Energy consumption location and fault prediction method of cooling and heating equipment under EMS energy consumption management system

Abstract

The present invention provides a method for locating and predicting energy consumption of heating and cooling equipment under an EMS energy consumption management system, including S1, collecting the original data of the target equipment status to construct a comprehensive evaluation data set, using the source data extraction function to match the indicators to the fuzzy set to form a fuzzy vector, and generating equipment-level time series data by clustering, sorting and identifying its important features; S2, training a health score prediction model and a fault classification model based on the time series data vector; S3, using the training model to perform health scoring and fault classification prediction on real-time data to generate early warning information. The present invention further screens important features using a random forest algorithm by defining fuzzy sets and source data extraction functions and extracting key features through clustering analysis, which not only improves the accuracy of target equipment performance evaluation, but also provides a more accurate basis for subsequent fault diagnosis, thereby simplifying the model and improving analysis efficiency, and bringing significant economic benefits to corporate partners.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial energy consumption identification and analysis, and in particular to a method for locating energy consumption and predicting faults of heating and cooling equipment under an EMS energy consumption management system. Background Art

[0002] With the advancement of industrial and smart city construction, smart energy management has become an important means to improve energy utilization efficiency and achieve energy conservation and emission reduction. In modern industrial and commercial facilities, heating and cooling equipment (such as air-conditioning systems, refrigeration units, heating equipment, etc.) are key operating infrastructure. These devices are in operation for a long time and often face complex usage environments (such as temperature, humidity, vibration, electrical fluctuations, etc.) and diverse operating requirements. Their stable operation directly affects production efficiency, energy consumption and user experience. However, due to the complex structure and changeable operating conditions of heating and cooling equipment, real-time monitoring of its operating status, rapid fault location and intelligent early warning have always been the core challenges faced by the industry. In response to this, although the existing energy management system has a certain degree of data collection, analysis and partial visualization functions, it still has the following defects and shortcomings:

[0003] 1) Limitations of data integration and traditional monitoring display: Existing systems often lack effective integration of factory floor plans, power distribution diagrams, and real-time energy consumption data, which makes it difficult for managers to intuitively understand energy consumption patterns from a spatial dimension and to quickly locate energy consumption hotspots and potential energy-saving points. Current energy consumption data is usually only presented in the form of static reports or simple charts. There is a lack of real-time updated visualization methods such as dynamic line charts, making it difficult to track energy consumption trends in real time and respond to energy usage anomalies in a timely manner. In layman's terms, most heating and cooling equipment data only rely on the threshold of a single parameter (such as temperature or current) to determine the operating status, making it difficult to capture the complex operating modes and potential failures of the equipment. Traditional monitoring equipment lacks intelligence and cannot achieve multi-dimensional analysis of equipment operation and fault prediction.

[0004] 2) Incomplete monitoring equipment status information: The key metering monitoring of heating and cooling equipment (such as air-conditioning systems, refrigeration units, and heating equipment) mainly focuses on total statistics. There is a lack of real-time monitoring and visual display of equipment working status and performance parameters, making it difficult to detect equipment failures or performance degradation at an early stage, making it difficult for managers to quickly identify changes in equipment status.

[0005] 3) Delay in alarm location and response: When an abnormality occurs in the energy system or equipment fails, the existing system can only provide text alarm information, lacking an intuitive mapping with the actual physical space. As a result, managers need to spend extra time locating the source of the problem when handling the alarm. At the same time, after the failure occurs, relying on manual experience and gradual investigation can easily lead to extended maintenance time. Some early hidden dangers cannot be discovered in time, resulting in increased equipment damage or even systemic failures. Especially in large buildings or industrial facilities, the distribution of heating and cooling equipment is usually relatively wide and large in number, making it difficult to achieve real-time centralized monitoring using traditional methods. At the same time, the alarm information is not linked enough with the factory floor plan and power distribution diagram, affecting the timeliness and accuracy of fault response.

[0006] Based on this, it is urgent to propose a cooling and heating equipment operation status monitoring and fault prediction solution based on multi-sensor data acquisition, deep learning modeling and intelligent early warning analysis, and to achieve comprehensive monitoring and predictive maintenance of the equipment operation status by building a complete monitoring, analysis and decision support system. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention aims to provide a method for locating energy consumption and predicting faults of heating and cooling equipment under an EMS energy consumption management system. In order to achieve the above-mentioned purpose, the present invention is implemented through the following technical scheme: a method for locating energy consumption and predicting faults of heating and cooling equipment under an EMS energy consumption management system, comprising:

[0008] S1. Device-level time series data generation:

[0009] Collecting raw data representing the state of the target equipment, matching and formatting the raw data according to five dimensions including equipment information, health status, fault diagnosis, warning information and maintenance suggestion related indicators, creating a comprehensive evaluation data set based on the matched and formatted data, and on the basis of the comprehensive evaluation data set, matching each related indicator to a corresponding fuzzy set by constructing a source data extraction function, determining the degree to which each related indicator belongs to different fuzzy sets, calculating the fuzzy coefficient of each fuzzy set, and obtaining a fuzzy vector quantifying the operating state of each target equipment;

[0010] The fuzzy vectors are clustered using a fuzzy C-means clustering algorithm to form different feature clusters that fully cover the original data and characterize the key features of the target device, importance scores are calculated for different feature clusters, all feature clusters are sorted according to the score results, and within each feature cluster, the indicator feature that has the greatest impact on the operating status of the target device is selected and identified by setting a correlation threshold, and recorded as an important feature;

[0011] The selected important features and the remaining features in this feature cluster are sorted according to the rule of first between clusters and then within clusters to form the final device-level time series data;

[0012] S2. Model training

[0013] Based on the device-level time series data, a time series data vector of each target device is constructed, and a health score prediction model and a fault classification model are trained respectively. A loss function is defined to measure the difference between the predicted health score and fault classification and the actual label, and then the Adam optimizer is used to optimize the two model parameters to minimize the loss function.

