Air conditioner outdoor unit multi-sensor online detection method and system based on evidence theory

Abstract

The invention belongs to the technical field of air conditioner detection, and relates to an air conditioner outdoor unit multi-sensor online detection method and system based on the evidence theory, and the method comprises the steps: marking all data, smaller than a preset distance average quantile, in historical time sequence working data of a plurality of air conditioner outdoor units as normal; marking all data not less than the quantile of the distance average value as abnormal; performing normalization processing on the marked historical time sequence working data and dividing the data into a training set and a test set; inputting the training set into a plurality of XGboost classifiers for training; testing the trained XGboost classifier by using a test set, and outputting a plurality of groups of test results; fusing the multiple groups of test results based on an evidence theory and then outputting a detection result corresponding to each time point; and whether the air conditioner outdoor unit breaks down or not is judged through the detection result and the historical abnormal time point number threshold value. According to the invention, online rapid fault detection of the air conditioner outdoor unit can be realized, and the accuracy of the detection result is improved.

Description

Technical Field

[0001] The present application belongs to the field of air conditioning detection technology, and more specifically, relates to an air conditioning outdoor unit multi-sensor online detection method and system based on evidence theory. Background Art

[0002] Existing online detection methods for air conditioner outdoor units use time series of collected physical quantities to determine the cooling and heating performance of the units and detect defective products. This method assumes that normal products exhibit consistent behavioral patterns in the measured physical quantities and can be clearly distinguished from the patterns of abnormal products. However, in complex production environments, the sampling frequency and data volume limitations of sensors lead to performance bottlenecks in accurately detecting fault categories in air conditioner outdoor units.

[0003] Currently, factory inspections for air conditioners primarily use the average values ​​of various physical quantities measured when multiple outdoor air conditioner units are operating at a stable temperature. This relies on a large, constant-temperature operating room and requires a long inspection period. The operating room plays a more critical role in summer and winter, when ambient temperatures deviate from the standard setting, than during normal times. However, this also increases resource consumption and costs. Furthermore, the long inspection time can lead to production bottlenecks during peak production seasons, impacting timely product delivery. Furthermore, this traditional inspection method, based solely on average data at a fixed ambient temperature, fails to fully reflect the performance of outdoor air conditioner units in complex and variable real-world environments, potentially leading to the release of underperforming products. The resulting returns and repairs of defective products increase production costs for manufacturers and increase energy consumption and financial losses for users. Summary of the Invention

[0004] In response to the defects of the existing technology, the purpose of this application is to provide an air-conditioning outdoor unit multi-sensor online detection method and system based on evidence theory, aiming to solve the problem that the existing air-conditioning outdoor unit detection method is difficult to adapt to the actual complex working environment of the production environment, and the detection accuracy is low, and the detection results are difficult to reflect the performance of the air-conditioning outdoor unit in a complex usage environment.

[0005] To achieve the above objectives, in a first aspect, the present application provides an air conditioner outdoor unit multi-sensor online detection method based on evidence theory, comprising: S1 marks the historical time series working data of multiple air conditioner outdoor units: marks all data with a distance less than a preset distance average quantile as normal, and marks all data with a distance not less than the distance average quantile as abnormal; normalizes the marked historical time series working data; and divides the normalized historical time series working data into a training set and a test set; S2 inputs the training set into multiple XGboost classifiers for training; uses the test set to test the trained XGboost classifiers and outputs multiple groups of test results; S3 fuses the multiple test results based on evidence theory and outputs the test results corresponding to each time point; S4 determines whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold value of the number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

[0006] This application proposes an ensemble learning algorithm that integrates multiple and diverse XGBoost classifiers for a single category. XGBoost combines multiple decision tree models, with each model learning the first-order and second-order gradient information between the outputs of all previous decision trees and the true value, thus avoiding overfitting caused by noise. By combining multiple XGBoost models with different structural parameters through evidence theory, the overfitting problem of individual models is alleviated, resulting in a method with high fault classification accuracy, low time complexity, and good generalization performance.

