Hydraulic power plant safety protection method and system

Abstract

The embodiment of the invention relates to the technical field of hydraulic power plant safety protection, and provides a hydraulic power plant safety protection method and system, and the method comprises the steps: carrying out the multi-level analysis of a factory monitoring image of a hydraulic power plant through employing an artificial intelligence technology based on deep learning, and extracting the deep features and shallow features of the factory monitoring image, the deep semantic information expression of the image is enhanced by performing feature selection optimization on the deep features of the image, so that whether a fire occurs or not is intelligently identified by performing joint perception on the deep and deep features of the image, and corresponding alarm prompt is performed when the fire is identified. The intelligent level of fire safety protection of the hydraulic power plant can be remarkably improved, the false alarm rate is reduced, the system can adapt to the complex and changeable factory environment, and the overall safety level is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of safety protection of hydropower plants, and in particular to a safety protection method and system for hydropower plants. Background Art

[0002] With the development of digital and intelligent technologies, the power generation industry, especially hydropower stations, is undergoing a profound transformation. Modern hydropower plants must not only ensure the efficiency and stability of power production, but also the safety of equipment and personnel. In this process, fire safety is a crucial link in the operation and management of hydropower stations.

[0003] Traditional fire safety supervision methods for hydropower plants mainly rely on hardware equipment such as smoke alarms and temperature detectors. Although this method can improve the fire safety of hydropower plants to a certain extent, it is easily affected by environmental factors and has problems such as slow response speed, limited monitoring range, and high false alarm rate. Especially in complex and changeable plant environments, such as extreme conditions such as high temperature, high humidity, and dust, traditional fire monitoring methods often fail to achieve ideal protection effects. Summary of the invention

[0004] The present disclosure aims to solve at least one of the problems existing in the prior art and provide a hydropower plant safety protection method and system.

[0005] In one aspect of the present disclosure, a hydropower plant safety protection method is provided, the hydropower plant safety protection method comprising:

[0006] Collect monitoring images of the hydropower plant;

[0007] Inputting the factory monitoring images into the security protection cloud processing center;

[0008] Utilizing the security protection cloud processing center, multi-level feature extraction is performed on the factory monitoring image to obtain a shallow feature map of the factory status and a deep feature map of the factory status;

[0009] Performing feature selection and enhancement on the deep feature map of the plant state to obtain an enhanced deep feature map of the plant state;

[0010] Determining whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map;

[0011] In response to detecting a fire, a fire safety protection alarm prompt is generated.

[0012] Optionally, multi-level feature extraction is performed on the factory monitoring image to obtain a shallow feature map of the factory state and a deep feature map of the factory state, including:

[0013] Performing noise reduction processing on the factory monitoring image to obtain a noise-reduced factory monitoring image;

[0014] The denoised factory monitoring image is input into a multi-scale extractor of factory status features based on a pyramid network to obtain a shallow feature map of the factory status and a deep feature map of the factory status.

[0015] Optionally, feature selection and enhancement are performed on the deep feature map of the plant state to obtain an enhanced deep feature map of the plant state, including:

[0016] Reshaping the deep feature map of the plant state to obtain a set of deep local feature vectors of the plant state;

[0017] Inputting the set of deep local feature vectors of the plant state into a feature sequence selection optimization module based on a feature energy distribution spectrum to obtain a selected set of deep local feature vectors of the plant state;

[0018] The set of the selected deep local feature vectors of the plant state is reshaped to obtain the enhanced deep feature map of the plant state.

[0019] Optionally, the set of deep local feature vectors of the plant state is input into a feature sequence selection optimization module based on a characteristic energy distribution spectrum to obtain a selected set of deep local feature vectors of the plant state, including:

[0020] Calculating the characteristic energy distribution spectrum vector of each deep local feature vector of the plant state in the set of deep local feature vectors of the plant state to obtain a set of deep local feature energy distribution spectrum vectors of the plant state;

[0021] Calculating the position mean vector of the set of deep local characteristic energy distribution spectrum vectors of the plant state as the center vector of the energy spectrum distribution;

[0022] Calculating the Mahalanobis distance between the energy spectrum distribution center vector and each plant state deep local characteristic energy distribution spectrum vector in the set of plant state deep local characteristic energy distribution spectrum vectors as the energy distribution spectrum span factor, and obtaining a set of energy distribution spectrum span factors;

[0023] Each energy distribution spectrum span factor in the set of energy distribution spectrum span factors is compared with a preset threshold, and the deep local feature vectors of the plant state corresponding to the energy distribution spectrum span factors less than the preset threshold are selected to form the set of the selected deep local feature vectors of the plant state.

[0024] Optionally, calculating the characteristic energy distribution spectrum vector of each plant state deep local feature vector in the set of plant state deep local feature vectors to obtain the set of plant state deep local feature energy distribution spectrum vectors includes:

[0025] Calculate the feature distribution energy synergistic representation vector between the i-th deep local feature vector of the plant area state and each other deep local feature vector of the plant area state in the set of deep local feature vectors of the plant area state in a manner of taking the maximum eigenvalue element by element, and obtain a set of feature distribution energy synergistic representation vectors;

[0026] The characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of characteristic distribution energy synergy representation vectors is calculated to obtain the characteristic energy distribution spectrum vector of the i-th plant state deep local characteristic vector composed of multiple characteristic distribution energy synergy factors.

