Cooling fan fault root cause diagnosis method based on multi-scale causal analysis

Abstract

The invention discloses a cooling fan fault root cause diagnosis method based on multi-scale causal analysis, and relates to the technical field of fault diagnosis. A plurality of variables are set corresponding to the cooling fan, and the method comprises the following steps: for each independent variable, obtaining first predicted values at a plurality of target moments according to historical information of the independent variable in a plurality of target time periods, obtaining a plurality of second prediction values at a plurality of target moments according to the historical information of the independent variable and the plurality of dependent variable groups of the independent variable in the plurality of target time periods; performing causal testing according to the true values, the first predicted values and the plurality of second predicted values of the respective variables at the plurality of target moments, and constructing a causal network based on a testing result; when the cooling fan breaks down, fault variables are determined, and fault root causes are traced according to the fault variables and the causal network; the two predicted values are obtained through a multi-scale space-time prediction model. According to the method, multi-scale information and causal analysis are combined, and the stability and accuracy of root cause diagnosis of the cooling fan can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a cooling fan fault root cause diagnosis method based on multi-scale causal analysis. Background Art

[0002] In recent years, causal analysis has received extensive attention in fault diagnosis and root cause identification in power systems. Cooling fans under heavy overload of transformers usually have complex dynamic behaviors, and the occurrence of faults often involves spatiotemporal correlations between multiple variables, while traditional correlation analysis methods are difficult to accurately reveal the causal relationship between variables.

[0003] In the field of fault diagnosis of key equipment in power systems, the state of transformer cooling fans is usually affected by multiple sensor signals, which may have different time scales and spatial distributions. For example, in intelligent manufacturing and power systems, the temperature, pressure, current, voltage and other signals collected by transformer cooling fan sensors may have different dynamic change patterns, and the fault propagation path often spans multiple subsystems, and the variables such as temperature, flow, and valve position have strong linkage, and some faults have nonlinear propagation characteristics of "hidden initiation-slow diffusion-mutation amplification". However, traditional causal analysis is mainly based on linear assumptions, and it is difficult to effectively model the causal relationship in cooling fans with nonlinear, time-varying characteristics and multi-scale spatiotemporal dependence. Therefore, how to effectively mine this causal relationship under heavy overload of the transformer to identify the root cause of the fault has become the focus of current research.

[0004] To this end, methods based on statistical models and deep learning are proposed in related technologies. However, the statistical model-based method has good interpretability on small-scale data sets, but is limited in the diagnosis of cooling fan faults under heavy transformer overload scenarios; while the deep learning-based method faces problems such as insufficient spatiotemporal information extraction and unstable fault root cause identification, making it difficult to effectively capture causal relationships across time scales and spatial ranges. Summary of the invention

[0005] The purpose of the present invention is to propose a cooling fan fault root cause diagnosis method based on multi-scale causal analysis to improve the stability and accuracy of cooling fan root cause diagnosis.

[0006] A method for diagnosing the root cause of a cooling fan fault based on multi-scale causal analysis is proposed in an embodiment of the present invention. A plurality of variables are set for the corresponding cooling fan. The method comprises: for each independent variable, a first prediction value at a plurality of target moments is obtained according to historical information of the independent variable in a plurality of target time periods, and a plurality of second prediction values ​​at a plurality of target moments are obtained according to historical information of the independent variable and a plurality of dependent variable groups of the independent variable in a plurality of target time periods; a causal test is performed according to the true value, the first prediction value and a plurality of second prediction values ​​of each independent variable at a plurality of target moments, and a causal network is constructed based on the test results; when a fault occurs in the cooling fan, a fault variable is determined, and the root cause of the fault is traced according to the fault variable and the causal network; wherein the plurality of target time periods correspond one-to-one to the plurality of target moments, the first prediction value and the second prediction value are obtained through a multi-scale spatiotemporal prediction model, the plurality of second prediction values ​​correspond one-to-one to the plurality of dependent variable groups, the independent variable is any variable among the plurality of variables, and the dependent variable group includes at least one variable other than the independent variable.

