A chiller fault prediction diagnosis method based on a GRU-SA model

Abstract

The present application relates to the technical field of fault prediction, and more particularly to a chiller fault prediction and diagnosis method based on a GRU-SA model.The method comprises the following steps: obtaining a multimodal chiller dataset; performing data preprocessing on the multimodal chiller dataset to generate a multimodal chiller preprocessed dataset; performing physical knowledge virtual feature enhancement on the multimodal chiller preprocessed dataset to generate a multimodal chiller virtual feature set; and constructing dynamic adversarial data samples for the multimodal chiller virtual feature set to generate multimodal chiller model input data; therefore, the present application uses technical means such as multimodal data processing, physical knowledge enhancement, dynamic adversarial data sample generation, GRU-SA model optimization, and cloud-edge collaborative reasoning to solve the problems of insufficient accuracy and poor robustness of traditional chiller fault diagnosis models, thereby improving the accuracy and real-time performance of fault prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault prediction, and in particular to a chiller fault prediction and diagnosis method based on a GRU-SA model. BACKGROUND

[0002] Traditional chiller fault diagnosis methods rely heavily on static data analysis, which cannot effectively cope with the dynamic changes in unit operating conditions. Traditional chiller fault diagnosis methods rely heavily on static data analysis, which cannot effectively cope with the dynamic changes in unit operating conditions. GRU (Gated Recurrent Unit) is a structure based on recurrent neural network (RNN), aiming to solve the problem of gradient vanishing and explosion encountered by traditional RNN in long sequence data processing. Compared with LSTM (Long Short-Term Memory), GRU structure is more simplified, containing two main gating mechanisms: reset gate and update gate. The reset gate controls how to combine the past information with the current input, while the update gate determines how the current state affects the future information transmission. This makes GRU able to effectively capture long-term dependencies when processing time series data, making it suitable for processing chiller fault diagnosis problems with time series characteristics. SA (Self-Attention, self-attention mechanism) is a mechanism widely used in deep learning, which can capture the global dependence of data by calculating the relative importance between different parts of the input data. In traditional recurrent neural networks, information transmission mainly depends on the order of time steps, while the self-attention mechanism can more flexibly assign weights to different parts of the data, so that it can better focus on the important parts for the current task when processing sequence data. The GRU-SA model combines GRU with the self-attention mechanism, which can not only capture temporal dependencies with GRU, but also capture complex relationships between data in a wider time dimension through the self-attention layer, improving the model's ability to handle dynamic changes and complex relationships. Although the GRU-SA model has shown good performance in many fields, in the prior art, fault diagnosis systems usually only analyze data through traditional statistical methods or based on a single modality. These methods cannot handle the complex relationships of multi-modal data, and cannot effectively utilize the advantages of GRU-SA model. In addition, the lack of dynamic adjustment and adaptive mechanism makes the existing methods have limited effect on high-dimensional, nonlinear time series data such as chillers, resulting in insufficient accuracy of fault prediction SUMMARY

[0003] Therefore, it is necessary to provide a chiller fault prediction and diagnosis method based on a GRU-SA model to solve at least one of the above technical problems.

[0004] To achieve the above object, a chiller fault prediction and diagnosis method based on a GRU-SA model, the method comprising the following steps:

[0005] Step S1: Obtain a multi-modal chiller dataset; perform data preprocessing on the multi-modal chiller dataset to generate a multi-modal chiller preprocessed dataset;

[0006] Step S2: Perform physical knowledge virtual feature enhancement on the multi-modal chiller preprocessed dataset to generate a multi-modal chiller virtual feature set; perform dynamic adversarial data sample construction on the multi-modal chiller virtual feature set to generate multi-modal chiller model input data;

[0007] Step S3: Input the multi-modal chiller model input data into a preset GRU convolution model, and perform feature correlation matrix calculation on the self-attention layer to generate a GRU-SA chiller preliminary model; perform self-adaptive parameter adjustment on the GRU-SA chiller preliminary model to generate a GRU-SA chiller fault judgment model;

[0008] Step S4: Cloud-deploy the GRU-SA chiller fault judgment model and perform edge-cloud collaborative reasoning to generate GRU-SA chiller fault judgment data; draw a contribution thermodynamic diagram from the GRU-SA chiller model output data, thereby completing the chiller fault prediction and diagnosis based on the GRU-SA model.

[0009] The application has the beneficial effects that, by acquiring and preprocessing the multi-modal water chiller unit dataset, the consistency and accuracy of the data are ensured, so that the subsequent analysis process can be based on high-quality data. The data preprocessing not only eliminates noise, but also converts the data into a standardized format, providing reliable input for feature enhancement and analysis. Then, the physical knowledge virtual feature enhancement technology is adopted to generate a multi-modal water chiller unit virtual feature set by fusing the physical characteristics of unit operation. This process provides more rich feature information for the model, enabling the model to capture deeper unit operation rules, especially in complex environments, effectively extracting potential patterns related to faults. The dynamic adversarial data sample construction method is used to extend and optimize the multi-modal water chiller unit virtual feature set. Through the generation of adversarial samples, the model can better cope with different disturbance factors, improving the ability to identify abnormal states and potential faults. This method enhances the robustness of the model, ensuring its stability and accuracy in actual operation. By inputting the constructed input data into the preset GRU convolutional model and combining the self-attention mechanism (SA) to calculate the feature correlation matrix, a GRU-SA water chiller unit preliminary model is generated. This model can further optimize feature weights through adaptive parameter adjustment, forming a fault judgment model with high precision. The trained GRU-SA water chiller unit fault judgment model is deployed in the cloud, and combined with edge computing for collaborative reasoning, further improving data processing efficiency and real-time performance. The collaborative reasoning of the cloud and the edge enables the model to achieve efficient data transmission and processing between different computing nodes, thereby quickly responding to fault prediction needs in actual operation. In addition, by drawing a contribution heat map, the output results of the model are visualized, intuitively showing the contribution of each feature to fault diagnosis, which helps decision-makers quickly locate the problem source. Therefore, through multi-modal data processing, physical knowledge enhancement, dynamic adversarial data sample generation, GRU-SA model optimization, and cloud-edge collaborative reasoning, the application solves the problems of insufficient accuracy and poor robustness of traditional water chiller unit fault diagnosis models, improving the precision and real-time performance of fault prediction.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: acquiring a multi-modal water chiller unit dataset, wherein the multi-modal water chiller unit dataset includes unit vibration signals, unit temperature signals, and unit pressure data;

[0012] Step S12: performing wavelet packet transform denoising on the unit vibration signals to generate unit wavelet transform denoising data; and performing data standardization processing on the unit wavelet transform denoising data to generate unit vibration standardized data;

[0013] Step S13: Time series drift processing is performed on the unit temperature signal and unit pressure data to generate unit temperature standardized data and unit pressure standardized data;

[0014] Step S14: Time window synchronization is performed on the unit vibration standardized data, unit temperature standardized data and unit pressure standardized data to generate a multi-modal cold water unit pre-processing data set.

[0015] The multi-modal cold water unit data set obtained by the present application includes vibration signals, temperature signals and pressure data, which reflect the operating state and performance changes of the unit and are important basis for fault diagnosis and performance evaluation. Since various signals are easily disturbed by noise during sensor acquisition, wavelet packet transform denoising processing is performed on the unit vibration signal. Wavelet packet transform can effectively remove high-frequency noise in the signal and retain effective information of the signal, thereby generating clear and representative unit vibration signal data, which provides a more accurate data basis for subsequent analysis. In addition, the denoised vibration signal data is standardized to effectively eliminate the differences between different dimensions and different amplitudes, so that the data is in a unified scale range, and the comparability of different features in model training is improved. In the processing of unit temperature signal and pressure data, time series drift processing eliminates the drift phenomenon caused by temperature and pressure measurement errors or environmental factors, ensuring the stability and consistency of these data. Standardizing these two types of data further eliminates the dimensional differences between different sensors, so that temperature and pressure data can be effectively compared and integrated with vibration signal data, thereby ensuring the uniformity and reliability of the data processing process. Finally, time window synchronization is adopted to unify and align the unit vibration standardized data, unit temperature standardized data and unit pressure standardized data to form a multi-modal cold water unit pre-processing data set. Time window synchronization ensures the consistency of different signals in time, so that subsequent time series-based analysis methods can effectively extract features and train models. In addition, through synchronous processing, the inconsistency of data caused by time offset is avoided, and the prediction accuracy and reliability of the model are improved.

[0016] Preferably, step S13 comprises the following steps:

[0017] Step S131: Perform two-dimensional linear projection on the unit temperature signal and unit pressure data to generate unit temperature-pressure two-dimensional projection; perform linear interpolation anomaly value correction on the unit temperature-pressure two-dimensional projection to generate unit temperature-pressure anomaly value processing data;

[0018] Step S132: Perform white correlation trend analysis on the unit temperature-pressure anomaly value processing data to generate unit temperature-pressure white correlation trend data;

[0019] Step S133: Time series drift processing is performed on the unit temperature-pressure white correlation trend data to generate unit temperature-pressure time series drift data; second-order difference standardization processing is performed on the unit temperature-pressure time series drift data to generate unit temperature standardized data and unit pressure standardized data.

