Cooling tower cross-domain fault diagnosis method based on multi-source confrontation field self-adaption

Abstract

The invention provides a cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain self-adaption, and relates to the field of transfer learning, and the method comprises the steps: collecting historical sensor time sequence data, and dividing a source domain and a target domain; calculating a heat transfer efficiency ratio, and constructing a coupling characteristic matrix; cross-domain time sequence alignment processing is carried out, feature similarity measurement is carried out, and aligned feature representation is output; dynamically adjusting the weight of the source domain according to the divergence and the water hardness information to obtain a target fusion feature; and inputting the target fusion features into a fault classifier obtained by training, and outputting fault diagnosis information and trend prediction information of the cooling tower. According to the method, by considering the thermodynamic parameters, the water hardness and other physical states of the cooling tower, the difference between different domains is made precise, the situation that a high-hardness source domain interferes with low-hardness target domain diagnosis is avoided, and a coupling feature matrix and domain invariant features are considered; the method enhances the recognition and diagnosis of the fault mode of the cooling tower, and improves the accuracy of the diagnosis of the composite fault of the cooling tower.

Description

Technical Field

[0001] The present application relates to the field of transfer learning technology, and in particular to a cooling tower cross-domain fault diagnosis method based on multi-source adversarial domain adaptation. Background Art

[0002] With the rapid development of technologies such as the internet, cloud computing, and big data, data centers, as core infrastructure for information storage, processing, and transmission, are expanding in number and scale, playing a vital role in the global economy and society. Cooling towers are a core component of data center cooling systems, responsible for dissipating the significant heat generated by servers and other equipment, maintaining a suitable temperature environment and ensuring stable operation. A cooling tower failure can cause the data center temperature to rise, impacting the performance of servers and other equipment, and even causing equipment overheating and damage, resulting in serious consequences such as data loss and service interruptions, bringing significant economic losses to businesses and society.

[0003] Most existing research on cooling tower fault diagnosis focuses on a single data center. This requires high data annotation requirements. Whenever a new data center's cooling tower is fault-diagnosed, data annotation and model retraining are necessary. This wastes significant time and resources, including manpower for data annotation, computing resources for model training, and time costs.

[0004] Specifically, the relevant technologies have the following defects: (1) Existing fault diagnosis models generally rely on the full amount of labeled data in the target domain. However, the acquisition of labeled data in industrial scenarios faces the problem of high cost, and in the cold start phase, there is not enough labeled data available for training fault diagnosis models; (2) Existing technologies are based on the assumption of single-source domain homogeneity, ignoring the distribution offset of sensor data caused by environmental conditions, load fluctuations, and equipment heterogeneity in different data centers; (3) Existing research is limited by the scale of single-domain data collection, and its training set only covers a limited number of fault types and steady-state conditions, and it is difficult to cover complex and diverse fault modes and operating conditions, which limits the model's ability to identify and diagnose diverse faults; (4) In existing multi-source domain fault diagnosis methods, most directly use data-driven methods to quantify domain differences, ignoring the domain differences caused by physical factors, resulting in inaccurate quantification of domain differences; (5) In existing multi-source domain fault diagnosis methods, the complex fault modes of cooling towers (such as vibration anomalies caused by scaling) may be identified as independent features for processing, resulting in inaccurate fault diagnosis. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, this application provides a cooling tower cross-domain fault diagnosis method based on multi-source adversarial domain adaptation, which solves the problems of current cooling tower fault diagnosis relying on labeled data, being limited by single-domain data, and insufficient fault identification and diagnosis capabilities.

[0006] To achieve the above objectives, this application is implemented through the following technical solutions: In the first aspect, an embodiment of the present application provides a cooling tower cross-domain fault diagnosis method based on multi-source adversarial domain adaptation, and the cooling tower cross-domain fault diagnosis method based on multi-source adversarial domain adaptation includes: collecting historical sensor time series data containing spatial position marks of cooling towers in multiple data centers, dividing the source domain and the target domain and performing timestamp alignment, sampling frequency unification and independent normalization processing on the data, learning to extract domain-invariant features and label information of the source domain and the target domain; extracting data representing thermodynamic and vibration coupling characteristics from the historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling feature matrix; based on the coupling feature matrix and domain-invariant features, performing cross-domain time series alignment processing through comparative learning, combining the difference in spray volume per unit area of ​​the cooling tower and the filler type to construct positive sample pairs, and combining the spray density to perform sample separation to construct negative sample pairs from different fault types, performing feature similarity measurement, and outputting the aligned feature representation; collecting circulating water conductivity through a circulating water conductivity sensor to determine water hardness information; Based on the aligned feature representation, feature fusion is performed by combining local temporal patterns with global dependencies. The source domain weights are dynamically adjusted according to the KL divergence and water hardness information to obtain the target fusion features. The target fusion features are input into the trained fault classifier to output fault diagnosis information and trend prediction information of the cooling tower.

[0007] According to the first aspect of the embodiment of the present application, after the target fusion feature is input into the fault classifier obtained by training and the fault diagnosis information and trend prediction information of the cooling tower are output, the cross-domain fault diagnosis method of the cooling tower based on multi-source adversarial domain adaptation can also specifically include the following steps: constructing a data distribution drift detection mechanism and a water quality parameter monitoring mechanism to monitor state changes, and triggering a preset model retraining strategy when the state change reaches a preset target threshold; the data distribution drift detection mechanism includes: using a sliding window mechanism to continuously cache the real-time sensor data stream of the target cooling tower; based on the cached data, regularly performing model parameter fine-tuning to adapt the model to the slow drift of equipment operation; deploying an autoencoder to learn the feature representation of the normal operation data of the target domain and calculate the reconstruction error of the real-time data; when the reconstruction error continues to exceed the preset first threshold, it is determined that the current data distribution deviates significantly from the distribution during model training, and it is determined that the state change reaches the preset target threshold. The water quality parameter monitoring mechanism includes: collecting the conductivity of the supplementary water through the supplementary water conductivity sensor, monitoring the water quality parameters based on the conductivity of the supplementary water and the conductivity of the circulating water, and determining the concentration multiple; When the absolute deviation between the current concentration factor and the concentration factor during model training exceeds a preset second threshold, it is determined that the state change has reached the preset target threshold. During the model retraining process, adaptive sampling is implemented to prioritize the collection of data from high concentration factor periods corresponding to scaling-sensitive periods to optimize data efficiency and the model's sensitivity to critical faults. The data sampling frequency or weight is also increased to allow the model to focus on learning more discriminative features under scaling-prone conditions.

[0008] According to the first aspect of the embodiment of the present application, the aforementioned method is based on the coupling feature matrix and the domain invariant features, performs cross-domain time series alignment processing through contrastive learning, combines the difference in the spray volume per unit area of ​​the cooling tower and the filler type to construct positive sample pairs, and combines the spray density to perform sample separation to construct negative sample pairs from different fault types, performs feature similarity measurement, and outputs the aligned feature representation. Specifically, the following steps may be included: input the coupling feature matrix and the domain invariant features, use the source domain pre-training model to predict the fault type, and screen high-confidence samples with a Softmax probability exceeding 0.9 as pseudo labels; the pseudo labels are used to combine with the original labels of the target domain to determine new target domain labels; determine time series data segments from different fields but belonging to the same fault type, and time series data segments with small differences in the spray volume per unit area. The samples with a probability greater than 0.5 and the same filler type are considered positive sample pairs; the filler types include film type and drip type; time series segments of different fault types are selected from the source domain and the target domain to determine cross-domain different faults; and time series segments of different fault types in the same domain are determined to be same-domain different faults; the spray density greater than 1.2 is determined to be high spray density, and the spray density less than 0.8 is determined to be low spray density; from the sample pairs constructed with cross-domain different faults and same-domain different faults, one sample with high spray density and the other with low spray density is removed to avoid the model using the spray density difference as a feature to distinguish faults, and negative sample pairs are obtained. Feature similarity measurement is performed based on the positive sample pairs and the negative sample pairs, and the contrast loss value is calculated based on the contrast loss function, and the aligned feature representation and contrast loss value are output.