[0014] S3. Warning information output

[0015] Through the trained model, the real-time data of the target equipment is evaluated for health scores and fault classification predictions, and early warning information is generated based on the prediction results.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. In order to solve the problems of incomplete data integration, equipment performance evaluation, and delayed alarm positioning and response in the prior art, the present invention adopts a combination of multidimensional data comprehensive analysis methods and fuzzy logic. In terms of data processing, by defining fuzzy sets and source data extraction functions, the EMS energy consumption management system can comprehensively consider a variety of sensor data including temperature, pressure, current, and vibration, and comprehensively evaluate the health status, maintenance recommendations, fault diagnosis, early warning information, and the integrity and accuracy of equipment information. For example, the health status of the equipment is divided into three fuzzy sets of high, medium, and low. The degree of membership of each device in different fuzzy sets can be calculated based on the relationship between sensor data and semantic parameters, thereby obtaining a comprehensive fuzzy coefficient to quantify the health status of the equipment;

[0018] 2. The present invention uses the extreme value method to perform homogeneous quantitative processing, eliminating the dimensional differences between indicators of different dimensions, so that data of different dimensions including health status, maintenance recommendations, and energy consumption can be compared and comprehensively evaluated under the same standard, thereby improving the flexibility and accuracy of subsequent data processing;

[0019] 3. In terms of fault warning, the present invention identifies abnormal patterns in the operation of equipment through a comprehensive analysis method of multidimensional data, extracts key features through cluster analysis, and further screens important features using a random forest algorithm to form a feature sequence, which not only improves the accuracy of target equipment performance evaluation, but also provides a more accurate basis for subsequent fault diagnosis. For example, when evaluating the performance of the combustion chamber under the EMS energy management system, it can be found through cluster analysis that the membership of temperature and combustion efficiency is the most important for distinguishing combustion chambers with different performances, while the influence of pressure and emission levels is relatively small. Therefore, the data can be reduced in dimensionality, and only the membership of temperature and combustion efficiency is retained, thereby simplifying the model and improving analysis efficiency, thereby bringing significant economic benefits to corporate partners. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In the drawings, the same reference numerals are used to refer to the same components. Among them:

[0021] Figure 1 It is a flow chart of a method for locating energy consumption and predicting faults of cooling and heating equipment under an EMS energy consumption management system proposed in one embodiment of the present invention;

[0022] Figure 2 A schematic diagram of the process of outputting warning information proposed in one embodiment of the present invention;

[0023] Figure 3 A schematic diagram of a main control panel of an EMS energy consumption management system proposed in one embodiment of the present invention;

[0024] Figure 4 The main control panel interface of the EMS energy consumption management system is divided into two parts, the left part is the power distribution diagram area, and the right part is the total incoming line energy consumption overview area, which is a schematic diagram showing the real-time operating status of each device circuit;

[0025] Figure 5 It is a schematic diagram of generating early warning information according to prediction results proposed in one embodiment of the present invention. DETAILED DESCRIPTION

[0026] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific implementation modes and drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as the whole of the present invention or as a limitation or restriction to the technical solution of the present invention.

[0027] The present invention is further described in detail below with reference to the accompanying drawings, but is not intended to limit the present invention.

[0028] like Figure 1 - Figure 2 As shown, as an embodiment of the present invention, the proposed method for locating and predicting energy consumption of heating and cooling equipment under the EMS energy consumption management system adopts a combination of multidimensional data comprehensive analysis methods and fuzzy logic. By defining fuzzy sets and source data extraction functions, the EMS energy consumption management system can comprehensively consider a variety of sensor data, comprehensively evaluate the health status, maintenance recommendations, fault diagnosis and other indicators of the equipment, and use the extreme value method to perform homogeneous quantitative processing to eliminate the differences between indicators of different dimensions, which can improve the flexibility and accuracy of data processing. Finally, in terms of fault warning, key features are extracted through multidimensional data comprehensive analysis and cluster analysis, and important features are screened using the random forest algorithm to form a feature sequence, thereby improving the accuracy of equipment performance evaluation and the accuracy of fault diagnosis, and bringing significant economic benefits to the enterprise. Including specific steps:

[0029] S1. Device-level time series data generation:

[0030] S1-1. Collect raw data that represents a comprehensive view of the target equipment status. The raw data includes data describing the equipment number, location, health score, fault classification, severity level, and warning information. It can be understood that the raw data set is the starting point for subsequent evaluation and decision-making methods, and the purpose is to provide a comprehensive view of the equipment status.

[0031] S1-2, match and format the original data according to five dimensions including equipment information, health status, fault diagnosis, warning information and maintenance recommendation indicators. It should be noted that this step is a key step in converting complex data sets into a form that is easy to analyze and understand. The purpose is to ensure the consistency and comparability of the original data and provide a standardized data foundation for subsequent feature assignment and model training. The specific implementation process includes:

[0032] S1-21. Integrate the data transmitted back to the EMS energy management system from multiple preset sensors, including at least temperature data, pressure data, current data, vibration data and equipment location information, and match the data with preset semantic parameters to achieve quantitative analysis of the operating status of the target equipment. It should be noted that semantic parameters usually refer to the standards or attributes used to interpret and evaluate data sets in a specific field. Among them,

[0033] For the device information R of the target device 1 have: (1), where L r , O r Score the accuracy of the rth device number and location information, aL 、a 0 is the weight of the device number and location information, i is the number of device information, and the semantic parameter can be understood as L 、a 0 relevant, i.e., if location information is critical to equipment management and scheduling, then a 0 should be assigned a "higher" value;

[0034] For the health status H of the target device 2 have: (2), where W T , W P , W I , W V are the weights of temperature, pressure, current and vibration data respectively, n is the preset number of sensors, T k , P k ,I k 、V k are the temperature, pressure, current and vibration sensor data of the kth sensor respectively. Similarly, the semantic parameters at this time can be understood as T , W P , W I , W V If temperature has the greatest impact on equipment health, then W T should be assigned a higher value;

[0035] Fault diagnosis for the target device 3 have: (3), where F a is the score of the ath fault category, S a is the score of the ath severity level, d F d S are the weights of fault classification and severity level, respectively, and p is the number of fault diagnoses. Similarly, the semantic parameters at this time can be understood as F d S If a fault with a higher severity level has a greater impact on the equipment, then d S should be assigned a higher value;

[0036] For the warning information W of the target device 4 have: (4), where W b is the score of the e-th warning information, t W is the weight of the warning information, e is the number of warning information, and the semantic parameter at this time can be understood as W If timely warning can avoid potential serious failure, then t W should be assigned a higher value;

[0037] Maintenance recommendations for this target device M 5 have: (5), where m is the number of maintenance recommendations, M j Score the jth maintenance suggestion, S m To maintain the suggestion weight, the semantic parameter at this time can be understood as m If certain maintenance recommendations can significantly improve the reliability of the equipment, then S m Should be assigned a higher value.

[0038] S1-3. Create a comprehensive evaluation data set based on the matched and formatted data. The specific operations are as follows:

[0039] S1-31. Integrate the multiple dimensional data of each target device into a table, where each row represents a target device, each column represents the data of an indicator dimension, and add time information; assign features to each column in the table to ensure that the data in each column is correctly identified and assigned a clear name and unit, so that in subsequent model training and evaluation, we can know what the data in each column represents and how to interpret the model results.