[0020] The application converts the unit temperature signal and the unit pressure data into a unit temperature-pressure two-dimensional projection by two-dimensional linear projection, which effectively maps the original data from a high-dimensional space to a low-dimensional space, simplifies the complexity of the data, and preserves the potential correlation between temperature and pressure. The data after two-dimensional linear projection can more intuitively reveal the relationship between temperature and pressure, providing a basis for subsequent data analysis. On this basis, linear interpolation anomaly value correction is performed on the unit temperature-pressure two-dimensional projection, which can eliminate abnormal values caused by sensor errors or external disturbances, thereby generating more stable and reliable data. This step significantly reduces the bias in the data and improves the data quality. White correlation trend analysis is performed on the unit temperature-pressure abnormal value processing data, which can effectively remove random fluctuations in temperature and pressure data by extracting the trend component of the data, so that the long-term trend in the data is clearly presented. White correlation trend analysis can eliminate periodicity and non-stationarity in the data, making subsequent data processing more accurate and providing more stable input for model prediction. In addition, white correlation trend analysis helps to reveal the change rule of unit temperature and pressure in a long time series, thereby identifying key change factors related to the equipment state. Time series drift processing further eliminates the drift phenomenon of temperature data over time, thereby ensuring the consistency and stability of the data in the entire time series. This processing step effectively removes the drift of the temperature signal caused by external environmental changes or equipment failure, so that the data can accurately reflect the performance of the unit under normal operating conditions. Then, the temperature data is subjected to second-order difference standardization processing, which can effectively eliminate the trend changes and periodic fluctuations in the temperature signal, convert it into standardized difference data, unify the dimension of the data, and enable the model to better learn the subtle changes in the temperature signal. In addition, this difference standardization processing also improves the stationarity of the data, providing more accurate prediction input for the model.

[0021] Preferably, the physical knowledge virtual feature enhancement on the multi-modal water chiller pre-processing data set comprises the following steps:

[0022] According to the unit vibration standardized data, a refrigeration cycle physical process correlation analysis is performed to generate unit cold cycle vibration analysis data; an indirect refrigeration cycle efficiency influence analysis is performed on the unit cold cycle vibration analysis data to generate a unit vibration theoretical energy efficiency ratio;

[0023] According to the unit temperature standardized data, temperature fluctuation mode identification potential equipment failure is generated, and unit temperature fluctuation mode data is generated; the unit temperature fluctuation mode data is subjected to least square regression physical constraint dimension reduction, and unit temperature supercooling degree deviation is generated;

[0024] According to the unit pressure standardized data, thermodynamic state analysis is performed, and unit pressure thermodynamic state curve is generated; the unit pressure thermodynamic state curve is subjected to abnormal value peak value analysis, and unit pressure thermodynamic abnormal peak value data is generated; temperature-pressure correlation analysis is performed on the unit pressure thermodynamic abnormal peak value data and the unit temperature supercooling degree deviation, and the unit temperature-pressure supercooling degree deviation is generated.

[0025] According to the unit temperature-pressure supercooling degree deviation and the unit vibration theoretical energy efficiency ratio, One-hot coding is performed, and physical prior processing is performed, and a multi-modal water chiller virtual feature set is generated.

[0026] The present application can effectively extract vibration characteristics closely related to the performance of the refrigeration cycle by correlating the standardized vibration data of the unit with the physical process of the refrigeration cycle, thereby generating unit cold cycle vibration analysis data. This analysis process reveals the internal relationship between unit vibration and its cold cycle process, providing a key basis for further performance evaluation and fault prediction. Then, through the indirect refrigeration cycle efficiency impact analysis of the unit cold cycle vibration analysis data, the unit vibration theoretical energy efficiency ratio is generated. This data reflects the energy efficiency performance of the unit in the actual operation process, and provides a quantitative index for judging the working efficiency of the unit. In the process of temperature and pressure data processing, first, the temperature fluctuation pattern recognition is performed on the unit temperature standardized data, which can accurately capture the potential equipment fault signals and generate unit temperature fluctuation pattern data. Through this pattern recognition, the system can discover abnormal trends in temperature changes in advance, and then identify the fault occurrence. In addition, after the unit temperature fluctuation pattern data is processed by the least square regression physical constraint dimension reduction, the unit temperature supercooling degree deviation is generated. This step effectively reduces the data dimension, eliminates the interference of noise, and makes the temperature data more accurately reflect the deviation of the temperature supercooling degree, thereby providing more accurate data input for unit fault warning. For the unit pressure standardized data, first, the thermodynamic state analysis is performed to generate the unit pressure thermodynamic state curve. This analysis reveals the thermodynamic characteristics of the unit under different working conditions and provides a basis for understanding the running efficiency of the unit. After the abnormal value peak analysis of the pressure thermodynamic state curve, the unit pressure thermodynamic abnormal peak data is generated. This data can highlight the extreme values or abnormal changes in the pressure data, providing an effective clue for fault detection. Further, the temperature-pressure correlation analysis is performed on the unit pressure thermodynamic abnormal peak data and the unit temperature supercooling degree deviation to generate the unit temperature-pressure supercooling degree deviation. This analysis helps to reveal the mutual influence between the unit temperature and pressure, especially the dynamic change pattern under abnormal conditions. Finally, the One-hot coding is performed on the unit temperature-pressure supercooling degree deviation and the unit vibration theoretical energy efficiency ratio, and the physical prior processing is combined to generate a multi-modal cold water chiller virtual feature set. This feature set integrates the key information of various sensor data, which can provide rich and efficient feature input for subsequent model training and fault prediction. Through the coding and physical processing of various features, it ensures that the model can better learn the potential laws in the operation process of the unit, further improving the identification accuracy of the system for faults.

[0027] Preferably, step S2 comprises the following steps:

[0028] Step S21: performing physical knowledge virtual feature enhancement on the multi-modal cold water chiller pre-processing data set to generate a multi-modal cold water chiller virtual feature set;

[0029] Step S22: Random noise disturbance is performed on the multi-modal virtual feature set of the water chiller to generate multi-modal random noise disturbance data; gradient descent is performed on the multi-modal random noise disturbance data to generate multi-modal gradient descent data;

[0030] Step S23: Data scrambling is performed on the multi-modal gradient descent data to generate multi-modal gradient descent scrambled data; time series disturbance analysis is performed on the multi-modal gradient descent scrambled data, thereby completing dynamic adversarial data sample construction to generate multi-modal water chiller model input data.

[0031] The present application enhances the physical feature expression of the unit data by using the prior knowledge in the physical field and systematically enhancing the physical feature expression of the unit data through physical knowledge virtual feature enhancement on the multi-modal water chiller pretreatment data set, and generates a multi-modal water chiller virtual feature set. This process effectively combines the physical characteristics of the unit operation with the sensor data, providing more rich feature information for subsequent model learning. These virtual feature sets can more accurately reflect the actual operation state of the unit, improving the accuracy of subsequent analysis. First, random noise disturbance is performed on the multi-modal water chiller virtual feature set to generate multi-modal random noise disturbance data. This step artificially introduces noise to simulate the interference factors encountered in actual operation, improving the adaptability of the data set to noise and abnormal changes. Noise disturbance not only enhances the diversity of data, but also enables the model to better cope with uncertainties in the real environment during the training process. Then, gradient descent is performed on these disturbance data to generate multi-modal gradient descent data. Gradient descent optimizes the parameters of the model, so that the data set continuously approaches the optimal solution during the training process, thereby improving the feature representation of the data and further improving the performance and efficiency of the model. Data scrambling is performed on the multi-modal gradient descent data to generate multi-modal gradient descent scrambled data. Data scrambling further enhances the complexity of the data set, enabling the model to handle more changes when processing these data, avoiding overfitting and improving the generalization ability. Subsequently, time series disturbance analysis is performed to complete dynamic adversarial data sample construction. Time series disturbance analysis not only considers the changes in time factors, but also simulates abnormal fluctuations and trend changes in data over time. Through these steps, multi-modal water chiller model input data is generated, fully considering various disturbances and abnormal situations in reality, further enhancing the diversity and robustness of the data set.

[0032] Preferably, step S3 comprises the following steps:

[0033] Step S31: input the multi-modal water chiller model input data into the preset GRU convolution model, the model includes a GRU layer with 64 hidden units, and the feature correlation matrix calculation of the self-attention layer is performed, the hidden state of the GRU layer and the self-attention layer is output, and a GRU-SA water chiller preliminary model is constructed, wherein the calculation formula of the feature correlation matrix calculation includes,

[0034] ;

[0035] wherein, , , are query, key, and value matrices respectively, is the feature dimension;

[0036] Step S32: expand the cross-entropy loss function of the GRU-SA water chiller preliminary model to generate a GRU-SA energy regularization loss function;

[0037] Step S33: perform adaptive parameter adjustment on the GRU-SA water chiller preliminary model and the GRU-SA energy regularization loss function to generate a GRU-SA water chiller fault judgment model, wherein the weight of the GRU-SA model is adjusted by the Adam optimizer, and the learning rate is set to 0.001.