[0009] According to the first aspect of the embodiment of the present application, the aforementioned method of extracting data representing thermodynamic and vibration coupling characteristics from historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling characteristic matrix, can specifically include the following steps: determining heat transfer efficiency index data in the historical sensor time series data, the heat transfer efficiency index data including: cooling tower inlet water temperature, cooling tower outlet water temperature, and circulating water mass flow rate; The preset heat transfer coefficient, effective area of ​​the packing and logarithmic mean temperature difference are obtained, and the heat transfer efficiency ratio is calculated and determined in combination with the heat transfer efficiency index data; the heat transfer efficiency change rate is calculated based on the heat transfer efficiency ratio, and combined with the obtained vibration acceleration of the characteristic frequency band of the fan bearing, a coupling characteristic matrix is ​​constructed; among them, the heat transfer efficiency ratio and the heat transfer efficiency change rate are also used to trigger the scaling fault warning; when the heat transfer efficiency ratio is less than 0.85 and the heat transfer efficiency change rate decreases by more than 0.5 percentage points per minute, the scaling fault warning is triggered.

[0010] According to the first aspect of the embodiment of the present application, the spatial position mark includes the water pipe segment position, the packing layer height and the radial coordinates of the fan; the historical sensor time series data includes the time series data corresponding to the vibration sensor, current sensor, water flow sensor, water pressure sensor, wind pressure sensor, temperature sensor, circulating water conductivity sensor and make-up water conductivity sensor; wherein, the time series data collected by the vibration sensor includes the cooling tower fan bearing vibration data and the reducer axial vibration data; the temperature sensor is used to measure the inlet and outlet temperatures of the cooling tower. According to the first aspect of the embodiment of the present application, the fault diagnosis information includes the fault type, spatial coordinate encoding and scaling probability; the trend prediction information includes: scaling rate and fault development trend extrapolated by the long short-term memory network (LSTM) hidden state, the fault development trend includes the vibration amplitude trend per unit time; the fault type includes: heat transfer efficiency reduction related to scaling, packing layer collapse, fan blade corrosion imbalance, water distributor pipe blockage, motor overload, and bearing wear; the spatial coordinate encoding includes: water pipe segment coordinates, packing layer height coordinates and fan radial coordinates corresponding to the spatial position mark.

[0011] According to the first aspect of the embodiment of the present application, the aforementioned KL divergence is used to quantify the difference between multiple source domains and the target domain, dynamically adjust the source domain weights, and is related to the total loss function of adversarial training. The aforementioned feature representation based on alignment, combined with local temporal patterns and global dependencies, performs feature fusion, and dynamically adjusts the source domain weights according to KL divergence and water hardness information to obtain target fusion features. Specifically, the following steps may be included: based on the aligned feature representation, local feature extraction is performed through a one-dimensional convolutional neural network, and the convolution kernel is used to capture local patterns including transient shocks in the vibration signal to obtain the first target feature; global feature extraction is performed based on the long short-term memory network LSTM to capture the long-term dependency of the signal to obtain the second target feature; the long-term dependency includes a gradual trend; the first target feature and the second target feature are feature spliced ​​to obtain a multi-source fusion feature; based on the multi-source fusion feature, the source domain weights are dynamically adjusted through KL divergence and water hardness information, and the fusion weights of the multi-source features are adjusted to obtain the target fusion feature.

[0012] According to the first aspect of the embodiment of the present application, the aforementioned historical sensor time series data containing spatial position marks of multiple data center cooling towers are collected, the source domain and the target domain are divided, and the data are timestamp aligned, the sampling frequency is unified and independently normalized, and the domain-invariant features and label information of the source domain and the target domain are extracted by learning. Specifically, the following steps may be included: collecting historical sensor time series data of multiple data center cooling towers, determining that the data of the data center cooling towers of a part of the historical sensor time series data are source domain data; determining that the data of the data center cooling towers of another part of the historical sensor time series data are target domain data; using linear interpolation to unify the time base of all data to complete timestamp alignment; all data The data is resampled to a unified frequency to align the sampling frequency, and the missing values ​​caused by frequency conversion are filled in the resampling process; when the original sampling frequency is lower than the target frequency, the data points are increased by interpolation; when the original sampling frequency is higher than the target frequency, the data points are reduced by downsampling; each type of sensor data is independently normalized to eliminate dimensional differences; the source domain data and target domain data after timestamp alignment, sampling frequency unification and independent normalization are input into the adversarial training model; forward propagation is performed based on the feature extraction network, gradient reversal layer, and domain classifier to obtain the domain prediction probability; the domain classifier is a 2-class network; based on the preset domain classification loss function , preset fault classification loss function The loss is calculated and weighted with the domain prediction probability to obtain the total loss; backpropagation is performed based on the total loss and the feature extraction network CNN, domain classifier and fault classifier are updated to extract domain-invariant features and label information, and the domain classification loss is minimized to accurately distinguish the source domain and target domain, and the fault classification loss is minimized to improve the fault classification accuracy; among them, the fault classifier includes a fully connected layer and a Softmax activation layer and is used to map the aggregated features to the fault category probability.

[0013] In the second aspect, an embodiment of the present application provides a cooling tower cross-domain fault diagnosis system based on multi-source confrontation field adaptation. The cooling tower cross-domain fault diagnosis system based on multi-source confrontation field adaptation includes: a first acquisition and analysis module, a feature extraction and analysis module, a cross-domain timing alignment module, a second acquisition and analysis module, a feature fusion module and a fault diagnosis prediction module.

[0014] Specifically, the first acquisition and analysis module is used to collect historical sensor time series data containing spatial position marks from cooling towers in multiple data centers, divide the source domain and target domain, and perform timestamp alignment, sampling frequency unification and independent normalization on the data, and learn to extract domain-invariant features and label information of the source domain and target domain; the feature extraction and analysis module is used to extract data representing the thermodynamic and vibration coupling characteristics from the historical sensor time series data to calculate the heat transfer efficiency ratio, and obtain the vibration acceleration of the fan bearing characteristic frequency band to jointly construct the coupling feature matrix; the cross-domain time series alignment module is used to perform cross-domain time series alignment processing through comparative learning based on the coupling feature matrix and domain-invariant features, combined with the unit area spray of the cooling tower. Positive sample pairs are constructed based on the difference in spraying volume and filler type, and negative sample pairs from different fault types are constructed by sample separation in combination with spraying density, feature similarity measurement is performed, and the aligned feature representation is output; the second acquisition and analysis module is used to collect circulating water conductivity through a circulating water conductivity sensor to determine water hardness information; the feature fusion module is used to perform feature fusion based on the aligned feature representation, combining local temporal patterns with global dependencies, and dynamically adjust the source domain weights according to the KL divergence and water hardness information to obtain the target fusion feature; the fault diagnosis and prediction module is used to input the target fusion feature into the trained fault classifier, and output fault diagnosis information and trend prediction information of the cooling tower.