[0040] Based on the above technical concept, it should be noted that in the data formatting stage, the original data needs to be converted into a unified and standardized format for subsequent processing. In the specific implementation, for each dimension of data, a clear label needs to be defined, and the format and unit of the data must be consistent. For example, the unit of quantitative analysis of health status is "points", the unit of quantitative analysis of maintenance recommendations is also "points", the unit of quantitative analysis of service life limit is "years", the unit of quantitative analysis of the number of failures is "times", and the unit of quantitative analysis of energy consumption is "kilowatt-hour / kWh".

[0041] Example: Assume the original data is as follows Table 1:

[0042] When converted into quantitative scores, the following Table 2 is obtained:

[0043] S1-32, create a comprehensive evaluation data set D for each target device, including a complete data record of all relevant indicators and time information,

[0044] D={(H 1 ,M 1 ,Y 1 ,F 1 ,E 1 ,W 1 ),(H 2,M 2 ,Y 2 ,F 2 ,E 2 ,W 2 ),...,(H n ,M n ,Y n ,F n ,E n ,W n )}(6), where H i is the quantitative analysis of the health status of the i-th device (points), M i Quantitative analysis of maintenance recommendations for the i-th equipment (points), Y i is the limited analysis of the service life of the i-th equipment (year), F i is the quantitative analysis of the number of failures of the i-th device (times), E i is the energy consumption quantitative analysis of the i-th equipment (kWh), W i is the quantitative analysis of the warning level of the i-th device (points), and n is the total number of target devices. The comprehensive evaluation data set D obtained based on the above example is as shown in Table 3:

[0045] Then, for the first target device, the elements in the comprehensive evaluation data set D can be expressed as: D1 = (90, 0, 5, 2, 1200, 10, 2024-4-11). Based on the concept of step S1-32, although the collected data has been formatted according to certain standards, the consistency and comparability of the data are ensured.

[0046] However, in actual applications, the data generated by heating and cooling equipment during operation are often accompanied by uncertainty and ambiguity. This is because the heating and cooling equipment monitoring system involves a variety of sensor data, such as temperature, pressure, current and vibration, which directly leads to the difficulty in processing multi-source heterogeneous data. For example, in a large commercial building, the production efficiency of a manufacturing plant is being evaluated. The heating and cooling equipment monitoring system collects data about the air-conditioning unit through multiple sensors, including temperature sensors, pressure sensors, current sensors and vibration sensors. Data, such as machine failure rate (times / month), production speed (pieces / hour) and energy consumption (kilowatt-hours / hour), are used to evaluate the health status of the air-conditioning unit and predict potential failures. If existing technologies are used, such as the monitoring system only relying on the threshold of a single parameter (temperature) to judge the operating status (essentially, the dimensions of these indicators are different and there are uncertainties, such as the machine failure rate may change due to maintenance), it will ignore abnormalities in other sensor data, such as increased vibration or abnormal current. In this case, even if the air-conditioning unit has shown signs of performance degradation, the system may not be able to issue an early warning in time, resulting in the failure being discovered only after it occurs. The common approach at present is to simply add or average the values ​​of these indicators. In this case, the high value of production speed will mask the negative impact of machine failure rate and energy consumption. In addition, if the energy consumption data of a certain month is missing, the manager / EMS energy consumption management system cannot effectively evaluate the production efficiency of that month.

[0047] Based on this, it is necessary to continue to perform the following steps: S1-4, based on the comprehensive evaluation data set, implement a multi-dimensional data comprehensive analysis method. Specifically including:

[0048] S1-41, first construct the source data extraction function G Q,P (X), match each relevant indicator Q to the corresponding fuzzy set P, and conduct a comprehensive assessment of the health status of the equipment. It can be understood that this process allows the EMS energy consumption management system to comprehensively consider all sensor data, such as vibration and current data, to identify abnormal patterns and issue early warnings in a timely manner.

[0049] S1-42. Use homogeneous quantification to eliminate the dimensional differences between indicators of different dimensions and ensure that different related indicators can be compared under a unified standard.

[0050] S1-43. Finally, by extracting the function from the source data and the quantified data, the fuzzy coefficient of each fuzzy set is calculated by determining the degree to which each relevant indicator belongs to different fuzzy sets, and a fuzzy vector that quantifies the operating status of each target device is obtained. It can be understood that this fuzzy vector provides a quantitative performance evaluation for each device, which helps to comprehensively evaluate the status or performance of the device and provide support for decision making and resource allocation.

[0051] As an embodiment of the present invention, the specific implementation of the above step S1-41 is as follows: based on the obtained comprehensive evaluation data set D, determine the fuzzy set P to which each relevant indicator Q belongs, and define each different fuzzy set to represent a different performance or state level of the target device, wherein when each relevant indicator reaches the optimal level, it means that the target device is in the best operating state, and the fuzzy set at this time represents the high performance state H PQ ,

[0052] H PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|H i ≥90,M i =0,Y i ≥5,F i =0,E i ≤1200,W i =10}(7), where H i ≥90, indicating that the health status score is at least 90 points, and the target device is in the best health state. i =0, indicating that the maintenance score is 0, no maintenance is required, Y i ≥5, the target device has been in use for at least 5 years, indicating that the target device has been running stably for a period of time. i =0, indicating no fault, E i ≤1200, indicating that the energy consumption does not exceed 1200kwh, and the target device has the lowest energy consumption, W i =10, indicating no warning. When some related indicators perform well, it means that the target device is in a good operating state. The fuzzy set at this time represents a good performance state G PQ ,G PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|70≤H i <90,M i ≥50,3≤Y i <5,F i =1,1000≤E i <1200,W i ≥5}(8), where 70≤H i <90, indicating that the target device is in good operating condition, M i ≥50, indicating that local maintenance is required, 3≤Y i<5, the target equipment is between 3 and 5 years old, which means the target equipment has been in operation for a period of time. i =1, indicating 1 fault occurred, 1000≤E i <1200, indicating that the energy consumption is between 1000 and 1200 kwh, and the target equipment has medium energy consumption. i ≥5, indicating that attention should be paid to the early warning. It should be noted that the energy consumption does not exceed 1200 kWh in H PQ is defined as “low energy consumption” in G PQ The energy consumption of medium-sized vehicles is between 1000 and 1200 kWh, which means that although the upper limit of energy consumption of both categories is 1200 kWh, the energy consumption of H PQ The energy consumption standard is "not to exceed", which means that PQ In the G PQ The energy consumption range starts from 1000 kWh, which indicates that even at the lower end of this range, the energy consumption is considered medium. When some related indicators need attention, it means that the target device is in a medium operating state. The fuzzy set at this time represents the medium performance state M PQ ,

