Temperature monitoring and alarm system and method for generating units of hydropower station

Abstract

The present application relates to the field of intelligent alarm technology, and specifically discloses a temperature monitoring alarm system and method for a hydropower station generator set, which uses artificial intelligence technology based on deep learning to monitor and analyze the operating power and ambient temperature of the hydropower station generator set, and infers the real-time temperature characteristics of the generator set in a normal working mode based on the cross-modal joint characteristics between the operating power and ambient temperature of the generator set, thereby intelligently judging whether the real-time temperature of the generator set is abnormal based on the characteristic difference between the inferred real-time temperature characteristics in the normal working mode and the actual real-time temperature characteristics of the generator set. In this way, the temperature monitoring accuracy and reliability of the hydropower station generator set can be improved, the false alarm rate can be reduced, and potential failures or performance degradations can be predicted, and corresponding measures can be taken in advance to ensure the stable and efficient operation of the generator set.

Description

Technical Field

[0001] The present application relates to the field of intelligent alarm technology, and more specifically, to a temperature monitoring alarm system and method for a generator set of a hydropower station. Background Art

[0002] During the power generation process, the temperature of some areas of the hydropower station generator set will gradually rise due to the rotating parts and the internal current. If the temperature exceeds a certain safety threshold, it may cause equipment failure or even cause shutdown accidents. With the continuous development of technology, the temperature monitoring system of the hydropower station generator set is also gradually upgraded and improved.

[0003] In the current power station monitoring system, temperature sensors are usually installed at the corresponding parts of the generator set to monitor the temperature of the equipment. The alarm upper limit and the shutdown upper limit are generally set in the temperature table. When the temperature exceeds the alarm upper limit, the monitoring system generates an alarm signal; when the temperature exceeds the shutdown upper limit, the monitoring system issues a shutdown command to protect the generator set. However, although this monitoring and alarm method that relies on a preset threshold is simple and intuitive, it is easily affected by environmental factors, prone to false alarms, and can only provide static temperature data. It lacks an in-depth understanding of the actual operating status of the generator set and cannot adapt to different working conditions of the equipment. At the same time, this alarm method can usually only provide an alarm after a problem occurs, and cannot predict potential failures or performance degradation. Therefore, an optimized temperature monitoring alarm system and method for hydropower station generator sets is expected. Summary of the invention

[0004] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a temperature monitoring alarm system and method for a hydropower station generator set, which uses artificial intelligence technology based on deep learning to monitor and analyze the operating power and ambient temperature of the hydropower station generator set, and infers the real-time temperature characteristics of the generator set in the normal working mode based on the cross-modal joint characteristics between the operating power and the ambient temperature of the generator set, so as to intelligently judge whether there is an abnormality in the real-time temperature of the generator set based on the characteristic difference between the inferred real-time temperature characteristics in the normal working mode and the actual real-time temperature characteristics of the generator set. In this way, the temperature monitoring accuracy and reliability of the hydropower station generator set can be improved, the false alarm rate can be reduced, and potential failures or performance degradations can be predicted, and corresponding measures can be taken in advance to ensure the stable and efficient operation of the generator set.

[0005] Accordingly, according to one aspect of the present application, a temperature monitoring and alarm method for a hydropower station generator set is provided, comprising:

[0006] Acquire the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station;

[0007] Performing time series association coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit association feature vector, an operating power time series implicit association feature vector and a real-time temperature time series implicit association feature vector;

[0008] Performing cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector to obtain a content-saliency ambient temperature-operation power time series implicit correlation feature matrix;

[0009] Based on the content, the environment temperature-operating power timing implicit correlation feature matrix is ​​significantly analyzed to perform real-time temperature semantic reasoning to obtain a real-time temperature semantic reasoning decoding representation vector;

[0010] Based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector, it is determined whether the real-time temperature of the hydropower station generator set is abnormal.

[0011] According to another aspect of the present application, a temperature monitoring and alarm system for a hydropower station generator set is provided, comprising:

[0012] The generator set operation monitoring module is used to obtain the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station;

[0013] A time series correlation coding module, used for performing time series correlation coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit correlation feature vector, an operating power time series implicit correlation feature vector and a real-time temperature time series implicit correlation feature vector;

[0014] A cross-modal joint encoding module, used for performing cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector to obtain a content-salient ambient temperature-operating power time series implicit correlation feature matrix;

[0015] A real-time temperature semantic reasoning module is used to perform real-time temperature semantic reasoning based on the content-signalized ambient temperature-operating power time series implicit correlation feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector;

[0016] The real-time temperature anomaly diagnosis module is used to determine whether the real-time temperature of the hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector.

