Power plant coal pile spontaneous combustion detection method and alarm system based on deep learning
Through the processing of hyperspectral and infrared imaging data and deep learning models, the problem of data noise and multimodal fusion in coal pile spontaneous combustion detection is solved, and more accurate spontaneous combustion risk assessment and automatic identification are achieved, which improves the accuracy of detection and robustness.
Patent Information
- Application Number
- CN202410862610.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-15
AI Technical Summary
The existing technology has data noise influence, temperature measurement inaccuracy, and multimodal data fusion lacks efficient feature extraction and fusion mechanisms in the detection of coal piles, resulting in insufficient detection accuracy and robustness, especially in complex environments. Key risk areas.
The coal pile is scanned regularly by hyperspectral imager and infrared imager, and the data is processed through sliding averaging method and NUC algorithm, combined with deep convolutional neural network for feature extraction and multimodal fusion, and an attention mechanism is introduced to construct the coal pile spontaneous combustion detection Model, automatically identify key areas and conduct spontaneous combustion risk assessment.
It improves the accuracy and timeliness of coal pile spontaneous combustion detection, enhances the judgment of spontaneous combustion risks in complex environments, achieves seamless connection from monitoring to decision-making, and reduces unnecessary panic and resource waste.
Smart Images

Figure CN118968278A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electric power safety technology, and in particular to a deep learning-based method and alarm system for detecting spontaneous combustion of coal piles in power plants. Background Art
[0002] With the rapid development of the power industry, coal, as one of the main energy sources, occupies a pivotal position in power plants. However, the problem of spontaneous combustion of coal piles has always been a major hidden danger that plagues the safety of power plant operations. During long-term storage, coal piles are very likely to cause spontaneous combustion, resulting in economic losses and even endangering personal safety. Therefore, fast and accurate coal pile spontaneous combustion detection technology has become a key link in ensuring the safety of power production. In this context, hyperspectral imaging technology and infrared thermal imaging technology have gradually been used in the early identification of coal pile spontaneous combustion due to their advantages of non-contact and large-area monitoring.
[0003] However, existing technologies still face several challenges when applying these two types of data for spontaneous combustion detection. First, the original hyperspectral data is easily disturbed by environmental factors and contains a lot of noise, which affects the accuracy of the analysis. Similarly, infrared imaging data is often affected by inherent equipment noise and external environmental factors, resulting in inaccurate temperature measurements. Secondly, although deep learning has made remarkable achievements in the field of image processing, in terms of multimodal data fusion, especially in the joint analysis of hyperspectral and infrared images, existing models often lack efficient and targeted feature extraction and fusion mechanisms, making it difficult to fully utilize the complementary information of the two modalities. In addition, most models fail to make full use of the attention mechanism, which makes it difficult to focus on key areas of spontaneous combustion risk in complex environments, reducing the accuracy and robustness of detection. Summary of the invention
[0004] In view of the above existing problems, the present invention is proposed.
[0005] Therefore, the present invention provides a deep learning-based method and alarm system for detecting spontaneous combustion of coal piles in power plants, which solve the technical problems of insufficient accuracy and real-time performance of spontaneous combustion detection of coal piles in the prior art, and difficulty in identifying key risk areas in complex environments.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In a first aspect, an embodiment of the present invention provides a method for detecting spontaneous combustion of coal piles in power plants based on deep learning, which includes: a method for detecting spontaneous combustion of coal piles in power plants based on deep learning, characterized in that: it includes: using a hyperspectral imager and an infrared imager to periodically scan the coal pile at regular intervals to collect hyperspectral data and infrared imaging data; using a sliding average method to reduce the influence of noise on the hyperspectral data to obtain accurate hyperspectral data; using a NUC algorithm to eliminate fixed pattern noise on the infrared imaging data to obtain accurate infrared imaging data; the processed hyperspectral data and infrared imaging data are subjected to feature extraction and multimodal fusion by a deep convolutional neural network to obtain hyperspectral features and temperature features before spontaneous combustion; introducing an attention mechanism so that the coal pile spontaneous combustion detection model can automatically identify key areas, construct a coal pile spontaneous combustion detection model, and learn the hyperspectral features and infrared imaging features before spontaneous combustion; inputting the collected real-time data into the coal pile spontaneous combustion detection model to perform spontaneous combustion risk assessment, and formulating strategies based on the assessment.