[0053] M PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|50≤H i <70,M i >0,1≤Y i <3,F i =2,1100≤E i <1300,W i >0}(9), where 50≤H i <70, indicating that the target device is operating within the preset acceptable range, M i >0, indicating maintenance is required, 1≤Y i <3, the target equipment is between 1 and 3 years old, indicating that the target equipment is new. i =2, indicating that two faults have occurred and attention is required, 1100≤E i <1300, indicating that the energy consumption is between 1100 and 1300 kwh, and the target equipment has a high energy consumption. i >0, indicating that there is a warning; when some related indicators decrease, it means that the target device is in a poor operating state, and the fuzzy set at this time represents a low performance state L PQ ,L PQ ={(Hi ,M i ,Y i ,F i ,E i ,W i )|H i <50,M i ≥100,Y i ≤1,F i ≥3,E i ≥1300,W i =0}, where H i <50, indicating that the target device is in a poor operating state, M i ≥100, indicating that a lot of maintenance is required, Y i ≤1, the target device is no more than 1 year old, indicating that the target device is new or frequently replaced. i ≥3, indicating that at least 3 failures occurred, and the target device failed frequently. i ≥1300, indicating that the energy consumption exceeds 1300kwh, and the target device has high energy consumption. i =0, indicating no warning; when each relevant indicator is at its lowest, it means that the target device is in a faulty operation state, and the fuzzy set at this time represents the fault state F PQ ,

[0054] F PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|H i <30,M i ≥100,Y i ≤7,F i ≥3,E i ≥1300,W i <10}(10), where H i <30, indicating that the target device is in a faulty operating state and has serious problems. i ≥100, indicating that immediate and comprehensive maintenance is required, Y i ≤7, the target device is no more than 7 years old, indicating that the target device is an aging device. i ≥3, indicating that at least 3 failures occurred, and the target device failed frequently. i ≥1300, indicating that the energy consumption exceeds 1300kwh, and the target device has high energy consumption. i <10, indicating a serious warning.

[0055] The specific implementation of the above step S1-42 is: for each relevant indicator Q, calculate its homogeneity quantization value Ω iq , in order to map all data into the interval [0,1] to reduce its impact on subsequent overall evaluation, (11), where X iq is the value of the qth related indicator of the i-th target device, min(X q ) and max(X q ) are the minimum and maximum values ​​of the qth related index respectively.

[0056] The specific operation of obtaining the fuzzy vector for quantifying the operating status of each target device in the above step S1-43 is as follows: First, according to the quantized Ω iq , calculate the fuzzy coefficient C i , (12), where C i is the fuzzy coefficient of the i-th target device, indicating the comprehensive membership degree of the target device in all fuzzy sets, G q (Ω iq ) is the membership function value of the qth related index in the fuzzy set, indicating the membership degree of the i-th target device in the fuzzy set. is the weight of the qth related index in the i-th target device, reflecting the importance of the related index in the quantitative analysis of the operating status of the target device. Secondly, all fuzzy coefficients are integrated into a fuzzy vector, C=(C 1 ,C 2 ,C 3 ,...,C i ).

[0057] Example: Data Table 2 is still obtained based on the above quantitative scoring.

[0058] a. Data formatting: Data on the health status of the equipment has been collected and the qualitative description has been converted into a quantitative score, such as the health status score of "excellent" = 90 points, "good" = 70 points, and "poor" = 30 points.

[0059] b. Determine the fuzzy set and source data extraction function G Q,P (X): Define three fuzzy sets, representing the health status of the equipment: high (H), medium (M), and low (L). Since the source data extraction function G Q,P (X) can be linear, so for the health status score, if 80 points and above are high (H), 60-79 points are medium (M), and 40-59 points are low (L).

[0060] c. Homogeneous quantification: In this example, there are three indicators: health status (H), maintenance recommendation (M), and energy consumption (E), and their dimensions are different. Therefore, the extreme value method is needed for homogeneous quantification. For example, if the highest score of health status is 100 and the lowest score is 0, the homogeneous quantification value of 90 points is (90-0) / (100-0)=0.9.

[0061] d. Calculate the fuzzy coefficient: For each target device, calculate the fuzzy coefficient of each fuzzy set. For example, for health status, if a device scores 90 points, then its source data extraction function G in the high (H) fuzzy set Q,P (X) value is 0.9. If the score of the maintenance recommendation is 0, then its information extraction function value in the low (L) fuzzy set is 1 (because it is exactly the upper limit of the low fuzzy set). If the energy consumption is 1200 kWh, assuming that the high (H) fuzzy set of energy consumption is above 1300 kWh, the medium (M) fuzzy set is 1000-1299 kWh, and the low (L) fuzzy set is below 999 kWh, then the information extraction function value of 1200 kWh in the medium (M) fuzzy set is 1.

[0062] e. Comprehensive evaluation: Assign weights to each fuzzy set. For example, the weight of the high (H) fuzzy set of health status is 0.5, the weight of the medium (M) fuzzy set is 0.3, and the weight of the low (L) fuzzy set is 0.2. Then for the health status, the fuzzy coefficient C is obtained. H =0.9*0.5=0.45. For maintenance recommendations, we get the fuzzy coefficient C M =1*0.2=0.2. For energy consumption, we get the fuzzy coefficient C E =1*0.2=0.2.

[0063] f. Based on this, we get the fuzzy vector: All fuzzy coefficients are integrated into a fuzzy vector, expressed as C = (C H ,C M , C E ) = (0.45, 0.15, 0.2). At this point, the fuzzy vector can be used as a quantitative indicator to evaluate / analyze the overall performance of the target equipment in the EMS energy management system.

[0064] Based on the above technical concept, the formed fuzzy vector contains multiple dimensions. Therefore, continuing to perform cluster analysis can help identify and reduce these dimensions and extract key features, such as Figure 3 - Figure 5As shown, there is a set of combustion chamber performance data, each combustion chamber is represented by four dimensions of features: temperature, pressure, combustion efficiency and emission level. When these features are converted into fuzzy vectors, each dimension has three possible fuzzy set memberships (such as "low temperature", "medium temperature" and "high temperature" for temperature). In this way, the fuzzy vector of each combustion chamber has 12 dimensions (4 features * 3 memberships of each feature). At this time, identifying which dimensions are most critical for distinguishing target devices of different groups, while reducing the complexity of the data and extracting the features that best represent the performance of the equipment are necessary analysis steps.