[0038] The application inputs the multi-modal water chiller model input data into the preset GRU convolution model as input, and combines the self-attention layer to perform feature correlation matrix calculation, thereby generating a GRU-SA water chiller preliminary model. In this process, the introduction of the GRU convolution model enables the long-term and short-term dependence relationship of the time series data to be effectively captured, thereby enhancing the hidden dynamic characteristics in the data. Meanwhile, the self-attention layer can automatically capture key features in the input of multi-modal data by calculating the correlation matrix between features, thereby reducing the interference of redundant information. In this way, the model can effectively extract the most representative information from complex multi-modal data, thereby further enhancing the accuracy and efficiency of water chiller fault prediction. The GRU-SA water chiller preliminary model is subjected to cross-entropy loss function expansion, and a GRU-SA energy regularization loss function is generated. The cross-entropy loss function is usually used for classification problems and can quantify the gap between the predicted value and the true label. The expansion of the cross-entropy loss function enables it to adapt to more types of data features, thereby further optimizing the loss function design of the model. In addition, after the introduction of the energy regularization loss function, the model complexity can be effectively controlled, thereby avoiding overfitting of the model during the training process. As a constraint term, the energy regularization can effectively limit the model parameters, thereby ensuring that the model has good generalization ability when processing complex data. The GRU-SA water chiller preliminary model and the GRU-SA energy regularization loss function are subjected to adaptive parameter adjustment, thereby generating a GRU-SA water chiller fault judgment model. The adaptive parameter adjustment enables the model to automatically optimize the model parameters during the training process according to the actual performance of the data, thereby improving the efficiency and accuracy of the training. This step further enhances the self-adjusting ability of the model, thereby enabling it to maintain high prediction accuracy under different operating conditions and data distributions. Through this optimization process, the finally generated GRU-SA water chiller fault judgment model can still efficiently and accurately perform fault diagnosis in complex and variable data environments, thereby having strong robustness and adaptability.

[0039] Preferably, the self-attention layer performs feature correlation matrix calculation, including the following steps:

[0040] Obtain the water chiller refrigeration drawing and real-time refrigeration working condition parameters;

[0041] Perform three-dimensional modeling on the water chiller refrigeration drawing to generate a water chiller refrigeration cycle physical structure, and perform refrigeration system topology node construction on the water chiller refrigeration cycle physical structure to generate a water chiller topology node;

[0042] According to the water chiller topology node and the real-time refrigeration working condition parameters, real-time working condition coding is performed to generate dynamic unit working condition driving coding data, and sparse matrix correlation matrix synthesis is performed on the dynamic unit working condition driving coding data to generate a water chiller sparse correlation matrix;

[0043] The physical constraint self-attention process calculation is performed on the sparse association matrix of the water chiller unit, and irrelevant noise mask processing is performed, so as to complete the feature association matrix calculation of the attention layer.

[0044] The present application provides sufficient physical and operating background for subsequent data processing by obtaining the water chiller unit refrigeration drawing and real-time refrigeration working condition parameters. By performing three-dimensional modeling on the water chiller unit refrigeration drawing, the refrigeration cycle physical structure of the unit can be accurately constructed, which provides a basic framework for subsequent topological node construction. By constructing the refrigeration system topological node of the water chiller unit refrigeration cycle physical structure, the connection relationship between each device in the system and its function can be clearly displayed, which provides important structural information for further data analysis and fault diagnosis. According to the water chiller unit topological node and the real-time refrigeration working condition parameters, the dynamic unit working condition driving coding data is generated. In this process, the introduction of real-time working condition parameters enables the model to dynamically adjust according to the actual operation and realize real-time response. The generated dynamic unit working condition driving coding data not only contains the current working state of the device, but also can reflect the working load and energy efficiency of each part under different working conditions, thereby providing accurate input data for further fault prediction and performance optimization. By synthesizing the sparse matrix association matrix of the dynamic unit working condition driving coding data, the water chiller unit sparse association matrix is generated. The introduction of the sparse matrix can effectively reduce the complexity of data storage and calculation, while retaining the key feature information, improving the efficiency of the model in processing large-scale data. The water chiller unit sparse association matrix quantifies the association relationship between each data, providing an effective data structure for subsequent feature calculation. The physical constraint self-attention process calculation is performed on the sparse association matrix of the water chiller unit, and irrelevant noise mask processing is performed, so as to complete the feature association matrix calculation of the attention layer. The physical constraint self-attention process calculation considers the physical characteristics of the water chiller unit to ensure that the model can identify features closely related to unit failure and performance from the data. Through irrelevant noise mask processing, redundant information and noise interference can be effectively filtered out, so that the finally generated feature association matrix can focus more on key features, improving the accuracy and stability of the model under complex working conditions.

[0045] Preferably, step S33 comprises the following steps:

[0046] Step S331: associating the GRU-SA water chiller unit preliminary model and the GRU-SA energy regularization loss function in the time dimension to generate GRU-SA water chiller unit time dimension data; associating the GRU-SA water chiller unit preliminary model and the GRU-SA energy regularization loss function in the feature dimension to generate GRU-SA water chiller unit feature dimension data;

[0047] Step S332: Multi-level association and aggregation of the GRU-SA chiller time dimension data and the GRU-SA chiller feature dimension data to generate GRU-SA chiller time-feature dimension data;

[0048] Step S333: Prior injection of fault type to the GRU-SA chiller time-feature dimension data to generate chiller fault type data; time sequence fault cause guarantee is performed on the chiller fault type data, and adaptive parameter adjustment of the refrigerant evaporator and the compressor suction pipe is performed to generate a GRU-SA chiller fault judgment model.

[0049] The present application generates GRU-SA chiller time dimension data by associating the GRU-SA chiller preliminary model and the GRU-SA energy regularization loss function in time dimension. The association processing in time dimension enables the model to capture the dynamic changes of the chiller at different time nodes, identify the potential trends and abnormal changes in the time series. This processing provides a key information basis for subsequent fault diagnosis, ensuring that the model can analyze the changes of the equipment running state from the perspective of time series. Then, the GRU-SA chiller preliminary model and the GRU-SA energy regularization loss function are associated in feature dimension to generate GRU-SA chiller feature dimension data. The association in feature dimension enables the model to comprehensively consider the mutual influence between multiple sensor data and physical characteristics, improving the model's ability to analyze the running state of the chiller from multiple dimensions. The GRU-SA chiller time dimension data and the GRU-SA chiller feature dimension data are associated and aggregated in multiple levels to generate GRU-SA chiller time-feature dimension data. Through multi-level association and aggregation, the model not only can capture the individual information in time and feature dimensions, but also can identify the interaction and internal relationship between them, thereby improving the model's processing capability for complex multi-dimensional data. This step effectively integrates time series features and various physical information of equipment running state, so that the finally generated time-feature dimension data can fully reflect the running condition of the chiller. By prior injection of fault type to the GRU-SA chiller time-feature dimension data, chiller fault type data is generated. This process combines the prior knowledge of fault type, providing the model with clear fault category information, thereby guiding the model to more accurately identify different types of faults in the fault diagnosis process. Then, time sequence fault cause guarantee is performed, and adaptive parameter adjustment of the refrigerant evaporator and the compressor suction pipe is performed, thereby further optimizing the fault diagnosis model. Time sequence fault cause guarantee ensures that the model can follow the physical cause and effect relationship when predicting faults, effectively eliminating unreasonable fault hypotheses, while adaptive parameter adjustment enables the model to flexibly adjust according to the actual operating environment, improving the model's robustness and accuracy under varying operating conditions.

[0050] Preferably, step S4 comprises the following steps:

[0051] Step S41: cloud deployment of the GRU-SA water chiller unit fault judgment model, generating a GRU-SA water chiller unit fault cloud judgment model; edge-cloud collaborative reasoning of the GRU-SA water chiller unit fault cloud judgment model, generating GRU-SA water chiller unit fault judgment data;

[0052] Step S42: drawing a contribution heat map for the GRU-SA water chiller unit model output data, generating a GRU-SA water chiller unit fault heat map;

[0053] Step S43: based on the GRU-SA water chiller unit fault heat map, water chiller unit fault prediction diagnosis evaluation is carried out, thereby completing the water chiller unit fault prediction diagnosis based on the GRU-SA model.

[0054] The application generates a GRU-SA water chiller fault cloud judgment model by cloud deployment of the GRU-SA water chiller fault judgment model. Cloud deployment provides powerful computing resources and storage capabilities for the model, enabling it to process large amounts of real-time data and perform deep learning analysis. This process ensures that the model can quickly respond to different equipment and working condition data in a distributed environment, and make timely fault judgments on the system. Subsequently, the GRU-SA water chiller fault judgment data is generated by edge-cloud collaborative reasoning of the GRU-SA water chiller fault cloud judgment model. Edge-cloud collaborative reasoning can distribute part of the computing tasks to edge devices for fast processing, reducing the cloud computing burden and enabling faster response to real-time data on the device side. At the same time, the cloud performs deep analysis and decision-making on the data, ensuring the accuracy and reliability of the overall reasoning results of the system. This collaborative reasoning architecture greatly improves the real-time performance and response speed of the fault diagnosis system in complex environments, ensuring that the water chiller can obtain accurate fault judgments in a timely manner. The GRU-SA water chiller fault heat map is generated by drawing a contribution heat map of the GRU-SA water chiller model output data. The generation of the heat map provides a visual fault analysis method, highlighting the fault contribution of key parts to help engineers quickly locate problem areas. The contribution heat map not only reflects the health status of each component of the water chiller, but also visually displays the contribution of different components to the overall system failure, making fault diagnosis more accurate and operable. Finally, based on the GRU-SA water chiller fault heat map, the water chiller fault prediction diagnosis evaluation is carried out, thereby completing the water chiller fault prediction diagnosis based on the GRU-SA model. Through the combination of fault heat map and prediction diagnosis evaluation, the system can predict the failure of the water chiller in advance, providing important decision support for maintenance personnel, thereby intervening in advance and reducing the probability of failure.