[0027] Optionally, calculating the characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of characteristic distribution energy synergy representation vectors to obtain the characteristic energy distribution spectrum vector of the i-th plant state deep local characteristic vector composed of multiple characteristic distribution energy synergy factors includes:

[0028] The square of the difference between the maximum eigenvalue of the feature distribution energy synergistic representation vector and its eigenmean, the feature variance of the feature distribution energy synergistic representation vector, and the weighted sum of the bias term are calculated, and then the feature variance of the feature distribution energy synergistic representation vector and the bias term are added and divided by the weighted sum to obtain the feature distribution energy synergy factor.

[0029] Optionally, determining whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map includes:

[0030] Inputting the plant state enhanced deep feature map and the plant state shallow feature map into a salient joint perception module based on an attention mechanism to obtain a plant state multi-scale joint perception salient feature map;

[0031] The multi-scale joint perception salient feature map of the plant state is input into a classifier-based fire recognition module to obtain a recognition result, and the recognition result is used to indicate whether a fire is detected.

[0032] Optionally, the plant state enhanced deep feature map and the plant state shallow feature map are input into a salient joint perception module based on an attention mechanism to obtain a plant state multi-scale joint perception salient feature map, including:

[0033] Reshaping the plant state enhanced deep feature map and the plant state shallow feature map to obtain a plant state enhanced deep feature shape reshaping matrix and a plant state shallow feature shape reshaping matrix;

[0034] Input the plant state enhanced deep feature shape reshaping matrix and the plant state shallow feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain a deep-shallow dependency matrix and a shallow-deep dependency matrix;

[0035] Inputting the deep-shallow dependency matrix and the shallow-deep dependency matrix into a random dropout module to obtain a pruned deep-shallow dependency matrix and a pruned shallow-deep dependency matrix;

[0036] Performing matrix multiplication of the pruned deep-shallow dependency matrix and the plant state shallow feature shape reshaping matrix to obtain a dependency-optimized plant state shallow feature matrix;

[0037] Performing matrix multiplication of the pruned shallow-deep dependency matrix and the plant state enhanced deep feature shape reshaping matrix to obtain a dependency optimized plant state enhanced deep feature matrix;

[0038] Reshaping the feature shapes of the dependency-optimized plant state enhancement deep feature matrix and the dependency-optimized plant state shallow feature matrix to obtain an optimized plant state enhancement deep feature map and an optimized plant state shallow feature map;

[0039] The weighted sum of the optimized plant state enhanced deep feature map and the optimized plant state shallow feature map is calculated to obtain the multi-scale joint perception salient feature map of the plant state.

[0040] Optionally, the plant state enhanced deep feature shape reshaping matrix and the plant state shallow feature shape reshaping matrix are input into a feature channel-by-channel interactive perception module to obtain a deep-shallow dependency matrix and a shallow-deep dependency matrix, including:

[0041] Multiplying the plant state enhanced deep feature shape reshaping matrix by the transposed matrix of the plant state shallow feature shape reshaping matrix to obtain a deep-shallow correlation representation matrix;

[0042] The deep-shallow correlation representation matrix is ​​divided by the scale of the shallow feature shape reshaping matrix of the plant state and then input into the softmax function to obtain the deep-shallow dependency matrix;

[0043] Calculate the transposed matrix of the plant state enhanced deep feature shape reshaping matrix and multiply it by the plant state shallow feature shape reshaping matrix to obtain the plant state shallow-deep correlation representation matrix;

[0044] The shallow-deep correlation representation matrix is ​​divided by the scale of the plant state enhanced deep feature shape reshaping matrix and then input into the softmax function to obtain the shallow-deep dependency matrix.

[0045] Another aspect of the present disclosure provides a hydropower plant safety protection system, the hydropower plant safety protection system comprising:

[0046] The plant monitoring module is used to collect plant monitoring images of the hydropower plant;

[0047] A data transmission module, used to input the factory monitoring images into a security protection cloud processing center;

[0048] A factory monitoring image feature extraction module is used to use the security protection cloud processing center to perform multi-level feature extraction on the factory monitoring image to obtain a factory state shallow feature map and a factory state deep feature map;

[0049] A feature selection and enhancement module is used to perform feature selection and enhancement on the deep feature map of the plant state to obtain an enhanced deep feature map of the plant state;

[0050] A fire identification module, used to determine whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map;

[0051] The alarm prompt generation module is used to generate a fire safety protection alarm prompt in response to detecting a fire.