[0007] According to one embodiment of the present invention, the multi-scale spatiotemporal prediction model includes multiple feature extractors, a splicing layer and a fully connected layer; wherein the multiple feature extractors are connected in sequence, the input end of the first feature extractor among the multiple feature extractors serves as the input end of the multi-scale spatiotemporal prediction model, and the multiple feature extractors are used to extract spatiotemporal information of different scales; the input end of the splicing layer is respectively connected to the output ends of the multiple feature extractors, the input end of the fully connected layer is connected to the output end of the splicing layer, and the output end of the fully connected layer serves as the output end of the multi-scale spatiotemporal prediction model.

[0008] According to one embodiment of the present invention, the feature extractor adopts a gated recurrent unit embedded in a graph convolutional neural network, the convolution kernels of multiple feature extractors have different sizes, and the convolution kernel of the previous feature extractor is smaller than the convolution kernel of the subsequent feature extractor.

[0009] According to one embodiment of the present invention, a causal test is performed based on the true value, the first predicted value and the multiple second predicted values ​​of the independent variable at multiple target moments, including: respectively calculating the residuals of the true value at each target moment and the first predicted value and each second predicted value to obtain the first residual and multiple second residuals at each target moment; for each dependent variable group, determining whether the dependent variable in the dependent variable group has a causal effect on the independent variable based on the first residuals at multiple target moments and the second residuals corresponding to the dependent variable group.

[0010] According to one embodiment of the present invention, determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable based on the first residuals at multiple target moments and the second residuals corresponding to the dependent variable group includes: calculating a first variance based on the first residuals at multiple target moments, and calculating a second variance based on the second residuals corresponding to the dependent variable group at multiple target moments; calculating a test value based on the first variance and the second variance; and determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable based on the test value.

[0011] According to one embodiment of the present invention, the test value is calculated by the following formula: Wherein, F represents the test value, represents the first variance, represents the second variance, p represents the number of dependent variables in the dependent variable group, N represents the number of target time periods, and L represents the number of samples in the historical information corresponding to the target time period.

[0012] According to one embodiment of the present invention, determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable based on the test value includes: if the test value is less than a preset threshold, determining that the dependent variables in the dependent variable group have a causal effect on the independent variable; if the test value is greater than or equal to the preset threshold, determining that the dependent variables in the dependent variable group have no causal effect on the independent variable.

[0013] According to one embodiment of the present invention, constructing a causal network based on the test results includes: if the dependent variables in the dependent variable group have a causal influence on the independent variables, then drawing the directed edges from each of the dependent variables in the dependent variable group to the independent variables respectively; and obtaining the causal network according to the directed edges corresponding to all independent variables.

[0014] According to one embodiment of the present invention, the dependent variable group is divided according to the circulation loop of the cooling fan, and the circulation loop includes at least two of a chilled water loop, a condensed water loop, an auxiliary hot water loop, a city water loop and an evaporator loop.

[0015] According to an embodiment of the present invention, the plurality of variables include a plurality of sensor variables and at least one calculated variable, wherein the calculated variable is calculated according to the at least one sensor variable.

[0016] A cooling fan fault root cause diagnosis method based on multi-scale causal analysis in an embodiment of the present invention, for each independent variable, obtains a first prediction value at multiple target moments according to the historical information of the independent variable in multiple target time periods, and obtains multiple second prediction values ​​at multiple target moments according to the historical information of the independent variable and multiple dependent variable groups of the independent variable in multiple target time periods; performs causal tests based on the true values, first prediction values ​​and multiple second prediction values ​​of each variable at multiple target moments, and constructs a causal network based on the test results; when a cooling fan fails, determines the fault variable, and traces the root cause of the fault according to the fault variable and the causal network; the two prediction values ​​are obtained through a multi-scale spatiotemporal prediction model. Therefore, by combining multi-scale information and causal analysis, the stability and accuracy of cooling fan root cause diagnosis can be improved.

