A method for detecting data anomalies and troubleshooting of a cold source system

Abstract

The application discloses a kind of cold source system data anomaly detection and troubleshooting method, it is characterized in that, including the following steps: obtaining target building cold source system topology structure diagram, the equipment type and quantity of target building cold source system are numbered and counted, collect target building cold source system historical operation data;The equipment historical operation data of target building is statistically analyzed, preliminary troubleshooting and outlier separation, then the normalization processing is carried out to credible data set;Based on multiple hidden layer auto-encoder, establish target area cold source system anomaly detection network model;Combining the target area cold source system anomaly detection network model established with target building cold source topology structure carries out real-time data anomaly detection and troubleshooting work.The application can detect the data of collection anomaly in time, and troubleshoot cold source system fault, realize long-term stable energy-saving application.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and HVAC automation control, and particularly relates to a cold source system data anomaly detection and fault troubleshooting method. BACKGROUND

[0002] With the continuous advancement of urbanization process in China, the energy consumption of large public buildings has reached 30% of the total energy consumption, so building energy saving has become the key to energy saving and low-carbon development in China. And the energy saving of central air conditioning cold source system is the key to building energy saving. The current energy saving strategy is to establish an energy consumption model from bottom to top, optimize and adjust the equipment parameters, and switch the mode control. Some institutions and research institutes have also developed various energy consumption simulation software, such as Energy Plus, Trnsys, etc. When all the input information parameters are correct, such software can accurately approximate the actual energy consumption. Then the best operation parameters can be obtained through various optimization methods to guide manual or automatic operation control.

[0003] However, the building cold source energy saving system is complex, the equipment is diverse, and various sensors are prone to produce abnormal data when collecting, and various equipment will also appear faults, which greatly affects the accuracy of the final simulation model and the effect of system energy saving control. The existing abnormal data and fault detection methods applied in practice are mostly based on rule judgment, but various rules are difficult to exhaust, time-consuming and laborious, and different application scenarios require different rules, which are not universal. In addition, under the same application scenario, the change of equipment characteristics caused by time change will also make part of the rules gradually invalid. Therefore, there is an urgent need for a convenient and fast multi-building universal cold source system data anomaly detection and fault troubleshooting method to detect abnormal data collected in time and troubleshoot cold source system faults, and realize long-term stable energy saving application. SUMMARY

[0004] The purpose of the present application is to provide a cold source system data anomaly detection and fault troubleshooting method. The present application can detect abnormal data collected in time and troubleshoot cold source system faults, and realize long-term stable energy saving application.

[0005] The technical scheme of the present application: a cold source system data anomaly detection and fault troubleshooting method, comprising the following steps:

[0006] S1, obtaining the topological structure diagram of the target building cold source system, counting the types and quantities of the equipment of the target building cold source system and numbering, and collecting the historical operation data of the target building cold source system;

[0007] S2, statistically analyzing the historical operation data of the equipment of the target building, preliminarily troubleshooting and separating outliers, and then normalizing the credible data set;

[0008] S3, establishing a target area cold source system anomaly detection network model based on a multi-hidden layer autoencoder;

[0009] S4, combining the established target area cold source system anomaly detection network model with the target building cold source topology to perform real-time data anomaly detection and troubleshooting.

[0010] In the foregoing cold source system data anomaly detection and troubleshooting method, the equipment of the target building cold source system includes various types of water chillers, cooling towers, variable frequency cooling water pumps, variable frequency chilled water pumps, temperature sensors, pressure sensors, and flow sensors, and each device is numbered.

[0011] In the foregoing cold source system data anomaly detection and troubleshooting method, the target building cold source system historical operation data includes the chilled water set outlet temperature, chilled water outlet temperature, chilled water return temperature, cooling water outlet temperature, cooling water return temperature, chilled water flow, cooling water flow, and chiller operating power of each chiller at each time node; the inlet and outlet pressures of each variable frequency cooling water pump and chilled water pump; the temperature at the inlet and outlet of the cooling tower system; the flow, temperature, and pressure of the end system inlet and outlet headers; and the indoor and outdoor environmental temperature and humidity.

[0012] In the foregoing cold source system data anomaly detection and troubleshooting method, the specific process of step S2 is: using the 3σ principle to statistically analyze the collected target building cold source system historical operation data, and regarding the values of the data characteristics collected outside three times the standard deviation from the average value as outliers. This process can be represented as: wherein, represents the i-th data characteristic collected at time t, μ i and σ i respectively represent the average value and standard deviation of the i-th data characteristic in the historical operation process at multiple times, and N represents the number of characteristics collected at a unit time;

[0013] Then, take the data of each time when all data characteristics are not outliers as credible data, and the data of the remaining times as outlier data.