[0052] Compared with the prior art, the present invention adopts artificial intelligence technology based on deep learning to perform multi-level analysis on the plant monitoring images of the hydropower plant, extracts the deep features and shallow features of the plant monitoring images, and optimizes the feature selection of the deep features of the images to enhance the deep semantic information expression of the images, and then through the joint perception of the deep and shallow features of the images, it is possible to intelligently identify whether a fire has occurred, and issue corresponding alarm prompts when a fire is identified. This can not only significantly improve the intelligence level of fire safety protection in the hydropower plant and reduce the false alarm rate, but also adapt to the complex and changeable plant environment and improve the overall safety level. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplifications do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0054] Figure 1A flow chart of a hydropower plant safety protection method provided in one embodiment of the present disclosure;

[0055] Figure 2 A data flow diagram of a hydropower plant safety protection method provided by another embodiment of the present disclosure;

[0056] Figure 3 A flow chart of step S4 in a hydropower plant safety protection method provided in another embodiment of the present disclosure;

[0057] Figure 4 A flowchart of step S42 in a hydropower plant safety protection method provided in another embodiment of the present disclosure;

[0058] Figure 5 A flowchart of step S5 in a hydropower plant safety protection method provided in another embodiment of the present disclosure;

[0059] Figure 6 A flowchart of step S51 in a hydropower plant safety protection method provided in another embodiment of the present disclosure;

[0060] Figure 7 A schematic structural diagram of a hydropower plant safety protection system provided in another embodiment of the present disclosure. DETAILED DESCRIPTION

[0061] In view of the problems existing in traditional fire safety supervision methods for hydropower plants, in order to significantly improve the intelligence level of fire safety protection in hydropower plants, reduce the false alarm rate, adapt to the complex and changeable plant environment, and improve the overall safety level, the present invention discloses a hydropower plant safety protection method and system, and its technical concept is: using deep learning-based artificial intelligence technology to perform multi-level analysis on plant monitoring images, extracting deep features and shallow features of the plant monitoring images, and optimizing feature selection of the deep features of the images to enhance the deep semantic information expression of the images, and then through joint perception of the deep and shallow features of the images, intelligently identify whether a fire has occurred, and issue corresponding alarm prompts when a fire is identified.

[0062] In order to make the purpose, technical scheme and advantages of the embodiments of the present disclosure clearer, the embodiments of the present disclosure will be described in detail below in conjunction with the accompanying drawings. However, it can be understood by those skilled in the art that in each embodiment of the present disclosure, many technical details are proposed in order to enable readers to better understand the present disclosure. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed for protection in the present disclosure can also be implemented. The division of the following embodiments is for the convenience of description and should not constitute any limitation on the specific implementation of the present disclosure. The various embodiments can be combined and referenced with each other without contradiction.

[0063] One embodiment of the present disclosure relates to a hydropower plant safety protection method, the process of which is as follows: Figure 1 As shown, the data flow is as follows Figure 2 shown.

[0064] Combined Figure 1 and Figure 2 The hydropower plant safety protection method provided in this embodiment includes the following steps:

[0065] Step S1, collecting monitoring images of the hydropower plant.

[0066] Specifically, step S1 can collect monitoring images of the hydropower plant through monitoring cameras installed at different locations in the hydropower plant. As a front-end sensing device for the safety protection of the hydropower plant, the monitoring camera can capture image information of the plant in real time, which helps to discover potential safety hazards in a timely manner. Compared with traditional smoke alarms and temperature detectors, camera monitoring not only has a wider coverage, but also can capture more visual detail information, providing a more reliable data source for fire identification.

[0067] Step S2, inputting the factory monitoring image into the security protection cloud processing center.

[0068] Specifically, step S2 can input the factory monitoring image into the security protection cloud processing center through a wireless communication module that is connected to the security protection cloud processing center, so as to utilize the powerful computing power and storage resources of the cloud to perform image processing and analysis on the factory monitoring image.

[0069] Step S3, using the security protection cloud processing center, performing multi-level feature extraction on the factory monitoring image to obtain a shallow feature map of the factory status and a deep feature map of the factory status.

[0070] Specifically, considering that there may be interference from various factors such as light changes and equipment vibration in the actual factory monitoring environment, which may cause noise in the factory monitoring image, thereby affecting the accuracy of subsequent fire identification, therefore, step S3 can first perform noise reduction processing on the factory monitoring image before performing multi-level feature extraction on the factory monitoring image to eliminate or reduce the impact of noise on image quality and improve the signal-to-noise ratio of the image, thereby obtaining a noise-reduced factory monitoring image, providing a clearer and more accurate image basis for subsequent image feature extraction and fire identification. For example, the noise reduction processing here can be, but is not limited to, processing the factory monitoring image using a denoising algorithm such as a Gaussian filter and a median filter.

[0071] In order to better adapt to the complex factory environment and achieve efficient and accurate fire identification, step S3 can use a multi-scale extractor of factory state features based on a pyramid network to perform multi-level feature extraction on the denoised factory monitoring image, so as to make full use of the shallow detail information and deep semantic information in the image and more comprehensively describe the factory state. Specifically, the pyramid network stacks multiple convolutional layers of different scales to form a hierarchical structure similar to the shape of a pyramid. As the number of network layers increases, the resolution of the feature map gradually decreases, but the semantic information contained is richer. By extracting image feature representations from different network levels, the local details and global structure of the image can be effectively captured, and the shallow feature map of the factory state and the deep feature map of the factory state can be obtained, thereby providing a comprehensive and in-depth feature representation for subsequent fire identification.