[0015] In a third aspect, an embodiment of the present application provides an electronic device comprising: a processor, a memory, and a program stored in the memory and runnable on the processor. When the program is executed by the processor, the cooling tower cross-domain fault diagnosis method based on multi-source confrontation field adaptation in the aforementioned first aspect is implemented.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a program or instruction. When the program or instruction is executed by a processor, the cooling tower cross-domain fault diagnosis method based on multi-source confrontation field adaptation in the aforementioned first aspect is implemented.

[0017] This application provides a cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation. Compared with the existing technology, it has the following advantages: This application effectively reduces the distribution difference between the source domain and the target domain through the adversarial training mechanism, so that the domain-invariant features generated by the model are more focused on fault discrimination information, significantly improving the generalization ability of the model in different geographical regions and heterogeneous data centers; this application performs cross-domain temporal alignment processing on the domain-invariant features of the source domain and the target domain based on contrastive learning, and can trigger the model to be retrained without the labeled data of the target domain, which can significantly reduce the dependence on manual labeling and reduce the labeling cost. By combining local temporal patterns with global dependencies to perform multi-level temporal feature fusion, and dynamically adjusting the source domain weights based on divergence and water hardness information, a multi-dimensional characterization of fault modes is achieved, and the accuracy of target domain fault diagnosis is improved. This application makes the differences between different domains precise by considering the thermodynamic parameters of the cooling tower, water hardness and other physical states, and avoids the high-hardness source domain interfering with the diagnosis of the low-hardness target domain. In the process of cross-domain time series alignment through contrastive learning, the coupling feature matrix and domain-invariant features are comprehensively considered; the difference in spray volume per unit area of ​​the cooling tower and the filler type are considered when constructing positive sample pairs; and the spray density is considered for sample separation when constructing negative sample pairs from different fault types, which enhances the recognition and diagnosis of cooling tower fault modes and improves the accuracy of cooling tower compound fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation provided by an embodiment of the present application; Figure 2 This is a schematic structural diagram of a cooling tower cross-domain fault diagnosis system based on multi-source confrontation domain adaptation provided by an embodiment of the present application; Figure 3 Schematic diagram of another cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation provided by an embodiment of the present application; Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0021] The embodiments of the present application provide a cooling tower cross-domain fault diagnosis method based on multi-source adversarial domain adaptation, which solves the problems existing in current cooling tower fault diagnosis, such as reliance on labeled data, limitation to single-domain data, and insufficient fault identification and diagnosis capabilities.

[0022] The technical solution in the embodiments of the present application is to solve the above technical problems, and the overall idea is as follows: With the rapid development of technologies such as the internet, cloud computing, and big data, data centers, as core infrastructure for information storage, processing, and transmission, are expanding in number and scale, playing a vital role in the global economy and society. Cooling towers are a core component of data center cooling systems, responsible for dissipating the significant heat generated by servers and other equipment, maintaining a suitable temperature environment and ensuring stable operation. A cooling tower failure can cause the data center temperature to rise, impacting the performance of servers and other equipment, and even causing equipment overheating and damage, resulting in serious consequences such as data loss and service interruptions, bringing significant economic losses to businesses and society.

[0023] Most existing research on cooling tower fault diagnosis focuses on a single data center. This requires high data annotation requirements. Whenever a new data center's cooling tower is fault-diagnosed, data annotation and model retraining are necessary. This wastes significant time and resources, including manpower for data annotation, computing resources for model training, and time costs.

[0024] Specifically, the relevant technologies have the following defects: (1) Existing fault diagnosis models generally rely on the full amount of labeled data in the target domain. However, the acquisition of labeled data in industrial scenarios faces the problem of high cost, and in the cold start phase, there is not enough labeled data available for training fault diagnosis models. (2) Existing technologies are based on the assumption of identical distribution of single-source domains, ignoring the distribution offset of sensor data caused by environmental conditions, load fluctuations, and equipment heterogeneity in different data centers. (3) Existing research is limited by the scale of single-domain data collection. Its training set only covers limited fault types and steady-state conditions, and it is difficult to cover complex and diverse fault modes and operating conditions, which limits the model's ability to identify and diagnose diverse faults. (4) In existing multi-source domain fault diagnosis methods, most directly use data-driven methods to quantify domain differences, ignoring the domain differences caused by physical factors, resulting in inaccurate quantification of domain differences; (5) In existing multi-source domain fault diagnosis methods, the complex fault modes of cooling towers (such as vibration anomalies caused by scaling) may be identified as independent features for processing, resulting in inaccurate fault diagnosis.

[0025] It should also be understood that adversarial domain adaptation is a method that uses adversarial training to align features between models in different domains. The core idea of ​​adversarial domain adaptation is to use gradient reversal layers and domain discriminators to enable the feature extractor to generate domain-invariant features, thereby narrowing the distribution gap between the source and target domains. This approach allows the model to adapt to the distribution of sensor data from different data center cooling towers, thereby improving its generalization and cross-domain adaptability.

[0026] Multi-source transfer learning is a method that learns knowledge from multiple source domains and transfers it to a target domain. However, data from different source domains differ in distribution and features. For example, in a cooling tower fault diagnosis scenario, the cooling towers in different data centers vary in environment, load, and equipment model, and the distribution of sensor data from each data center is also different. Therefore, it is necessary to dynamically adjust the weights based on the similarity between the source and target domains, assigning greater weights to source domains with high similarity and reducing the weights to those with low similarity. This can reduce the interference of source domains with large differences on target domain learning and allow the model to better focus on transferring knowledge related to the target domain. KL divergence is one of the commonly used methods for dynamically adjusting weights. It can effectively quantify the differences in feature distribution between different source domains and the target domain, providing a basis for weight allocation. This method enables the model to more rationally utilize information from multiple source domains, thereby improving the model's performance in the target domain.

[0027] Therefore, in this context, it is particularly necessary to explore a cross-domain cooling tower fault diagnosis method. A cross-domain cooling tower fault diagnosis method can achieve knowledge transfer and sharing between different data centers, effectively avoiding repeated data labeling and model training, improving diagnostic efficiency, reducing costs, and better adapting to the rapid development and diversified needs of data centers.

[0028] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0029] The following first introduces a cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation provided by an embodiment of the present application.

[0030] The present invention provides a flow chart of a cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation. Figure 1 and Figure 3 The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation may include the following steps S110-S160.

[0031] S110. Collect historical sensor time series data containing spatial location tags from cooling towers in multiple data centers, divide the source domain and target domain, align the data with timestamps, unify the sampling frequency, and independently normalize the data, and learn to extract domain-invariant features and label information from the source domain and target domain.

[0032] S120. Extract data representing thermodynamic and vibration coupling characteristics from historical sensor time series data to calculate the heat transfer efficiency ratio, and obtain the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling characteristic matrix.

[0033] S130. Based on the coupled feature matrix and domain-invariant features, cross-domain time series alignment is performed through contrastive learning. Positive sample pairs are constructed by combining the difference in spray volume per unit area of ​​the cooling tower and the filler type. Sample separation is performed based on the spray density to construct negative sample pairs from different fault types. Feature similarity is measured and the aligned feature representation is output.

[0034] S140. Collect the conductivity of the circulating water through a circulating water conductivity sensor to determine water hardness information.

[0035] S150. Based on the aligned feature representation, feature fusion is performed by combining local temporal patterns with global dependencies. The source domain weights are dynamically adjusted according to the KL divergence and water hardness information to obtain the target fusion features.