[0055] Preferably, step S43 comprises the following steps:

[0056] Step S431: obtaining historical water chiller operation data;

[0057] Step S432: extracting key faults from the GRU-SA water chiller fault heat map to generate GRU-SA water chiller fault key data;

[0058] Step S433: comparing the GRU-SA water chiller fault key data with the historical water chiller operation data for fault diagnosis to generate GRU-SA water chiller fault diagnosis comparison data; performing fault judgment evaluation on the GRU-SA water chiller fault diagnosis comparison data to generate GRU-SA water chiller evaluation data;

[0059] Step S434: Based on the GRU-SA water chiller unit evaluation data, the fault judgment report is constructed, and the GRU-SA water chiller unit evaluation report is generated, so as to complete the fault prediction diagnosis of the water chiller unit based on the GRU-SA model.

[0060] The present application provides rich background information for subsequent fault analysis and evaluation by obtaining historical water chiller unit operation data. Historical data can not only reflect the running state of the equipment at different time periods, but also provide long-term trends and change rules for the fault diagnosis model, and provide baseline data for the system to judge whether the current state is abnormal. Then, the GRU-SA water chiller unit fault heat map is extracted to generate the GRU-SA water chiller unit fault key data. In this process, the fault heat map helps to identify and calibrate the most critical fault area in the system, and extracts the key data related to unit performance loss, abnormal fluctuation, system failure, etc. This step can effectively narrow the analysis range and ensure the focus and accuracy of subsequent fault judgment. The GRU-SA water chiller unit fault key data and the historical water chiller unit operation data are compared for fault diagnosis to generate the GRU-SA water chiller unit fault diagnosis comparison data. By comparing historical data and current fault data, the system can identify the similarities and differences between the current fault and historical abnormalities, so as to judge whether it is a similar fault or a system state change. This comparative analysis not only helps to identify the fault type, but also provides more actual working condition data for the model, improving the accuracy and reliability of fault diagnosis. Subsequently, the GRU-SA water chiller unit fault diagnosis comparison data is evaluated for fault judgment to generate the GRU-SA water chiller unit evaluation data. The evaluation data based on multiple dimensions of comparison and analysis can provide detailed fault diagnosis results to help decision-makers respond quickly, thereby effectively preventing the expansion and spread of equipment failure. Finally, the GRU-SA water chiller unit evaluation data is used to construct the fault judgment report to generate the GRU-SA water chiller unit evaluation report. The report not only provides comprehensive fault diagnosis information, but also includes detailed evaluation results, recommended maintenance solutions, equipment state prediction, etc., providing scientific basis and decision support for equipment management and maintenance. Through this process, the system can provide effective support for fault early warning, maintenance plan arrangement and optimization control strategy, which helps to improve the reliability of the water chiller unit, reduce sudden failures, and prolong the service life of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A step flowchart of a water chiller unit fault prediction diagnosis method based on a GRU-SA model;

[0062] Figure 2 A detailed implementation step flowchart of step S2 in the method; Figure 1 A detailed implementation step flowchart of step S2 in the method;

[0063] Figure 3 To Figure 1 Detailed implementation step flow diagram of step S3 in the method;

[0064] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0065] The technical method of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0066] In addition, the accompanying drawings are only schematic illustrations of the present application and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and thus repeated description thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities, which do not necessarily have to correspond to physically or logically independent entities. The functional entities can be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0067] It should be understood that although the terms "first", "second" and the like can be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of the exemplary embodiments, a first element can be referred to as a second element, and similarly a second element can be referred to as a first element. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0068] To achieve the above-mentioned purpose, please refer to Figures 1 to 3 A GRU-SA model-based chiller fault prediction and diagnosis method, the method comprising the following steps:

[0069] Step S1: Obtain a multi-modal chiller data set; perform data preprocessing on the multi-modal chiller data set to generate a multi-modal chiller preprocessed data set;

[0070] Step S2: Perform physical knowledge virtual feature enhancement on the multi-modal chiller preprocessed data set to generate a multi-modal chiller virtual feature set; construct dynamic adversarial data samples on the multi-modal chiller virtual feature set to generate multi-modal chiller model input data;

[0071] Step S3: input the multi-modal water chiller unit model input data as input to the preset GRU convolution model, and perform feature correlation matrix calculation on the self-attention layer to generate a GRU-SA water chiller unit preliminary model; perform adaptive parameter adjustment on the GRU-SA water chiller unit preliminary model to generate a GRU-SA water chiller unit fault judgment model;

[0072] Step S4: cloud deployment of the GRU-SA water chiller unit fault judgment model is performed, and edge-cloud collaborative reasoning is performed to generate GRU-SA water chiller unit fault judgment data; contribution thermodynamic diagram is drawn for the GRU-SA water chiller unit model output data, thereby completing the water chiller unit fault prediction diagnosis based on the GRU-SA model.

[0073] The present application has the beneficial effect that by acquiring and preprocessing the multi-modal water chiller dataset, the consistency and accuracy of the data are ensured, so that the subsequent analysis process can be based on high-quality data. Data preprocessing not only eliminates noise, but also converts data into a standardized format, providing reliable input for feature enhancement and analysis. Then, the physical knowledge virtual feature enhancement technology is adopted to generate a multi-modal water chiller virtual feature set by fusing the physical characteristics of the unit operation. This process provides more rich feature information for the model, enabling the model to capture deeper unit operation rules, especially in complex environments, effectively extracting potential patterns related to faults. The dynamic adversarial data sample construction method is used to extend and optimize the multi-modal water chiller virtual feature set. Through the generation of adversarial samples, the model can better cope with different disturbance factors, improving the ability to identify abnormal states and potential faults. This method enhances the robustness of the model, ensuring its stability and accuracy in actual operation. By inputting the constructed input data into the preset GRU convolutional model and combining the self-attention mechanism (SA) to calculate the feature correlation matrix, a GRU-SA water chiller preliminary model is generated. This model can further optimize feature weights through adaptive parameter adjustment, forming a fault judgment model with high precision. The trained GRU-SA water chiller fault judgment model is deployed in the cloud, and combined with edge computing for collaborative reasoning, further improving data processing efficiency and real-time performance. The collaborative reasoning of the cloud and the edge enables the model to achieve efficient data transmission and processing between different computing nodes, thereby quickly responding to fault prediction needs in actual operation. In addition, by drawing a contribution heat map, the output results of the model are visualized, intuitively showing the contribution of each feature to fault diagnosis, which helps decision-makers quickly locate the problem source. Therefore, through multi-modal data processing, physical knowledge enhancement, dynamic adversarial data sample generation, GRU-SA model optimization, and cloud-edge collaborative reasoning, the present application solves the problems of insufficient accuracy and poor robustness of traditional water chiller fault diagnosis models, improving the precision and real-time performance of fault prediction.

[0074] In the embodiment of the present application, as shown in the reference Figure 1 The GRU-SA model-based water chiller fault prediction and diagnosis method includes the following steps:

[0075] Step S1: Acquire a multi-modal water chiller dataset; perform data preprocessing on the multi-modal water chiller dataset to generate a multi-modal water chiller preprocessed dataset;

[0076] In the embodiments of the present application, a multi-modal water chiller data set is obtained, which contains data from multiple sensors or measuring devices, such as vibration signals, temperature signals, pressure data, etc. of the chiller. These data often have different acquisition frequencies, noise characteristics and different physical units, and must be effectively processed to provide effective information for subsequent analysis. In the data preprocessing stage, multiple data cleaning and conversion methods are used to standardize the original data to eliminate inconsistencies and ensure that the data is analyzed on the same basis. Specifically, for vibration signals, wavelet packet transform denoising technology can be used to remove high-frequency noise and low-frequency drift and extract valuable vibration features; for temperature and pressure signals, time series drift correction and standardization can be performed to ensure consistency at different time points and eliminate systematic errors caused by temperature changes, equipment operation, etc. Further, data preprocessing also includes missing value filling, outlier detection and processing, data smoothing, etc. Missing value filling methods can use mean interpolation, linear interpolation or multiple interpolation techniques, and select according to the distribution characteristics and missing mechanism of the data. Outlier detection can combine statistical methods and physical models to improve data reliability by identifying data points deviating from the normal range. Data smoothing methods such as sliding window averaging can reduce random fluctuations in the data and improve signal stability.

[0077] Step S2: performing physical knowledge virtual feature enhancement on the multi-modal water chiller preprocessed data set to generate a multi-modal water chiller virtual feature set; constructing dynamic adversarial data samples on the multi-modal water chiller virtual feature set to generate multi-modal water chiller model input data;

[0078] In the embodiments of the present application, by introducing physical knowledge of unit working principle, thermodynamic model and system state, the preprocessed data can be supplemented to generate a multi-modal virtual feature set of water chillers. For example, for temperature, pressure and vibration signals of the unit, physical formulas can be embedded in the data based on thermodynamic equilibrium, refrigeration cycle efficiency and mechanical vibration model, etc. Through mathematical derivation and simulation, these original data are converted into features with higher recognition and explanatory power. This enhancement process not only improves the physical relevance of the data, but also ensures the physical reasonableness of the model input features, so that the model can understand and capture complex patterns in the data at a deeper level. The process of constructing dynamic adversarial data samples aims to enhance the robustness and generalization ability of the model by adversarial data augmentation of the generated virtual feature set. Dynamic adversarial sample construction uses perturbation technology, usually combined with gradient generated adversarial samples (GAN) or optimization-based adversarial training methods, to perturb the original data and generate challenging samples. These adversarial samples can explore the data space, so that the model can not only learn the features of the regular data in the training process, but also cope with noise, bias and other abnormal situations

[0079] Step S3: input the multi-modal water chiller model input data into the preset GRU convolution model, and perform feature correlation matrix calculation on the self-attention layer to generate a GRU-SA water chiller preliminary model; and perform self-adaptive parameter adjustment on the GRU-SA water chiller preliminary model to generate a GRU-SA water chiller fault judgment model;