[0072] That is to say, in step S3, multi-level feature extraction is performed on the factory monitoring image to obtain a shallow feature map of the factory status and a deep feature map of the factory status, which may include: denoising the factory monitoring image to obtain a denoised factory monitoring image; inputting the denoised factory monitoring image into a multi-scale extractor of factory status features based on a pyramid network to obtain the shallow feature map of the factory status and the deep feature map of the factory status.

[0073] Step S4, performing feature selection and enhancement on the deep feature map of the factory status to obtain an enhanced deep feature map of the factory status.

[0074] Specifically, as the number of network layers increases, some redundant information or noise interference may be introduced into the deep feature map of the plant state, which will not only increase the computational complexity, but also have a negative impact on the accuracy of fire identification. Therefore, in order to further enhance the deep semantic feature representation capability of the image and improve the accuracy of fire identification, step S4 further performs feature selection enhancement on the deep feature map of the plant state.

[0075] Exemplary, combined Figure 3 , step S4 comprises:

[0076] Step S41, reshape the plant state deep feature map to obtain a set of plant state deep local feature vectors, wherein each plant state deep local feature vector corresponds to a different spatial position in the plant state deep feature map.

[0077] Step S42, inputting the set of deep local feature vectors of the plant state into a feature sequence selection optimization module based on the characteristic energy distribution spectrum to obtain a selected set of deep local feature vectors of the plant state.

[0078] Specifically, step S42 inputs the set of deep local feature vectors of the plant state into the feature sequence selection optimization module based on the characteristic energy distribution spectrum for processing, and performs feature screening and enhancement by evaluating the importance of each local feature representation. Among them, the feature sequence selection optimization module first calculates the characteristic energy distribution spectrum vector of each deep local feature vector of the plant state to map it to the energy distribution spectrum space, revealing the distribution of each deep local feature vector of the plant state at different energy levels. Then, the position mean vector of the sequence of characteristic energy distribution spectrum vectors is further calculated to obtain the energy spectrum distribution center vector, and this is used as a reference to evaluate the closeness of each deep local feature vector of the plant state to the overall distribution by calculating the difference between the energy spectrum distribution center vector and each characteristic energy distribution spectrum vector, and generate the corresponding energy distribution spectrum span factor to describe the discrete degree of the feature vector. Then, the energy distribution spectrum span factor is compared with the preset threshold, and the deep local feature vector of the plant state corresponding to the energy distribution spectrum span factor less than the preset threshold is selected to form a new set, thereby achieving effective filtering of redundant information and noise interference, and only retaining the feature representation important for fire identification.

[0079] That is to say, exemplary, combined Figure 4 , step S42 specifically includes:

[0080] Step S421, calculating the characteristic energy distribution spectrum vector of each deep local feature vector of the plant state in the set of deep local feature vectors of the plant state, and obtaining the set of deep local feature energy distribution spectrum vectors of the plant state.

[0081] Exemplarily, step S421 specifically includes: calculating the characteristic distribution energy synergy representation vector between the i-th deep local feature vector of the plant state and each other deep local feature vector of the plant state in the set of deep local feature vectors of the plant state by taking the maximum eigenvalue element by element, to obtain a set of characteristic distribution energy synergy representation vectors; calculating the characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of characteristic distribution energy synergy representation vectors, to obtain the characteristic energy distribution spectrum vector of the i-th deep local feature vector of the plant state composed of multiple characteristic distribution energy synergy factors.

[0082] Exemplarily, calculating the characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of the characteristic distribution energy synergy representation vectors to obtain the characteristic energy distribution spectrum vector of the i-th plant state deep local characteristic vector composed of multiple characteristic distribution energy synergy factors further includes: calculating the square of the difference between the maximum eigenvalue of the characteristic distribution energy synergy representation vector and its characteristic mean, the characteristic variance of the characteristic distribution energy synergy representation vector, and the weighted sum of the bias terms, and then adding the characteristic variance of the characteristic distribution energy synergy representation vector and the bias term and dividing the result by the weighted sum to obtain the characteristic distribution energy synergy factor.

[0083] Step S422, calculating the positional mean vector of the set of deep local characteristic energy distribution spectrum vectors of the plant state as the center vector of the energy spectrum distribution.

[0084] Step S423, calculate the Mahalanobis distance between the energy spectrum distribution center vector and each plant state deep local characteristic energy distribution spectrum vector in the set of plant state deep local characteristic energy distribution spectrum vectors as the energy distribution spectrum span factor, and obtain a set of energy distribution spectrum span factors.

[0085] Step S424, compare each energy distribution spectrum span factor in the set of energy distribution spectrum span factors with a preset threshold, and select the deep local feature vectors of the plant state corresponding to the energy distribution spectrum span factors less than the preset threshold to form the set of selected deep local feature vectors of the plant state.