[0014] The credible data set is normalized and used as input features in the subsequent model training process.

[0015] In the foregoing cold source system data anomaly detection and troubleshooting method, the target area cold source system anomaly detection network model includes an encoder, a hidden layer, and a decoder.

[0016] The encoder is used for encoding the normalized high-dimensional input feature X of the target building cold source system into a low-dimensional vector; the encoding process from the input layer to the hidden layer is represented as: h=g(X)=σ(W1X+b1), wherein h is the output of the encoder, σ represents an activation function, W1 and b are respectively the weight and bias in the encoding process;

[0017] The hidden layer has L layers in common;

[0018] The decoder restores the hidden variable h of the hidden layer into a high-dimensional feature, and the process is represented as X R =f(h)=σ(W2X+b2), wherein X R represents the output feature reconstructed by decoding, W2 and b2 represent the weight and bias in the decoding process;

[0019] The target building cold source system anomaly detection network model judges data anomalies by comparing the difference between the input X and the output X R , and the optimization objective function of the model is represented as:

[0020] In the foregoing cold source system data anomaly detection and troubleshooting method, the specific implementation method of step S4 is as follows: the input feature of the target area cold source system feature collected in real time is put into the target building cold source system anomaly detection network model trained, the reconstructed feature at this moment is obtained, and whether the data is abnormal and whether troubleshooting is needed is judged by comparing the reconstructed feature and the input feature; that is, whether the whole system is in a normal condition is determined by comparing the size of the real-time acquisition data Loss and the system threshold t s ;

[0021] When Loss<t s , it is considered that the whole system is basically in normal operation, and no troubleshooting work is performed;

[0022] When Loss>t s , the reconstruction error of each corresponding feature device is calculated respectively, and the device threshold t i is set respectively, which is used to judge the correctness of the acquisition data and the abnormality of each device;

[0023] Finally, after the abnormal device is determined, the fault area can be quickly locked by combining the topological structure diagram of the region cold source system, and targeted troubleshooting work is performed.

[0024] The beneficial effects of the present application are as follows:

[0025] The application screens reliable data by statistically analyzing historical data and applying the 3 sigma principle, and then trains a multi-hidden layer autoencoder anomaly detection model, thereby excluding part of outlier abnormal data, improving the data quality of the input autoencoder, making the model have stronger reconstruction accuracy, and the method belongs to unsupervised learning, without marking the data, convenient and fast.

[0026] The application models based on sensor and equipment operation data, and the building cold source system relationship of the target area is analyzed through network structure learning; therefore, the method is suitable for various types of buildings, has strong universality, and does not need to design different expert rules according to different scenes.

[0027] The abnormal detection and fault troubleshooting method of the application can quickly lock abnormal equipment and area, accurately troubleshoot faults, and can assist in establishing a high-precision energy consumption simulation model, and better energy-saving effect is obtained.

[0028] The application can detect abnormal data in time, troubleshoot cold source system faults, and realize long-term stable energy-saving application. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 is a specific step flowchart of the application;

[0030] Figure 2 is a target area cold source system topology diagram in the embodiment of the application;

[0031] Figure 3 is a target area cold source system anomaly detection network model diagram established by the application. DETAILED DESCRIPTION

[0032] The specific embodiments of the application are further described in detail below in combination with the drawings.

[0033] The application is further described below in combination with the drawings and embodiments, but is not used as the basis for limiting the application.

[0034] Example. A cold source system data anomaly detection and fault troubleshooting method, attached Figure 1 is a flowchart of the method, mainly including the following steps:

[0035] S1: Obtain the topology structure diagram of the target building cold source system, count the types and quantities of devices of the system, and collect system historical operation data. Further, the devices include various types of water chillers, cooling towers, variable frequency cooling water pumps, variable frequency chilled water pumps, temperature sensors, pressure sensors, and flow sensors, and the devices need to be numbered. The system operation history data includes the chilled water set outlet temperature, chilled water outlet temperature, chilled water return temperature, cooling water outlet temperature, cooling water return temperature, chilled water flow, cooling water flow, and chiller operating power of each chiller at each time node; the inlet and outlet pressures of each variable frequency cooling water pump and chilled water pump; the temperatures at the inlet and outlet of the cooling tower system; the flow, temperature, and pressure of the inlet and outlet pipes of the terminal system; and the indoor and outdoor environment temperature and humidity. The amount of historical operation data can be determined according to the type of building and the region, and in this example, the air conditioning system of an office building in Hangzhou is basically started in summer, so the historical operation data of June, July, and August is collected, and the data is shown in Table 1. Figure 2 The topology structure diagram of the cold source system of the target region in this example is shown, which includes two centrifugal water chillers CM (1-2) , three screw water chillers SM (1-3) , seven variable frequency cooling water pumps P cw(1-7) , nine chilled water pumps P chw(1-9) , nine cooling towers CT (1-9) , and 34 pressure sensors, 24 temperature sensors, and 12 flow sensors.