[0007] Furthermore, in step S1, the method for obtaining the distance quantile from the mean is: S101 calculates the distance between each sample data and all other sample data in the historical time series working data of each air conditioner outdoor unit. The calculation formula is:

[0008] in, dij For the i Sample data and j The spacing of sample data, T is the length of the time series, and each sample data is represented as , xit for t The moment i Sample data, xjt for t The moment j Sample data; S102 sorts the distances from small to large, and uses the following formula to obtain the previous K The average of the distances:

[0009] in, For the front K The average value of the distances, di(k) For the k distance; S103 sorts the average values ​​of all air conditioner outdoor units from small to large and compares them with the experience qualification rate.p The corresponding mean value is used as the distance mean quantile q .

[0010] Furthermore, in step S1, the instance normalization method is used to normalize the marked historical time series work data. The instance normalization formula is:

[0011] in, For historical time series working data i The first time point j The value of the feature, The sample data for the air conditioner outdoor unit is j Features in all n The mean at each time point, The sample data for the air conditioner outdoor unit is j Features in all n The standard deviation at each time point, ε is a small amount to prevent the mean square error from being zero; i is the time series dimension, and ; j is the feature dimension, and .

[0012] Furthermore, in step S2, the step of training the XGboost classifier includes: S201 inputs the training set into each of the XGboost classifiers and outputs a real number vector; S202 converts the real number vector into a probability distribution vector using a softmax function; S203 calculates the loss based on the probability distribution vector using a cross entropy loss function; iterates the training using the loss until a specified number of iterations is reached, and then the training ends.

[0013] Furthermore, in step S203, the cross entropy loss function is:

[0014] in, p ij To be the first i The time point is judged as j The probability of each category is [0,1]; N is the number of all time points in the air conditioner outdoor unit detection process, K is the number of all fault categories; is the indicator function, when the true label 1 if yes, 0 otherwise.

[0015] Furthermore, in step S3, the method for fusing the multiple test results based on evidence theory is: S301 outputs probability distributions of multiple groups of test results and calculates evidence distances between the probability distributions; S302 Obtaining similarity measures of adjacent probability distributions Sim i,j and the support of each probability distribution Sup i , based on the evidence distance, similarity measure Sim i,j and support Sup i Assign a value to the weighted average probability; S303 uses a fusion algorithm to perform an iterative fusion operation on the weighted average probability assignment, and uses the hypothesis corresponding to the maximum weighted average probability value obtained after the fusion operation as the detection result of a single time point; the number of iterative fusion operations is one less than the number of XGboost classifiers.

[0016] In a second aspect, the present application provides a system for implementing the aforementioned multi-sensor online detection method for an air conditioner outdoor unit, comprising: The data processing module is used to mark the historical time series working data of multiple air conditioner outdoor units: mark all data with a distance less than a preset distance average quantile as normal, and mark all data with a distance not less than the distance average quantile as abnormal; normalize the marked historical time series working data; and divide the normalized historical time series working data into a training set and a test set; A model training module is used to input the training set into multiple XGboost classifiers for training; use the test set to test the trained XGboost classifiers and output multiple groups of test results; A test result output module is used to fuse the multiple test results based on evidence theory and output the test result corresponding to each time point; The fault judgment module is used to judge whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

[0017] In a third aspect, the present application provides an electronic device, comprising: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method described in the first aspect.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in any possible implementation of the first aspect.

[0019] In a fifth aspect, the present application provides a computer program product, which, when executed on a processor, enables the processor to execute the method described in any possible implementation of the first aspect.

[0020] It can be understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here.

[0021] In general, the above technical solutions conceived by this application have the following beneficial effects compared with the existing technologies: (1) This application utilizes the multi-sensor collection system in the factory's existing air-conditioning detection system to collect various historical data on the trolley based on the operating characteristics of the air-conditioning outdoor unit and the physical parameters involved in the refrigerant circulation, and labels and trains multiple XGBoost sub-classifiers (i.e., the aforementioned XGboost classifier). Then, through the integration of multiple classifiers using a fusion algorithm based on evidence theory, it achieves rapid online fault detection of large quantities of air-conditioning outdoor units, greatly improves the accuracy of the detection results, and reduces manual participation.