[0017] Compared with the prior art, the temperature monitoring alarm system and method of the hydropower station generator set provided by the present application uses artificial intelligence technology based on deep learning to monitor and analyze the operating power and ambient temperature of the hydropower station generator set, and infers the real-time temperature characteristics of the generator set in the normal working mode based on the cross-modal joint characteristics between the operating power and ambient temperature of the generator set, thereby intelligently judging whether the real-time temperature of the generator set is abnormal based on the characteristic difference between the inferred real-time temperature characteristics in the normal working mode and the actual real-time temperature characteristics of the generator set. In this way, the temperature monitoring accuracy and reliability of the hydropower station generator set can be improved, the false alarm rate can be reduced, and potential failures or performance degradations can be predicted, and corresponding measures can be taken in advance to ensure the stable and efficient operation of the generator set. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 The present invention is a flow chart of a method for temperature monitoring and alarming of a generator set of a hydropower station according to an embodiment of the present application.

[0020] Figure 2 Schematic diagram of the architecture of a temperature monitoring and alarm method for a generator set of a hydropower station according to an embodiment of the present application.

[0021] Figure 3 The present invention is a flow chart of cross-modal joint encoding and feature manifestation of the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector in the temperature monitoring and alarm method of the hydropower station generator set according to an embodiment of the present application to obtain a content-salient ambient temperature-operating power time series implicit correlation feature matrix.

[0022] Figure 4 This is a flowchart for determining whether the real-time temperature of a hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector in the temperature monitoring and alarm method of the hydropower station generator set according to an embodiment of the present application.

[0023] Figure 5 The present invention is a flowchart of the training steps in the temperature monitoring and alarm method of a hydropower station generator set according to an embodiment of the present application.

[0024] Figure 6 It is a block diagram of a temperature monitoring and alarm system for a generator set of a hydropower station according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0026] As mentioned in the above background technology, in the current power station monitoring system, the temperature monitoring alarm is usually based on the preset temperature threshold. Although this method is simple and intuitive, it is easily affected by environmental factors, prone to false alarms, and can only provide static temperature data. It lacks an in-depth understanding of the actual operating status of the generator set and cannot adapt to different working conditions of the equipment. At the same time, this alarm method can usually only provide an alarm after the problem occurs, and cannot predict potential failures or performance degradation. In view of the above technical problems, the technical concept of this application is to use artificial intelligence technology based on deep learning to monitor and analyze the operating power and ambient temperature of the generator set of the hydropower station, and infer the real-time temperature characteristics of the generator set in the normal working mode based on the cross-modal joint characteristics between the operating power and ambient temperature of the generator set, so as to intelligently judge whether the real-time temperature of the generator set is abnormal based on the characteristic difference between the inferred real-time temperature characteristics under the normal working mode and the actual real-time temperature characteristics of the generator set. In this way, the temperature monitoring accuracy and reliability of the generator set of the hydropower station can be improved, the false alarm rate can be reduced, and potential failures or performance degradation can be predicted, and corresponding measures can be taken in advance to ensure the stable and efficient operation of the generator set.

[0027] Figure 1 The present invention is a flow chart of a method for temperature monitoring and alarming of a generator set of a hydropower station according to an embodiment of the present application. Figure 2 FIG. 1 is a schematic diagram of the architecture of a temperature monitoring and alarm method for a hydropower station generator set according to an embodiment of the present application. Figure 1 and Figure 2As shown, the temperature monitoring and alarm method for the generator set of a hydropower station according to the embodiment of the present application includes the following steps: S110, acquiring the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station; S120, respectively performing time series correlation coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature to obtain an ambient temperature time series implicit correlation feature vector, a motion power time series implicit correlation feature vector and a real-time temperature time series implicit correlation feature vector; S130 , cross-modal joint encoding and feature manifestation are performed on the ambient temperature time series implicit association feature vector and the motion power time series implicit association feature vector to obtain a content-salient ambient temperature-operating power time series implicit association feature matrix; S140, real-time temperature semantic reasoning is performed based on the content-salient ambient temperature-operating power time series implicit association feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector; S150, based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector, it is determined whether the real-time temperature of the hydropower station generator set is abnormal.