[0036] S2: Statistically analyze the historical operation data of the devices of the target region, preliminarily troubleshoot faults and separate outliers, and then normalize the credible data set. The operation condition of the building cold source system has strong periodicity, so the historical operation data of each device of the cold source system can be approximately considered to follow a normal distribution. The collected historical operation data is first statistically analyzed by using the 3 σ principle, it is considered that the probability of the data feature being outside three standard deviations from the average value is very small, and it can be considered as an outlier, which can be represented as: wherein represents the i-th data feature collected at time t, μ i and σ i respectively represent the average value and the standard deviation of the i-th data feature in the historical operation process at multiple times, and N represents the number of features collected at a unit time. Then, the data at each time when all data features are not outliers is taken as credible data, and the data at the remaining times is taken as outlier data. The purpose of this step is to preliminarily screen the data by statistical law, and further, the credible data set is normalized as the input feature in the subsequent model training process.

[0037] S3: Establishing a target area cold source system anomaly detection network model based on a multi-hidden layer autoencoder. Figure 3 The target area cold source system anomaly detection neural network model established in the present example is shown in the figure. The algorithm model mainly includes three parts: an encoder, a hidden layer, and a decoder. The encoder is used to encode the normalized high-dimensional input X of the cold source system into a low-dimensional vector, so that the neural network learns the most useful features. In addition, the meaning sequence of each feature in the input data X must be recorded. The features can be sorted by device order number or by sensor number to facilitate subsequent troubleshooting work. The encoding process from the input layer to the hidden layer can be represented as: h = g(X) = σ(W1X + b1), where h is the output of the encoder, σ represents the activation function, W1 and b are the weights and biases in the encoding process, respectively. The hidden layer has L layers, which are used to further learn the encoded hidden variable features to improve the generalization ability of the model. The number of hidden layers is determined by the size of the target area cold source system. The larger the cold source system, the more the number of hidden layers. In the present example, three hidden layers are adopted, and the activation function is ReLU function. The decoder restores the hidden variables h of the hidden layer to high-dimensional features. The process can be represented as X R = f(h) = σ(W2X + b2), where X R represents the output features reconstructed by decoding, W2 and b2 represent the weights and biases in the decoding process, respectively, and R e LU activation function is selected to increase the nonlinearity of the network model. The model judges data anomalies by comparing the differences between the input X and the output X R , so the optimization objective function of the model can be expressed as After determining the network structure and optimization objective, the best anomaly detection model is obtained by repeatedly iterating in the historical training set. In addition, as the service life increases, the aging and wear of various devices will cause the system characteristics to change to some extent, so the model needs to be updated and iterated at regular intervals.

[0038] S4: Real-time data anomaly detection and fault troubleshooting based on the established model and the target area cold source topology structure. The device fault troubleshooting includes the cold machine, water pump, cooling tower, water pipe, valve, etc. in the system. The specific implementation method is: the real-time collected model features are put into the trained anomaly detection model to obtain the reconstructed features at this moment, and the data is judged to be abnormal or not by comparing the reconstructed features and the input features, and whether fault troubleshooting is needed. It is worth noting that due to different importance levels of device faults in practice, some small anomalies do not affect the normal operation of the system. In order to avoid frequent abnormal alarms, the system threshold t s is set in the present example, i.e. the Loss of the real-time collected data is compared with ts The size of Loss determines whether the whole system is in normal condition. When Loss < t s , it is considered that the whole system is basically in normal operation, and no troubleshooting work is performed; when Loss > t s , the reconstruction errors of the corresponding characteristic devices are respectively calculated , and the device threshold t i is respectively set to judge the correctness of the collected data and the abnormal condition of each device. Finally, after the abnormal device is determined, the fault area can be quickly locked in combination with the topological structure diagram of the cold source system in the region, and targeted troubleshooting work is performed, such as whether the temperature sensor and the pipeline, valve and device connected with the temperature sensor are normally operated when the reconstruction error of the temperature sensor is greater than the threshold set by itself, so as to realize the long-term stable operation of the cold source system.