[0017] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of a cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to an embodiment of the present invention; Figure 2 It is a framework diagram of a cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to an embodiment of the present invention; Figure 3 The present invention is a GCGRU structural unit diagram of an embodiment; Figure 4 is an architecture diagram of a multi-scale spatiotemporal prediction model according to an embodiment of the present invention; Figure 5 is a framework diagram of a cooling fan according to an embodiment of the present invention; Figure 6 is a cause-effect matrix diagram of a cooling fan according to an example of the present invention; Figure 7 FIG. 1 is a diagram showing the root causes of a cooling fan failure according to an example of the present invention. DETAILED DESCRIPTION

[0019] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0020] In the related technologies, causal analysis mainly includes methods based on statistical models and methods based on deep learning. However, the statistical model-based method is limited in the diagnosis of cooling fan faults under heavy transformer overload scenarios, and the deep learning-based method faces problems such as insufficient spatiotemporal information extraction and unstable fault root cause identification, making it difficult to effectively capture causal relationships across time scales and spatial ranges.

[0021] To this end, the present invention proposes a cooling fan fault root cause diagnosis method based on multi-scale causal analysis. The method integrates spatiotemporal feature extraction, multi-scale modeling and causal analysis technology, which can enhance the causal relationship learning ability across variables and improve the stability and accuracy of causal analysis, so as to improve the operation safety and maintenance efficiency of cooling fans and power systems under heavy overload of transformers.

[0022] In an embodiment of the present invention, a plurality of variables are provided corresponding to the cooling fan.

[0023] Among them, the multiple variables may include multiple sensor variables (that is, variables that can be directly collected by sensors, such as the evaporator inlet water temperature, evaporator outlet water temperature, evaporator water flow rate, etc. of the cooling fan) and at least one calculated variable, and the calculated variable is calculated based on at least one sensor variable. For example, the cooling capacity is calculated based on the inlet and outlet water temperature difference and the chilled water flow rate.

[0024] Figure 1 It is a flow chart of a cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to an embodiment of the present invention.

[0025] like Figure 1 As shown in FIG. 1 , the cooling fan fault root cause diagnosis method based on multi-scale causal analysis includes: S11, for each independent variable, a first prediction value at multiple target moments is obtained based on the historical information of the independent variable in multiple target time periods, and multiple second prediction values ​​at multiple target moments are obtained based on the historical information of the independent variable and multiple dependent variable groups of the independent variable in multiple target time periods.

[0026] Among them, multiple target time periods correspond one-to-one to multiple target moments, the first prediction value and the second prediction value are obtained through a multi-scale spatiotemporal prediction model, the multiple second prediction values ​​correspond one-to-one to multiple dependent variable groups, the independent variable is any variable among the multiple variables, and the dependent variable group includes at least one variable other than the independent variable.

[0027] like Figure 2 As shown, represents the time series data of the ith variable, Indicates the number of variables. When k=1, represents the first predicted value of the i-th variable at the target time; when k>1, represents the second predicted value of the k-1th dependent variable group corresponding to the i-th variable at the target time ( Figure 2 The dependent variable group contains variables As an example, it should be noted that for the independent variable , the variables in the dependent variable group do not include , that is, i≠j).

[0028] In one embodiment, the dependent variable group is divided according to the circulation loop of the cooling fan, and the circulation loop includes at least two of a chilled water loop, a condensed water loop, an auxiliary hot water loop, a city water loop, and an evaporator loop.

[0029] Exemplarily, for each loop, for each variable set corresponding to the loop, each other variable set in the loop can be divided into a dependent variable group, and the variables set in other loops can be divided into a dependent variable group. For example, a1 variables are set corresponding to the chilled water loop, a2 variables are set corresponding to the condensed water loop, a2 variables are set corresponding to the auxiliary hot water loop, a4 variables are set corresponding to the city water loop, and a5 variables are set corresponding to the evaporator loop. For each variable in the chilled water loop, the other a1-1 variables are divided into a1-1 dependent variable groups (that is, each dependent variable group includes one variable), a2 variables corresponding to the condensed water loop are divided into a dependent variable group, a3 variables corresponding to the auxiliary hot water loop are divided into a dependent variable group, a4 variables corresponding to the city water loop are divided into a dependent variable group, and a5 variables corresponding to the evaporator loop are divided into a group. For the condensed water loop, the auxiliary hot water loop, the city water loop and the evaporator loop, the grouping method of the dependent variables is similar to that of the chilled water loop.