[0086] In other words, step S42 specifically includes: using the following feature selection formula to process the set of deep local feature vectors of the plant state to obtain a selected set of deep local feature vectors of the plant state, wherein the feature selection formula is:

[0087] X i，j =max(X i , X j )

[0088]

[0089] P i ={e i，1 , e i，2 , ..., e i，n}

[0090]

[0091]

[0092] X′={V i |Dist(c,Pi )≤θ}

[0093] Among them, X i and X j denote the ith and ith deep local feature vectors of the plant state in the set of the deep local feature vectors of the plant state, respectively; max(·, ·) denotes taking the maximum eigenvalue of the two feature vectors element by element; X i，j The characteristic distribution energy synergy representation vector between the i-th plant state deep local feature vector and the j-th plant state deep local feature vector, μ, σ 2 and max(X i，j ) represent the characteristic mean, characteristic variance and maximum eigenvalue of the characteristic distribution energy cooperative representation vector, λ is the bias term, e i，1 、e i，2 、e i，j and e i，n represents the characteristic distribution energy synergy factor between the i-th deep local feature vector of the plant state and the first, second, th and n-th deep local feature vector of the plant state in the set of the deep local feature vector of the plant state, P i represents the deep local characteristic energy distribution spectrum vector of the i-th plant state, P represents the set of deep local characteristic energy distribution spectrum vectors of the plant state, the value of n is the number of deep local characteristic vectors of the plant state, c represents the center vector of the energy spectrum distribution, S i，c represents the covariance matrix between the deep local characteristic energy distribution spectrum vector of the i-th plant state and the center vector of the energy spectrum distribution, Dist(·, ·) represents the distance metric function, D i represents the span factor of the i-th energy distribution spectrum, θ represents the preset threshold, X′ represents the set of deep local feature vectors of the selected plant state, V i Represents the selected deep local feature vector of the i-th plant state.

[0094] Step S43, reshaping the selected set of deep local feature vectors of the factory state to obtain the enhanced deep feature map of the factory state.

[0095] Step S5, determining whether a fire is detected based on the significant joint features of the plant status enhanced deep feature map and the plant status shallow feature map.

[0096] Specifically, since the plant status enhanced deep feature map and the plant status shallow feature map describe the deep semantic information and shallow detail information of the image from different levels respectively, in order to further comprehensively utilize the complementary information of the two to improve the accuracy of fire identification, a salient joint perception module based on the attention mechanism can be introduced to jointly perceive the plant status enhanced deep feature map and the plant status shallow feature map, so as to optimize and fuse the features of the two based on the correlation dependency between the two, thereby significantly improving the feature fusion effect and expression ability.

[0097] Therefore, if Figure 5 As shown, step S5 specifically includes:

[0098] Step S51, inputting the plant state enhanced deep feature map and the plant state shallow feature map into a salient joint perception module based on an attention mechanism to obtain a plant state multi-scale joint perception salient feature map.

[0099] Specifically, the salient joint perception module based on the attention mechanism can first reshape the plant state enhanced deep feature map and the plant state shallow feature map into feature matrices so that their dimensions and shapes adapt to the input requirements of the attention mechanism. Then, based on the dual attention interaction mechanism, by calculating the bidirectional dependency between the two sets of feature matrices, the interaction between the features is quantified, and the deep-shallow and shallow-deep dependency matrices are generated, which serve as the guidance for the feature attention optimization fusion. At the same time, in order to reduce the model's sensitivity to noise and prevent overfitting, the two sets of dependency matrices are further subjected to random inactivation processing, and the generalization ability of the model is enhanced by selectively ignoring or suppressing part of the feature interaction. Furthermore, based on the two sets of dependency matrices after random inactivation processing, the two sets of feature matrices after reshaping are weighted optimized, and then restored to the original feature map form by reshaping and weighted fusion, to generate the salient feature map of multi-scale joint perception of the plant state, thereby achieving high-precision feature representation of the plant environment state.

[0100] For example, Figure 6 As shown, step S51 specifically includes:

[0101] Step S511, reshaping the plant state enhanced deep feature map and the plant state shallow feature map to obtain a plant state enhanced deep feature shape reshaping matrix and a plant state shallow feature shape reshaping matrix.

[0102] Step S512, input the factory state enhanced deep feature shape reshaping matrix and the factory state shallow feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain a deep-shallow dependency matrix and a shallow-deep dependency matrix.

[0103] Specifically, step S512 further includes: multiplying the plant state enhanced deep feature shape reshaping matrix by the transposed matrix of the plant state shallow feature shape reshaping matrix to obtain a deep-shallow association representation matrix; dividing the deep-shallow association representation matrix by the scale of the plant state shallow feature shape reshaping matrix and inputting it into a softmax function to obtain the deep-shallow dependency matrix; calculating the transposed matrix of the plant state enhanced deep feature shape reshaping matrix and multiplying it by the plant state shallow feature shape reshaping matrix to obtain a plant state shallow-deep association representation matrix; dividing the shallow-deep association representation matrix by the scale of the plant state enhanced deep feature shape reshaping matrix and inputting it into a softmax function to obtain the shallow-deep dependency matrix.