[0065] Therefore, it is necessary to perform the following steps S1-5. Through cluster analysis, it can be found that the membership of temperature and combustion efficiency is the most important for distinguishing the performance of the combustion chamber, while the influence of pressure and emission level is relatively small. Therefore, in the subsequent calculation, as long as the data is reduced in dimension and only the membership of temperature and combustion efficiency is retained, the model can be simplified and the efficiency of analysis can be improved.

[0066] In one embodiment of the present invention, step S1-5 includes:

[0067] S1-51. Use the fuzzy C-means clustering algorithm to cluster the fuzzy vectors to form different feature clusters X that fully cover the original data and characterize the key features and behavior patterns of the target device ik ;

[0068] S1-52, respectively calculate the importance scores for different feature clusters, sort all feature clusters according to the score results, and select and identify the indicator features that have the greatest impact on the operating status of the target device within each feature cluster by setting a correlation threshold, and record them as important features;

[0069] S1-53, sort the selected important features and the remaining features in this feature cluster according to the rule of first between clusters and then within clusters to form the final device-level time series data.

[0070] The specific implementation of the above step S1-51 is as follows: Step 1, select the number of clusters K, and randomly initialize k cluster centers U, U = {U 1 ,U 2 ,U 3 ,...,U k}; Step 2, for each fuzzy vector C i , calculate its belonging to each cluster center U k The membership degree V ij , (13), where m is the fuzzification parameter, and m>1, which is used to control the fuzziness of the membership degree, ||C i −U l|| is the Euclidean distance between the i-th fuzzy vector and the l-th cluster center. The smaller the distance, the greater the membership degree. Step 3, according to the membership degree update formula, (14), update the cluster center, and repeat the above steps 2 and 3 until the change of the cluster center is less than the preset threshold or the maximum number of iterations is reached, where n is the total number of fuzzy vectors; step 4, according to the clustering results, assign the fuzzy vectors to the corresponding cluster centers to form different feature clusters X ik , at this time, different feature clusters X ik It contains the fuzzy vector with the highest membership degree to a specific cluster center, and based on this clustering result, the combustion chambers with high performance and the combustion chambers with low performance can be identified.

[0071] Example: Assuming k=2 cluster centers are selected, the algorithm will group the combustion chambers according to the similarity of membership. For example, the clustering results may show that combustion chambers A and B belong to the "high performance" category, while combustion chamber C belongs to the "low performance" category. At this point, it can be identified that combustion chamber C needs to be improved, and the combustion process needs to be adjusted or the equipment needs to be maintained. It can be understood that at this time, the different feature clusters X ik The fuzzy vector containing the highest membership degree with a specific cluster center can be understood as considering all performance indicators to evaluate the performance of the combustion chamber. The fuzzy logic proposed in the present invention allows the administrator / EMS energy consumption management system to handle uncertainty. By converting performance indicators into membership degrees, performance data can be processed more flexibly. Then, through cluster analysis, the combustion chamber with degraded performance can be identified, and early warnings can be issued in time to prevent the occurrence of failures.

[0072] As an embodiment of the present invention, in order to evaluate the above clustering effect, it is also necessary to select the optimal cluster number K for verification, that is, calculate the Davies-Bouldin Index (DBI) index of each cluster, select the cluster number k that minimizes the DBI index value as the optimal cluster number K∗, and then perform the following steps S1-52 based on the optimal cluster number K∗ determined after verification, and use the random forest algorithm to evaluate each feature cluster X ik Reflects the contribution of the performance or status of the target device, thereby ensuring that each feature cluster X ik Accurately reflect the actual performance or status of the target device. That is, the final data expected to be obtained is the feature cluster X to which each target device belongs. ik , each feature cluster X ik It can fully cover the equipment number, location, health score, fault classification, severity level and warning information in the original data set. ik It is a simplified representation of the high-dimensional features of the original data after cluster analysis, representing the key features (important features) of the target device.

[0073] The specific implementation of S1-52 includes:

[0074] S1-521, feature cluster X ik Evaluation of the random forest algorithm: First, train the random forest model without replacing any feature clusters in the random forest model, and record the number of out-of-bag error samples E1 of the random forest model Ci , taking this as the baseline error rate, and then replacing each feature cluster X one by one in the random forest model ik The data is randomly disrupted to randomly shuffle the feature cluster X ik , and keep other feature clusters X ik The data remains unchanged, destroying the original recorded information; again, train the random forest model and record the new number of out-of-bag error samples E2 Ci ; Finally, if the error rate of the model increases after permutation, it indicates that the feature cluster X ik The data makes an important contribution to the predictive ability of the model and can accurately reflect the actual performance or status of the target device.

[0075] S1-522, feature cluster X ik Importance score calculation and feature selection: First, according to the permutation results, by comparing the number of out-of-bag error samples E1 before permutation Ci And the number of out-of-bag error samples after replacement E2 Ci , using the formula V I (X ik )=E1 Ci −E2 Ci , calculate the importance score, where the importance score V I (X ik ) is higher, indicating that the feature cluster X ik The greater the contribution of the data to the actual performance or status of the target device; secondly, according to this score V I (X ik ) for all feature clusters X ik Sort and in each feature cluster X ik Calculate the correlation between a single feature and classification information P xy ; Again, set the correlation threshold yh, only when the correlation between a single feature and the classification information P xy When the correlation threshold yh is exceeded, the feature is considered important; finally, from each feature cluster X ik Select the features that have an impact on the contribution as important features, among which the correlation P xy The calculation formula is: (15), where P xyis the Pearson correlation coefficient between variables x and y, which is used to measure the linear correlation between two variables. Its value is between -1 and 1. P xy 1 indicates a completely positive correlation, -1 indicates a completely negative correlation, and 0 indicates no linear correlation. x and y are respectively ik A dataset with single features and classification information. and is the mean of variables x and y, , , x i and i are the i-th observation values ​​of variables x and y, respectively. It is the covariance calculation between variables x and y, which indicates the degree of joint change of the two variables. The classification information is the target device performance indicator used to distinguish different categories or states.

[0076] So far, through steps S1-521~S1-522, we can not only identify the feature cluster X that contributes significantly to the classification performance ik , and can also be in each feature cluster X ik The key features (important features) are further screened out, which helps to improve the prediction accuracy of subsequent prediction models and reduce the complexity of the models.

[0077] The specific implementation of the above steps S1-53 is as follows: first, for each target device i, at time step t, the selected important features and the remaining features are sorted according to the rule of first between clusters and then within clusters; second, the sliding average filter denoising is used to eliminate short-term fluctuations and random noise; finally, according to the sorting results, the final device-level time series data X is formed. processed,t ={d t,1 ,d t,2 ,d t,3 ,d t,I}, where d t,I =[L i ,T t,i smooth ,P t,i smooth ,I t,i smooth ,V t,i smooth ], where L i is the location and / or number and / or room number information of the target device i, T t,i smooth ,P t,i smooth ,I t,i smooth ,V t,i smoothare the smoothed values ​​of the temperature, pressure, current, and vibration sensor data of the target device i at time step t, respectively.

