameter relationship between the incineration process parameter conversion feature and the image associated feature is determined, and the mapping parameter relationship is subjected to normalization processing to obtain a normalized relationship, and then the image associated feature is subjected to weighting processing based on the normalized relationship to output the first associated feature.

[0121] The embodiment of the present application also discloses a tail gas data analysis system of a waste incineration device.

[0122] Reference Figure 2 The tail gas data analysis system of the waste incineration device comprises:

[0123] A data acquisition module is configured to acquire incineration process parameter data and tail gas image data generated by a plurality of waste incineration devices when incinerating a same batch of waste, wherein the incineration process parameter data is time series data of various incineration process parameters, and the tail gas image data is image data formed by image acquisition on a tail gas discharge port of the waste incineration device.

[0124] A first correlation mining module is configured to, for each waste incineration device, take the waste incineration device as a target device, perform first correlation mining on target tail gas image data corresponding to the target device based on non-target tail gas image data corresponding to a non-target device, and obtain corresponding image correlation features.

[0125] A second correlation mining module is configured to perform second correlation mining on the image correlation features based on the incineration process parameter data, and obtain corresponding correlation parameter image features.

[0126] A tail gas analysis result generation module is configured to analyze the correlation parameter image features corresponding to each waste incineration device respectively, and obtain tail gas analysis results, wherein the tail gas analysis results are used to represent state abnormality of each waste incineration device and configuration rationality of corresponding incineration process parameters.

[0127] The tail gas data analysis system of the waste incineration device can implement any one of the tail gas data analysis methods of the waste incineration device, and the specific working process of the tail gas data analysis system of the waste incineration device can refer to the corresponding process in the tail gas data analysis method of the waste incineration device.

[0128] The embodiment of the present application also discloses a computer device.

[0129] Reference Figure 3 The computer device comprises a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor implements any one of the tail gas data analysis methods of the waste incineration device when executing the computer program.

[0130] The embodiment of the present application also discloses a computer readable storage medium.

[0131] The computer readable storage medium stores a computer program capable of being loaded by a processor and executing any one of the tail gas data analysis methods of the waste incineration device.

[0132] Among them, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used or combined with an instruction execution system, device or instrument; The program code contained on the computer readable medium can be transmitted by any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any appropriate combination of the above.

[0133] The above are the preferred embodiments of the present application, not to limit the protection scope of the present application, any feature disclosed in the specification (including abstract and drawings) can be replaced by other equivalent or similar purpose alternative features, unless specifically described. That is, unless specifically described, each feature is only an example of a series of equivalent or similar features.

Claims

1. A method of exhaust gas data analysis for a waste incineration plant, characterized in that, The method comprises the following steps: obtaining waste incineration process parameter data and tail gas image data generated by a plurality of waste incineration devices incinerating the same batch of waste, wherein the waste incineration process parameter data is time series data of each waste incineration process parameter, and the tail gas image data is image data obtained by image acquisition of a tail gas discharge port of the waste incineration device; for each waste incineration device, taking the waste incineration device as a target device, and based on non-target tail gas image data corresponding to a non-target device, performing first association mining on target tail gas image data corresponding to the target device to obtain corresponding image association features; based on the waste incineration process parameter data, performing second association mining on the image association features to obtain corresponding associated parameter image features; respectively analyzing the associated parameter image features corresponding to each waste incineration device to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent the state abnormality of each waste incineration device and the configuration rationality of the corresponding waste incineration process parameters; the step of performing second association mining on the image association features based on the waste incineration process parameter data to obtain corresponding associated parameter image features comprises: performing semantic embedding on the waste incineration process parameter data to obtain corresponding waste incineration process parameter embedding features; performing semantic mining on the waste incineration process parameter embedding features to obtain waste incineration process parameter semantic features corresponding to the waste incineration process parameter data; based on the waste incineration process parameter semantic features, performing second association mining on the image association features to obtain corresponding associated parameter image features; the step of performing semantic mining on the waste incineration process parameter embedding features to obtain waste incineration process parameter semantic features corresponding to the waste incineration process parameter data comprises: loading the waste incineration process parameter embedding features into a deep mining unit of a tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping, and the second mapping branch includes linear mapping and Sigmoid activation function; performing self-attention processing on the waste incineration process parameter embedding features to obtain waste incineration process parameter attention features corresponding to the waste incineration process parameter embedding features; performing mapping operation on the waste incineration process parameter attention features through the first mapping branch to obtain waste incineration process parameter candidate features corresponding to the waste incineration process parameter attention features; performing mapping operation on the waste incineration process parameter attention features through the second mapping branch to obtain waste incineration process parameter gating features corresponding to the waste incineration process parameter attention features; performing element-wise multiplication on the waste incineration process parameter candidate features and the waste incineration process parameter gating features to obtain waste incineration process parameter semantic features corresponding to the waste incineration process parameter data.

2. The exhaust gas data analysis method of a waste incineration apparatus according to claim 1, characterized by, the step of performing first association mining on the target tail gas image data corresponding to the target device based on the non-target tail gas image data corresponding to the non-target device to obtain corresponding image association features comprises: The target tail gas image data corresponding to the target device is semantically encoded to obtain target image semantic features, and the non-target tail gas image data corresponding to each non-target device is semantically encoded to obtain non-target image semantic features; The first association mining unit in the tail gas analysis network is used to perform association mining on the target image semantic features based on the non-target image semantic features, and a corresponding association weight distribution is obtained; Based on the association weight distribution, the non-target image semantic features are weighted and summed to obtain a reference background feature; The reference background feature and the target image semantic feature are spliced to obtain a corresponding image association feature.