[0030] To reduce the amount of calculation, for each variable in each loop, the dependent variable group can be divided according to the types of other variables except the variable. Taking the chilled water loop as an example, for a certain temperature variable in the loop, other temperature variables can be divided into a dependent variable group, flow variables can be divided into a dependent variable group, valve position variables (or opening variables) can be divided into a dependent variable group, and so on.

[0031] In some embodiments of the present invention, a multi-scale spatiotemporal prediction model includes multiple feature extractors, a splicing layer and a fully connected layer; wherein the multiple feature extractors are connected in sequence, the input end of the first feature extractor among the multiple feature extractors serves as the input end of the multi-scale spatiotemporal prediction model, and the multiple feature extractors are used to extract spatiotemporal information of different scales; the input end of the splicing layer is respectively connected to the output ends of the multiple feature extractors, the input end of the fully connected layer is connected to the output end of the splicing layer, and the output end of the fully connected layer serves as the output end of the multi-scale spatiotemporal prediction model.

[0032] For example, the feature extractor uses a Gated Recurrent Unit (GCGRU) embedded in a graph convolutional neural network, and its structure is as follows: Figure 3 As shown. The GCGRU structure introduces a graph convolutional neural network (GCN) based on the classic gated recurrent unit (GRU) to fully integrate the structural dependencies and timing dynamic characteristics between devices. It is especially suitable for modeling systems with spatial topological structures (such as cooling fan groups) and multivariate coupling characteristics. Among them, the GCN convolution kernels of multiple feature extractors have different sizes, and the convolution kernel of the previous feature extractor is smaller than the convolution kernel of the subsequent feature extractor.

[0033] See also Figure 3 , express The node features input at the moment, express The hidden state of the moment, express The hidden state of the moment, represents the sigmoid activation function, express Activation function, represents element-wise multiplication, Represents element-wise addition.

[0034] First, using GCN based And the existing topological matrix A (the adjacency matrix of the graph, representing the connection relationship between nodes), we get . Then, through nonlinear transformation according to and Calculate the update gate Ut and reset gate Rt respectively. Ut and Rt represent the gating factors for retaining the old state information and the degree of forgetting in the current state, respectively. , , , Respectively represent calculation , The weight matrix used is, , Respectively represent calculation , The bias term used. Then, through the nonlinear transformation According to the reset gate Rt The modulation result, and X t , calculate the candidate hidden state ,in, , Representation calculation The weight matrix used is, Representation calculation Finally, according to the update gate Determines the current hidden state The integration ratio of new and old information, completion status Update, where .

[0035] In one embodiment, if Figure 4 As shown in the figure, the multi-scale spatiotemporal prediction model includes three feature extractors. The three feature extractors use graph convolution gated recurrent units with different convolution kernel sizes to extract short-term local dependencies, medium-term dependencies, and long-term global dependencies in turn, and through the hierarchical feature fusion strategy, the integrity of subsequent causal analysis can be improved.

[0036] Specifically, since the operating variables of the cooling fan may have different time scales and spatial distributions, including spatiotemporal characteristics of different scales, the predictive ability between variables depends on this information. In order to better extract the complex spatiotemporal patterns between variables, a multi-scale spatiotemporal prediction model is designed. First, GCGRU is introduced, that is, GCN is used to replace the linear transformation in the traditional GRU, so as to enhance the modeling ability of complex topological structure data. GCGRU can combine the spatial feature extraction ability of GCN with the time series modeling ability of GRU to more comprehensively learn the spatiotemporal interaction relationship between cooling fan variables under heavy overload of transformers. Compared with traditional GRU, GCGRU explicitly introduces graph convolution operations in the calculation, so that the model can extract spatial features based on topological structure data.