[0080] In the embodiment of the present application, the multi-modal chiller unit model input data is input to the preset GRU (Gated Recurrent Unit) convolution model. The core of this process is to input the time series features into the GRU model, and the GRU effectively captures the long-term dependencies in the time series data through its internal gating mechanism (such as the reset gate and the update gate). Since the data of the chiller unit has significant time series characteristics, such as the trend of changes in temperature, pressure and vibration signals, etc., the GRU model can better understand the dynamic characteristics and trend changes in the data through its recursive structure, especially when dealing with nonlinear, non-stationary time series data. The convolution operation further enhances the model's expression ability in spatial features, effectively extracting local patterns and important features from the input data, so that the model can not only capture temporal dependencies, but also process different types of data features. Subsequently, a self-attention mechanism is introduced to calculate the feature correlation matrix. The self-attention mechanism can weight and aggregate the relationships between different time steps of the input data, calculating the importance of each feature in the entire sequence, thereby generating a feature correlation matrix. This mechanism helps to strengthen the model's focus on key features and improve its understanding of the relationships between different dimensional features, further improving prediction accuracy. After the preliminary model is generated, adaptive parameter adjustment of the GRU-SA chiller unit preliminary model is another important step. Adaptive parameter adjustment dynamically adjusts the weights and hyperparameters of the model through optimization algorithms (such as gradient descent, Adam optimizer, etc.) to minimize the loss function and improve the model's generalization ability. This process can automatically adjust the learning rate, network structure and other hyperparameters of the model according to the distribution of the data and the task requirements, thereby continuously optimizing the model's performance during training, avoiding overfitting and improving its prediction ability on unknown data. After adaptive adjustment, the final GRU-SA chiller unit fault judgment model can more accurately identify the fault patterns of the chiller unit in actual application, providing efficient fault prediction and diagnosis capabilities.

[0081] Step S4: The GRU-SA chiller unit fault judgment model is deployed in the cloud, and edge-cloud collaborative reasoning is performed to generate GRU-SA chiller unit fault judgment data; the contribution thermodynamic map of the GRU-SA chiller unit model output data is drawn, thereby completing the chiller unit fault prediction and diagnosis based on the GRU-SA model.

[0082] In the embodiment of the present application, the GRU-SA water chiller fault judgment model is deployed in the cloud. This process uploads the trained GRU-SA model to the cloud platform for real-time calculation and inference. Cloud deployment can fully utilize the advantages of cloud computing resources, such as powerful computing and storage capabilities, to achieve large-scale data processing and model inference. After deployment, the GRU-SA model can receive data input from the water chiller in real time and perform fault judgment in the cloud. At the same time, through the edge-cloud collaborative inference mechanism, part of the computing tasks are allocated to the edge devices for local inference, reducing the data transmission delay and the computing pressure of the cloud, thereby improving the response speed and efficiency of the system. Edge devices usually have lower computing resources and are mainly used for preliminary data preprocessing and simple inference, while the cloud is responsible for complex computing tasks and large-scale model inference. The collaborative work of the two can achieve efficient and low-latency fault prediction. The generated GRU-SA water chiller fault judgment data contains the fault types and their severity that occur during the operation of the water chiller. Based on these output data, the next step is to draw a contribution heat map for the GRU-SA water chiller model output data. The contribution heat map is generated by analyzing the weights of each feature in the model output data, showing the influence of each feature on the final fault judgment. The heat map can visually display the contribution of different features (such as temperature, pressure, vibration, etc.) to the fault prediction result, helping engineers understand the root cause of the fault and providing guidance for subsequent fault prevention and maintenance. In this way, the heat map not only improves the interpretability of the model, but also provides data support for the fault diagnosis and maintenance decision of the water chiller. Finally, by combining the GRU-SA water chiller fault prediction diagnosis with the heat map, a complete GRU-SA model-based water chiller fault prediction diagnosis system is formed.

[0083] Preferably, step S1 comprises the following steps:

[0084] Step S11: Obtain a multi-modal water chiller dataset, wherein the multi-modal water chiller dataset comprises unit vibration signals, unit temperature signals, and unit pressure data;

[0085] Step S12: Perform wavelet packet transform denoising on the unit vibration signals to generate unit wavelet transform denoising data; and perform data standardization processing on the unit wavelet transform denoising data to generate unit vibration standardized data;

[0086] Step S13: Perform time series drift processing on the unit temperature signals and the unit pressure data to generate unit temperature standardized data and unit pressure standardized data;

[0087] Step S14: synchronizing the unit vibration standardized data, the unit temperature standardized data and the unit pressure standardized data in a time window to generate a multi-modal water chiller preprocessing data set.

[0088] In the embodiment of the present application, a multi-modal water chiller data set is obtained, which contains unit vibration signals, unit temperature signals and unit pressure data, providing basic data for subsequent analysis and modeling. The multi-modal data set means data from different physical quantities, such as vibration, temperature and pressure, which have complementarity and diversity in reflecting the running state of the water chiller and can reveal more system characteristics. The wavelet packet transform denoising processing is performed on the unit vibration signals. The wavelet packet transform is a time-frequency analysis method that can effectively decompose signals and extract features of different frequency bands, thereby removing noise and improving signal quality. The wavelet transform denoising technology decomposes and reconstructs the signal in multiple levels by selecting an appropriate wavelet basis function, filters out high-frequency noise while retaining the key features of the signal. The signal after wavelet packet transform is standardized to ensure the dimension of different characteristic data is uniform and to eliminate the dimensional differences between variables, so that the data can be comparable and consistent in subsequent analysis. This standardization process includes linear transformation of the data to adjust it to a uniform standard range (such as mean value of 0 and variance of 1), thereby enhancing the training effect of the model. The unit temperature signals and unit pressure data need to be processed for time series drift. Time series drift refers to the trend or drift of a signal over time, which can affect signal analysis and model prediction effect. By removing or adjusting this drift trend, the periodic changes and instantaneous fluctuations of the signal can be better captured, thereby improving the accuracy of subsequent processing. The drift-adjusted data is standardized to ensure that the temperature and pressure signals also have a uniform scale and standard, facilitating subsequent multi-modal data fusion and model training. The standardized data of unit vibration, temperature and pressure signals need to be processed in a time window. The purpose of time window synchronization is to ensure that multi-modal signals are aligned at the same time point, facilitating multi-variable analysis and modeling. During the operation of the water chiller, the changes in vibration, temperature and pressure data are interrelated, but these signals usually have different sampling frequencies and time series, so time window synchronization technology is needed to align the multi-modal data at each time point consistently to provide complete, multi-dimensional system state data sets for subsequent analysis.

[0089] Preferably, step S13 comprises the following steps:

[0090] Step S131: performing two-dimensional linear projection on the unit temperature signals and the unit pressure data to generate a unit temperature-pressure two-dimensional projection; performing linear interpolation anomaly correction on the unit temperature-pressure two-dimensional projection to generate unit temperature-pressure anomaly processing data;

[0091] Step S132: white correlation trend analysis is performed on the unit temperature-pressure abnormal value processing data to generate unit temperature-pressure white correlation trend data;

[0092] Step S133: time series drift processing is performed on the unit temperature-pressure white correlation trend data to generate unit temperature-pressure time series drift data; second-order difference standardization processing is performed on the unit temperature-pressure time series drift data to generate unit temperature standardization data and unit pressure standardization data.

[0093] In the embodiment of the present application, the unit temperature signal and the unit pressure data are projected in two dimensions, which is a process of mapping high-dimensional data onto a two-dimensional plane. Through linear projection, the original multi-dimensional data can be compressed into low-dimensional data that is easy to analyze, while the main information of the original data is preserved. The core technology of this step is matrix operation in linear algebra, which can visualize and present the change trend and relationship of each variable in two-dimensional space through weight coefficients, thereby reducing the complexity of data and facilitating subsequent analysis. Next, in the process of linear interpolation abnormal value correction of the unit temperature-pressure two-dimensional projection, linear interpolation technology is used to fill or correct the abnormal values in the data. Abnormal values are caused by noise, sensor failure or collection errors, etc., which will have a negative impact on subsequent data analysis and modeling. Linear interpolation is to calculate the value of the missing data point through the linear relationship between the known data points, which can ensure the integrity of the data while minimizing the distortion of the original data trend. This process ensures the continuity and consistency of the data. White correlation trend analysis is performed on the corrected unit temperature-pressure data. White correlation analysis aims to identify potential noise or trend components in the data. In time series data, there are usually white noise (i.e., random fluctuations of data) and systematic trend components. Through white correlation analysis, these trend components can be effectively separated from the data, thereby enhancing the extraction of effective information in the data. This process helps to more accurately describe the internal relationship between unit temperature and pressure data, remove irrelevant noise, and improve analysis quality. Time series drift processing is performed on the unit temperature-pressure white correlation trend data. Time series drift refers to the systematic change of data over time, which usually affects the stability of data and the accuracy of prediction. Through drift processing, these time-dependent change trends can be removed, making the data more stable and easy to analyze. Then, second-order difference standardization processing is performed on the temperature time series drift data. This process uses the second-order difference method to remove the trend components in the data and emphasizes the fluctuation part of the data, so that the data is more stable and suitable for subsequent statistical modeling or machine learning analysis. Second-order difference standardization processing converts the change rate of data into a unit standard deviation form, ensuring that temperature and pressure data are compared and integrated on the same dimension, improving the consistency and operability of data.