[0028] In the temperature monitoring and alarm method for the generator set of the hydropower station mentioned above, the step S110 obtains the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station. It should be understood that under normal circumstances, there is a certain correlation and change law between the temperature of the generator set and the ambient temperature and the operating power of the generator set. Specifically, the heat generated by the generator set will be different if the operating power is different; at the same time, the change of ambient temperature will directly affect the heat dissipation conditions of the generator set, thereby affecting its temperature. Therefore, in the technical solution of the present application, it is expected to monitor and analyze the ambient temperature and the operating power of the generator set to predict the temperature change law of the generator set under the normal working mode, and then judge whether the generator set is in a normal operating state based on the comparative analysis results between it and the actual temperature change pattern, so as to realize abnormal temperature monitoring and early warning of the generator set.

[0029] In the temperature monitoring and alarm method of the above-mentioned hydropower station generator set, the step S120 performs time series correlation coding on the time series of the ambient temperature, the time series of the operating power, and the time series of the real-time temperature to obtain the time series implicit correlation feature vector of the ambient temperature, the time series implicit correlation feature vector of the motion power, and the time series implicit correlation feature vector of the real-time temperature. In a specific example of the present application, the encoding method of performing time series correlation coding on the time series of the ambient temperature, the time series of the operating power, and the time series of the real-time temperature is to pass the time series of the ambient temperature, the time series of the operating power, and the time series of the real-time temperature through a sequence encoder based on a Bi-LSTM model to obtain the time series implicit correlation feature vector of the ambient temperature, the time series implicit correlation feature vector of the motion power, and the time series implicit correlation feature vector of the real-time temperature. It should be understood that the ambient temperature, the real-time temperature of the generator set, and the operating power data have time series dependence and correlation in the time dimension. Therefore, in order to capture the temporal association patterns in these time series data, in the technical solution of the present application, a sequence encoder based on the Bi-LSTM (bidirectional long short-term memory network) model is introduced to perform temporal encoding on the time series of the ambient temperature, the time series of the operating power, and the time series of the real-time temperature. Specifically, the Bi-LSTM model has a powerful nonlinear feature extraction capability. Compared with the traditional recurrent neural network, the Bi-LSTM model is bidirectional and can simultaneously consider the forward and backward information in the time series, thereby capturing the long-term dependencies in the data. Here, the sequence encoder based on the Bi-LSTM model processes the forward and backward information of the time series data through two LSTM networks respectively, and controls the transmission of information flow based on the gating mechanism to mine the intrinsic association dependencies in the input time series data, thereby capturing the temporal implicit association patterns of the ambient temperature, operating power, and real-time temperature data, so as to reveal the temporal variation characteristics of the ambient temperature, the operating power of the generator set, and the real-time temperature, and provide important information support for the subsequent abnormal temperature warning of the generator set.

[0030] In the temperature monitoring and alarm method for the generator set of the hydropower station, the step S130 performs cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector to obtain a content-salient ambient temperature-operation power time series implicit correlation feature matrix. Figure 3 The present invention is a flow chart of performing cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector in the temperature monitoring and alarm method of the hydropower station generator set according to the embodiment of the present application to obtain a content-salient ambient temperature-operation power time series implicit correlation feature matrix. Figure 3As shown, the step S130 includes: S131, passing the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector through a cross-modal joint encoder based on the CLIP model to obtain an ambient temperature-operating power time series implicit correlation feature matrix; S132, passing the ambient temperature-operating power time series implicit correlation feature matrix through a foreground-like attention network based on gated logic to obtain the content-saliency ambient temperature-operating power time series implicit correlation feature matrix.

[0031] Specifically, in step S131, the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector are passed through a cross-modal joint encoder based on the CLIP model to obtain an ambient temperature-operating power time series implicit correlation feature matrix. It should be understood that since the ambient temperature and the operating power of the generator set belong to different modes, direct correlation fusion may be difficult and cannot fully reveal the intrinsic correlation between the two. Therefore, in the technical solution of the present application, a cross-modal joint encoder based on the CLIP (Contrastive Language-Image Pre-training) model is further introduced to capture the cross-modal correlation characteristics between ambient temperature and operating power. Specifically, the CLIP model is a cross-modal pre-training model based on contrastive learning, which can learn the potential correlation and mutual influence between different modal data. Here, the cross-modal joint encoder based on the CLIP model can map the implicit correlation feature vector of the ambient temperature time series and the implicit correlation feature vector of the motion power time series to a shared latent feature space for comparison and matching, and capture the similarities and differences between the two based on a contrastive learning mechanism, revealing the cross-modal correlation relationship between ambient temperature and operating power, thereby more comprehensively understanding the operating status of the generator set and providing more comprehensive and accurate feature information for subsequent real-time temperature semantic reasoning.