[0036] S160: Input the target fusion features into the trained fault classifier, and output the fault diagnosis information and trend prediction information of the cooling tower.

[0037] The above is a specific implementation method of the cross-domain fault diagnosis method for cooling towers based on multi-source adversarial domain adaptation provided by the embodiment of the present application. It can be understood that the present application effectively reduces the distribution difference between the source domain and the target domain through the adversarial training mechanism, so that the domain-invariant features generated by the model are more focused on fault discrimination information, significantly improving the generalization ability of the model in different geographical regions and heterogeneous data centers; the present application performs cross-domain temporal alignment processing on the domain-invariant features of the source domain and the target domain based on contrastive learning, and can trigger the model to be retrained without the labeled data of the target domain, which can significantly reduce the dependence on manual labeling and reduce the labeling cost. The present application combines local temporal patterns with global dependencies to perform multi-level temporal feature fusion, and dynamically adjusts the source domain weights based on divergence and water hardness information, thereby realizing multi-dimensional characterization of fault modes and improving the accuracy of target domain fault diagnosis. The present application can quickly predict fault development trends by collecting multi-sensor data in real time, thereby improving the overall operation and maintenance efficiency of the cooling tower of the new data center. It should be noted that this application, by considering the thermodynamic parameters of the cooling tower, water hardness and other physical states, makes the differences between different domains more precise, avoiding the interference of the high-hardness source domain with the diagnosis of the low-hardness target domain. In the process of cross-domain time series alignment through contrastive learning, the coupling feature matrix and domain-invariant features are comprehensively considered; the difference in the spray volume per unit area of ​​the cooling tower and the type of filler are considered when constructing positive sample pairs; and the spray density is considered for sample separation when constructing negative sample pairs from different fault types. This enhances the recognition and diagnosis of cooling tower fault modes and improves the accuracy of cooling tower complex fault diagnosis.

[0038] In one example, the spatial position mark includes the water pipe segment position, filler layer height, and fan radial coordinate; The historical sensor time series data includes the time series data of the corresponding vibration sensors, current sensors, water flow sensors, water pressure sensors, wind pressure sensors, temperature sensors, circulating water conductivity sensors and make-up water conductivity sensors; among them, the time series data collected by the vibration sensor includes the vibration data of the cooling tower fan bearing and the axial vibration data of the reducer; the temperature sensor is used to measure the inlet and outlet temperatures of the cooling tower.

[0039] It is understandable that this application first performs multi-source sensor data collection and domain partitioning. Multiple data centers (≥3) are selected, and all historical cooling tower data from each data center is used as a source domain. This domain includes the time series data from the aforementioned various sensors and is annotated with fault type, location, and timestamp. All historical cooling tower data from an independent data center is selected as a target domain. This domain also includes the time series data from the aforementioned various sensors, with a small number of fault annotations. The frequency range of the data collected by the vibration sensor is 5Hz-2000Hz.

[0040] In another example, the process of determining water hardness information based on the conductivity of circulating water satisfies the expression: Where H represents the water hardness information, and EC represents the electrical conductivity of the circulating water.

[0041] In some embodiments, after inputting the target fusion features into the trained fault classifier and outputting the fault diagnosis information and trend prediction information of the cooling tower, the cross-domain fault diagnosis method for cooling towers based on multi-source adversarial domain adaptation may further include the following steps: S170. Build a data distribution drift detection mechanism and a water quality parameter monitoring mechanism to monitor state changes, and trigger a preset model retraining strategy when the state change reaches a preset target threshold.

[0042] The aforementioned data distribution drift detection mechanism may specifically include the following steps: S171, using a sliding window mechanism to continuously cache the real-time sensor data stream of the target cooling tower; S172. Based on the cached data, regularly perform model parameter fine-tuning to adapt the model to the slow drift of device operation; S173. Deploy an autoencoder to learn the feature representation of normal operating data in the target domain and calculate the reconstruction error of the real-time data; S174. When the reconstruction error continues to exceed a preset first threshold, determine that the current data distribution significantly deviates from the distribution during model training, and determine that the state change reaches a preset target threshold.

[0043] In the embodiment of the present application, it can be understood that when the reconstruction error continues to exceed the preset first threshold, it indicates that the current data distribution has significantly deviated from the distribution during model training, triggering model retraining.

[0044] The aforementioned water quality parameter monitoring mechanism may specifically include the following steps: S175. Collecting the conductivity of the supplementary water through a supplementary water conductivity sensor, monitoring water quality parameters based on the supplementary water conductivity and the circulating water conductivity, and determining the concentration ratio; S176: When the absolute deviation between the current concentration factor and the concentration factor during model training exceeds a preset second threshold, determining that the state change reaches a preset target threshold; S177. During model retraining, implement adaptive sampling to prioritize data collection during high-concentration periods corresponding to scaling-sensitive periods to optimize data efficiency and the model's sensitivity to critical faults; and increase data sampling frequency or weights to allow the model to focus on learning more discriminative features under scaling-prone conditions.

[0045] In the embodiments of this application, it is understood that to ensure that the diagnostic model can continuously adapt to changes in equipment status and new operating conditions after deployment in the target data center, this method adopts an online adaptive update strategy; this application performs two-dimensional joint monitoring based on a data distribution drift detection mechanism and a water quality parameter monitoring mechanism. The window size of the sliding window mechanism can be set to 1 hour, and the calculation of the concentration factor CT satisfies the expression: Where, represents the conductivity of circulating water, Indicates the conductivity of the make-up water.

[0046] Regarding the water quality parameter monitoring mechanism, the triggering condition for determining that the state change reaches the preset target threshold and triggering the preset model retraining strategy satisfies the expression: Where, Indicates the current concentration multiple. This represents the concentration factor used during model training, with a second threshold of 1.5. If the absolute deviation between the current concentration factor and the concentration factor used during model training exceeds the preset second threshold, it indicates a significant change in water quality (such as a change in drainage strategy or a water source switch), which could significantly affect scaling trends and heat transfer efficiency, necessitating model retraining to adapt to the new water quality conditions.

[0047] It should be noted that this application optimizes the online update mechanism and adds concentration multiple monitoring and adaptive sampling strategies in order to ensure that the system can dynamically adapt to the working status of the cooling tower equipment during actual operation and adjust the monitoring frequency in a timely manner.

[0048] In some embodiments, the aforementioned collecting of historical sensor time series data containing spatial location tags from cooling towers of multiple data centers, dividing the source domain and the target domain, and performing timestamp alignment, sampling frequency unification, and independent normalization processing on the data, and learning to extract domain-invariant features and label information of the source domain and the target domain, that is, the aforementioned S110 may specifically include the following steps: S210: Collect historical sensor time series data of multiple data center cooling towers, determine that data of a portion of the historical sensor time series data of the data center cooling towers is source domain data; and determine that data of another portion of the historical sensor time series data of the data center cooling towers is target domain data. S220, using linear interpolation to unify the time base for all data to complete timestamp alignment; S230, resampling all data to a uniform frequency to align the sampling frequencies, and filling in missing values ​​caused by frequency conversion during the resampling process; wherein, if the original sampling frequency is lower than the target frequency, adding data points by an interpolation method; if the original sampling frequency is higher than the target frequency, reducing data points by a downsampling method; S240, independently normalizing each type of sensor data to eliminate dimensional differences; S250, inputting the source domain data and target domain data after timestamp alignment, sampling frequency unification and independent normalization into the adversarial training model; S260, performing forward propagation based on the feature extraction network, the gradient reversal layer, and the domain classifier to obtain a domain prediction probability; wherein the domain classifier is a binary classification network; S270, based on the preset domain classification loss function , preset fault classification loss function Perform loss calculation and weighting on the domain prediction probability to obtain the total loss; S280, backpropagation is performed based on the total loss to update the feature extraction network CNN, domain classifier, and fault classifier to extract domain-invariant features and label information, and the domain classification loss is minimized to accurately distinguish the source domain and target domain, and the fault classification loss is minimized to improve the fault classification accuracy; The fault classifier includes a fully connected layer and a Softmax activation layer, which is used to map aggregated features to fault class probabilities. This application uses the Adam optimizer, with a learning rate that decays with each training round. This application jointly trains the fault classifier based on contrastive loss, classification loss, and adversarial loss.