[0094] Preferably, the physical knowledge virtual feature enhancement on the multi-modal water chiller unit preprocessing data set comprises the following steps:

[0095] According to the unit vibration standardized data, a physical process correlation analysis of the refrigeration cycle is performed to generate unit cold cycle vibration analysis data; an indirect refrigeration cycle efficiency influence analysis is performed on the unit cold cycle vibration analysis data to generate a unit vibration theoretical energy efficiency ratio;

[0096] According to the unit temperature standardized data, a temperature fluctuation pattern recognition potential equipment fault is performed to generate unit temperature fluctuation pattern data; a least square regression physical constraint dimension reduction is performed on the unit temperature fluctuation pattern data to generate a unit temperature supercooling degree deviation;

[0097] According to the unit pressure standardized data, a thermodynamic state analysis is performed to generate a unit pressure thermodynamic state curve; an abnormal value peak analysis is performed on the unit pressure thermodynamic state curve to generate unit pressure thermodynamic abnormal peak data; a temperature-pressure correlation analysis is performed on the unit pressure thermodynamic abnormal peak data and the unit temperature supercooling degree deviation to generate a unit temperature-pressure supercooling degree deviation.

[0098] According to the unit temperature-pressure supercooling degree deviation and the unit vibration theoretical energy efficiency ratio, One-hot coding is performed, and physical prior processing is performed to generate a multi-modal water chiller unit virtual feature set.

[0099] In the embodiments of the present application, by performing standardized data correlation analysis on the unit vibration, the relationship between the physical model and the vibration data is applied, and the cold cycle characteristics of the unit can be extracted from the vibration data. This correlation analysis usually relies on physical modeling and signal processing technology, and by establishing the mathematical relationship between the refrigeration cycle and the vibration, the influence of the unit vibration characteristics on the refrigeration efficiency is revealed, so as to generate the unit cold cycle vibration analysis data. This process realizes the quantitative correlation between the vibration data and the refrigeration performance through mathematical modeling. The indirect refrigeration cycle efficiency influence analysis is performed, and the influence of vibration on the efficiency of the water chiller is evaluated by calculating the theoretical energy efficiency ratio of the vibration data. This analysis process relies on thermodynamic and kinetic models, and combines vibration characteristics to derive the theoretical energy efficiency ratio of the unit, so as to obtain the efficiency performance of the unit. This technology can provide effective prediction of the running state of the unit. In terms of unit temperature standardized data, first, temperature fluctuation pattern recognition is performed, and the fluctuation pattern in the temperature data is recognized through a machine learning algorithm. These patterns can reveal potential equipment failures, such as temperature instability, which is a signal of cooling system failure. The identification of temperature fluctuation patterns is completed through time series analysis or cluster analysis, and then the unit temperature fluctuation pattern data is generated. The unit temperature fluctuation data is processed by dimensionality reduction through least squares regression physical constraints to reduce the complexity of the feature space and make the data more concise and easy to analyze. This dimensionality reduction method effectively extracts the temperature supercooling degree deviation of the unit by retaining the main physical characteristics and removing redundant data. This technical method is based on a regression model to map complex temperature data to a low-dimensional space, which is convenient for subsequent evaluation and analysis. In the processing of the unit pressure standardized data, the unit pressure thermodynamic state curve is generated through thermodynamic state analysis. Thermodynamic state analysis usually uses state parameters such as pressure, temperature and volume to generate the working state curve of the equipment according to the thermodynamic equation, so as to reflect the running efficiency and performance of the unit. When there is an abnormality in the pressure data, potential abnormal phenomena can be identified through peak analysis, such as signs of system overload or failure. Through the analysis of abnormal peak data, abnormal conditions in the pressure data can be detected and corrected in a timely manner. By performing temperature-pressure correlation analysis on the unit pressure thermodynamic abnormal peak data and the temperature supercooling degree deviation, the mutual influence between temperature and pressure can be revealed, so as to analyze the influence of the combination of the two on the running state of the unit. This analysis helps to reveal the comprehensive performance of the unit and improve the accuracy of fault prediction. Finally, the temperature-pressure supercooling degree deviation and the vibration energy efficiency ratio are One-hot encoded, and combined with physical prior processing, a multi-modal water chiller virtual feature set is generated. One-hot encoding, as a common feature engineering technique, converts discrete features of different categories into binary vectors, which is used here to standardize multiple features so that they are suitable for machine learning models.

[0100] As an example of the present application, reference is made toFigure 2 As shown, in this example, the step S2 includes:

[0101] Step S21: Perform physical knowledge virtual feature enhancement on the multi-modal water chiller preprocessing data set to generate a multi-modal water chiller virtual feature set;

[0102] Step S22: Perform random noise disturbance on the multi-modal water chiller virtual feature set to generate multi-modal random noise disturbance data; perform gradient descent on the multi-modal random noise disturbance data to generate multi-modal gradient descent data;

[0103] Step S23: Perform data scrambling on the multi-modal gradient descent data to generate multi-modal gradient descent scrambled data; perform time series disturbance analysis on the multi-modal gradient descent scrambled data, thereby completing dynamic adversarial data sample construction to generate multi-modal water chiller model input data.

[0104] In the embodiment of the present application, the multi-modal water chiller preprocessing data set is subjected to physical knowledge virtual feature enhancement. The physical knowledge virtual feature enhancement is based on the physical model and expert knowledge of the water chiller, and utilizes known physical laws and system behavior characteristics to generate virtual features that can supplement the original data set. These virtual features enhance the expression ability of the data set by mapping the working principle of the water chiller, enabling it to cover a wider range of working scenarios and system states, which helps the model to make accurate predictions and diagnoses in actual applications. By performing random noise disturbance on the generated virtual feature set, the diversity of the data is enhanced. Noise disturbance is a common enhancement method that blurs the precise features of the original data by adding random noise within a certain range, enabling the model to adapt to a wider variety of input scenarios during training. This method helps to improve the robustness of the model to data fluctuations and external disturbances, thereby avoiding overfitting and improving prediction performance on unknown data. Subsequently, the multi-modal random noise disturbance data is further optimized through the gradient descent algorithm, which is a widely used optimization method that can effectively minimize the error function, helping the model to better learn the underlying patterns in the data. In this way, the disturbed data is guided towards a state that is more consistent with the actual situation, further enhancing the practicality of the data set. Further data scrambling is performed on the multi-modal gradient descent data, so that the data is subjected to certain disturbance and reconstruction before being input into the model. This process enhances the complexity and nonlinearity of the data set by nonlinearly scrambling the data, enabling the model to better capture complex patterns. On this basis, time series disturbance analysis is performed on the data, which is mainly used to process dynamic changes in time series data, analyze and capture the changing trends of the data in the time dimension, thereby improving the time dependence and time sequence of the data set.

[0105] Preferably, step S3 includes the following steps:

[0106] Step S31: inputting the multi-modal water chiller model input data into a preset GRU convolution model, the model containing a GRU layer with 64 hidden units and performing feature correlation matrix calculation of a self-attention layer, outputting a hidden state fused by the GRU layer and the self-attention layer, and constructing a GRU-SA water chiller preliminary model, wherein the calculation formula of the feature correlation matrix calculation includes,

[0107] ;

[0108] wherein, , , are query, key and value matrices respectively, is a feature dimension;

[0109] Step S32: performing cross-entropy loss function expansion on the GRU-SA water chiller preliminary model to generate a GRU-SA energy regularization loss function;

[0110] Step S33: performing self-adaptive parameter adjustment on the GRU-SA water chiller preliminary model and the GRU-SA energy regularization loss function to generate a GRU-SA water chiller fault judgment model, wherein the weight of the GRU-SA model is adjusted by an Adam optimizer, and the learning rate is set to 0.001.

[0111] In the embodiment of the application, at the data level, the technical means of step S3 realizes time sequence feature extraction and model optimization through multi-stage data processing: first, the multi-modal water chiller model input data (containing multi-source time sequence signals such as vibration, temperature and pressure) are input as a three-dimensional tensor (dimension: time step x feature dimension x modal number). The tensor input is preset to a GRU convolution model, the core of which is a GRU layer containing 64 hidden units, which realizes dynamic modeling of time dependence through gate mechanism (reset gate and update gate) wherein is an update gate, is a candidate state, realizes dynamic modeling of time dependence. At the same time, the self-attention layer maps the hidden state sequence output by the GRU into a query matrix Q, a key matrix K and a value matrix S, and calculates a feature correlation matrix through a scaled dot-product attention mechanism:

[0112]

[0113] wherein For the feature dimension (typical value is 64), the softmax function normalizes the attention weight, and highlights the feature contribution of the key time step. The fusion of the GRU layer and the self-attention layer is realized by splicing or weighted summation, generating a hidden state output that has both local time sequence dependence and global feature correlation, and constructing a GRU-SA preliminary model. In step S32, based on the classification output of the preliminary model, the standard cross-entropy loss function is extended to an energy regularization loss function. This regularization term suppresses the overfitting tendency of the model to noisy data by constraining the parameter norm, and improves the generalization ability. Step S33 adopts an adaptive parameter adjustment mechanism: the GRU-SA model weights are iteratively updated by the Adam optimizer (combined with momentum and adaptive learning rate), and the parameter update rule is:

[0114]

[0115] where is a fixed learning rate, and are the bias correction estimates of the first and second moments of the gradient, is a small constant (usually ). The entire GRU convolution kernel weight, attention projection matrix and fully connected layer parameter are finally generated, and the GRU-SA fault diagnosis model that can accurately capture the fault features is generated. The whole process realizes the closed loop from the original multi-modal data to the high-dimensional feature representation, and then to the parameter optimization guided by the regularization loss.