[0032] Specifically, in step S132, the ambient temperature-operating power timing implicit correlation feature matrix is ​​passed through a foreground-like attention network based on gating logic to obtain the content-salient ambient temperature-operating power timing implicit correlation feature matrix. It should be understood that, considering that the generator set may be affected by various factors during actual operation, such as load changes, environmental factors, etc., there may be some redundant information and noise in the ambient temperature-operating power timing implicit correlation feature matrix. Therefore, in order to more accurately infer the real-time temperature characteristics of the generator set under the normal working mode, it is necessary to further perform feature selection and filtering on the ambient temperature-operating power timing implicit correlation feature matrix. Based on this, in the technical solution of the present application, a foreground-like attention network based on gating logic is used to process the ambient temperature-operating power timing implicit correlation feature matrix. Specifically, the foreground-like attention network based on gating logic uses an attention mechanism to learn the feature importance of each local position in the ambient temperature-operating power timing implicit correlation feature matrix, controls the reinforcement and suppression of features through gating logic, and dynamically adjusts the attention weights of different features, so that the model focuses on important information related to the operating status of the generator set, while suppressing noise interference and redundant information therein, thereby achieving content saliency of the ambient temperature-operating power timing implicit correlation feature matrix, so as to further improve the accuracy and robustness of subsequent real-time temperature semantic reasoning.

[0033] In a specific example of the present application, the step S132 includes: calculating the attention weight probability matrix of the ambient temperature-operating power timing implicit correlation feature matrix; multiplying the attention weight probability matrix and the ambient temperature-operating power timing implicit correlation feature matrix by position point to obtain the content-salient ambient temperature-operating power timing implicit correlation feature matrix.

[0034] More specifically, calculating the attention weight probability matrix of the ambient temperature-operating power timing implicit correlation feature matrix includes: processing the ambient temperature-operating power timing implicit correlation feature matrix using the following foreground attention formula to obtain the attention weight probability matrix; wherein the foreground attention formula is: ;in, The ambient temperature-operating power timing implicit correlation characteristic matrix is The eigenvalues ​​of the positions, Indicates The exponential function operation with base , Indicates masking processing. represents a predetermined threshold, is the attention weight probability matrix The eigenvalue of the location.

[0035] In the temperature monitoring and alarm method for the generator set of the hydropower station mentioned above, the step S140 performs real-time temperature semantic reasoning based on the content-significant ambient temperature-operating power timing implicit correlation feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector. In a specific example of the present application, the implementation method of performing real-time temperature semantic reasoning based on the content-significant ambient temperature-operating power timing implicit correlation feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector is to pass the content-significant ambient temperature-operating power timing implicit correlation feature matrix through a real-time temperature semantic reasoner based on a decoder to obtain the real-time temperature semantic reasoning decoding representation vector. That is, by utilizing the powerful decoding and reasoning ability of the decoder model, the real-time temperature information of the generator set in the normal working mode is inferred according to the feature pattern of the content-significant ambient temperature-operating power timing implicit correlation feature matrix. Specifically, the decoder-based real-time temperature semantic reasoner uses internal complex nonlinear transformations to deeply learn the cross-modal correlation patterns contained in the implicit correlation feature matrix of the ambient temperature-operating power time series, fully understand the thermal characteristics and heat dissipation conditions of the generator set itself, and combine the working characteristics and physical laws of the generator set learned during the training process, so as to achieve accurate prediction and semantic reasoning of the real-time temperature of the generator set, and provide an intuitive and effective reference basis for subsequent temperature anomaly warnings.

[0036] In the temperature monitoring and alarm method for the hydropower station generator set, the step S150 determines whether the real-time temperature of the hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector. Figure 4 This is a flow chart of determining whether the real-time temperature of the hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector in the temperature monitoring and alarm method of the hydropower station generator set according to the embodiment of the present application. Figure 4 As shown, the step S150 includes: S151, calculating the hash similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector; S152, based on the comparison between the hash similarity and a predetermined threshold, determining whether the real-time temperature of the hydropower station generator set is abnormal.