[0022] (2) This application significantly improves the accuracy and reliability of air-conditioning outdoor unit detection. By deeply mining and effectively labeling time series historical data, it can fully capture the characteristic information of air-conditioning outdoor units at different operating stages, providing a rich and accurate data basis for subsequent model training. In the optimization process of the XGBoost basic classification model, the parameters of the model are fine-tuned according to the specific needs of air-conditioning outdoor unit detection, such as reasonably setting the tree depth, learning rate, and number of iterations according to the characteristics of the data, so that it can better fit the data, thereby improving the accuracy and stability of the model and laying a solid foundation for subsequent fusion processing.

[0023] (3) This application proposes to use evidence theory and DS fusion rules to fuse the probability distributions output by multiple XGBoost models with different structural parameters to obtain the final classification results at each time point, thereby enhancing the model's generalization ability for data under different working conditions. Since XGBoost models with different structural parameters can learn and characterize data features from different perspectives, integrating the outputs of these models through evidence theory and DS fusion rules can effectively avoid the limitations and overfitting problems of a single model; when faced with complex and changing working conditions, such as different ambient temperatures, humidity, loads, etc., the fused model can comprehensively consider multiple factors and more accurately judge the operating status of the air conditioner outdoor unit, thereby improving the reliability and adaptability of the model in practical applications, enabling it to better cope with various complex situations and reduce the occurrence of misjudgments and missed judgments.

[0024] (4) This application determines whether the air-conditioning outdoor unit is faulty based on establishing judgment thresholds for each fault category (i.e., the thresholds for the number of historical abnormal time points) from the historical detection process. By analyzing and statistics a large amount of historical detection data, the characteristic threshold ranges corresponding to different fault categories are determined. These thresholds can serve as an important basis for judging whether the air-conditioning outdoor unit is faulty and the type of fault. The fault threshold setting method based on historical data can quickly and effectively judge the real-time operating data of the air-conditioning outdoor unit, discover potential fault hazards in a timely manner, provide strong support for the maintenance and repair of the air-conditioning outdoor unit, reduce the failure rate of the air-conditioning outdoor unit, extend the service life of the outdoor unit, and improve production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a multi-sensor online detection method for an air conditioner outdoor unit based on evidence theory provided in an embodiment of the present application; Figure 2 This is a schematic diagram of DS fusion rules provided in an embodiment of the present application; Figure 3 1 is a schematic diagram comparing the confusion matrix of the XGBoost algorithm provided in the embodiment of the present application without combining evidence theory and the confusion matrix of the present application; Figure 4 It is a distribution histogram of the number of abnormal time points in each sample in the training set provided by the embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0027] The term "and / or" as used herein describes an association between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. The symbol " / " as used herein indicates that the related objects are in an "or" relationship, for example, A / B means either A or B.

[0028] The terms "first" and "second" in this specification and claims are used to distinguish different objects rather than to describe a specific order of objects. For example, "first response message" and "second response message" are used to distinguish different response messages rather than to describe a specific order of response messages.

[0029] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0030] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more, for example, multiple processing units means two or more processing units, etc.; multiple elements means two or more elements, etc.

[0031] The embodiments of the present application are described below in conjunction with the drawings in the embodiments of the present application.

[0032] This embodiment provides an online detection method for air conditioner outdoor unit with multiple sensors based on evidence theory, such as Figure 1 As shown, the following steps are included: S1 labels the historical time series working data of multiple air conditioner outdoor units. The labeling method is as follows: all data in the historical time series working data that are less than a preset distance quantile from the average value are marked as normal, and all data that are not less than a preset distance quantile from the average value are marked as abnormal; the labeled historical time series working data are normalized; and the normalized historical time series working data are divided into a training set and a test set; S2 inputs the training set into multiple XGboost classifiers for training; uses the test set to test the trained XGboost classifiers and outputs multiple sets of test results; S3 integrates multiple test results based on evidence theory and outputs the test results corresponding to each time point; S4 determines whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

[0033] In step S1, the method for collecting the aforementioned historical time series working data is: according to the multi-sensor data acquisition system of the existing factory air-conditioning outdoor unit online detection system, the power, pressure, current and other operating parameters of the air-conditioning outdoor unit are collected during operation, and the collected data are sent to the host computer database and saved as historical time series working data.