[0037] In the multi-scale spatiotemporal prediction model, see Figure 4 , designed three GCGRU feature extractors of different scales, using GCN convolution kernels of different sizes to extract spatiotemporal information at different levels, Figure 4 In and are the true value and predicted value of the variable, respectively. First, the first layer of GCGRU uses the smallest scale (such as 1×1) convolution kernel, which is used to capture local short-term causal patterns, that is, to identify the dependencies of variables in a short time window, while considering the influence of their local topological structure. The features extracted by this layer are not only used as the input of the next layer of GCGRU, but also stored in the final concatenation layer to ensure that features of different scales can be fully integrated. Then, the second layer of GCGRU uses a medium-scale (such as 3×3) convolution kernel to further expand the time window and spatial receptive field to capture a wider range of spatiotemporal dependency information. In this process, the local short-term information extracted by the first layer of GCGRU is combined with the new features of the second layer of GCGRU, thereby enhancing the model's ability to learn multi-level causal relationships. Finally, the third layer of GCGRU uses the largest scale (such as 5×5) convolution kernel, which is responsible for extracting global long-term dependencies to ensure that the model can learn the remote causal influences between variables. Its mathematical expression is: in, Indicates that the convolution kernel is × The spatial features extracted by GCN, Indicates The output of layer GCGRU at time t, It is an adjacency matrix, which represents the topological connection relationship between the sensors where the variables are located.

[0038] Three-layer The output feature information is sent to the concatenation layer to form a complete multi-scale spatiotemporal pattern representation, which is formulated as follows: Among them, ⊕ represents the feature concatenation operation, Represents the final fused multi-scale spatiotemporal features. The features are then input into the fully connected layer for prediction of future time steps, and the formula is as follows: in, represents the predicted value of variable x at time t+1, and W represents the weight matrix.

[0039] For example, the multi-scale spatiotemporal prediction model can be pre-trained, and the process may include: 1) data acquisition and preprocessing, which can obtain the time series data corresponding to the sensor variables of the cooling fan under heavy overload of the transformer, and perform corresponding preprocessing on the data, such as obtaining the time series data corresponding to the calculation variables, denoising the time series data corresponding to the sensor variables, dividing the training set and the test set, etc.; 2) model training, constructing Figure 4 The multi-scale spatiotemporal prediction model shown, for each independent variable, regards the independent variable as a prediction target, inputs the training set of the independent variable (including two types, one is the historical information of the independent variable, and the other is the historical information of the independent variable and the historical information of the dependent variable in the dependent variable group of the independent variable) into the multi-scale spatiotemporal prediction model, predicts the independent variable, and finally obtains a multi-scale spatiotemporal prediction model for all variables.

[0040] Specifically, for each independent variable Train the corresponding multi-scale spatiotemporal prediction model , in order to analyze the variables Is it right With causality, two multi-scale spatiotemporal prediction models are set up: The first multi-scale spatiotemporal prediction model uses only the historical information of independent variables for prediction, which is expressed as follows: The second multi-scale spatiotemporal prediction model adds historical information of the dependent variable for prediction, which is expressed as follows: in, Indicates the length of the historical time window, that is, the length of the target time period. and Respectively means using only its own historical information and adding variables As the predicted value of auxiliary information at the target time t, Representation variables in the past The historical information within time steps, Representation variables in the past The historical information within a time step.

[0041] It should be noted that multiple multi-scale spatiotemporal prediction models can be trained for each variable, or one multi-scale spatiotemporal prediction model can be trained for all variables. The former is more targeted and more accurate, while the latter is more universal and easier to use and store.

[0042] S12, performing a causal test according to the true value of each variable at multiple target moments, the first predicted value, and multiple second predicted values, and constructing a causal network based on the test results.

[0043] In some embodiments of the present invention, a causal test is performed based on the true value, the first predicted value and the multiple second predicted values ​​of the independent variable at multiple target moments, including: respectively calculating the residuals of the true value at each target moment and the first predicted value and each second predicted value to obtain the first residual and multiple second residuals at each target moment; for each dependent variable group, determining whether the dependent variable in the dependent variable group has a causal effect on the independent variable based on the first residuals at multiple target moments and the second residuals corresponding to the dependent variable group.

[0044] In one embodiment, based on the first residuals at multiple target moments and the second residuals corresponding to the dependent variable group, determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable includes: calculating a first variance based on the first residuals at multiple target moments, and calculating a second variance based on the second residuals corresponding to the dependent variable group at multiple target moments; calculating a test value based on the first variance and the second variance; and determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable based on the test value.

[0045] Exemplarily, the test value is calculated by the following formula: Among them, F represents the test value, represents the first variance, represents the second variance, p represents the number of dependent variables in the dependent variable group, N represents the number of target time periods, and L represents the number of samples in the historical information corresponding to the target time period.