[0037] Specifically, the step S151 calculates the hash similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector. It should be understood that by calculating the similarity between the real-time temperature semantic reasoning decoding representation vector (the predicted temperature feature obtained based on cross-modal reasoning of ambient temperature and unit operating power) and the real-time temperature time series implicit association feature vector (the real-time temperature feature actually monitored), the degree of closeness between the unit temperature and the actual temperature under the ideal state can be evaluated, thereby judging whether the generator set is operating in a normal working state, that is, whether the real-time temperature of the generator set is abnormal. Here, the hash similarity calculation method is used to measure the feature consistency between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector. Specifically, hash similarity is an efficient similarity measurement method. The input real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector are mapped to a low-dimensional hash space through a hash function to obtain a fixed-length binary hash code, and the similarity between the hash codes corresponding to the feature vectors is calculated in the hash space to quantify the similarity between the original feature vectors, thereby providing key indicators for the real-time temperature anomaly assessment of the generator set.

[0038] Specifically, the step S152 determines whether the real-time temperature of the generator set of the hydropower station is abnormal based on the comparison between the hash similarity and the predetermined threshold. That is, when the hash similarity is higher than the predetermined threshold, it means that the real-time temperature of the generator set is close to the predicted temperature obtained based on cross-modal reasoning, and it is considered that the generator set is in normal working condition and the temperature is normal; when the hash similarity is lower than the predetermined threshold, it is considered that there is a large difference between the real-time temperature of the generator set and the predicted temperature, and there may be an abnormal situation, and further abnormal warning is required to remind the operation and maintenance personnel to conduct inspections and maintenance in time to avoid possible failures or accidents. In this way, the automatic monitoring and abnormal warning of the real-time temperature of the generator set of the hydropower station are realized, which can effectively improve the operating safety and reliability of the generator set and reduce the failure rate and maintenance cost.

[0039] It should be understood that before using the above neural network model, it needs to be trained. That is to say, in the temperature monitoring and alarm method of the hydropower station generator set of the present application, it also includes a training step of training the sequence encoder based on the Bi-LSTM model, the cross-modal joint encoder based on the CLIP model, the foreground-like attention network based on the gated logic, and the real-time temperature semantic reasoner based on the decoder.

[0040] Figure 5 Flow chart of the training steps in the temperature monitoring and alarm method of the hydropower station generator set according to the embodiment of the present application. Figure 5 As shown, the training step S200 includes: S210, obtaining training data, the training data including the time series of training environment temperature, the time series of training real-time temperature of the monitored hydropower station generator set and the time series of training operating power, as well as the true value of hash similarity; S220, passing the time series of the training environment temperature, the time series of the training operating power and the time series of the training real-time temperature through the sequence encoder based on the Bi-LSTM model to obtain the training environment temperature time series implicit association feature vector, the training motion power time series implicit association feature vector and the training real-time temperature time series implicit association feature vector; S230, passing the training environment temperature time series implicit association feature vector and the training motion power time series implicit association feature vector through the cross-modal joint encoder based on the CLIP model to obtain the training environment temperature-operating power time series implicit association feature matrix; S240, passing the training environment temperature-operating power time series implicit association feature matrix through the class foreground annotation based on the gating logic The intention network is used to obtain the training content-significant ambient temperature-operating power timing implicit correlation feature matrix; S250, the training content-significant ambient temperature-operating power timing implicit correlation feature matrix is ​​passed through the decoder-based real-time temperature semantic reasoner to obtain the training real-time temperature semantic reasoning decoding representation vector; S260, the training hash similarity between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature timing implicit correlation feature vector is calculated; S270, the difference between the training hash similarity and the true value of the hash similarity is calculated as the hash similarity loss function value; S280, the distribution loss function value based on the training hash similarity is calculated; S290, the Bi-LSTM model-based sequence encoder, the CLIP model-based cross-modal joint encoder, the gated logic-based foreground attention network and the decoder-based real-time temperature semantic reasoner are trained with the weighted sum of the hash similarity loss function value and the distribution loss function value as the loss function value.

[0041] In an optimized embodiment of the present application, considering that the training real-time temperature semantic reasoning decoding representation vector expresses the semantic decoding representation of the full-time domain cross-modal attention features of the training environment temperature and the training motion power, and the training real-time temperature temporal implicit association feature vector expresses the temporal association feature representation of the training real-time temperature, in the process of mapping the feature similarity domains based on hash similarity under different feature expression dimensions, the difference in feature expression will cause noise outside the feature similarity domain distribution, thereby reducing the calculation accuracy of the hash similarity between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature temporal implicit association feature vector.