[0078] Example: Assume that in the data of the above combustion chambers A, B, and C, it has been found through random forest evaluation that the feature clusters of combustion efficiency and temperature have a significant impact on the classification performance. Therefore, this feature is selected as an important feature. When forming the feature sequence, the feature differences between different combustion chambers are first considered, and then the importance of the features in the same combustion chamber is considered. You can get: the feature sequence of combustion chamber A: [combustion efficiency-high efficiency, temperature-high temperature, emission-low emission, pressure-medium pressure], the feature sequence of combustion chamber B: [combustion efficiency, temperature, emission, pressure], the feature sequence of combustion chamber C: [combustion efficiency, temperature, emission, pressure]. Based on the above data, it can be understood that in the feature cluster X ik In addition to the combustion efficiency and temperature that have been identified as important features, the remaining features include emissions, pressure and other features that may affect the performance of the equipment. Although these remaining features are not identified as key features in the random forest evaluation, they still contain valuable information and may contribute to the evaluation of equipment performance in some cases. Therefore, through the multidimensional data comprehensive analysis method proposed in the present invention, abnormal patterns in the operation of the equipment are identified, which can ensure that the final feature sequence contains not only the most critical information for the performance evaluation of the target equipment, but also secondary information that is not identified as a key feature but contains valuable information, thereby greatly improving the interpretability of the data and the predictive ability of the model.

[0079] In one embodiment of the present invention, the proposed method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system further includes step S2, model training.

[0080] S2-1. Construct the time series data vector of each target device based on the device-level time series data. Specifically, input the device-level time series data X processed,t , construct the time series data vector D for each device t,i ={d t-n+1,i ,d t-n+2,i ,...,d t,I}, where n represents the length of the time window.

[0081] S2-2, extract time series features, by inputting device-level time series data D t,i , use LSTM to extract time series features: h t,i =LSTM(D t,i ;θ), where h t,i is the hidden state feature vector, which represents the comprehensive characteristics of device i in the time series, and θ is the LSTM model parameter, including: the forget gate weight W fand bias b f ; Input gate weight W i and bias b i ; Candidate unit weight W c and bias b c ; Output gate weight W o and bias b o Among them, the specific calculation h t,i =LSTM(D t,i ;θ) The process includes: inputting the X of the current time step t , the hidden state h of the previous time step t-1 and memory state c t-1 , calculate the hidden state h of the current time step t and memory state c t Understandably, X t is the input of the current time step (device-level data vector); h t-1 is the hidden state of the previous time step; c t-1 is the memory state of the previous time step; h t is the hidden state of the current time step; c t is the memory state of the current time step.

[0082] S2-3, perform forget gate and input gate calculations, and combine the forget gate and input gate outputs, update the memory state, and then perform output gate calculations to obtain the output gate output o t， The value range is [0,1], and the hidden state h t , and then get the time series hidden state feature h t,i .

[0083] S2-4. Time series hidden state features h extracted using LSTM model t,i , combined with the historical fault records and health score labels of the target device, the health score prediction model and the fault classification model are trained respectively. The health score prediction model uses a regression method to predict the health score of the target device; the fault classification model uses a classification method to determine the fault category of the target device.

[0084] By inputting the hidden state feature h of the time series t,i and device location information L i , calculate the health score of device i: , where Device health score, range [0,1], The weight matrix calculated for the health score, is the bias term of health score, [h t,i ,L i ] is the concatenated vector of time series features and location information.

[0085] By inputting the time series hidden state feature h t,i and device location information L i , calculate the probability distribution of device i belonging to each fault category: (16), where W j 、b j are the weight matrix and bias term of fault classification, C is the total number of fault categories, P ij is the probability that device i belongs to fault category j.

[0086] S2-5. The difference between the predicted health score and fault classification and the actual label is measured by defining the loss function, which is:

[0087] By entering the health score and the fault classification probability P ij Define the loss function L: (17), where N is the total number of training samples, S i is the true health score manually labeled by the manager, y ij is the real fault classification label (1 indicates fault category match, 0 indicates mismatch), and ℷ is the weight balance parameter. The Adam optimizer is then used to optimize the two model parameters to minimize the loss function. Specifically, by inputting the loss value L, the Adam optimizer is used to update the model parameter θ, and the update amplitude is dynamically adjusted according to the gradient information, thereby optimizing each parameter to minimize the loss function L: (18), where is the learning rate (initial value is 0.001), is the first-order momentum of the gradient (the exponentially weighted mean of the gradient), The second-order momentum of the gradient (the exponentially weighted mean of the squared gradients), It is a numerical stability parameter, usually taken as 10 to the power of -8.

[0088] At this point, the updated model parameters θ are output t+1 , through continuous iterative optimization, the model parameters θ will gradually converge to the optimal solution, and then be determined as the trained model parameters for prediction calculation. At this time, the trained model parameters Includes: LSTM parameters {W f ,b f ,W i ,b i ,W c ,b c ,W o ,b o}、Health score prediction parameters , fault classification parameters {W j ,b j}.

[0089] In one embodiment of the present invention, the proposed method for locating energy consumption and predicting faults of heating and cooling equipment under the EMS energy consumption management system also includes step S3, outputting early warning information, performing health scoring and fault classification prediction on the real-time data of the target equipment through the trained model, evaluating the operating status of the target equipment, and generating early warning information based on the prediction results. And fault classification results , generate equipment warning information: (19), where The health score is a warning when it is lower than the set threshold. and The severity level of the warning, in this embodiment, the severity level of the warning includes: Level 1 warning: ∈[0.8,1.0], the device is in normal state, no operation is required; Second level warning: ∈[0.5,1.0], indicating that the device may have a slight abnormality and needs attention or inspection to eliminate possible faults. Level 3 warning: ∈[0,0.5], indicating that there is a serious problem with the device and it needs to be dealt with immediately.

[0090] As the second aspect of the present invention, a system for locating energy consumption and predicting faults of heating and cooling equipment under an EMS energy consumption management system is proposed. When implemented specifically, it includes: a data acquisition and processing module, which is responsible for collecting data in real time from various sensors (including at least temperature sensors, pressure sensors, current sensors and vibration sensors) installed on multiple target devices; a model training module: using a deep learning model based on LSTM to train a machine learning model using historical data to learn the normal operating mode and potential fault mode of the equipment; a prediction monitoring and early warning module: running the trained model in real time, analyzing and predicting the newly collected data, and evaluating the health status of the equipment in real time and predicting potential faults. Based on the prediction results of the model, the system will generate early warning information to prompt managers and maintenance teams to pay attention to possible equipment problems; a user interface module: providing an intuitive user interface so that managers can easily view the real-time data, historical trends, early warning information and operation suggestions of the equipment.