[0008] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning described in the present invention, wherein: the hyperspectral data includes image data with high resolution and multi-band characteristics, and the hyperspectral data performs spatial imaging of the target while recording detailed spectral information of each pixel point;
[0009] The infrared imaging data is obtained by capturing the surface temperature of an object through an infrared imager, thereby reflecting the ambient temperature.
[0010] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning of the present invention, the sliding average method used for the hyperspectral data is as follows:
[0011] By defining a fixed-size window, the window slides point by point on the data sequence. For each window position, the average value of all data points in the window is calculated, and then this average value is assigned to the point at the center of the window. The specific expression is:
[0012]
[0013] Among them, Ssmooth(i) is the smoothed hyperspectral value, N is a positive integer, S(j) is the jth value in the original hyperspectral data, i is the index position of the hyperspectral data, and j is the index of each point in the sliding average window.
[0014] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning described in the present invention, the feature extraction performed by the deep convolutional neural network is specifically as follows:
[0015] A subnetwork capable of extracting hyperspectral features is constructed through CNN, local feature detection is performed, and hyperspectral images are processed to extract hyperspectral features related to chemical composition and material state. The specific expression is:
[0016] Fhyp = CNNhyp(Ssmooth);
[0017] Among them, CNNhyp is a hyperspectral convolutional neural sub-network, and Fhyp is a hyperspectral feature;
[0018] Then, another CNN is used to construct a subnetwork that can extract infrared thermal imaging features. Through infrared thermal imaging technology, the features of temperature distribution and abnormal areas are extracted. The specific expression is:
[0019] Ftherm=CNNtherm(Icorrected);
[0020] Among them, CNNtherm is the infrared thermal imaging convolutional neural sub-network, Ftherm is the infrared thermal imaging feature, and Icorrected is the infrared imaging data corrected by the NUC algorithm.
[0021] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning described in the present invention, the multimodal fusion is to perform weighted fusion of the hyperspectral features and infrared imaging features extracted by the two sub-networks through the fusion layer, assign different weights to the features of each mode, and perform weighted summation to reflect the importance of the two modes. The specific expression is:
[0022] F fused =αFhyp+βFtherm;
[0023] Among them, F fused is the final fused feature vector, α is the hyperspectral feature weight coefficient, β is the infrared imaging feature weight coefficient, and Fhyp represents the hyperspectral feature.
[0024] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning described in the present invention, the attention mechanism is introduced to construct a coal pile spontaneous combustion detection model, which is specifically as follows:
[0025] Through the multi-head attention mechanism MultiHeadAttention, the fusion feature F fused Perform deep processing while maintaining the training stability of the model, and finally output an enhanced feature vector. The specific expression is:
[0026] F′=MultiHeadAttention(Q=F fused , K=F fused , V=F fused)+F fused ;
[0027] Among them, F′ is the enhanced feature vector obtained after the multi-head attention mechanism, Q is the query vector obtained by linearly transforming each position of the input sequence itself, K is the key vector used to match the query vector, and V is the information content value vector;
[0028] The input features are flattened through the multi-head attention mechanism, and the optimized feature vector is generated after linear transformation and ReLU activation. The specific expression is:
[0029] Z = ReLU(W z ·Flatten(F′)+·b z );
[0030] Among them, Z is the input of the final output layer, and the activation function introduces nonlinearity into the neural network. z is a weight matrix, b z is the bias vector, and Flatten is the operation of converting the multi-dimensional F′ into a one-dimensional vector;
[0031] The model finally outputs the probability of coal pile, which is expressed as:
[0032] P ignite =σ(W p ·Z+b p );
[0033] Among them, σ is the sigmoid activation function, W p is the linear transformation of the weight matrix from the previous layer to the current layer, b p is the bias term, P ignite Probability of spontaneous combustion.
[0034] As a preferred solution of the power plant coal pile spontaneous combustion detection method based on deep learning described in the present invention, the spontaneous combustion risk assessment is specifically as follows:
[0035] Define T1 as the dividing point between low risk and medium risk, and T2 as the dividing point between medium risk and high risk. These two thresholds are used to classify the level of spontaneous combustion risk.