[0034] In step S1, a database is first established for all the time series curves of historical data. Assuming that there are N time series curves of air conditioner outdoor unit samples (hereinafter referred to as sample data), the sample data of each air conditioner outdoor unit is represented as , T is the length of the time series, i Indicates the i If there are sample data, the method to obtain the distance quantile from the mean is: S101 calculates the distance between each sample data and all other sample data in the historical time series working data of each air conditioner outdoor unit. The calculation formula is: (1) in, dij For the i Sample data and j The spacing of sample data, T is the length of the time series, and each sample data is represented as , xit for t The moment i Sample data, xjt for t The moment j Sample data; S102 sorts the distances from small to large and uses the following formula to obtain the previous K The average of the distances: (2) in, is the average of the first K distances, di(k) is the kth distance; S103 sorts the average values ​​of all air conditioner outdoor units from small to large and compares them with the experience qualification rate. p The corresponding mean value is used as the distance from the mean quantile q ,Right now .

[0035] In step S1, the instance normalization method is used to normalize the labeled historical time series work data. That is, the labeled sample data is instance normalized according to each detection process to eliminate the difference in the numerical range of each data dimension, accelerate the convergence speed of model training, and enhance the generalization ability of the model. Assume that the input sample data is: (3) Each sample has n time points and M feature dimensions, so the instance normalization formula is: (4) in, x ij For historical time series working data i The first time point j The value of the feature, μ j This is the sample data of the air conditioner outdoor unit j Features in all n The mean at each time point, σ j This is the sample data of the air conditioner outdoor unit j Features in all n The standard deviation at each time point, ε To prevent the mean square error from being zero, i is the time series dimension, and ; j is the feature dimension, and .

[0036] In the aforementioned step S2, the steps of training multiple XGboost classifiers using normalized data include: S201 inputs the training set into each XGboost classifier and outputs a real vector; specifically, the XGboost classifier outputs a real vector for each time point (i.e., sample data) , corresponding to the scores of K categories; S202 uses the softmax function to convert the real number vector into a probability distribution vector P =( p1 , p2 ,..., pK ),in, pK represents the K-th probability distribution vector; the softmax function is: (5) in, h j is the real number vector output by the model for a sample The score corresponding to the jth category, h K is the real number vector output by the model for a sample Corresponding to K The score of a category is a real value. The larger the value, the more likely the sample data is to belong to the corresponding category.

[0037] S203 calculates the loss based on the probability distribution vector using the cross entropy loss function to drive model optimization; iterates the training using the loss until the specified number of iterations is reached, and the training ends. The cross entropy loss function is: (6) in, p ij To be the first i The time point is judged as j The probability of each category is [0,1]; N is the number of all time points in the air conditioner outdoor unit detection process, K is the number of categories; is the indicator function, when the true label 1 if yes, 0 otherwise.

[0038] The first and second derivatives of the loss function with respect to each sample are calculated, and a greedy algorithm is used to iteratively select the features and feature splits with the largest gains to construct XGboost subtrees. Each iteration constructs a new subtree, and each subtree is fitted based on the predictions of the previous subtree and the first and second derivatives of the loss function at the current iteration. Training terminates after the specified number of iterations. After multiple iterations, the predictions of all subtrees are scaled by the learning rate and accumulated as the output of the overall XGboost model.