[0046] Exemplarily, determining whether the dependent variables in the dependent variable group have a causal effect on the independent variable is based on the test value, including: if the test value is less than a preset threshold, determining that the dependent variables in the dependent variable group have a causal effect on the independent variable; if the test value is greater than or equal to the preset threshold, determining that the dependent variables in the dependent variable group have no causal effect on the independent variable.

[0047] Specifically, if Figure 4 As shown, a causal analysis is performed based on the predicted values ​​obtained by the multi-scale spatiotemporal prediction model.

[0048] First, calculate the true value by and the first predicted value The first residual , and the true value and the second predicted value (Add dependent variable The second residual of the prediction : Afterwards, compare and The variance difference between the dependent variables Is it right Has a predictive contribution. That is, if we add After being used as input, the residual variance is significantly reduced, so it is considered right There is a causal relationship. First variance and the second variance The calculation is as follows: Then, by Evaluate and Is there a significant difference? The probability value corresponding to the value is less than the significance level (i.e. the preset threshold, such as 0.05), then reject the null hypothesis (i.e. right No causal effect), yes reason.

[0049] Exemplarily, when performing a causal test, for a dependent variable group containing multiple dependent variables, if it is detected that the dependent variable group has a causal effect on the independent variable, in order to improve accuracy, the dependent variable group can be further divided. For example, if the number of dependent variables in the dependent variable group is less than the set value, such as 3, 4, etc., each dependent variable in the dependent variable group can be divided into a dependent variable group; if the number of dependent variables in the dependent variable group is greater than or equal to the set value, the dependent variable group can be divided according to the variable type. Afterwards, the independent variable is predicted and causally tested based on the newly divided dependent variable group until a single dependent variable that has a causal relationship with the independent variable is determined. It should be noted that if the dependent variables in the dependent variable group are of the same type and the number is greater than or equal to the set value, the dependent variables in the dependent variable group can be randomly grouped, such as every K dependent variables in the dependent variable group are divided into a group, K is a preset value.

[0050] In some embodiments of the present invention, a causal network is constructed based on the test results, including: if the dependent variables in the dependent variable group have a causal influence on the independent variables, then the directed edges from each dependent variable in the dependent variable group to the independent variables are drawn respectively; and the causal network is obtained according to the directed edges corresponding to all independent variables.

[0051] Specifically, after completing the causal test between all variable pairs (i.e., independent variables and dependent variable groups), a causal network can be constructed, which includes nodes and directed edges. Among them, the nodes are the monitoring variables of the cooling fan (such as temperature, pressure, flow, etc.), and the directed edges are the variables that are monitored by the cooling fan. Detected If there is a causal relationship, draw a line in the causal network The directed edges of , represent causal relationships.

[0052] S13, when a cooling fan fails, the fault variable is determined, and the root cause of the fault is traced based on the fault variable and the causal network.

[0053] Specifically, when a cooling fan fails, the fault variable is first determined. The fault variable can be a variable that deviates significantly from the normal state (for example, the difference between the variable value and the normal state value is greater than the difference threshold and lasts for a preset time). Then, the root cause of the fault (i.e., the root cause) is traced based on the propagation path in the causal network.

[0054] In an embodiment of the present invention, Figure 2 As shown in the figure, the root cause diagnosis method of cooling fan fault based on multi-scale causal analysis is implemented by two parts, the first part is the multi-scale spatiotemporal prediction model, and the second part is the causal analysis module. For the multi-scale spatiotemporal prediction model, GCGRU is first introduced, that is, the graph convolutional network (GCN) is used to replace the linear transformation in the traditional gated recurrent unit (GRU), so that the spatiotemporal dependency relationship can be more effectively modeled, so as to more accurately capture the spatiotemporal dependency pattern in the cooling fan. By designing GCGRU feature extractors of various scales, that is, GCN convolution kernels of various sizes are used to extract spatiotemporal information of different scales; then the spatiotemporal information of different scales is spliced ​​in the splicing layer to form a complete multi-scale spatiotemporal pattern representation; finally, the spliced ​​information is input into the fully connected layer for prediction. For the causal analysis module, the prediction relationship between multiple variables of the cooling fan is modeled, and the causal relationship is identified, so that when the cooling fan fails, the propagation path of the fault can be tracked, so as to determine the root cause location of the fault.