[0036] When the spontaneous combustion probability P(y|x) predicted by the model is less than T1, it is considered to be a low risk;
[0037] When the spontaneous combustion probability P(y|x) predicted by the model is greater than or equal to T1 and less than T2, it is considered to be a medium risk;
[0038] When the spontaneous combustion probability P(y|x) predicted by the model is less than T2, it is considered to be a high risk;
[0039] When the risk is low, maintain routine monitoring frequency and maintenance operations. When the risk is medium, increase the monitoring frequency and conduct physical inspections of local areas to confirm whether there is local overheating. When the risk is high, prepare emergency supplies and personnel, develop an evacuation plan, and take physical cooling measures to reduce internal heat accumulation.
[0040] In the second aspect, the present invention provides a power plant coal pile spontaneous combustion alarm system based on deep learning, including a data acquisition module, a data preprocessing module, a feature extraction and fusion module, an attention mechanism module and a spontaneous combustion risk assessment module; the data acquisition module is used to use a hyperspectral imager and an infrared imager to periodically scan the coal pile at regular intervals to collect hyperspectral data and infrared imaging data; the data preprocessing module is used to use a sliding average method to reduce the influence of noise on the hyperspectral data to obtain accurate hyperspectral data, and use a NUC algorithm to eliminate fixed pattern noise on the infrared imaging data to obtain accurate infrared imaging data; the feature extraction and fusion module is used to extract features and fuse multimodal data from the processed hyperspectral data and infrared imaging data through a deep convolutional neural network to obtain hyperspectral features and temperature features before spontaneous combustion; the attention mechanism module is used to introduce an attention mechanism so that the coal pile spontaneous combustion detection model can automatically identify key areas, construct a coal pile spontaneous combustion detection model, learn the hyperspectral features before spontaneous combustion and the infrared imaging features; the spontaneous combustion risk assessment module is used to input the collected real-time data into the coal pile spontaneous combustion detection model to perform spontaneous combustion risk assessment, and formulate strategies based on the assessment.
[0041] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for detecting spontaneous combustion of coal piles in power plants based on deep learning as described in the first aspect of the present invention is implemented.
[0042] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for detecting spontaneous combustion of coal piles in power plants based on deep learning as described in the first aspect of the present invention.
[0043] The beneficial effects of the present invention are as follows: by using a hyperspectral imager and an infrared imager to periodically scan the coal pile, continuous monitoring of the coal pile status is ensured, and small changes in the early stage of spontaneous combustion are captured in time. The data captured by the hyperspectral imager and the infrared imager are combined to provide a basis for subsequent analysis, thereby improving the comprehensiveness and timeliness of monitoring; the hyperspectral data is denoised by a sliding average method to reduce the influence of random noise and make the collected data more accurate; features are extracted from the hyperspectral and infrared data respectively by a deep convolutional neural network, and then the complementary information between the two types of data is integrated through multimodal fusion, so that the model can make more accurate judgments on the risk of spontaneous combustion in complex situations; by introducing an attention mechanism, the model can automatically focus on the key areas of spontaneous combustion risk in the coal pile, thereby improving the detection targeting and avoiding excessive attention to irrelevant areas; by setting a risk level threshold, the natural probability can be quickly converted into a specific response strategy, achieving a seamless connection from monitoring to decision-making, and the division of low, medium and high risk levels makes emergency measures more reasonable and avoids unnecessary panic and waste of resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0045] Figure 1 This is a flow chart of the method for detecting spontaneous combustion of coal piles in power plants based on deep learning in Example 1. DETAILED DESCRIPTION
[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0047] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0048] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0049] Example 1, reference Figure 1, which is the first embodiment of the present invention, and provides a method for detecting spontaneous combustion of coal piles in power plants based on deep learning, comprising the following steps:
[0050] S1. Use a hyperspectral imager and an infrared imager to periodically scan the coal pile and collect hyperspectral data and infrared imaging data.
[0051] Furthermore, the hyperspectral imager is set to scan the surface of the coal pile every 4 hours, which enables in-depth analysis of the coal pile. It can not only provide spatial image information, but also go deep into the spectral level of the material, capturing spontaneous combustion risk signals that are difficult to detect with the naked eye and traditional imaging technology.