[0039] In step S3, multiple XGboost classifiers with different structural parameters are combined using evidence theory. The probability distribution results after fusion based on evidence theory are used as the basis for judging whether the air conditioner is normal or faulty, and the corresponding results are obtained. Specifically, the method of fusing multiple groups of test results based on evidence theory is as follows: S301 outputs the probability distribution of multiple groups of test results and calculates the evidence distance between the probability distributions. The evidence distance is used to describe the differences between the evidence bodies. S302 Obtaining similarity measures of adjacent probability distributions Sim i,j and the support of each probability distribution Sup i , based on evidence distance, similarity measure Sim i,j and support Sup i Assign a value to the weighted average probability; S303 uses a fusion algorithm to perform an iterative fusion operation on the weighted average probability assignment, and uses the hypothesis corresponding to the maximum weighted average probability value obtained after the fusion operation as the detection result of a single time point; the number of iterative fusion operations is one less than the number of XGboost classifiers.

[0040] Specifically,m 1 and m 2 is the same identification framework The probability assignment of the two XGboost classifier outputs under the recognition framework It contains the aforementioned XGBoost models with different structural parameters and N mutually exclusive and exhaustive hypotheses. In the air conditioner outdoor unit fault detection task, it assumes two states: normal and fault. Define m 1 and m The distance between 2 is: (7) Where, d BPA(m1,m2) is the evidence distance, is a scalar product and can be expressed as the following formula: (8) in , i,j=1,2,...,N ; A i ,B j For XGboost classifier model A and XGboost classifier model B, the state i and j In the air conditioner outdoor unit fault detection task, since the state only includes fault and normal, if A i ,B j The corresponding assumptions are different. is an empty set, so the previous formula can be simplified to: (9) Define the evidence body as the probability distribution of the output of the XGboost classifier model, then the aforementioned similarity measure Sim i,j is defined as: (10) Because the similarity between two evidence bodies is a mutual relationship, a similarity matrix can be constructed based on the multiple evidence bodies output by multiple XGboost classifiers. From this similarity matrix, we can clearly see the degree of mutual support or correlation between the two evidence bodies. The similarity matrix is ​​defined as: (11) Where D is the number of XGboost classifiers.

[0041] Define each body of evidence m i The support degree Sup supported by other evidence bodies is: (12) Define the credibility of each evidence body Crd as: (13) The improved evidence probability assignment is: (14) The final weighted probability is subjected to D-1 DS rule fusion (Dempster-Shafer rule fusion). The specific fusion process is as follows: Figure 2 As shown in , after the iteration is completed, the final probability distribution can be obtained, and the hypothesis corresponding to the maximum probability is taken as the classification result of a single time point.

[0042] This embodiment selects power, refrigerant pressure and temperature as the data sources collected by the sensor based on the inherent characteristics of the air conditioner, the on-site detection equipment and the important operating status of the air conditioner outdoor unit. When the detection time reaches the specified time, the classification results of each time point are calculated. Finally, whether the air conditioner outdoor unit is faulty is judged based on whether the number of abnormal time points in the entire detection process exceeds the abnormal time point threshold determined in the historical data.

[0043] The following uses the data sample of air conditioner outdoor units in July of a production line as the data source to demonstrate the practical application of the above method. First, the operating data of 13,426 air conditioner outdoor units produced in that month were sampled with a sampling frequency of 1 Hz and a sampling time of 110 seconds. The factory experience pass rate is 0.99, which is the experience pass rate in the previous step. p .

[0044] The data is normalized by the instance normalization operation mentioned in the above step S1. The distance mean quantile is calculated by the Euclidean distance calculation formula q =7.079, all time points of the samples whose average distance is less than q are marked as normal, and all time points of the samples whose average distance is greater than q are marked as normal. q The data of all time points of the samples are marked as abnormal. The labeled data of 3426 air conditioner outdoor units are used as the training set, and the labeled data of the remaining air conditioner outdoor units are used as the test set.

[0045] The labeled data is used to train XGboost classifiers of multiple different structural models to obtain trained XGboost classifiers. In this embodiment, the number D of XGboost classifiers is 6.