[0104] Step S513: input the deep-shallow dependency matrix and the shallow-deep dependency matrix into a random dropout module to obtain a pruned deep-shallow dependency matrix and a pruned shallow-deep dependency matrix.

[0105] Step S514, performing matrix multiplication on the pruned deep-shallow dependency matrix and the plant state shallow feature shape reshaping matrix to obtain a dependency-optimized plant state shallow feature matrix.

[0106] Step S515, performing matrix multiplication on the pruned shallow-deep dependency matrix and the plant state enhanced deep feature shape reshaping matrix to obtain a dependency optimized plant state enhanced deep feature matrix.

[0107] Step S516, reshaping the feature shapes of the dependency-optimized plant state enhanced deep feature matrix and the dependency-optimized plant state shallow feature matrix to obtain an optimized plant state enhanced deep feature map and an optimized plant state shallow feature map.

[0108] Step S517, calculating the weighted sum of the optimized factory state enhanced deep feature map and the optimized factory state shallow feature map to obtain the multi-scale joint perception salient feature map of the factory state.

[0109] In other words, step S51 specifically includes: processing the plant state enhanced deep feature map and the plant state shallow feature map according to the following joint perception formula to obtain a plant state multi-scale joint perception significant feature map, wherein the joint perception formula is:

[0110] M1=Reshape(F1)

[0111] M2=Reshape(F2)

[0112]

[0113]

[0114] W 21 = Dropout(M 21 )

[0115] W 12 = Dropout(M 12 )

[0116]

[0117]

[0118] F1'=Reshape(M1')

[0119] F2'=Reshape(M2')

[0120] F c =αF1'+βF2'

[0121] Wherein, F1 represents the plant state enhanced deep feature map, F2 represents the plant state shallow feature map, reshape(·) represents feature shape reshaping, M1 represents the plant state enhanced deep feature shape reshaping matrix, M2 represents the plant state shallow feature shape reshaping matrix, represents matrix multiplication, (·) T represents the transpose of the matrix, S represents the scale of the plant state enhanced deep feature shape reshaping matrix, that is, the width of the plant state enhanced deep feature shape reshaping matrix multiplied by the height, and the scale of the plant state enhanced deep feature shape reshaping matrix and the plant state shallow feature shape reshaping matrix are the same, Softmax is a normalized exponential function, M 21 represents the deep-shallow dependency matrix, M 12 represents the shallow-deep dependency matrix, Dropout(·) represents random dropout processing, W 21 represents the pruned deep-shallow dependency matrix, W 12 represents the pruned shallow-deep dependency matrix, M1′ represents the dependency optimization plant state enhanced deep feature matrix, M2′ represents the dependency optimization plant state shallow feature matrix, F1′ represents the optimized plant state enhanced deep feature map, F2′ represents the optimized plant state shallow feature map, α and β represent different weight coefficients respectively, F c A multi-scale joint perception salient feature map representing the factory state.

[0122] Step S52, inputting the multi-scale joint perception salient feature map of the plant state into a classifier-based fire recognition module to obtain a recognition result, wherein the recognition result is used to indicate whether a fire is detected.

[0123] Specifically, after training and optimization, the fire recognition module based on the classifier can effectively learn the correlation mapping relationship between image features and fire, and can accurately determine whether there is a fire situation based on the feature representation of the multi-scale joint perception significant feature map of the plant state. This embodiment does not limit the specific type of the classifier, as long as the fire recognition module based on the classifier can determine whether there is a fire situation based on the feature representation of the multi-scale joint perception significant feature map of the plant state.

[0124] Step S6, in response to detecting a fire, generating a fire safety protection alarm prompt.

[0125] Compared with the prior art, the hydropower plant safety protection method provided in the embodiment of the present disclosure adopts artificial intelligence technology based on deep learning to perform multi-level analysis on the plant monitoring images of the hydropower plant, extract the deep features and shallow features of the plant monitoring images, and optimize the feature selection of the deep features of the images to enhance the deep semantic information expression of the images, and then through the joint perception of the deep and shallow features of the images, it is possible to intelligently identify whether a fire has occurred, and issue corresponding alarm prompts when a fire is identified. This can not only significantly improve the intelligence level of fire safety protection in hydropower plants and reduce the false alarm rate, but also adapt to the complex and changeable plant environment and improve the overall safety level.

[0126] Another embodiment of the present disclosure relates to a hydropower plant safety protection system 100, such as Figure 7 As shown, including:

[0127] The plant monitoring module 110 is used to collect plant monitoring images of the hydropower plant.

[0128] The data transmission module 120 is used to input the factory monitoring image into the security protection cloud processing center.

[0129] The factory monitoring image feature extraction module 130 is used to use the security protection cloud processing center to perform multi-level feature extraction on the factory monitoring image to obtain a shallow feature map of the factory status and a deep feature map of the factory status.