[0042] Therefore, in the process of model training, at each iteration, for example, when gradient backpropagation is performed based on the difference between the true value of the hash similarity between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector and the training hash similarity, a distribution loss function based on the training hash similarity is further introduced. Among them, the calculation process of the distribution loss function based on the training hash similarity is specifically as follows: calculate each iteration by subtracting the training hash similarity to obtain the training hash similarity difference; determine the feature mean and feature variance of all feature values ​​of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector; calculate the class cross entropy difference in the form of class cross entropy by multiplying each feature value of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector and the training hash similarity, and then subtracting the product of the training hash similarity difference and the feature mean; and calculate the class cross entropy difference between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector. The squares of the class cross entropy differences corresponding to each eigenvalue in the feature union of the eigenvectors are summed and divided by the size of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector to obtain a first loss term, wherein the size of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector is the sum of the lengths of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector; the sum of the squares of the training hash similarity and the training hash similarity difference is multiplied by the feature variance to obtain a second loss term; the second loss term is subtracted from the first loss term to obtain the distribution loss function value based on the training hash similarity.

[0043] Here, the distribution loss function based on the training hash similarity sets the approximate expected variance characteristics of the out-of-distribution noise of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector relative to the Bernoulli sampling instance of the feature set by using the similarity probability distribution characteristics expressed by the training hash similarity and the training hash similarity difference of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector within the feature similarity domain.

[0044] Therefore, by using the distribution loss function based on the training hash similarity to perform distribution feature noise measurement based on the similar probability distribution characteristics expressed by the training hash similarity and the training hash similarity difference, the statistical noise distribution can be used to perform probability distribution mapping in the feature similarity domain based on each eigenvalue of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector. The possible sampling noise features can be degraded by using the probabilistic cross random selection characteristics of the expected variance of the feature union of the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector, thereby realizing distribution noise reduction of self-supervised learning in the process of updating model parameters to improve the calculation accuracy of the training hash similarity between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicitly associated feature vector.

[0045] In summary, the temperature monitoring and alarm method of the hydropower station generator set according to the embodiment of the present application is explained, which uses artificial intelligence technology based on deep learning to monitor and analyze the operating power and ambient temperature of the hydropower station generator set, and infers the real-time temperature characteristics of the generator set in the normal working mode based on the cross-modal joint characteristics between the operating power and ambient temperature of the generator set, so as to intelligently judge whether the real-time temperature of the generator set is abnormal based on the characteristic difference between the inferred real-time temperature characteristics in the normal working mode and the actual real-time temperature characteristics of the generator set. In this way, the temperature monitoring accuracy and reliability of the hydropower station generator set can be improved, the false alarm rate can be reduced, and potential failures or performance degradations can be predicted, and corresponding measures can be taken in advance to ensure the stable and efficient operation of the generator set.

[0046] Figure 6 FIG. 1 is a block diagram of a temperature monitoring and alarm system for a hydropower station generator set according to an embodiment of the present application. Figure 6As shown, the temperature monitoring and alarm system 100 of the hydropower station generator set according to the embodiment of the present application includes: a generator set operation monitoring module 110, which is used to obtain the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the monitored hydropower station generator set, as well as the time series of the operating power of the monitored hydropower station generator set; a time series association coding module 120, which is used to perform time series association coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit association feature vector, a motion power time series implicit association feature vector and a real-time temperature time series implicit association feature vector; a cross-modal joint coding module 120, which is used to perform time series association coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit association feature vector, a motion power time series implicit association feature vector and a real-time temperature time series implicit association feature vector; 30, used for cross-modal joint encoding and feature manifestation of the ambient temperature time series implicit association feature vector and the motion power time series implicit association feature vector to obtain a content-significant ambient temperature-operating power time series implicit association feature matrix; a real-time temperature semantic reasoning module 140, used for real-time temperature semantic reasoning based on the content-significant ambient temperature-operating power time series implicit association feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector; a real-time temperature anomaly diagnosis module 150, used for determining whether the real-time temperature of the hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector.

[0047] Here, those skilled in the art will appreciate that the specific operations of each module in the temperature monitoring alarm system for the hydropower station generator set have been described in detail above. Figures 1 to 5 The temperature monitoring alarm method for a hydropower station generator set has been described in detail, and therefore, its repeated description will be omitted.