[0046] The test set contains sample data from 10,000 air conditioner outdoor units, of which 128 were identified as abnormal by production personnel. Predictions are performed using the six trained XGboost classifiers. The predictions from the six XGboost classifiers are combined according to the aforementioned evidence theory and output as the final prediction result.

[0047] The data of each time point of each air conditioner outdoor unit corresponds to a prediction result, which is the category of the time point, that is, normal or abnormal. The samples are classified according to the number threshold of abnormal time points at all time points. First, the training data is predicted to determine the threshold of the number of fault points. Figure 4 As shown in the figure, among the 3426 training samples, most air conditioner outdoor units contain less than 5 abnormal time points. Therefore, the abnormal time point threshold (i.e., time point failure threshold) is determined to be 5. If the number of abnormal points in each sample is not less than 5, it is judged to be a faulty unit. If the number of abnormal points in each sample is less than 5, it is judged to be a normal unit.

[0048] like Figure 3 Figure a) shows the confusion matrix of the XGBoost algorithm without combining evidence theory. Figure 3 Figure b) shows the confusion matrix obtained in this application. Table 1 below shows the accuracy of the normal and faulty machines in the test set after the evidence theory combination, and compares the detection accuracy of the XGBoost model without the evidence theory combination: Table 1 Comparison of detection accuracy between the XGBoost method and the method of this application

[0049] above Figure 3 The results in Table 1 show that the XGBoost model combined with evidence theory in this application has improved the accuracy of detecting normal machines and faulty machines.

[0050] The following describes the system provided by the present application for implementing the multi-sensor online detection method for the air conditioner outdoor unit as described above. The detection system described below and the detection method described above can be referenced to each other.

[0051] The multi-sensor online detection system for air conditioner outdoor units includes: The data processing module is used to mark all data with a distance less than a preset distance average quantile in the historical time series working data of multiple air conditioner outdoor units as normal, and mark all data with a distance not less than a preset distance average quantile as abnormal; normalize the marked historical time series working data; and divide the normalized historical time series working data into a training set and a test set; The model training module is used to input the training set into multiple XGboost classifiers for training to obtain the trained XGboost classifier model; use the test set to test the trained XGboost classifier and output multiple sets of test results; The test result output module is used to fuse multiple test results based on evidence theory and output the test results corresponding to each time point; The fault judgment module is used to judge whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

[0052] It is understandable that the detailed functional implementation of each of the above units / modules can be found in the introduction of the aforementioned method embodiment, and will not be repeated here.

[0053] It should be understood that the above-mentioned device is used to execute the method in the above-mentioned embodiment. The implementation principle and technical effect of the corresponding program module in the device are similar to those described in the above-mentioned method. The working process of the device can refer to the corresponding process in the above-mentioned method and will not be repeated here.

[0054] Based on the method in the above embodiment, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method in the above embodiment.

[0055] Based on the method in the above embodiment, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method in the above embodiment.

[0056] It is understood that the processor in the embodiments of the present application may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0057] The method steps in the embodiments of the present application can be implemented by hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, mobile hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC.

[0058] The above embodiments can be implemented in whole or in part using software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. A computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that integrates one or more available media. Available media can include magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0059] It will be understood that the various numerical numbers involved in the embodiments of the present application are merely distinctions for the convenience of description and are not intended to limit the scope of the embodiments of the present application.

[0060] It is easy for those skilled in the art to understand that the above is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. An online multi-sensor detection method for air conditioner outdoor units based on evidence theory is characterized by: include: S1 marks the historical time series working data of multiple air conditioner outdoor units: all data whose value is less than a preset distance quantile from the average value are marked as normal, and all data whose value is not less than the distance quantile from the average value are marked as abnormal; Normalize the marked historical time series work data; divide the normalized historical time series work data into a training set and a test set; S2 inputs the training set into multiple XGboost classifiers for training; Using the test set to test the trained XGboost classifier, output multiple groups of test results; S3 fuses the multiple test results based on evidence theory and outputs the test results corresponding to each time point; S4 determines whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold value of the number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