[0130] The feature selection and enhancement module 140 is used to perform feature selection and enhancement on the deep feature map of the plant status to obtain an enhanced deep feature map of the plant status.

[0131] The fire identification module 150 is used to determine whether a fire is detected based on the significant joint features of the plant status enhanced deep feature map and the plant status shallow feature map.

[0132] The alarm prompt generating module 160 is used to generate a fire safety protection alarm prompt in response to detecting a fire.

[0133] The specific implementation method of the hydropower plant safety protection system provided in the embodiment of the present disclosure can be found in the description of the hydropower plant safety protection method provided in the embodiment of the present disclosure, and will not be repeated here.

[0134] Compared with the prior art, the hydropower plant safety protection system provided in the embodiment of the present disclosure adopts artificial intelligence technology based on deep learning to perform multi-level analysis on the plant monitoring images of the hydropower plant, extract the deep features and shallow features of the plant monitoring images, and optimize the feature selection of the deep features of the images to enhance the deep semantic information expression of the images, and then through the joint perception of the deep and shallow features of the images, it can intelligently identify whether a fire has occurred, and issue corresponding alarm prompts when a fire is identified. This can not only significantly improve the intelligence level of fire safety protection in hydropower plants and reduce the false alarm rate, but also adapt to the complex and changeable plant environment and improve the overall safety level.

[0135] The basic principles of the technical solutions provided by the present disclosure are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present disclosure. In addition, the specific details of the above embodiments are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present disclosure to being implemented by adopting the above specific details.

[0136] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the module division is only a logical function division, and there may be other division methods in actual implementation. The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0137] In addition, each functional module in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0138] Those skilled in the art will appreciate that the above-mentioned embodiments are specific embodiments for implementing the present disclosure, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present disclosure.

[0139] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units included in the system can also be implemented by one unit through software or hardware.

[0140] Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present disclosure and are not intended to limit it. Although the present disclosure is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present disclosure can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present disclosure.

Claims

1. A hydropower plant safety protection method, characterized in that: The hydropower plant safety protection method comprises: Collect monitoring images of the hydropower plant; Inputting the factory monitoring images into the security protection cloud processing center; Utilizing the security protection cloud processing center, multi-level feature extraction is performed on the factory monitoring image to obtain a shallow feature map of the factory status and a deep feature map of the factory status; Performing feature selection and enhancement on the deep feature map of the plant state to obtain an enhanced deep feature map of the plant state; Determining whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map; In response to detecting a fire, a fire safety protection alarm prompt is generated.

2. The hydropower plant safety protection method according to claim 1, characterized in that: Perform multi-level feature extraction on the factory monitoring image to obtain a shallow feature map of the factory state and a deep feature map of the factory state, including: Performing noise reduction processing on the factory monitoring image to obtain a noise-reduced factory monitoring image; The denoised factory monitoring image is input into a multi-scale extractor of factory status features based on a pyramid network to obtain a shallow feature map of the factory status and a deep feature map of the factory status.

3. The hydropower plant safety protection method according to claim 1, characterized in that: The plant state deep feature map is subjected to feature selection and enhancement to obtain a plant state enhanced deep feature map, including: Reshaping the deep feature map of the plant state to obtain a set of deep local feature vectors of the plant state; Inputting the set of deep local feature vectors of the plant state into a feature sequence selection optimization module based on a feature energy distribution spectrum to obtain a selected set of deep local feature vectors of the plant state; The set of the selected deep local feature vectors of the plant state is reshaped to obtain the enhanced deep feature map of the plant state.

4. The hydropower plant safety protection method according to claim 3, characterized in that: The set of deep local feature vectors of the plant state is input into a feature sequence selection optimization module based on a feature energy distribution spectrum to obtain a selected set of deep local feature vectors of the plant state, including: Calculating the characteristic energy distribution spectrum vector of each deep local feature vector of the plant state in the set of deep local feature vectors of the plant state to obtain a set of deep local feature energy distribution spectrum vectors of the plant state; Calculating the position mean vector of the set of deep local characteristic energy distribution spectrum vectors of the plant state as the center vector of the energy spectrum distribution; Calculating the Mahalanobis distance between the energy spectrum distribution center vector and each plant state deep local characteristic energy distribution spectrum vector in the set of plant state deep local characteristic energy distribution spectrum vectors as the energy distribution spectrum span factor, and obtaining a set of energy distribution spectrum span factors; Each energy distribution spectrum span factor in the set of energy distribution spectrum span factors is compared with a preset threshold, and the deep local feature vectors of the plant state corresponding to the energy distribution spectrum span factors less than the preset threshold are selected to form the set of the selected deep local feature vectors of the plant state.