3. The exhaust gas data analysis method of a waste incineration apparatus according to claim 2, characterized by, The step of using the first association mining unit in the tail gas analysis network to perform association mining on the target image semantic features based on the non-target image semantic features to obtain a corresponding association weight distribution comprises: The non-target image semantic features and the target image semantic features are loaded into the first association mining unit, wherein the first association mining unit includes a first linear mapping layer, a second linear mapping layer and a third linear mapping layer; The target image semantic features are linearly mapped by the first linear mapping layer to generate query features; The non-target image semantic features are linearly mapped by the second linear mapping layer to generate key features; The non-target image semantic features are linearly mapped based on the third linear mapping layer to generate value features; The similarity distribution between the query features and the key features is determined, and the similarity distribution is scaled and normalized to generate a corresponding association weight distribution.

4. The exhaust gas data analysis method of a waste incineration apparatus according to claim 1, characterized by, The step of performing second association mining on the image association feature based on the incineration process parameter data to obtain a corresponding association parameter image feature comprises: The incineration process parameter data and the image association feature are loaded into the second association mining unit of the tail gas analysis network, wherein the second association mining unit includes a first association branch and a second association branch; The image association feature is associated mined based on the incineration process parameter semantic feature by the first association branch to output a first association feature, wherein the first association branch is built-in with a first semantic space conversion matrix, and the first semantic space conversion matrix is used to convert the incineration process parameter semantic feature from the current semantic space to the semantic space where the image association feature is located; The incineration process parameter semantic feature is associated mined based on the image association feature by the second association branch to output a second association feature, wherein the second association branch is built-in with a second semantic space conversion matrix, and the second semantic space conversion matrix is used to convert the image association feature from the current semantic space to the semantic space where the incineration process parameter semantic feature is located; The first association feature, the second association feature and the target feature are weighted or spliced to synthesize a corresponding association parameter image feature, wherein the target feature is the image association feature or the incineration process parameter semantic feature.

5. The exhaust gas data analysis method of a waste incineration device according to claim 4, characterized by, The step of performing correlation mining on the image correlation features based on the incineration process parameter semantic features through the first correlation branch and outputting first correlation features comprises: Performing semantic space conversion on the incineration process parameter semantic features through a first semantic space conversion matrix of the first correlation branch to obtain corresponding incineration process parameter conversion features; Determining a mapping parameter relationship between the incineration process parameter conversion features and the image correlation features and performing normalization processing on the mapping parameter relationship to obtain a normalized relationship; Performing weighting processing on the image correlation features based on the normalized relationship and outputting first correlation features.

6. A flue gas data analysis system for a waste incineration plant, characterized in that Comprise: A data acquisition module is configured to acquire incineration process parameter data and tail gas image data generated by multiple waste incineration devices during incineration of the same batch of waste, wherein the incineration process parameter data is time series data of various incineration process parameters, and the tail gas image data is image data obtained by image acquisition of a tail gas discharge port of the waste incineration device; A first correlation mining module is configured to, for each waste incineration device, take the waste incineration device as a target device, perform first correlation mining on target tail gas image data corresponding to the target device based on non-target tail gas image data corresponding to non-target devices, and obtain corresponding image correlation features; A second correlation mining module is configured to perform second correlation mining on the image correlation features based on the incineration process parameter data to obtain corresponding correlation parameter image features; A tail gas analysis result generation module is configured to analyze the correlation parameter image features corresponding to each waste incineration device respectively to obtain tail gas analysis results, wherein the tail gas analysis results are used to represent state abnormality of each waste incineration device and configuration rationality of corresponding incineration process parameters; The step of performing second correlation mining on the image correlation features based on the incineration process parameter data to obtain corresponding correlation parameter image features comprises: Performing semantic embedding on the incineration process parameter data to obtain corresponding incineration process parameter embedding features; Performing semantic mining on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data; Performing second correlation mining on the image correlation features based on the incineration process parameter semantic features to obtain corresponding correlation parameter image features; The step of performing semantic mining on the incineration process parameter embedding features to obtain incineration process parameter semantic features corresponding to the incineration process parameter data comprises: The incineration process parameter embedding features are loaded into a deep mining unit of a tail gas analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both the first mapping branch and the second mapping branch comprise linear mapping, and the second mapping branch comprises linear mapping and a Sigmoid activation function; Performing self-attention processing on the incineration process parameter embedding features to obtain incineration process parameter attention features corresponding to the incineration process parameter embedding features; The incineration process parameter attention feature is mapped through the first mapping branch to obtain an incineration process parameter candidate feature corresponding to the incineration process parameter attention feature; The incineration process parameter attention feature is mapped through the second mapping branch to obtain an incineration process parameter gating feature corresponding to the incineration process parameter attention feature; The incineration process parameter candidate feature and the incineration process parameter gating feature are multiplied element by element to obtain an incineration process parameter semantic feature corresponding to the incineration process parameter data.

7. A computer device, characterized by A memory and a processor are included, the memory has a computer program stored thereon, the computer program being executable on the processor, and the processor implements the method in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, A computer program is stored, which can be loaded and executed by a processor to perform the method in any one of claims 1 to 5.

Images