[0048] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention 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 invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

[0049] 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 by the present invention, 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.

[0050] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.

[0051] 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 stated in the system claims can also be implemented by one unit through software or hardware.

[0052] 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 invention and are not intended to limit it. Although the present invention has been 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 invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A temperature monitoring and alarm method for a hydropower station generator set, It is characterized in that include: Acquire the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station; Performing time series association coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit association feature vector, an operating power time series implicit association feature vector and a real-time temperature time series implicit association feature vector; Performing cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector to obtain a content-saliency ambient temperature-operation power time series implicit correlation feature matrix; Based on the content, the environment temperature-operating power timing implicit correlation feature matrix is ​​significantly analyzed to perform real-time temperature semantic reasoning to obtain a real-time temperature semantic reasoning decoding representation vector; Determine whether the real-time temperature of the generator set of the hydropower station is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector; The ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector are cross-modal jointly encoded and feature-manifested to obtain a content-salience ambient temperature-operation power time series implicit correlation feature matrix, including: The ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector are passed through a cross-modal joint encoder based on a CLIP model to obtain an ambient temperature-operation power time series implicit correlation feature matrix; The environment temperature-operating power timing implicit correlation feature matrix is ​​passed through a foreground-like attention network based on gated logic to obtain the content-saliency environment temperature-operating power timing implicit correlation feature matrix; The process of passing the ambient temperature-operating power timing implicit correlation feature matrix through a foreground-like attention network based on gated logic to obtain the content-saliency ambient temperature-operating power timing implicit correlation feature matrix includes: Calculate the attention weight probability matrix of the ambient temperature-operating power time series implicit correlation feature matrix; Multiply the attention weight probability matrix and the ambient temperature-operating power time series implicit correlation feature matrix by position points to obtain the content saliency ambient temperature-operating power time series implicit correlation feature matrix; The calculation of the attention weight probability matrix of the ambient temperature-operating power time series implicit correlation feature matrix includes: The ambient temperature-operating power time series implicit correlation feature matrix is ​​processed using the following foreground attention formula to obtain the attention weight probability matrix; wherein the foreground attention formula is: ;in, The ambient temperature-operating power timing implicit correlation characteristic matrix is The eigenvalues ​​of the positions, Indicates The exponential function operation with base , Indicates masking processing. represents a predetermined threshold, is the attention weight probability matrix The eigenvalue of the location.

2. The temperature monitoring and alarm method for a hydropower station generator set according to claim 1, It is characterized in that The time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature are respectively subjected to time series association coding to obtain an ambient temperature time series implicit association feature vector, an operating power time series implicit association feature vector and a real-time temperature time series implicit association feature vector, including: The time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature are respectively passed through a sequence encoder based on a Bi-LSTM model to obtain an implicit correlation feature vector of the ambient temperature time series, an implicit correlation feature vector of the operating power time series and an implicit correlation feature vector of the real-time temperature time series.

3. The temperature monitoring and alarm method for a hydropower station generator set according to claim 2, It is characterized in that Based on the content, the environment temperature-operating power time series implicit correlation feature matrix is ​​significantly analyzed to perform real-time temperature semantic reasoning to obtain a real-time temperature semantic reasoning decoding representation vector, including: The content-salient ambient temperature-operating power timing implicit correlation feature matrix is ​​passed through a decoder-based real-time temperature semantic reasoner to obtain the real-time temperature semantic reasoning decoding representation vector.

4. The temperature monitoring and alarm method for a hydropower station generator set according to claim 3, It is characterized in that Determining whether the real-time temperature of the generator set of the hydropower station is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector, including: Calculating the hash similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector; Based on the comparison between the hash similarity and a predetermined threshold, it is determined whether the real-time temperature of the generator set of the hydropower station is abnormal.

5. The temperature monitoring and alarm method for a hydropower station generator set according to claim 4, It is characterized in that It also includes a training step: training the sequence encoder based on the Bi-LSTM model, the cross-modal joint encoder based on the CLIP model, the foreground-like attention network based on gated logic, and the real-time temperature semantic reasoner based on the decoder.