[0049] In the embodiments of this application, it is understood that this application selects multiple data centers and determines all historical cooling tower data from each data center as a source domain, covering time series data of vibration, current, water flow, and pressure sensors, and annotating fault type, location, and timestamp. This application also selects all historical cooling tower data from one or more independent data centers as a target domain, covering time series data of vibration, current, water flow, and pressure sensors, and requiring a small number of fault annotations to be retained.

[0050] This application uses an adversarial training mechanism to effectively align the feature distributions of different data centers. This allows the model to fully consider the differences in multi-source domain data during learning and generate domain-invariant features, thereby better adapting to the distribution characteristics of the target data center and improving diagnostic performance. Adversarial training involves forward propagation, loss calculation, backpropagation, and parameter updates.

[0051] In the forward propagation, the feature extraction network generates features based on the source domain data and the target domain data through a 3-layer convolution + maximum pooling feature extraction network CNN. ; The gradient reversal layer involves forward propagation and back propagation; in the forward propagation, the features are directly output ,Right now ;In backpropagation, the gradient is multiplied by the inversion coefficient (Initial value 1.0, can be adjusted dynamically); the domain classifier is a two-classification network (2-layer fully connected), and the input of the domain classifier is the feature , output domain prediction probability .

[0052] In loss calculation, the domain classification loss function Satisfies the expression: Where, ,in, is the number of samples in the source domain, is the number of samples in the target domain; : Domain label (1 indicates source domain, 0 indicates target domain); : Domain classifier for samples The predicted probability of .

[0053] Fault classification loss function Satisfies the expression: Where, is the number of samples in the source domain, For the The true category index of the sample (such as Indicates "pipeline blocked"), Represents the true category index in the probability distribution of the model output The probability value of .

[0054] The total loss function satisfies the expression: Where, It is an adversarial loss weight with an initial value of 1.0 and is dynamically adjusted based on the validation set.

[0055] It should be noted that this application dynamically adjusts the migration weight based on the similarity between the source domain and the target domain, so that the model can more reasonably utilize information from multiple source domains, further improving the generalization ability and adaptability of the model. Through data fusion across data centers, the model can more accurately identify the type of fault, locate the fault location, and predict the development trend of the fault, thereby improving the accuracy and reliability of fault diagnosis. Under the multi-source adversarial domain adaptive framework, after aligning the feature distribution using adversarial training, a small amount of labeled data in the target data center can quickly deploy a high-precision diagnostic model, greatly reducing the labeling cost and cold start difficulty, and improving the applicability of the model in actual industrial scenarios.

[0056] In some embodiments, the aforementioned step S120 may specifically include the following steps: extracting data representing thermodynamic and vibration coupling characteristics from historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining vibration acceleration in characteristic frequency bands of fan bearings to jointly construct a coupling characteristic matrix. S310, determining heat transfer efficiency index data in the historical sensor time series data, where the heat transfer efficiency index data includes: cooling tower inlet water temperature, cooling tower outlet water temperature, and circulating water mass flow rate; S320, obtaining a preset heat transfer coefficient, effective filler area, and logarithmic mean temperature difference, and calculating and determining a heat transfer efficiency ratio in combination with the heat transfer efficiency index data; S330, calculating the heat transfer efficiency change rate based on the heat transfer efficiency ratio, and constructing a coupling characteristic matrix based on the obtained vibration acceleration of the fan bearing characteristic frequency band; Among them, the heat transfer efficiency ratio and the heat transfer efficiency change rate are also used to trigger the scaling fault warning; when the heat transfer efficiency ratio is less than 0.85 and the heat transfer efficiency change rate decreases by more than 0.5 percentage points per minute, the scaling fault warning is triggered.

[0057] In the embodiments of the present application, it is understood that in order to enhance the cross-domain discrimination capability of complex faults (such as structural vibration anomalies), the present application introduces thermodynamic and vibration coupling characteristics for fault detection. The heat transfer efficiency ratio is calculated in real time based on the strong coupling characteristics of the cooling tower heat transfer efficiency and mechanical vibration. The calculation process of the heat transfer efficiency ratio satisfies the expression: Where, represents the heat transfer efficiency ratio, Indicates the cooling tower inlet water temperature, Indicates the cooling tower outlet water temperature. represents the mass flow of circulating water, K represents the heat transfer coefficient, A represents the effective area of ​​the packing, represents the logarithmic mean temperature difference.

[0058] The aforementioned coupling characteristic matrix satisfies the expression: Where, represents the coupling characteristic matrix, represents the rate of change of heat transfer efficiency, is the vibration acceleration of the fan bearing in the characteristic frequency band. More specifically, Refers to the effective value of the vibration acceleration of the fan bearing in the frequency band of 200Hz-500Hz; The calculation steps are as follows: perform a 200Hz-500Hz bandpass filter on the original vibration signal collected by the vibration sensor of the fan bearing to retain the signal components in this frequency band, and calculate the root mean square value of the filtered signal as the effective value of the vibration acceleration in this frequency band. A fault is detected only when the combined values ​​of the three variables change. For example, a scaling fault warning is triggered when the heat transfer efficiency ratio is less than 0.85 and the rate of change in heat transfer efficiency decreases by more than 0.5 percentage points per minute. This application combines the cooling tower's thermodynamic efficiency with vibration signals to form a new coupled feature matrix, which is particularly critical for fault diagnosis and can significantly improve the accuracy of early warnings for scaling and fan failures.

[0059] In some embodiments, the aforementioned cross-domain time series alignment processing is performed through contrastive learning based on the coupling feature matrix and domain-invariant features, and positive sample pairs are constructed in combination with the difference in the spray volume per unit area of ​​the cooling tower and the filler type. The samples are separated in combination with the spray density to construct negative sample pairs from different fault types, and feature similarity measurement is performed to output the aligned feature representation. That is, the aforementioned S130 may specifically include the following steps: S410: Input the coupled feature matrix and domain-invariant features, use the source domain pre-trained model to predict the fault type, and select high-confidence samples with a Softmax probability exceeding 0.9 as pseudo labels. The pseudo labels are combined with the original labels of the target domain to determine new target domain labels. S420: Determine as positive sample pairs time series data segments from different fields but belonging to the same fault type, and samples with a difference of less than 0.5 in spray volume per unit area and the same filler type; wherein filler types include film type and drip type; S430: Select time sequence segments of different fault types from the source domain and the target domain to determine cross-domain different faults; and determine time sequence segments of different fault types in the same domain as same-domain different faults; S440, determining that a spray density greater than 1.2 is a high spray density, and determining that a spray density less than 0.8 is a low spray density; S450: From the sample pairs constructed from different faults across different domains and different faults within the same domain, remove the case where one sample has high spray density and the other has low spray density, so as to avoid the model using the difference in spray density as a feature to distinguish faults, and obtain negative sample pairs. S460: perform feature similarity measurement based on the positive sample pair and the negative sample pair, calculate the contrast loss value based on the contrast loss function, and output the aligned feature representation and contrast loss value.