[0052] The infrared imager is placed on the surface of the coal pile to capture these infrared radiations every 4 hours. The sensors in the detector are sensitive to infrared radiations of different wavelengths, and when they receive the radiation, they convert it into electrical signals. The processed electrical signals are converted into visual image data, which reflects the temperature in the environment according to the different colors;
[0053] S2. Use the sliding average method to reduce the impact of noise on hyperspectral data, and use the NUC algorithm to eliminate fixed pattern noise on infrared imaging data to obtain accurate infrared imaging data.
[0054] Furthermore, by defining a sliding window size N, the window is like a moving "filter", starting from the starting position of the data sequence, and gradually sliding backward by one unit (usually a time point in the time series or an index position in the data sequence), and at each stop position, taking the arithmetic average of all data values within the window range (usually symmetrical, that is, from the first N / 2 data points to the last N / 2 data points). The calculated average value replaces the data point at the center of the window in the original sequence, forming a smoothed sequence, which can reduce noise interference and improve data accuracy. The specific expression is:
[0055]
[0056] Among them, Ssmooth(i) is the smoothed hyperspectral value, N is a positive integer representing the size of the sliding average window, S(j) is the jth value in the original hyperspectral data, i is the index position of the hyperspectral data, and j is the index of each point in the sliding average window;
[0057] The NUC algorithm analyzes the temperature information of relatively static areas in the image, calculates and applies corrected infrared imaging data in real time, enables the system to continuously self-correct, adapt to environmental changes, eliminate grayscale deviations and response time differences between each pixel, and ensure consistency and accuracy of temperature measurement;
[0058] S3. The processed hyperspectral data and infrared imaging data are subjected to feature extraction and multimodal fusion by a deep convolutional neural network to obtain the hyperspectral features and temperature features before spontaneous combustion.
[0059] Furthermore, a sub-network capable of extracting hyperspectral features is first constructed through a convolutional neural network (CNN), local feature detection is performed, and hyperspectral images are processed to extract hyperspectral features related to chemical composition and material state. The hyperspectral feature extraction sub-network constructed by the convolutional neural network (CNN) greatly improves the automation and intelligence level of spontaneous combustion detection. The specific expression is:
[0060] Fhyp = CNNhyp(Ssmooth);
[0061] Among them, CNNhyp() is a hyperspectral convolutional neural sub-network, and Fhyp is a hyperspectral feature;
[0062] Another CNN is used to construct a sub-network that can extract infrared thermal imaging features. Through infrared thermal imaging technology, the temperature distribution pattern on the surface of the coal pile can be accurately identified and analyzed. This is crucial for detecting local overheating areas and can effectively identify potential problem points. The specific expression is:
[0063] Ftherm=CNNtherm(Icorrected);
[0064] Among them, CNNtherm is the infrared thermal imaging convolutional neural sub-network, Ftherm is the characteristics of temperature distribution and abnormal areas, and Icorrected is the infrared imaging data corrected by the NUC algorithm.
[0065] S4. Introduce the attention mechanism so that the coal pile spontaneous combustion detection model can automatically identify key areas, build a coal pile spontaneous combustion detection model, and learn the hyperspectral features and infrared imaging features before spontaneous combustion.
[0066] Furthermore, multimodal fusion is to perform weighted fusion of the hyperspectral features and infrared imaging features extracted by the two sub-networks through the fusion layer, assign different weights to the features of each modality, and perform weighted summation to improve the accuracy and effectiveness of data analysis. The specific expression is:
[0067] F fused =αFhyp+βFtherm;
[0068] Among them, F fused is the final fused feature vector, α is a weight parameter, β is another weight parameter, Fhyp represents the hyperspectral feature, and Ftherm represents the infrared thermal imaging feature.