5. The hydropower plant safety protection method according to claim 4, characterized in that: Calculating the characteristic energy distribution spectrum vector of each deep local feature vector of the plant state in the set of deep local feature vectors of the plant state to obtain the set of deep local feature energy distribution spectrum vectors of the plant state, including: Calculate the feature distribution energy synergistic representation vector between the i-th deep local feature vector of the plant area state and each other deep local feature vector of the plant area state in the set of deep local feature vectors of the plant area state in a manner of taking the maximum eigenvalue element by element, and obtain a set of feature distribution energy synergistic representation vectors; The characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of characteristic distribution energy synergy representation vectors is calculated to obtain the characteristic energy distribution spectrum vector of the i-th plant state deep local characteristic vector composed of multiple characteristic distribution energy synergy factors.

6. The hydropower plant safety protection method according to claim 5, characterized in that: Calculating the characteristic distribution energy synergy factor of each characteristic distribution energy synergy representation vector in the set of characteristic distribution energy synergy representation vectors to obtain the characteristic energy distribution spectrum vector of the i-th plant state deep local characteristic vector composed of multiple characteristic distribution energy synergy factors, including: The square of the difference between the maximum eigenvalue of the feature distribution energy synergistic representation vector and its eigenmean, the feature variance of the feature distribution energy synergistic representation vector, and the weighted sum of the bias term are calculated, and then the feature variance of the feature distribution energy synergistic representation vector and the bias term are added and divided by the weighted sum to obtain the feature distribution energy synergy factor.

7. The hydropower plant safety protection method according to claim 1, characterized in that: Determining whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map includes: Inputting the plant state enhanced deep feature map and the plant state shallow feature map into a salient joint perception module based on an attention mechanism to obtain a plant state multi-scale joint perception salient feature map; The multi-scale joint perception salient feature map of the plant state is input into a classifier-based fire recognition module to obtain a recognition result, and the recognition result is used to indicate whether a fire is detected.

8. The hydropower plant safety protection method according to claim 7, characterized in that: The plant state enhanced deep feature map and the plant state shallow feature map are input into a salient joint perception module based on an attention mechanism to obtain a plant state multi-scale joint perception salient feature map, including: Reshaping the plant state enhanced deep feature map and the plant state shallow feature map to obtain a plant state enhanced deep feature shape reshaping matrix and a plant state shallow feature shape reshaping matrix; Input the plant state enhanced deep feature shape reshaping matrix and the plant state shallow feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain a deep-shallow dependency matrix and a shallow-deep dependency matrix; Inputting the deep-shallow dependency matrix and the shallow-deep dependency matrix into a random dropout module to obtain a pruned deep-shallow dependency matrix and a pruned shallow-deep dependency matrix; Performing matrix multiplication of the pruned deep-shallow dependency matrix and the plant state shallow feature shape reshaping matrix to obtain a dependency-optimized plant state shallow feature matrix; Performing matrix multiplication of the pruned shallow-deep dependency matrix and the plant state enhanced deep feature shape reshaping matrix to obtain a dependency optimized plant state enhanced deep feature matrix; Reshaping the feature shapes of the dependency-optimized plant state enhancement deep feature matrix and the dependency-optimized plant state shallow feature matrix to obtain an optimized plant state enhancement deep feature map and an optimized plant state shallow feature map; The weighted sum of the optimized plant state enhanced deep feature map and the optimized plant state shallow feature map is calculated to obtain the multi-scale joint perception salient feature map of the plant state.

9. The hydropower plant safety protection method according to claim 8, characterized in that: The plant state enhanced deep feature shape reshaping matrix and the plant state shallow feature shape reshaping matrix are input into the feature channel-by-channel interactive perception module to obtain a deep-shallow dependency matrix and a shallow-deep dependency matrix, including: Multiplying the plant state enhanced deep feature shape reshaping matrix by the transposed matrix of the plant state shallow feature shape reshaping matrix to obtain a deep-shallow correlation representation matrix; The deep-shallow correlation representation matrix is ​​divided by the scale of the shallow feature shape reshaping matrix of the plant state and then input into the softmax function to obtain the deep-shallow dependency matrix; Calculate the transposed matrix of the plant state enhanced deep feature shape reshaping matrix and multiply it by the plant state shallow feature shape reshaping matrix to obtain the plant state shallow-deep correlation representation matrix; The shallow-deep correlation representation matrix is ​​divided by the scale of the plant state enhanced deep feature shape reshaping matrix and then input into the softmax function to obtain the shallow-deep dependency matrix.

10. A safety protection system for a hydropower plant, characterized in that: The hydropower plant safety protection system includes: The plant monitoring module is used to collect plant monitoring images of the hydropower plant; A data transmission module, used to input the factory monitoring images into a security protection cloud processing center; A factory monitoring image feature extraction module is used to use the security protection cloud processing center to perform multi-level feature extraction on the factory monitoring image to obtain a factory state shallow feature map and a factory state deep feature map; A feature selection and enhancement module is used to perform feature selection and enhancement on the deep feature map of the plant state to obtain an enhanced deep feature map of the plant state; A fire identification module, used to determine whether a fire is detected based on the significant joint features of the plant state enhanced deep feature map and the plant state shallow feature map; The alarm prompt generation module is used to generate a fire safety protection alarm prompt in response to detecting a fire.

Images