[0091] The technical scope of the present invention is not limited to the contents in the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system is characterized by: Includes steps: S1. Device-level time series data generation: Collecting raw data representing the state of the target equipment, matching and formatting the raw data according to five dimensions including equipment information, health status, fault diagnosis, warning information and maintenance suggestion related indicators, creating a comprehensive evaluation data set based on the matched and formatted data, and on the basis of the comprehensive evaluation data set, matching each related indicator to a corresponding fuzzy set by constructing a source data extraction function, determining the degree to which each related indicator belongs to different fuzzy sets, calculating the fuzzy coefficient of each fuzzy set, and obtaining a fuzzy vector quantifying the operating state of each target equipment; The fuzzy vectors are clustered using a fuzzy C-means clustering algorithm to form different feature clusters that fully cover the original data and characterize the key features of the target device, importance scores are calculated for different feature clusters, all feature clusters are sorted according to the score results, and within each feature cluster, the indicator feature that has the greatest impact on the operating status of the target device is selected and identified by setting a correlation threshold, and recorded as an important feature; The selected important features and the remaining features in this feature cluster are sorted according to the rule of first between clusters and then within clusters to form the final device-level time series data; S2. Model training: constructing a time series data vector for each target device based on the device-level time series data, respectively training a health score prediction model and a fault classification model, and measuring the difference between the predicted health score and fault classification and the actual label by defining a loss function, and then optimizing the two model parameters using the Adam optimizer to minimize the loss function; S3. Warning information output: Through the trained model, the real-time data of the target equipment is evaluated for health and fault classification, and warning information is generated based on the prediction results.

2. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1 is characterized in that: The specific operations for dimension matching of original data are as follows: Integrate the data including at least temperature data, pressure data, current data, vibration data and device location information sent back by multiple preset sensors, match the data with preset semantic parameters, and realize quantitative analysis of the operating status of the target device, wherein: The device information R1 of the target device includes: , where L r , O r Score the accuracy of the rth device number and location information, a L , a0 is the weight of the device number and location information, i is the number of device information; The health status H2 of the target device is: , where W T , W P , W I , W V are the weights of temperature, pressure, current and vibration data respectively, n is the preset number of sensors, T k , P k ,I k 、V k are the temperature, pressure, current and vibration sensor data of the kth sensor respectively; The fault diagnosis F3 for the target device is: , where F a is the score of the ath fault category, S a is the score of the ath severity level, d F d S are the weights of fault classification and severity level, respectively, and p is the number of fault diagnoses; The warning information W4 for the target device includes: , where W b is the score of the e-th warning information, t W is the weight of the warning information, e is the number of warning information; The maintenance recommendations M5 for this target device are: , where m is the number of maintenance recommendations, M j Score the jth maintenance suggestion, S m weights are suggested for maintenance; The specific operations for constructing a comprehensive evaluation data set are as follows: Integrate multiple dimension data of each target device into a table, where each row represents a target device, each column represents data of an indicator dimension, and add time information; Assign characteristics to each column in the table, ensuring that the data in each column is correctly identified and assigned a clear name and unit. At this point, a comprehensive evaluation data set is created for each target device, including a complete data record of all relevant indicators and time information.

3. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1 is characterized in that: Construct source data extraction function G Q,P (X), the specific operation to reflect the relationship characteristics between each relevant indicator and the fuzzy set is as follows: Based on the obtained comprehensive evaluation data set, the fuzzy set P to which each relevant indicator Q belongs is determined, and each different fuzzy set is defined to represent a different performance or status level of the target device, wherein: When each relevant indicator reaches the optimal state, it means that the target device is in the best operating state, and the fuzzy set at this time represents the high performance state H PQ ,H PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|H i ≥90,M i =0,Y i ≥5,F i =0,E i ≤1200,W i =10}, where H i ≥90, indicating that the target device is in the best health state, M i =0, indicating no maintenance is required, Y i ≥5, the target device has been in use for at least five years, indicating that the target device has been running stably for a period of time. i =0, indicating no fault, E i ≤1200, indicating that the energy consumption does not exceed 1200kwh, and the target device has the lowest energy consumption, W i =10, indicating no warning; When some related indicators perform well, it means that the target device is in a good operating state. The fuzzy set at this time represents the good performance state G PQ ,G PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|70≤H i <90,M i ≥50,3≤Y i <5,F i =1,1000≤E i <1200,W i ≥5}, where 70≤H i <90, indicating that the target device is in good operating condition, M i ≥50, indicating that local maintenance is required, 3≤Y i <5, the target equipment is between 3 and 5 years old, which means the target equipment has been in operation for a period of time. i =1, indicating 1 fault occurred, 1000≤E i <1200, indicating that the energy consumption is between 1000 and 1200 kwh, and the target equipment has medium energy consumption. i ≥5, indicating that attention should be paid to early warning; When some related indicators need attention, it means that the target device is in a medium operating state. The fuzzy set at this time represents the medium performance state M PQ ,M PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|50≤H i <70,M i >0,1≤Y i <3,F i =2,1100≤E i <1300,W i >0}, where 50≤H i <70, indicating that the target device is operating within the preset acceptable range, M i >0, indicating maintenance is required, 1≤Y i <3, the target equipment is between 1 and 3 years old, indicating that the target equipment is new. i =2, indicating that two faults have occurred and attention is required, 1100≤E i <1300, indicating that the energy consumption is between 1100 and 1300 kwh, and the target equipment has a high energy consumption. i >0, indicating that there is a warning; When some related indicators decrease, it means that the target device is in a poor operating state, and the fuzzy set at this time represents a low performance state L PQ ,L PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|H i <50,M i ≥100,Y i ≤1,F i ≥3,E i ≥1300,W i =0}, where H i <50, indicating that the target device is in a poor operating state, M i ≥100, indicating that a lot of maintenance is required, Y i ≤1, the target device is no more than 1 year old, indicating that the target device is new or frequently replaced. i ≥3, indicating that at least 3 failures occurred, and the target device failed frequently. i ≥1300, indicating that the energy consumption exceeds 1300kwh, and the target device has high energy consumption. i =0, indicating no warning; When each relevant index is at its lowest, it means that the target device is in a faulty operation state, and the fuzzy set at this time represents the fault state F PQ ,F PQ ={(H i ,M i ,Y i ,F i ,E i ,W i )|H i <30,M i ≥100,Y i ≤7,F i ≥3,E i ≥1300,W i <10}, where H i <30, indicating that the target device is in a faulty operating state and has serious problems. i ≥100, indicating that immediate and comprehensive maintenance is required, Y i ≤7, the target device is no more than 7 years old, indicating that the target device is an aging device. i ≥3, indicating that at least 3 failures occurred, and the target device failed frequently. i ≥1300, indicating that the energy consumption exceeds 1300kwh, and the target device has high energy consumption. i <10, indicating a serious warning.

4. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 3 is characterized in that: After constructing the source data extraction function and before determining the fuzzy coefficient of each fuzzy set, it is also necessary to use homogeneous quantization to eliminate the dimensional differences between indicators of different dimensions to ensure that different related indicators can be compared under a unified standard. Then, the fuzzy vector that quantifies the operating status of each target device is obtained through the source data extraction function and the homogeneous quantized data, where: The specific operation of the homogeneous quantization process is as follows: For each relevant indicator Q, calculate its homogeneous quantization value Ω iq , in order to map all data into the interval [0,1] to reduce its impact on subsequent overall evaluation, , where X iq is the value of the qth related indicator of the i-th target device, min(X q ) and max(X q ) are the minimum and maximum values ​​of the qth related index respectively; The specific operation of obtaining the fuzzy vector to quantify the operating status of each target device is as follows: First, according to the quantized Ω iq , calculate the fuzzy coefficient C i , , where C i is the fuzzy coefficient of the i-th target device, indicating the comprehensive membership degree of the target device in all fuzzy sets, G q (Ω iq ) is the membership function value of the qth related index in the fuzzy set, indicating the membership degree of the i-th target device in the fuzzy set. is the weight of the qth related index in the i-th target device, reflecting the importance of the related index in the quantitative analysis of the operating status of the target device. Secondly, all fuzzy coefficients are integrated into a fuzzy vector, C=(C1,C2,C3,...,C i ).

5. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1 or 4, characterized in that: The fuzzy vectors are clustered using the fuzzy C-means clustering algorithm to form different feature clusters X ik The specific operations are as follows: First, select the number of clusters K and randomly initialize k cluster centers U, U = {U1, U2, U3, ..., U k }; Secondly, for each fuzzy vector C i , calculate its belonging to each cluster center U k The membership degree V ij , , where m is the fuzzification parameter, and m>1, which is used to control the fuzziness of the membership degree, ||C i −U l || is the Euclidean distance between the i-th fuzzy vector and the l-th cluster center. The smaller the distance, the greater the degree of membership; Again, according to the membership update formula, , update the cluster center, and repeat the above steps until the change of the cluster center is less than the preset threshold or the maximum number of iterations is reached, where n is the total number of fuzzy vectors; finally, according to the clustering results, the fuzzy vectors are assigned to the corresponding cluster centers to form different feature clusters X ik , at this time, different feature clusters X ik Contains the fuzzy vector with the highest degree of membership to a specific cluster center.

6. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 5 is characterized in that: For different feature clusters X ik Before calculating the importance scores, it is necessary to first calculate the DBI index of each clustering result and select the cluster number k that minimizes the value of the DBI index as the optimal cluster number K∗ to verify the clustering result. Then, based on the optimal cluster number K∗ determined after verification, the random forest algorithm is used to evaluate each feature cluster X. ik Reflects the contribution of the performance or status of the target device, thereby ensuring that each of the feature clusters X ik Accurately reflect the actual performance or status of the target device.

7. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 6 is characterized in that: The feature cluster X ik The random forest algorithm evaluation operation is as follows: First, train the random forest model without replacing any feature clusters in the random forest model, and record the number of out-of-bag error samples E1 of the random forest model Ci , taking this as the baseline error rate, and then replacing each feature cluster X one by one in the random forest model ik The data is randomly disrupted to randomly shuffle the feature cluster X ik , and keep other feature clusters X ik The data remains unchanged, destroying the original recorded information; again, train the random forest model and record the new number of out-of-bag error samples E2 Ci ; Finally, if the error rate of the model increases after permutation, it indicates that the feature cluster X ik The data makes an important contribution to the predictive ability of the model and can accurately reflect the actual performance or status of the target device.

8. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1 or 7, characterized in that: The importance scores of different feature clusters are calculated respectively, and the specific operations to identify important features are as follows: First, according to the permutation result, by comparing the number of out-of-bag error samples E1 before permutation Ci And the number of out-of-bag error samples after replacement E2 Ci , using the formula V I (X ik )=E1 Ci −E2 Ci , calculate the importance score, where the importance score V I (X ik ) is higher, indicating that the feature cluster X ik The greater the contribution of the data to the actual performance or status of the target device; secondly, according to this score V I (X ik ) for all feature clusters X ik Sort and in each feature cluster X ik Calculate the correlation between a single feature and classification information P xy ; Again, set the correlation threshold yh, only when the correlation between a single feature and the classification information P xy When the correlation threshold yh is exceeded, the feature is considered important. Finally, from each feature cluster X ik The features that have an impact on the contribution are selected as important features.

9. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1, characterized in that: The specific operations to form the final device-level time series data are as follows: First, for each target device i, at time step t, the selected important features and the remaining features are sorted according to the rule of first between clusters and then within clusters; secondly, the sliding average filter denoising is used to eliminate short-term fluctuations and random noise; finally, according to the sorting results, the final device-level time series data X is formed. processed,t ={d t,1 ,d t,2 ,d t,3 ,d t,I },in, d t,I =[L i ,T t,i smooth ,P t,i smooth ,I t,i smooth ,V t,i smooth ], where L i is the location and / or number and / or room number information of the target device i, T t,i smooth ,P t,i smooth ,I t,i smooth ,V t,i smooth are the smoothed values ​​of the temperature, pressure, current, and vibration sensor data of the target device i at time step t, respectively.

10. The method for locating energy consumption and predicting faults of cooling and heating equipment under the EMS energy consumption management system according to claim 1, characterized in that: In step S2, the specific operations for training the health score and fault classification model are as follows: First, the constructed device-level time series data is input into the LSTM model. By setting the internal forget gate, input gate, and output gate structure, the data of each time step is weighted and updated to extract the time series characteristics of the device operation status and generate a feature vector that reflects the device operation mode and potential fault characteristics. Secondly, the time series features extracted by the LSTM model are combined with the historical fault records and health score labels of the target device to train the health score prediction model and the fault classification model respectively. The health score prediction model uses a regression method to predict the health score of the target device; the fault classification model uses a classification method to determine the fault category of the target device.

Images