[0069] Furthermore, the attention mechanism is introduced to build a coal pile spontaneous combustion detection model, as follows:
[0070] Through the multi-head attention mechanism, the fusion feature F fused Perform deep processing while maintaining the training stability of the model, and finally output an enhanced feature vector to provide more accurate input for subsequent spontaneous combustion probability prediction. The specific expression is:
[0071] F′=MultiHeadAttention(Q=F fused , K=F fused , V=F fused )+F fused ;
[0072] Among them, F′ is the enhanced feature vector after being processed by the multi-head attention mechanism and combined with the original fusion features, Q is the vector obtained by linearly transforming each position of the input sequence itself, K is used to match the query vector, and V is the actual value or information content. MultiHeadAttention() is a powerful tool for sequence data processing;
[0073] It should be noted that the multi-head attention mechanism can capture different aspects or features of the data by dividing the input features into multiple "heads" and processing them independently, and then combining the results. Next, these processed features are linearly transformed to adjust the feature dimension and scale. The "ReLU" activation function is then applied, which can introduce nonlinearity to help the spontaneous combustion detection model learn more complex patterns and enhance the expressiveness of the spontaneous combustion detection model. This series of operations ultimately generates an optimized feature vector that combines the diversity and deep nonlinear relationships from the original features, providing a richer and more sophisticated input representation for subsequent tasks. The specific expression is:
[0074] Z = ReLU(W z Flatten(F′)+b z );
[0075] Among them, Z is the input of the final output layer, and the activation function introduces nonlinearity into the neural network. z is a weight matrix, b z is a bias vector used to adjust the output after linear transformation. Flatten is an operation that converts a multi-dimensional input tensor (enhanced feature vector F′) into a one-dimensional vector.
[0076] The spontaneous combustion detection model finally outputs the probability of coal pile, which is expressed as:
[0077] P ignite =σ(W p ·Z+bp );
[0078] Among them, the σ() function will map any real number to the interval (0,1), W p is the linear transformation of the weight matrix from the previous layer to the current layer, b p is the bias term, P ignite Probability of spontaneous combustion.
[0079] S5. Input the collected real-time data into the coal pile spontaneous combustion detection model to conduct spontaneous combustion risk assessment and formulate strategies based on the assessment.
[0080] Furthermore, the spontaneous combustion risk assessment is as follows: define T1 (T1 = 0.5) as the dividing point between low risk and medium risk, and T2 (T1 = 1) as the dividing point between medium risk and high risk. The two thresholds are used to divide the level of spontaneous combustion risk. When the spontaneous combustion probability P(y|x) predicted by the model is less than T1, it is considered as low risk. When the spontaneous combustion probability P(y|x) predicted by the model is greater than or equal to T1 and less than T2, it is considered as medium risk. When the spontaneous combustion probability P(y|x) predicted by the model is less than T2, it is considered as high risk.
[0081] The specific strategies formulated based on the assessment are as follows: When the risk is low, maintain regular monitoring frequency and maintenance operations; when the risk is medium, increase the monitoring frequency, conduct physical inspections of local areas, and confirm whether there is local overheating; when the risk is high, prepare emergency supplies and personnel, formulate an evacuation plan, and use ventilation and water injection to physically cool down, reduce internal heat accumulation, and prevent spontaneous combustion.
[0082] This embodiment also provides a power plant coal pile spontaneous combustion alarm system based on deep learning, including a data acquisition module, a data preprocessing module, a feature extraction and fusion module, an attention mechanism module and a spontaneous combustion risk assessment module; the data acquisition module is used to use a hyperspectral imager and an infrared imager to periodically scan the coal pile at regular intervals to collect hyperspectral data and infrared imaging data; the data preprocessing module is used to use a sliding average method to reduce the influence of noise on the hyperspectral data to obtain accurate hyperspectral data, and use a NUC algorithm to eliminate fixed pattern noise on the infrared imaging data to obtain accurate infrared imaging data; the feature extraction and fusion module is used to extract features and perform multimodal fusion on the processed hyperspectral data and infrared imaging data through a deep convolutional neural network to obtain hyperspectral features and temperature features before spontaneous combustion;
[0083] The attention mechanism module is used to introduce the attention mechanism so that the coal pile spontaneous combustion detection model can automatically identify key areas, build a coal pile spontaneous combustion detection model, and learn the hyperspectral features and infrared imaging features before spontaneous combustion;
[0084] The spontaneous combustion risk assessment module is used to input the collected real-time data into the coal pile spontaneous combustion detection model to conduct spontaneous combustion risk assessment and formulate strategies based on the assessment.