[0055] The present invention proposes a cooling fan fault root cause diagnosis method based on multi-scale causal analysis, which can effectively improve the root cause diagnosis capability of cooling fans. By introducing the GCGRU structure, the spatiotemporal dependency between system variables is fully explored, and the multi-scale fusion strategy is used to integrate information at different time and space scales to enhance the model's ability to capture complex dynamic changes. In addition, combined with causal analysis, the causal relationship between system variables is modeled to accurately identify the source of the fault.

[0056] Combine the following Figure 5-Figure 7 The beneficial effects of the cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to the embodiment of the present invention are described as follows: like Figure 5 As shown, the cooling fan consists of five circulation loops, namely, chilled water loop, condensing water loop, auxiliary hot water loop, city water loop and evaporator loop, which contain 48 sensor variables and 16 calculated variables. Figure 5The middle bypass represents the condenser water pipeline, which is not converted into evaporator water. Without loss of generality, taking the evaporator water reduction fault as an example, through the relevant monitoring method, the variables shown in Table 1 are obtained.

[0057] Table 1 Based on the cooling fan fault root cause diagnosis method based on multi-scale causal analysis of the present invention, a causal analysis is performed on the selected variable set to reveal the propagation path of the abnormal evaporator water flow fault and identify the root dependent variable. The results are as follows: Figure 6 , Figure 7 When the evaporator outlet water temperature is 1 ( ) is abnormal, the first thing that affects the evaporator inlet water temperature is ( ), resulting in changes in heat exchange efficiency and further affecting the evaporator water flow rate ( ). Fluctuations in water flow are transmitted to the evaporator valve position ( ), triggering regulatory actions, which may aggravate system instability. At the same time, as the core of building water supply, Directly affects the building inlet water temperature ( ), thereby changing the building outlet water temperature ( ), causing building load fluctuations. Eventually, the abnormal heat exchange of the evaporator will affect the hot water system, causing the hot water inlet temperature ( ) and outlet temperature ( ) deviates from the normal range, affecting the overall heating effect, thus forming a fault chain from the cooling system to the building water supply to the hot water circulation.

[0058] It can be seen that the results obtained by the cooling fan fault root cause diagnosis method based on multi-scale causal analysis proposed in the present invention are highly consistent with the expert knowledge analysis, which verifies the accuracy and reliability of the method in diagnosing the root cause of cooling fan faults.

[0059] In summary, the present invention proposes a cooling fan fault root cause diagnosis method based on multi-scale causal analysis, which combines multi-scale spatiotemporal information and causal analysis to improve the accuracy of cooling fan root cause diagnosis. Specifically: the multi-scale spatiotemporal prediction model extracts local and global spatiotemporal information in time series data through GCGRU structures of different scales, that is, it captures spatial dependencies of different scales through graph convolution kernels of different sizes, so that it is more suitable for the operation data of the cooling fan under heavy overload of the transformer, thereby improving the stability and robustness of fault prediction; causal analysis is used to model the interaction between the cooling fan variables under heavy overload of the transformer, and the propagation path of the fault can be accurately identified. Therefore, by analyzing the prediction ability of the historical information of different variables for the current state through the prediction model, a causal relationship network is constructed, the fault propagation path is tracked, and the root cause location of the fault is finally determined. While ensuring the prediction accuracy, the interpretability of the cooling fan fault diagnosis can be improved, providing strong support for the intelligent operation and maintenance of the power system. In addition, this method fully considers the industry characteristics such as the coupling relationship between variables during the operation of the cooling fan, the complex temporal and spatial characteristics, and the hidden abnormal propagation path. A causal analysis mechanism that adapts to the structure and dynamic response of the cooling fan is specially designed. Therefore, this method is highly specific.