[0060] In the embodiments of the present application, it is understood that the present application utilizes contrastive learning to construct positive and negative sample pairs. Through the cosine similarity metric and contrastive loss function, the temporal features of similar cross-domain faults are forced to cluster in the embedding space, and the features of different faults are separated, thereby improving the model's robustness to domain differences and its ability to discriminate fault features. In the present application, only a small amount of labeled data is required for the target domain. By generating pseudo-labels through the source domain pre-training model and combining it with contrastive learning, the cost of acquiring labeled data in industrial scenarios can be greatly reduced, solving the problem of insufficient data in the cold start phase.

[0061] To improve the quality of positive samples and fault differentiation, this application selects samples with a spray volume per unit area difference of less than 0.5L / (min·m²) and the same filler type (filler types include film type and drip type). This ensures that the positive samples not only match the fault type, but also have a high degree of similarity in key operating conditions (spray density) and core equipment structure (filler type), reducing feature offsets caused by differences in physical conditions and improving the effectiveness of comparative learning.

[0062] When constructing negative sample pairs, we actively separate samples with high and low spray densities, thereby strengthening the model's ability to distinguish between these two easily confused faults. Spray density is the total spray water volume of a cooling tower divided by its effective spraying area. High spray density (>1.2 L / min·m²) accelerates scaling, while low spray density (<0.8 L / min·m²) triggers cavitation. By adding spray density as a new constraint to screen positive and negative samples, we can effectively resolve the confusion between scaling and cavitation faults in traditional methods. Fault differentiation based on spray density can effectively enhance the robustness of fault diagnosis models and avoid misjudgments.

[0063] In other words, negative sample pairs must come from different fault types. Furthermore, if one sample has high spray density and the other has low spray density, a negative sample pair is not allowed (even if the fault types are different). Otherwise, negative sample pairs are allowed if both samples have high, low, or intermediate spray densities, or if one sample is in the intermediate range and the other is in the high / low range. This prevents the model from using spray density differences as a distinguishing feature, allowing it to focus on the true fault characteristics. Furthermore, by separating the spray density ranges, the model can more clearly distinguish between scaling and cavitation faults.

[0064] It should be noted that this application uses contrastive learning to perform cross-domain temporal alignment on the domain-invariant features and coupling feature matrices of the source and target domains to align the temporal features of the same fault across domains to enhance the model's robustness to domain differences. The source domain temporal features are the temporal features extracted from the domain-invariant features: ( is the number of source domain samples, Tis the time step, D is the feature dimension); The target domain temporal features are the temporal features after domain-invariant feature extraction: ( is the number of samples in the target domain); the label information is the fault type label of the source domain data and a small number of target domain labels.

[0065] Furthermore, cross-domain identical faults are time series segments of the same fault type selected from both the source and target domains (the target domain's fault labels are derived from a minimal number of original labels and previously generated pseudo-labels); cross-domain different faults are time series segments of different fault types selected from both the source and target domains; and same-domain different faults are time series segments of different fault types within the same domain. Cosine similarity can be used to measure feature similarity. After alignment, the time series features of the same fault across domains are closer in the embedding space, while features of different faults are farther apart. The contrast loss value can be used to jointly optimize the model.

[0066] In one example, the contrastive loss function Satisfies the expression: Where, is the batch size, that is, the number of samples processed simultaneously in one training iteration, refers to the anchor sample, that is, the sample of the source domain, is the positive sample in the target domain that has the same fault as the anchor sample, that is, the corresponding sample in the target domain; refers to negative samples (samples with different faults or different domains), s is a similarity function, It refers to the temperature parameter (usually set to 0.1-1.0), which is used to control the discrimination of similarity; M is the number of negative samples.

[0067] In some embodiments, the aforementioned KL divergence is used to quantify the differences between multiple source domains and the target domain, dynamically adjust the source domain weights, and is related to the total loss function of adversarial training. The aforementioned feature representation based on alignment combines local temporal patterns with global dependencies to perform feature fusion, and dynamically adjusts the source domain weights based on KL divergence and water hardness information to obtain the target fused features. Specifically, the aforementioned S150 may include the following steps: S510, based on the aligned feature representation, performing local feature extraction through a one-dimensional convolutional neural network, using a convolution kernel to capture local patterns including transient impacts in the vibration signal, to obtain a first target feature; S520, performing global feature extraction based on a long short-term memory network (LSTM) to capture long-term dependencies of the signal and obtain a second target feature; the long-term dependencies include gradual trends; S530, performing feature splicing on the first target feature and the second target feature to obtain a multi-source fusion feature; S540: Based on the multi-source fusion features, the source domain weights are dynamically adjusted through KL divergence and water hardness information, and the fusion weights of the multi-source features are adjusted to obtain the target fusion features.

[0068] In the embodiments of the present application, it is understood that the present application combines a one-dimensional convolutional neural network to extract local transient features of vibration signals with an LSTM to capture global long-term dependencies of signals such as current and water flow. By concatenating features, a fused feature containing multi-dimensional information is generated, improving the comprehensiveness of fault pattern recognition. By performing both local and global feature extraction on the aligned feature representations, the model can cover complex and diverse fault modes. For example, motor overload may manifest as a long-term, slow rise in the current signal (a global, gradual trend captured by the LSTM), accompanied by an increase in high-frequency transient shocks in the motor vibration signal (a local, bursty feature captured by the one-dimensional CNN). These two together constitute the coupled fault signature of motor overload. For another example, a pipe blockage may cause a sudden drop in water flow (a localized sudden change in the flow sensor), while simultaneously triggering an abnormal increase in water pump pressure (a long-term trend change in the pressure sensor), accompanied by a change in the pump's vibration pattern. The coupled effect of these three factors fully reflects the nature of the blockage fault.

[0069] This application dynamically adjusts the source domain weights through KL divergence and water hardness information to achieve optimal fusion of multi-source features and obtain target fusion features. By introducing water hardness information to optimize weight distribution, the physical differences between the multi-source domain and the target domain (such as water hardness) can be accurately quantified. Water hardness is considered in the multi-source domain fault diagnosis of cooling towers mainly because different water hardness has a significant impact on the heat transfer efficiency of cooling towers, such as scale formation and equipment corrosion. Dynamic hardness correction is performed based on water quality differences in different regions and weight distribution is combined with KL divergence, which significantly improves the accuracy of cross-domain transfer learning. In particular, adding a sensitivity coefficient for hardness differences as part of the dynamic adjustment can effectively alleviate the interference caused by water quality differences on the diagnostic model.

[0070] In one example, fault diagnosis information includes fault type, spatial coordinate encoding, and scaling probability; trend prediction information includes: scaling rate and fault development trend extrapolated through the long short-term memory network (LSTM) hidden state, and the fault development trend includes the vibration amplitude trend per unit time; fault types include: scaling-related heat transfer efficiency decrease, packing layer collapse, fan blade corrosion imbalance, water distributor pipe blockage, motor overload, and bearing wear; spatial coordinate encoding includes: water pipe segment coordinates corresponding to spatial position marks, packing layer height coordinates, and fan radial coordinates.