[0085] This embodiment also provides a computer device, which is suitable for the case of a method for detecting spontaneous combustion of coal piles in power plants based on deep learning, and includes: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the method for detecting spontaneous combustion of coal piles in power plants based on deep learning as proposed in the above embodiment.
[0086] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.
[0087] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the artificial intelligence visual detection method based on linear discrimination as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.
[0088] In summary, the present invention ensures continuous monitoring of the state of the coal pile by: using a hyperspectral imager and an infrared imager to periodically scan the coal pile, timely capturing small changes in the early stage of spontaneous combustion, combining the data captured by the hyperspectral imager and the infrared imager to provide a basis for subsequent analysis, and improving the comprehensiveness and timeliness of monitoring; denoising the hyperspectral data by a sliding average method to reduce the impact of random noise and make the collected data more accurate; extracting features from hyperspectral and infrared data respectively by a deep convolutional neural network, and then integrating the complementary information between the two types of data through multimodal fusion, so that the model can more accurately judge the risk of spontaneous combustion in complex situations; by introducing an attention mechanism, the model can automatically focus on the key areas of spontaneous combustion risk in the coal pile, improve the detection targeting, and avoid excessive attention to irrelevant areas; by setting a risk level threshold, the natural probability can be quickly converted into a specific response strategy, achieving a seamless connection from monitoring to decision-making, and the division of low, medium and high risk levels makes emergency measures more reasonable and avoids unnecessary panic and waste of resources.
[0089] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A method for detecting spontaneous combustion of coal piles in power plants based on deep learning, characterized in that: include, Use a hyperspectral imager and an infrared imager to periodically scan the coal pile and collect hyperspectral data and infrared imaging data; The sliding average method is used on the hyperspectral data to reduce the influence of noise and obtain accurate hyperspectral data; The NUC algorithm is used to eliminate fixed pattern noise from infrared imaging data to obtain accurate infrared imaging data; The processed hyperspectral data and infrared imaging data are subjected to feature extraction and multimodal fusion by deep convolutional neural networks to obtain the hyperspectral features and temperature features before spontaneous combustion; The attention mechanism is introduced to enable the coal pile spontaneous combustion detection model to automatically identify key areas, build a coal pile spontaneous combustion detection model, and learn the hyperspectral features and infrared imaging features before spontaneous combustion; The collected real-time data is input into the coal pile spontaneous combustion detection model to conduct spontaneous combustion risk assessment, and strategies are formulated based on the assessment.
2. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 1, characterized in that: The hyperspectral data includes image data with high resolution and multi-band characteristics. The hyperspectral data performs spatial imaging of the target while recording detailed spectral information of each pixel point; The infrared imaging data is obtained by capturing the surface temperature of an object through an infrared imager, thereby reflecting the ambient temperature.
3. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 2, characterized in that: The specific use of the sliding average method for hyperspectral data is as follows: By defining a fixed-size window, the window slides point by point on the data sequence. For each window position, the average value of all data points in the window is calculated, and then this average value is assigned to the point at the center of the window. The specific expression is: Among them, Ssmooth(i) is the smoothed hyperspectral value, N is a positive integer, S(j) is the jth value in the original hyperspectral data, i is the index position of the hyperspectral data, and j is the index of each point in the sliding average window.
4. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 3, characterized in that: The deep convolutional neural network performs feature extraction as follows: A subnetwork capable of extracting hyperspectral features is constructed through CNN, local feature detection is performed, and hyperspectral images are processed to extract hyperspectral features related to chemical composition and material state. The specific expression is: Fhyp = CNNhyp(Ssmooth); Among them, CNNhyp is a hyperspectral convolutional neural sub-network, and Fhyp is a hyperspectral feature; Then, another CNN is used to construct a subnetwork that can extract infrared thermal imaging features. Through infrared thermal imaging technology, the features of temperature distribution and abnormal areas are extracted. The specific expression is: Ftherm=CNNtherm(Icorrected); Among them, CNNtherm is the infrared thermal imaging convolutional neural sub-network, Ftherm is the infrared thermal imaging feature, and Icorrected is the infrared imaging data corrected by the NUC algorithm.
5. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 4, characterized in that: The multimodal fusion is to perform weighted fusion of the hyperspectral features and infrared imaging features extracted by the two sub-networks through the fusion layer, assign different weights to the features of each modality, and perform weighted summation to reflect the importance of the two modalities. The specific expression is: F fused =αFhyp+βFtherm; Among them, F fused is the final fused feature vector, α is the hyperspectral feature weight coefficient, β is the infrared imaging feature weight coefficient, and Fhyp represents the hyperspectral feature.
6. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 5, characterized in that: The attention mechanism is introduced to construct a coal pile spontaneous combustion detection model, which is as follows: Through the multi-head attention mechanism MultiHeadAttention, the fusion feature F fused Perform deep processing while maintaining the training stability of the model, and finally output an enhanced feature vector. The specific expression is: F′=MultiHeadAttention(Q=F fused ,K=F fused ,V=F fused )+F fused ; Among them, F′ is the enhanced feature vector obtained after the multi-head attention mechanism, Q is the query vector obtained by linearly transforming each position of the input sequence itself, K is the key vector used to match the query vector, and V is the information content value vector; The input features are flattened through the multi-head attention mechanism, and the optimized feature vector is generated after linear transformation and ReLU activation. The specific expression is: Z=ReLU(W z Flat(F′)+b z ); Among them, Z is the input of the final output layer, and the activation function introduces nonlinearity into the neural network. z is a weight matrix, b z is the bias vector, and Flatten is the operation of converting the multi-dimensional F′ into a one-dimensional vector; The model finally outputs the probability of coal pile, which is expressed as: P ignite =σ(W p ·Z+b p ); Among them, σ is the sigmoid activation function, W p is the linear transformation of the weight matrix from the previous layer to the current layer, b p is the bias term, P ignite Probability of spontaneous combustion.
7. The method for detecting spontaneous combustion of coal piles in power plants based on deep learning according to claim 6, characterized in that: The spontaneous combustion risk assessment is as follows: Define T1 as the dividing point between low risk and medium risk, and T2 as the dividing point between medium risk and high risk. These two thresholds are used to classify the level of spontaneous combustion risk. When the spontaneous combustion probability P(y|x) predicted by the model is less than T1, it is considered to be a low risk; When the spontaneous combustion probability P(y|x) predicted by the model is greater than or equal to T1 and less than T2, it is considered to be a medium risk; When the spontaneous combustion probability P(y|x) predicted by the model is less than T2, it is considered to be a high risk; When the risk is low, maintain routine monitoring frequency and maintenance operations. When the risk is medium, increase the monitoring frequency and conduct physical inspections of local areas to confirm whether there is local overheating. When the risk is high, prepare emergency supplies and personnel, develop an evacuation plan, and take physical cooling measures to reduce internal heat accumulation.
8. A power plant coal pile spontaneous combustion alarm system based on deep learning, based on the power plant coal pile spontaneous combustion detection method based on deep learning according to any one of claims 1 to 7, characterized in that: Including data acquisition module, data preprocessing module, feature extraction and fusion module, attention mechanism module and spontaneous combustion risk assessment module; A data acquisition module is used to periodically scan the coal pile using a hyperspectral imager and an infrared imager to collect hyperspectral data and infrared imaging data; The data preprocessing module is used to use the sliding average method to reduce the noise impact on the hyperspectral data to obtain accurate hyperspectral data, and to use the NUC algorithm to eliminate fixed pattern noise on the infrared imaging data to obtain accurate infrared imaging data; The feature extraction and fusion module is used to extract features and fuse multimodal data from the processed hyperspectral data and infrared imaging data through a deep convolutional neural network to obtain the hyperspectral features and temperature features before spontaneous combustion. The attention mechanism module is used to introduce the attention mechanism so that the coal pile spontaneous combustion detection model can automatically identify key areas, build a coal pile spontaneous combustion detection model, and learn the hyperspectral features and infrared imaging features before spontaneous combustion; The spontaneous combustion risk assessment module is used to input the collected real-time data into the coal pile spontaneous combustion detection model to conduct spontaneous combustion risk assessment and formulate strategies based on the assessment.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the power plant coal pile spontaneous combustion detection method based on deep learning as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the power plant coal pile spontaneous combustion detection method based on deep learning as described in any one of claims 1 to 7 are implemented.
Citation Information
Cited By
Coal spontaneous combustion early warning method based on deep learning and physical constraint
CN120823700A