6. The temperature monitoring and alarm method for a hydropower station generator set according to claim 5, It is characterized in that The training step comprises: Acquire training data, the training data including a time series of training environment temperature, a time series of training real-time temperature of a monitored hydropower station generator set, a time series of training operating power, and a true value of hash similarity; The time series of the training environment temperature, the time series of the training running power and the time series of the training real-time temperature are respectively passed through the sequence encoder based on the Bi-LSTM model to obtain the training environment temperature time series implicit correlation feature vector, the training exercise power time series implicit correlation feature vector and the training real-time temperature time series implicit correlation feature vector; Passing the training environment temperature time series implicit correlation feature vector and the training motion power time series implicit correlation feature vector through the cross-modal joint encoder based on the CLIP model to obtain a training environment temperature-operation power time series implicit correlation feature matrix; The training environment temperature-operating power timing implicit correlation feature matrix is ​​passed through the gated logic-based foreground-like attention network to obtain the training content-saliency environment temperature-operating power timing implicit correlation feature matrix; The training content significant ambient temperature-operating power timing implicit correlation feature matrix is ​​passed through the decoder-based real-time temperature semantic reasoner to obtain a training real-time temperature semantic reasoning decoding representation vector; Calculating the training hash similarity between the training real-time temperature semantic reasoning decoding representation vector and the training real-time temperature time series implicit association feature vector; Calculating a difference between the training hash similarity and the true value of the hash similarity as a hash similarity loss function value; Calculating a distribution loss function value based on the training hash similarity; The Bi-LSTM model-based sequence encoder, the CLIP model-based cross-modal joint encoder, the gated logic-based foreground attention network, and the decoder-based real-time temperature semantic reasoner are trained using the weighted sum of the hash similarity loss function value and the distribution loss function value as the loss function value.

7. A temperature monitoring and alarm system for a hydropower station generator set. It is characterized in that include: The generator set operation monitoring module is used to obtain the time series of the ambient temperature collected by the temperature sensor and the time series of the real-time temperature of the generator set of the monitored hydropower station, as well as the time series of the operating power of the generator set of the monitored hydropower station; A time series correlation coding module, used for performing time series correlation coding on the time series of the ambient temperature, the time series of the operating power and the time series of the real-time temperature respectively to obtain an ambient temperature time series implicit correlation feature vector, an operating power time series implicit correlation feature vector and a real-time temperature time series implicit correlation feature vector; A cross-modal joint encoding module, used for performing cross-modal joint encoding and feature explicitation on the ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector to obtain a content-salient ambient temperature-operating power time series implicit correlation feature matrix; A real-time temperature semantic reasoning module is used to perform real-time temperature semantic reasoning based on the content-signalized ambient temperature-operating power time series implicit correlation feature matrix to obtain a real-time temperature semantic reasoning decoding representation vector; A real-time temperature anomaly diagnosis module, used to determine whether the real-time temperature of the hydropower station generator set is abnormal based on the feature similarity between the real-time temperature semantic reasoning decoding representation vector and the real-time temperature time series implicit association feature vector; The ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector are cross-modal jointly encoded and feature-manifested to obtain a content-salience ambient temperature-operation power time series implicit correlation feature matrix, including: The ambient temperature time series implicit correlation feature vector and the motion power time series implicit correlation feature vector are passed through a cross-modal joint encoder based on a CLIP model to obtain an ambient temperature-operation power time series implicit correlation feature matrix; The environment temperature-operating power timing implicit correlation feature matrix is ​​passed through a foreground-like attention network based on gated logic to obtain the content-saliency environment temperature-operating power timing implicit correlation feature matrix; The process of passing the ambient temperature-operating power timing implicit correlation feature matrix through a foreground-like attention network based on gated logic to obtain the content-saliency ambient temperature-operating power timing implicit correlation feature matrix includes: Calculate the attention weight probability matrix of the ambient temperature-operating power time series implicit correlation feature matrix; Multiply the attention weight probability matrix and the ambient temperature-operating power time series implicit correlation feature matrix by position points to obtain the content saliency ambient temperature-operating power time series implicit correlation feature matrix; The calculation of the attention weight probability matrix of the ambient temperature-operating power time series implicit correlation feature matrix includes: The ambient temperature-operating power time series implicit correlation feature matrix is ​​processed using the following foreground attention formula to obtain the attention weight probability matrix; wherein the foreground attention formula is: ;in, The ambient temperature-operating power timing implicit correlation characteristic matrix is The eigenvalues ​​of the positions, Indicates The exponential function operation with base , Indicates masking processing. represents a predetermined threshold, is the attention weight probability matrix The eigenvalue of the location.

Images