[0116] Preferably, the self-attention layer performs feature correlation matrix calculation, including the following steps:

[0117] Obtain the refrigeration drawing of the water chiller and the real-time refrigeration working condition parameters;

[0118] Three-dimensional modeling is performed on the refrigeration drawing of the water chiller to generate a physical structure of the refrigeration cycle of the water chiller; and the refrigeration system topology nodes are constructed on the physical structure of the refrigeration cycle of the water chiller to generate water chiller topology nodes;

[0119] According to the water chiller topology nodes and the real-time refrigeration working condition parameters, real-time working condition coding is performed to generate dynamic unit working condition driving coding data; and sparse matrix correlation matrix synthesis is performed on the dynamic unit working condition driving coding data to generate a water chiller sparse correlation matrix;

[0120] The water chiller sparse correlation matrix is subjected to physical constraint self-attention process calculation, and irrelevant noise mask processing is performed, so as to complete the feature correlation matrix calculation of the attention layer.

[0121] In the embodiment of the present application, by acquiring the refrigeration drawing of the water chiller unit and the real-time working condition parameters, combining the engineering design drawing and the real-time running data, basic data support is provided for subsequent modeling and analysis. On this basis, the three-dimensional modeling technology is used to process the refrigeration drawing of the water chiller unit, and the refrigeration cycle physical structure of the water chiller unit is generated. The core of this step is to convert the two-dimensional design drawing into a three-dimensional space model, which is convenient for detailed physical process analysis and topology structure construction. The refrigeration system topology node of the water chiller unit is established, and this process is modeled by connecting each component (such as cooling tower, compressor, evaporator, etc.) of the water chiller unit, forming a complete topology network. This topology node construction provides a systematic basis for the subsequent real-time working condition coding, and then the real-time working condition coding is generated by combining the real-time refrigeration working condition parameters with the topology node. The coding process converts the running state of the unit into a digital signal suitable for processing by a machine learning model, embodying the real-time nature of the dynamic interaction between the machine and the system. The generated dynamic unit working condition driving coding data is synthesized by a sparse matrix correlation matrix to generate a sparse correlation matrix of the water chiller unit. The sparse matrix is used to describe the non-zero interaction relationship between the nodes in the system, and through the synthesis of the sparse matrix, the calculation complexity can be effectively reduced, and the useful relationship information in the system is retained, laying a foundation for subsequent model optimization. The sparse correlation matrix of the water chiller unit is calculated by a physical constraint self-attention process. The self-attention mechanism calculates the relative importance weight between each feature in the input data, so that the model can effectively capture long-distance dependence relationships and dynamic interactions between different features in the system. The introduction of physical constraints in this stage ensures that the model can follow the physical laws when calculating feature correlations, avoiding unreasonable predictions or judgments in actual application of the model.

[0122] Preferably, step S33 comprises the following steps:

[0123] Step S331: associating the GRU-SA water chiller unit preliminary model and the GRU-SA energy regularization loss function in the time dimension to generate GRU-SA water chiller unit time dimension data; associating the GRU-SA water chiller unit preliminary model and the GRU-SA energy regularization loss function in the feature dimension to generate GRU-SA water chiller unit feature dimension data;

[0124] Step S332: multi-level association and aggregation of the GRU-SA water chiller unit time dimension data and the GRU-SA water chiller unit feature dimension data to generate GRU-SA water chiller unit time-feature dimension data;

[0125] Step S333: Injecting the GRU-SA chiller time-feature dimension data with fault type priori, generating chiller fault type data; performing time sequence fault cause and effect guarantee on the chiller fault type data, and performing refrigerant evaporator and compressor suction pipe adaptive parameter adjustment, generating a GRU-SA chiller fault judgment model.

[0126] In the embodiments of the present application, the time effectiveness of the model is modeled by combining time series data, and the time features are extracted according to the dynamic change law of the time series, thereby providing data support with time sequence correlation for subsequent analysis. At the same time, the feature dimensions are associated, and the feature data of the aggregation model is further mined to find the internal relationship between the multi-dimensional features, thereby providing more comprehensive data support for complex system behavior modeling. The combination of time dimension and feature dimension enables the model to capture the behavior patterns at different time points and different feature dimensions during the operation of the chiller, thereby improving the prediction accuracy. By associating and aggregating the time dimension data and the feature dimension data at multiple levels, the time-feature dimension data of the GRU-SA chiller is generated. This process is a hierarchical processing of data, which deeply integrates the association relationship of different dimensions through a hierarchical structure, so that the model can more comprehensively understand the dynamic changes inside the system. This aggregation process not only optimizes the efficiency of information transmission, but also enhances the expression ability of the model, which can consider both time variation and feature characteristics, and improves the fitting ability of the model to complex system behavior. Through the prior injection of fault type, chiller fault type data is generated. This technical means uses prior knowledge to inject different types of fault information into the model training process, which helps the model to better identify and distinguish different fault types during the learning process, and provides more accurate guidance for subsequent fault prediction.

[0127] As an example of the present application, reference is made to Fig. 1, which shows a schematic diagram of a chiller system according to an embodiment of the present application. In this example, the step S4 comprises: Figure 3

[0128] Step S41: Cloud deployment of the GRU-SA chiller fault judgment model, generating a GRU-SA chiller fault cloud judgment model; edge-cloud collaborative reasoning of the GRU-SA chiller fault cloud judgment model, generating GRU-SA chiller fault judgment data;

[0129] Step S42: Drawing a contribution heat map of the GRU-SA chiller model output data, generating a GRU-SA chiller fault heat map;

[0130] Step S43: Based on the GRU-SA chiller fault heat map, performing chiller fault prediction and diagnosis evaluation, thereby completing the GRU-SA model-based chiller fault prediction and diagnosis. ​

[0131] In the embodiment of the present application, the GRU-SA water chiller fault judgment model is deployed to the cloud and the process of edge-cloud collaborative reasoning. The technical means of cloud deployment mainly migrates complex computing tasks and model processing to the cloud platform to utilize the high-performance computing resources of cloud computing for efficient data processing and analysis. The advantage of cloud deployment is that it can dynamically expand computing capacity to ensure the processing needs of large-scale data, and edge-cloud collaborative reasoning can effectively allocate computing tasks according to real-time requirements and data size through the division of labor between edge devices and the cloud. Preliminary data processing and reasoning are performed on the edge side, and complex computing tasks and large-scale data processing are concentrated in the cloud, reducing latency and optimizing resource utilization efficiency. The technical means of drawing a contribution heat map of the GRU-SA water chiller model output data. The heat map drawing technology visually displays each index of the model output through color coding, helping to identify the fault area and severity of the water chiller. Through contribution analysis of the model output data, the influence of each feature on fault judgment can be effectively determined, and then through heat map visualization, the location and cause of the fault are intuitively displayed, thereby providing maintenance personnel with accurate fault positioning and priority decision basis. The evaluation of fault prediction and diagnosis is performed by analyzing the heat map, combining historical fault data and changes in the operating environment to evaluate the fault risk and warning level of the current water chiller.

[0132] Preferably, step S43 comprises the following steps:

[0133] Step S431: Obtain historical water chiller operation data;

[0134] Step S432: Extract key faults from the GRU-SA water chiller fault heat map to generate GRU-SA water chiller fault key data;

[0135] Step S433: Compare the GRU-SA water chiller fault key data with the historical water chiller operation data for fault diagnosis to generate GRU-SA water chiller fault diagnosis comparison data; perform fault judgment evaluation on the GRU-SA water chiller fault diagnosis comparison data to generate GRU-SA water chiller evaluation data;

[0136] Step S434: Construct a fault judgment report based on the GRU-SA water chiller evaluation data to generate a GRU-SA water chiller evaluation report, thereby completing the fault prediction and diagnosis of the water chiller based on the GRU-SA model.

[0137] In the embodiment of the present application, historical chiller operation data is obtained, and this process includes the collection of various types of operation data of the chiller equipment, such as temperature, pressure, vibration and other multi-dimensional physical signals. These data provide a rich basis for subsequent fault prediction and diagnosis. Step S432 generates GRU-SA chiller fault key data by extracting key faults of the GRU-SA chiller fault thermogram. This process filters out fault modes or abnormal behaviors with high impact from large-scale operation data by applying the self-attention mechanism and thermogram analysis technology. The key technology in this step is the generation of the thermogram and the automatic identification of the focus, which can highlight the most likely to fail parts and provide target areas for further analysis. The GRU-SA chiller fault key data is compared with the historical chiller operation data for fault diagnosis to generate fault diagnosis comparison data. The core technology of this process is data comparison and analysis, which uses a machine learning model to compare the extracted fault data with the historical data, identifies similarities and differences, and judges whether the equipment is currently in a fault state, and then infers the specific nature and location of the fault. In addition, through the comparison data, data basis can be provided for the next step of fault judgment evaluation, and the severity and risk of the fault are quantitatively evaluated by generating GRU-SA chiller evaluation data. The technical means used in this step include weight adjustment of model output, deep fusion analysis of historical data and existing data, etc. Through the construction of a fault judgment report based on the evaluation data, a GRU-SA chiller evaluation report is generated, and the fault prediction and diagnosis are completed. The generation process of the report combines the evaluation results with the actual operation scene to form a diagnostic report with operational and early warning functions. This process combines data visualization and report generation technology to generate a specific evaluation report, which not only helps operation and maintenance personnel to identify fault types, but also provides targeted maintenance suggestions, thereby improving the intelligent level and response speed of equipment management.