2. The multi-sensor online detection method for an air conditioner outdoor unit according to claim 1, wherein: In step S1, the method for obtaining the distance quantile from the mean is: S101 calculates the distance between each sample data and all other sample data in the historical time series working data of each air conditioner outdoor unit. The calculation formula is: in, dij For the i Sample data and j The spacing of sample data, T is the length of the time series, and each sample data is represented as , xit for t The moment i Sample data, xjt for t The moment j Sample data; S102 sorts the distances from small to large, and uses the following formula to obtain the previous K The average of the distances: in, For the front K The average value of the distances, di(k) For the k distance; S103 sorts the average values ​​of all air conditioner outdoor units from small to large and compares them with the experience qualification rate. p The corresponding mean value is used as the distance mean quantile q .

3. The multi-sensor online detection method for an air conditioner outdoor unit according to claim 1, wherein: In step S1, the instance normalization method is used to normalize the marked historical time series work data. The instance normalization formula is: in, For historical time series working data i The first time point j The value of the feature, The sample data for the air conditioner outdoor unit is j Features in all n The mean at each time point, The sample data for the air conditioner outdoor unit is j Features in all n The standard deviation at each time point, ε is a small amount to prevent the mean square error from being zero; i is the time series dimension, and ; j is the feature dimension, and .

4. The multi-sensor online detection method for an air conditioner outdoor unit according to claim 1, wherein: In step S2, the steps of training the XGboost classifier include: S201 inputs the training set into each of the XGboost classifiers and outputs a real number vector; S202 converts the real number vector into a probability distribution vector using a softmax function; S203 calculates the loss based on the probability distribution vector using a cross entropy loss function; iterates the training using the loss until a specified number of iterations is reached, and then the training ends.

5. The multi-sensor online detection method for an air conditioner outdoor unit according to claim 4, characterized in that: In step S203, the cross entropy loss function is: in, p ij To be the first i The time point is judged as j The probability of each category is [0,1]; N is the number of all time points in the air conditioner outdoor unit detection process, K is the number of all fault categories; is the indicator function, when the true label 1 if yes, 0 otherwise.

6. The multi-sensor online detection method for an air conditioner outdoor unit according to claim 1, wherein: In step S3, the method for fusing the multiple test results based on evidence theory is: S301 outputs probability distributions of multiple groups of test results and calculates evidence distances between the probability distributions; S302 Obtaining similarity measures of adjacent probability distributions Sim i,j and the support of each probability distribution Sup i , based on the evidence distance, similarity measure Sim i,j and support Sup i Assign a value to the weighted average probability; S303 uses a fusion algorithm to perform an iterative fusion operation on the weighted average probability assignment, and uses the hypothesis corresponding to the maximum weighted average probability value obtained after the fusion operation as the detection result of a single time point. The number of iterative fusion operations is one less than the number of XGboost classifiers.

7. A system for implementing the multi-sensor online detection method for an air conditioner outdoor unit according to any one of claims 1 to 6, characterized in that: include: A data processing module is used to mark the historical time series working data of multiple air-conditioning outdoor units: all data whose value is less than a preset distance quantile from the average value is marked as normal, and all data whose value is not less than the distance quantile from the average value is marked as abnormal; Normalize the marked historical time series work data; divide the normalized historical time series work data into a training set and a test set; A model training module is used to input the training set into multiple XGboost classifiers for training; Using the test set to test the trained XGboost classifier, output multiple groups of test results; A test result output module is used to fuse the multiple test results based on evidence theory and output the test result corresponding to each time point; The fault judgment module is used to judge whether the number of time points corresponding to abnormal results in the detection results is not less than the threshold number of historical abnormal time points: if so, the air conditioner outdoor unit is faulty; if not, the air conditioner outdoor unit is normal.

8. An electronic device, characterized in that: include: at least one memory for storing a computer program; At least one processor is used to execute the program stored in the memory. When the program stored in the memory is executed, the processor is used to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed on a processor, the processor is caused to execute the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that When the computer program product is run on a processor, the processor is enabled to perform the method according to any one of claims 1 to 6.

Images