[0071] In some embodiments, the present application provides a cooling tower cross-domain fault diagnosis system 600 based on multi-source confrontation domain adaptation, such as Figure 2 As shown, the cooling tower cross-domain fault diagnosis system 600 based on multi-source confrontation domain adaptation may include the following modules: The first acquisition and analysis module 610 is configured to collect historical sensor time series data containing spatial location tags from cooling towers in multiple data centers, divide the data into source and target domains, align the data with timestamps, unify the sampling frequency, and independently normalize the data, and learn to extract domain-invariant features and label information from the source and target domains. Feature extraction and analysis module 620, for extracting data representing thermodynamic and vibration coupling characteristics from historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling feature matrix; Cross-domain time series alignment module 630 is used to perform cross-domain time series alignment processing through contrastive learning based on the coupling feature matrix and domain-invariant features. It combines the difference in spray volume per unit area of ​​the cooling tower and the filler type to construct positive sample pairs, and uses spray density to perform sample separation to construct negative sample pairs from different fault types. It then performs feature similarity measurement and outputs the aligned feature representation. The second collection and analysis module 640 is used to collect the conductivity of the circulating water through the circulating water conductivity sensor to determine the water hardness information; Feature fusion module 650 is used to perform feature fusion based on the aligned feature representation, combining local temporal patterns with global dependencies, and dynamically adjust the source domain weights according to KL divergence and water hardness information to obtain target fused features; The fault diagnosis and prediction module 660 is used to input the target fusion features into the trained fault classifier and output the fault diagnosis information and trend prediction information of the cooling tower.

[0072] According to an embodiment of the present application, any multiple modules among the first acquisition and analysis module 610, the feature extraction and analysis module 620, the cross-domain timing alignment module 630, the second acquisition and analysis module 640, the feature fusion module 650, and the fault diagnosis and prediction module 660 can be combined into a single module for implementation, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module.

[0073] Figure 2 Each module in the system shown has the function of implementing each step in the aforementioned cooling tower cross-domain fault diagnosis method based on multi-source confrontation field adaptation, and can achieve its corresponding technical effects. For the sake of concise description, it will not be repeated here.

[0074] In some embodiments, the present application provides an electronic device, the structural diagram of the electronic device is as follows Figure 4 shown.

[0075] The electronic device may include a processor 710 and a memory 720 storing computer program instructions.

[0076] Specifically, the processor 710 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0077] The memory 720 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 720 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 720 may include removable or non-removable (or fixed) media. Where appropriate, the memory 720 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 720 is a non-volatile solid-state memory.

[0078] The memory 720 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory 720 includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it may perform the operations described in any one of the cooling tower cross-domain fault diagnosis methods based on multi-source confrontation domain adaptation in the above-mentioned embodiments.

[0079] The processor 710 reads and executes computer program instructions stored in the memory 720 to implement any one of the cooling tower cross-domain fault diagnosis methods based on multi-source confrontation domain adaptation in the above embodiments.

[0080] In one example, the electronic device may further include a communication interface 730 and a bus 700. Figure 4 As shown, the processor 710 , the memory 720 , and the communication interface 730 are connected via a bus 700 and communicate with each other.

[0081] The communication interface 730 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application. Bus 700 includes hardware, software or both, and the parts of online data flow metering equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 700 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0082] In addition, in conjunction with the cross-domain fault diagnosis method for cooling towers based on multi-source adversarial domain adaptation in the above-mentioned embodiments, embodiments of the present application may provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the cross-domain fault diagnosis methods for cooling towers based on multi-source adversarial domain adaptation in the above-mentioned embodiments is implemented.

[0083] It should be understood that the present application is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present application is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present application.

[0084] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. Programs or code segments can be stored in machine-readable media, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROMs, flash memories, erasable ROMs (EROMs), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0085] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps. In other words, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0086] Aspects of the present disclosure have been described above with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit. It is also understood that each box in the block diagram and / or flowchart and the combination of the boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0087] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation, characterized in that: include: We collect historical sensor time series data with spatial location tags from cooling towers in multiple data centers, divide the data into source and target domains, align the data's timestamps, unify the sampling frequency, and independently normalize the data. We then learn to extract domain-invariant features and label information from the source and target domains. Extracting data representing thermodynamic and vibration coupling characteristics from the historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling characteristic matrix; Based on the coupling feature matrix and the domain-invariant features, cross-domain temporal alignment is performed through contrastive learning. Positive sample pairs are constructed by combining the difference in spray volume per unit area of ​​the cooling tower and the filler type. Sample separation is performed based on spray density to construct negative sample pairs from different fault types. Feature similarity is measured and the aligned feature representation is output. The conductivity of circulating water is collected by a circulating water conductivity sensor to determine the water hardness information; Based on the aligned feature representation, feature fusion is performed by combining local temporal patterns with global dependencies, and the source domain weight is dynamically adjusted according to the KL divergence and the water hardness information to obtain the target fusion feature; The target fusion features are input into the trained fault classifier, and fault diagnosis information and trend prediction information of the cooling tower are output.

2. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to claim 1 is characterized in that: After inputting the target fusion feature into the trained fault classifier and outputting the fault diagnosis information and trend prediction information of the cooling tower, the cross-domain fault diagnosis method for the cooling tower based on multi-source adversarial domain adaptation further includes: Build a data distribution drift detection mechanism and a water quality parameter monitoring mechanism to monitor state changes and trigger a preset model retraining strategy when the state change reaches the preset target threshold; The data distribution drift detection mechanism includes: A sliding window mechanism is used to continuously cache the real-time sensor data stream of the target cooling tower; Based on cached data, model parameter fine-tuning is performed regularly to adapt the model to the slow drift of device operation; Deploy an autoencoder to learn the feature representation of normal operating data in the target domain and calculate the reconstruction error of real-time data; When the reconstruction error continues to exceed a preset first threshold, it is determined that the current data distribution significantly deviates from the distribution during model training, and it is determined that the state change reaches a preset target threshold; The water quality parameter monitoring mechanism includes: collecting the conductivity of the supplementary water through a supplementary water conductivity sensor, monitoring water quality parameters based on the supplementary water conductivity and the circulating water conductivity, and determining the concentration ratio; When the absolute deviation between the current concentration factor and the concentration factor during model training exceeds a preset second threshold, determining that the state change reaches a preset target threshold; During the model retraining process, adaptive sampling is implemented to prioritize the collection of data from high-concentration periods corresponding to scaling-sensitive periods to optimize data efficiency and the model's sensitivity to critical faults. The data sampling frequency or weight is also increased to allow the model to focus on learning more discriminative features under scaling-prone conditions.

3. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to claim 1 is characterized in that: Based on the coupling feature matrix and the domain-invariant features, cross-domain time series alignment is performed through contrastive learning. Positive sample pairs are constructed by combining the difference in spray volume per unit area of ​​the cooling tower and the filler type. Sample separation is performed based on spray density to construct negative sample pairs from different fault types. Feature similarity measurement is performed, and the aligned feature representation is output, including: The coupling feature matrix and the domain-invariant features are input, and the source domain pre-trained model is used to predict the fault type. High-confidence samples with a Softmax probability exceeding 0.9 are selected as pseudo labels. The pseudo labels are combined with the original labels of the target domain to determine new target domain labels. Time series data segments from different fields but belonging to the same fault type, as well as samples with a difference of less than 0.5 in spray volume per unit area and the same filler type are determined as positive sample pairs; wherein the filler types include film type and drip type; Select time series segments of different fault types from the source and target domains to identify cross-domain different faults; and identify time series segments of different fault types within the same domain as same-domain different faults; Determine that the spray density is greater than 1.2 as high spray density, and determine that the spray density is less than 0.8 as low spray density; From the sample pairs constructed from the cross-domain different faults and the same-domain different faults, the case where one sample has the high spray density and the other has the low spray density is eliminated to avoid the model using the spray density difference as a feature to distinguish faults, and a negative sample pair is obtained. A feature similarity measurement is performed based on the positive sample pair and the negative sample pair, and a contrast loss value is calculated based on a contrast loss function, and the aligned feature representation and contrast loss value are output.

4. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to claim 1 is characterized in that: The method extracts data representing the thermodynamic and vibration coupling characteristics from the historical sensor time series data to calculate the heat transfer efficiency ratio, and obtains the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling characteristic matrix, including: Determining heat transfer efficiency index data in the historical sensor time series data, the heat transfer efficiency index data including: cooling tower water inlet temperature, cooling tower water outlet temperature, and circulating water mass flow rate; Obtaining the preset heat transfer coefficient, filler effective area, and logarithmic mean temperature difference, and calculating and determining the heat transfer efficiency ratio in combination with the heat transfer efficiency index data; Calculating the heat transfer efficiency change rate based on the heat transfer efficiency ratio, and constructing a coupling characteristic matrix based on the obtained vibration acceleration of the fan bearing characteristic frequency band; Among them, the heat transfer efficiency ratio and the heat transfer efficiency change rate are also used to trigger a scaling fault warning; when the heat transfer efficiency ratio is less than 0.85 and the heat transfer efficiency change rate decreases by more than 0.5 percentage points per minute, the scaling fault warning is triggered.

5. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to any one of claims 1 to 4, characterized in that: The spatial position mark includes the water pipe segment position, the filler layer height and the fan radial coordinate; The historical sensor time series data includes time series data corresponding to the vibration sensor, current sensor, water flow sensor, water pressure sensor, wind pressure sensor, temperature sensor, circulating water conductivity sensor and make-up water conductivity sensor; wherein, the time series data collected by the vibration sensor includes the cooling tower fan bearing vibration data and the reducer axial vibration data; the temperature sensor is used to measure the inlet and outlet temperatures of the cooling tower; The fault diagnosis information includes the fault type, spatial coordinate encoding, and scaling probability; the trend prediction information includes scaling rate and fault development trend extrapolated through the long short-term memory network (LSTM) hidden state, and the fault development trend includes the vibration amplitude trend per unit time; The types of failures mentioned include: decreased heat transfer efficiency related to scaling, packing layer collapse, fan blade corrosion and imbalance, water distributor pipe blockage, motor overload, and bearing wear; The spatial coordinate code includes: water pipe segment coordinates, filler layer height coordinates, and fan radial coordinates corresponding to the spatial position mark.

6. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to any one of claims 1 to 4, characterized in that: The KL divergence is used to quantify the differences between multiple source domains and target domains, dynamically adjust the source domain weights, and is related to the total loss function of adversarial training; The feature representation after alignment is combined with the local temporal pattern and the global dependency to perform feature fusion, and the source domain weight is dynamically adjusted according to the KL divergence and the water hardness information to obtain the target fusion feature, including: Based on the aligned feature representation, local feature extraction is performed using a one-dimensional convolutional neural network, and a convolution kernel is used to capture local patterns including transient impacts in the vibration signal to obtain a first target feature; Performing global feature extraction based on the long short-term memory network (LSTM) to capture the long-term dependency of the signal and obtain the second target feature; the long-term dependency includes a gradual trend; Performing feature splicing on the first target feature and the second target feature to obtain a multi-source fusion feature; Based on the multi-source fusion feature, the source domain weight is dynamically adjusted through the KL divergence and the water hardness information, and the fusion weight of the multi-source feature is adjusted to obtain the target fusion feature.

7. The cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation according to any one of claims 1 to 4, characterized in that: The method collects historical sensor time series data containing spatial location tags from multiple data center cooling towers, divides the source domain and target domain, and performs timestamp alignment, sampling frequency unification, and independent normalization on the data, and learns to extract domain-invariant features and label information of the source domain and target domain, including: Collecting historical sensor time series data of multiple data center cooling towers, determining that data of a portion of the historical sensor time series data of the data center cooling towers is source domain data; and determining that data of another portion of the historical sensor time series data of the data center cooling towers is target domain data; Linear interpolation is used to unify the time base of all data to complete timestamp alignment; All data are resampled to a uniform frequency to align the sampling frequencies, and missing values ​​caused by frequency conversion are filled in during the resampling process. When the original sampling frequency is lower than the target frequency, data points are added through interpolation; when the original sampling frequency is higher than the target frequency, data points are reduced through downsampling. Each type of sensor data is independently normalized to eliminate dimensional differences; Input the source domain data and target domain data after timestamp alignment, sampling frequency unification and independent normalization into the adversarial training model; Perform forward propagation based on the feature extraction network, gradient reversal layer, and domain classifier to obtain domain prediction probability; wherein the domain classifier is a binary classification network; Based on the preset domain classification loss function , preset fault classification loss function Perform loss calculation and weighting processing on the domain prediction probability to obtain the total loss; Backpropagation is performed based on the total loss to update the feature extraction network CNN, domain classifier, and fault classifier to extract domain-invariant features and label information, and the domain classification loss is minimized to accurately distinguish the source domain from the target domain, and the fault classification loss is minimized to improve the fault classification accuracy; The fault classifier includes a fully connected layer and a Softmax activation layer and is used to map aggregated features to fault category probabilities.

8. A cooling tower cross-domain fault diagnosis system based on multi-source confrontation domain adaptation, characterized in that: include: The first acquisition and analysis module is used to collect historical sensor time series data containing spatial location tags from cooling towers in multiple data centers, divide the data into source and target domains, align the data's timestamps, unify the sampling frequency, and independently normalize the data. It then learns to extract domain-invariant features and label information from the source and target domains. a feature extraction and analysis module for extracting data representing the thermodynamic and vibration coupling characteristics from the historical sensor time series data to calculate the heat transfer efficiency ratio, and obtaining the vibration acceleration of the fan bearing characteristic frequency band to jointly construct a coupling feature matrix; A cross-domain time series alignment module is configured to perform cross-domain time series alignment processing through contrastive learning based on the coupling feature matrix and the domain-invariant features, construct positive sample pairs based on the difference in spray volume per unit area of ​​the cooling tower and the filler type, and construct negative sample pairs from different fault types by sample separation based on spray density, perform feature similarity measurement, and output the aligned feature representation; The second collection and analysis module is used to collect the conductivity of the circulating water through the circulating water conductivity sensor to determine the water hardness information; A feature fusion module is used to perform feature fusion based on the aligned feature representation, combining local temporal patterns with global dependencies, and dynamically adjust source domain weights according to KL divergence and the water hardness information to obtain target fused features; The fault diagnosis and prediction module is used to input the target fusion features into the trained fault classifier and output the fault diagnosis information and trend prediction information of the cooling tower.

9. An electronic device, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for cross-domain fault diagnosis of a cooling tower based on multi-source confrontation domain adaptation as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the cooling tower cross-domain fault diagnosis method based on multi-source confrontation domain adaptation as described in any one of claims 1 to 7 is implemented.

Images