[0060] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0061] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0062] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0063] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0064] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0065] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0067] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A cooling fan fault root cause diagnosis method based on multi-scale causal analysis, characterized in that: A plurality of variables are provided for the corresponding cooling fan, and the method comprises: For each independent variable, a first prediction value at a plurality of target moments is obtained according to historical information of the independent variable in a plurality of target time periods, and a plurality of second prediction values ​​at a plurality of target moments are obtained according to historical information of the independent variable and a plurality of dependent variable groups of the independent variable in a plurality of the target time periods; Performing a causal test according to the true value of each of the independent variables at the plurality of target moments, the first predicted value, and the plurality of second predicted values, and constructing a causal network based on the test results; When the cooling fan fails, a fault variable is determined, and a root cause of the fault is traced based on the fault variable and the causal network; Among them, multiple target time periods correspond one-to-one to multiple target moments, the first prediction value and the second prediction value are obtained through a multi-scale spatiotemporal prediction model, multiple second prediction values ​​correspond one-to-one to multiple dependent variable groups, the independent variable is any variable among the multiple variables, and the dependent variable group includes at least one variable other than the independent variable.

2. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 1 is characterized in that: The multi-scale spatiotemporal prediction model includes multiple feature extractors, concatenation layers and fully connected layers; wherein, The multiple feature extractors are connected in sequence, and the input end of the first feature extractor among the multiple feature extractors serves as the input end of the multi-scale spatiotemporal prediction model. The multiple feature extractors are used to extract spatiotemporal information of different scales; the input end of the splicing layer is respectively connected to the output ends of the multiple feature extractors, the input end of the fully connected layer is connected to the output end of the splicing layer, and the output end of the fully connected layer serves as the output end of the multi-scale spatiotemporal prediction model.

3. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 2 is characterized in that: The feature extractor adopts a gated recurrent unit embedded in a graph convolutional neural network, and the convolution kernels of the multiple feature extractors have different sizes, and the convolution kernel of the previous feature extractor is smaller than the convolution kernel of the subsequent feature extractor.

4. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 1 is characterized in that: Performing a causal test according to the actual value of the independent variable at the plurality of target moments, the first predicted value, and the plurality of second predicted values, comprises: Residuals of the true value of each target moment and the first predicted value and each second predicted value are calculated respectively to obtain a first residual and a plurality of second residuals of each target moment; For each of the dependent variable groups, it is determined whether the dependent variables in the dependent variable group have a causal effect on the independent variables based on the first residuals at the plurality of target moments and the second residuals corresponding to the dependent variable group.

5. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 4 is characterized in that: The determining whether the dependent variable in the dependent variable group has a causal effect on the independent variable according to the first residuals at the plurality of target moments and the second residuals corresponding to the dependent variable group comprises: Calculating a first variance based on first residuals at a plurality of the target moments, and calculating a second variance based on second residuals corresponding to the dependent variable group at a plurality of the target moments; Calculate a test value based on the first variance and the second variance; Whether the dependent variable in the dependent variable group has a causal effect on the independent variable is determined according to the test value.

6. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 5 is characterized in that: The test value is calculated by the following formula: Wherein, F represents the test value, represents the first variance, represents the second variance, p represents the number of dependent variables in the dependent variable group, N represents the number of target time periods, and L represents the number of samples in the historical information corresponding to the target time period.

7. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 6 is characterized in that: The determining, according to the test value, whether the dependent variable in the dependent variable group has a causal effect on the independent variable comprises: If the test value is less than a preset threshold, it is determined that the dependent variable in the dependent variable group has a causal effect on the independent variable; If the test value is greater than or equal to the preset threshold, it is determined that the dependent variable in the dependent variable group has no causal effect on the independent variable.

8. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 6, characterized in that: The constructing of a causal network based on the test results includes: If the dependent variables in the dependent variable group have a causal effect on the independent variable, then a directed edge from each dependent variable in the dependent variable group to the independent variable is drawn respectively; The causal network is obtained according to the directed edges corresponding to all independent variables.

9. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to any one of claims 1 to 8, characterized in that: The dependent variable group is divided according to a circulation loop of the cooling fan, wherein the circulation loop includes at least two of a chilled water loop, a condensed water loop, an auxiliary hot water loop, a city water loop, and an evaporator loop.

10. The cooling fan fault root cause diagnosis method based on multi-scale causal analysis according to claim 9, characterized in that: The plurality of variables include a plurality of sensor variables and at least one calculated variable, wherein the calculated variable is calculated according to the at least one sensor variable.

Images