[0138] Therefore, from any viewpoint, the embodiments should be considered as being exemplary and non-limiting, the scope of the present application being defined by the appended claims and not by the above description, and it is intended that all changes which come within the meaning and range of equivalency of the claims are embraced therein.

[0139] The foregoing is considered as a specific implementation of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A GRU-SA model-based chiller fault prediction and diagnosis method, characterized in that, The method comprises the following steps: Step S1: obtaining a multi-modal water chiller data set; Step S1: obtaining a multi-modal water chiller data set; Step S2: performing physical knowledge virtual feature enhancement on the multi-modal water chiller pre-processing data set to generate a multi-modal water chiller virtual feature set; Step S2: performing physical knowledge virtual feature enhancement on the multi-modal water chiller pre-processing data set to generate a multi-modal water chiller virtual feature set; The physical knowledge virtual feature enhancement on the multi-modal water chiller pre-processing data set comprises the following steps: According to the unit vibration standardized data, the physical process correlation analysis of the refrigeration cycle is carried out to generate the unit cold cycle vibration analysis data; According to the unit vibration standardized data, the physical process correlation analysis of the refrigeration cycle is carried out to generate the unit cold cycle vibration analysis data; According to the unit temperature standardized data, the temperature fluctuation mode is identified to generate the unit temperature fluctuation mode data; the least square regression physical constraint dimension reduction is performed on the unit temperature fluctuation mode data to generate the unit temperature supercooling degree deviation; According to the unit pressure standardized data, the thermodynamic state analysis is carried out to generate the unit pressure thermodynamic state curve; the abnormal value peak value analysis is carried out on the unit pressure thermodynamic state curve to generate the unit pressure thermodynamic abnormal peak value data; the temperature-pressure correlation analysis is carried out on the unit pressure thermodynamic abnormal peak value data and the unit temperature supercooling degree deviation to generate the unit temperature-pressure supercooling degree deviation; According to the unit temperature-pressure supercooling degree deviation and the unit vibration theoretical energy efficiency ratio, One-hot coding is performed, and physical prior processing is performed to generate a multi-modal water chiller virtual feature set; Step S3: inputting the multi-modal water chiller model input data into a preset GRU convolution model, and performing feature correlation matrix calculation on the self-attention layer to generate a GRU-SA water chiller preliminary model; performing self-adaptive parameter adjustment on the GRU-SA water chiller preliminary model to generate a GRU-SA water chiller fault judgment model; 2. The GRU-SA model-based chiller fault prediction diagnosis method of claim 1, wherein, Step S4: cloud deployment of the GRU-SA water chiller fault judgment model, and edge-cloud collaborative reasoning to generate GRU-SA water chiller fault judgment data; contribution thermodynamic diagram drawing of the GRU-SA water chiller model output data. Step S1 comprises the following steps: Step S11: obtaining a multi-modal water chiller data set, wherein the multi-modal water chiller data set comprises unit vibration signals, unit temperature signals and unit pressure data; Step S12: performing wavelet packet transform denoising on the unit vibration signals to generate unit wavelet transform denoising data; performing data standardization processing on the unit wavelet transform denoising data to generate unit vibration standardized data; Step S13: performing time series drift processing on the unit temperature signals and the unit pressure data to generate unit temperature standardized data and unit pressure standardized data; 3. The GRU-SA model-based chiller fault prediction diagnosis method of claim 2, wherein, Step S14: synchronizing the unit vibration standardized data, the unit temperature standardized data and the unit pressure standardized data in a time window to generate a multi-modal water chiller pre-processing data set. Step S13 comprises the following steps: Step S131: two-dimensional linear projection is performed on the unit temperature signal and the unit pressure data to generate a unit temperature-pressure two-dimensional projection; linear interpolation anomaly value correction is performed on the unit temperature-pressure two-dimensional projection to generate unit temperature-pressure anomaly value processing data; Step S132: autocorrelation trend analysis is performed on the unit temperature-pressure anomaly value processing data to generate unit temperature-pressure autocorrelation trend data; Step S133: time sequence drift processing is performed on the unit temperature-pressure autocorrelation trend data to generate unit temperature-pressure time sequence drift data; second-order difference standardization processing is performed on the unit temperature-pressure time sequence drift data to generate unit temperature standardization data and unit pressure standardization data.

4. The GRU-SA model-based chiller fault prediction diagnosis method of claim 1, wherein, Step S2 includes the following steps: Step S21: physical knowledge virtual feature enhancement is performed on the multi-modal cold water unit preprocessing data set to generate a multi-modal cold water unit virtual feature set; Step S22: random noise disturbance is performed on the multi-modal cold water unit virtual feature set to generate multi-modal random noise disturbance data; gradient descent is performed on the multi-modal random noise disturbance data to generate multi-modal gradient descent data; Step S23: data scrambling is performed on the multi-modal gradient descent data to generate multi-modal gradient descent scrambled data; time series disturbance analysis is performed on the multi-modal gradient descent scrambled data to complete dynamic adversarial data sample construction, thereby generating multi-modal cold water unit model input data.

5. The GRU-SA model-based chiller fault prediction diagnosis method of claim 1, wherein, Step S3 includes the following steps: Step S31: the multi-modal cold water unit model input data is input into a preset GRU convolution model, the model includes a GRU layer with 64 hidden units, and feature correlation matrix calculation of a self-attention layer is performed, the output is a hidden state of the fusion of the GRU layer and the self-attention layer, and a GRU-SA cold water unit preliminary model is constructed, wherein the calculation formula of the feature correlation matrix calculation includes, wherein, are respectively query, key, value matrices, is the feature dimension; Step S32: cross-entropy loss function expansion is performed on the GRU-SA cold water unit preliminary model to generate a GRU-SA energy regularization loss function; Step S33: adaptive parameter adjustment is performed on the GRU-SA cold water unit preliminary model and the GRU-SA energy regularization loss function to generate a GRU-SA cold water unit fault judgment model, wherein the weight of the GRU-SA model is adjusted by an Adam optimizer, and the learning rate is set to 0.

001.

6. The GRU-SA model-based chiller fault prediction diagnosis method of claim 1, wherein, The self-attention layer performs feature correlation matrix calculation, including the following steps: Obtain the cold water unit refrigeration drawing and real-time refrigeration working condition parameters; Three-dimensional modeling is performed on the cold water unit refrigeration drawing to generate a cold water unit refrigeration cycle physical structure; refrigeration system topology node construction is performed on the cold water unit refrigeration cycle physical structure to generate a cold water unit topology node; Real-time working condition coding is performed on the cold water unit topology node and the real-time refrigeration working condition parameters to generate dynamic unit working condition driving coding data; sparse matrix correlation matrix synthesis is performed on the dynamic unit working condition driving coding data to generate a cold water unit sparse correlation matrix; Physical constraint self-attention process calculation is performed on the cold water unit sparse correlation matrix, and irrelevant noise mask processing is performed, thereby completing the feature correlation matrix calculation of the attention layer.

7. The GRU-SA model-based chiller fault prediction diagnosis method according to claim 5, characterized in that, Step S33 Comprising the following steps: Step S331: Time dimension association is performed on the GRU-SA water chiller preliminary model and the GRU-SA energy regularization loss function to generate GRU-SA water chiller time dimension data; feature dimension association is performed on the GRU-SA water chiller preliminary model and the GRU-SA energy regularization loss function to generate GRU-SA water chiller feature dimension data; Step S332: The GRU-SA water chiller time dimension data and the GRU-SA water chiller feature dimension data are associated and aggregated in multiple levels to generate GRU-SA water chiller time-feature dimension data; Step S333: The GRU-SA water chiller time-feature dimension data is subjected to prior injection of fault types to generate water chiller fault type data; the water chiller fault type data is subjected to time sequence fault cause and effect guarantee, and the refrigerant evaporator and the compressor suction pipe are subjected to adaptive parameter adjustment to generate a GRU-SA water chiller fault judgment model.

8. The GRU-SA model-based chiller fault prediction diagnosis method of claim 1, wherein, Step S4 comprises the following steps: Step S41: The GRU-SA water chiller fault judgment model is deployed on the cloud to generate a GRU-SA water chiller fault cloud judgment model; the GRU-SA water chiller fault cloud judgment model is subjected to edge-cloud collaborative reasoning to generate GRU-SA water chiller fault judgment data; Step S42: The contribution thermodynamic diagram of the GRU-SA water chiller model output data is drawn to generate a GRU-SA water chiller fault thermodynamic diagram; Step S43: Based on the GRU-SA water chiller fault thermodynamic diagram, water chiller fault prediction and diagnosis evaluation are performed to generate a GRU-SA water chiller evaluation report.

9. The GRU-SA model-based chiller fault prediction diagnosis method of claim 8, wherein, Step S43 comprises the following steps: Step S431: Historical water chiller operation data is acquired; Step S432: The GRU-SA water chiller fault thermodynamic diagram is subjected to key fault extraction to generate GRU-SA water chiller fault key data; Step S433: The GRU-SA water chiller fault key data and the historical water chiller operation data are subjected to fault diagnosis comparison to generate GRU-SA water chiller fault diagnosis comparison data; the GRU-SA water chiller fault diagnosis comparison data are subjected to fault judgment evaluation to generate GRU-SA water chiller evaluation data; Step S434: Based on the GRU-SA water chiller evaluation data, a fault judgment report is constructed to generate a GRU-SA water chiller evaluation report.

Images