[0039] The above examples are used to explain and illustrate the present application, but not to limit the present application, and any modification and change made to the present application within the spirit and protection scope of the claims of the present application all fall into the protection scope of the present application.

Claims

1. A method for detecting and troubleshooting data anomalies in a cold source system, characterized in that, Includes the following steps: S1. Obtain the topology diagram of the target building's cooling system, count and number the equipment types and quantities of the target building's cooling system, and collect historical operating data of the target building's cooling system. S2. Perform statistical analysis on the historical operating data of the equipment in the target building to separate outliers, and then normalize the reliable dataset. The specific process is as follows: Using The principle involves statistically analyzing the historical operating data of the target building's cooling system. Outliers are defined as data points whose characteristics deviate from the average by more than three standard deviations. This process is expressed as follows: ,in, express The first time collected at the current time Each data feature and These represent the first time step in the historical process at multiple points in time. The mean and standard deviation of each data feature This represents the number of features collected per unit time. Next, the data at each time point where none of the data features are outliers are taken as reliable data, and the data at the remaining time points are outliers; The trusted dataset is normalized and used as input features in the subsequent model training process. S3. Establish a target area cold source system anomaly detection network model based on a multi-hidden-layer autoencoder; the target area cold source system anomaly detection network model includes an encoder, hidden layers, and a decoder; The encoder is used to normalize the high-dimensional input features of the target building's cooling system. Encode into a low-dimensional vector; the encoding process from the input layer to the hidden layer is represented as: ,in For the encoder output, This represents the activation function. and These are the weights and biases in the encoding process, respectively. There are a total of hidden layers layer; The decoder will store the hidden variables of the hidden layer. The process of reducing to high-dimensional features is represented as follows: ,in This represents the output features reconstructed through decoding. and This represents the weights and biases during the decoding process; The target building cooling system anomaly detection network model compares the input... and output The differences between them are used to determine data anomalies, and the model's optimization objective function is expressed as: ; The target area cold source system anomaly detection network model includes an encoder, a hidden layer, and a decoder; The encoder is used to normalize the high-dimensional input features of the target building's cooling system. Encode into a low-dimensional vector; the encoding process from the input layer to the hidden layer is represented as: ,in For the encoder output, This represents the activation function. and These are the weights and biases in the encoding process, respectively. There are a total of hidden layers layer; The decoder will store the hidden variables of the hidden layer. The process of reducing to high-dimensional features is represented as follows: ,in This represents the output features reconstructed through decoding. and This represents the weights and biases during the decoding process; The target building cooling system anomaly detection network model compares the input... and output The differences between them are used to determine data anomalies, and the model's optimization objective function is expressed as: ; S4. Combine the established target area cold source system anomaly detection network model with the target building cold source topology to perform real-time data anomaly detection and fault diagnosis.

2. The method for detecting and troubleshooting abnormal data in a cold source system according to claim 1, characterized in that, The equipment in the target building's cooling system includes: various chillers, cooling towers, variable frequency cooling water pumps, variable frequency chilled water pumps, temperature sensors, pressure sensors, and flow sensors, and each piece of equipment is numbered.

3. The method for detecting and troubleshooting abnormal data in a cold source system according to claim 1, characterized in that: The historical operating data of the target building's cooling system includes, at each time point: the set outlet temperature of chilled water, the outlet temperature of chilled water, the return temperature of chilled water, the outlet temperature of cooling water, the return temperature of cooling water, the flow rate of chilled water, the flow rate of cooling water, and the operating power of the chillers; the inlet and outlet pressures of each variable frequency cooling water pump and chilled water pump; the temperature at the inlet and outlet of the cooling tower system; the flow rate, temperature, and pressure of the inlet and outlet main pipes of the terminal system; and the indoor and outdoor ambient temperature and humidity.

4. The method for detecting and troubleshooting abnormal data in a cold source system according to claim 1, characterized in that, The specific implementation method of step S4 is as follows: The real-time collected features of the target area's cooling system, i.e., the input features, are input into the trained target building cooling system anomaly detection network model to obtain the reconstructed features at that moment. The reconstructed features are compared with the input features to determine whether the data is abnormal and whether troubleshooting is necessary; that is, by comparing the real-time collected data... and system threshold The size determines whether the entire system is in a normal state; when The system is then considered to be operating normally, and no troubleshooting is performed. when At that time, the reconstruction error of each corresponding feature device is calculated separately. And set device thresholds respectively. It is used to determine the correctness of the collected data and the abnormality of each device; Finally, after identifying the malfunctioning equipment, the faulty area can be quickly located by combining the topology diagram of the cold source system in that area, allowing for targeted troubleshooting.

Images