Anomaly detection system, anomaly detection method for anomaly detection system, and recording medium for anomaly detection system

Abstract

Provided is an abnormality detection system capable of detecting an abnormality of an air conditioning device composed of an outdoor unit and an indoor unit, determining a cause of the abnormality, and determining which device has the abnormality. A plurality of measured values related to a plurality of devices are acquired from a plurality of sensors, measured values of sensors related to a device type are selected, a statistical value is calculated based on the selected values, a feature quantity set representing a position within a feature quantity space is calculated based on the calculated plurality of statistical values for each of a plurality of device types, in a case of learning, the feature quantity set is registered as normal data in the feature quantity space, in a case of a monitoring state after learning, based on a deviation of the feature quantity set from the normal data, it is determined whether or not there is an abnormality, in a case where it is determined that there is an abnormality, from among the plurality of device types, a prescribed device type is determined as a device type in which the abnormality has occurred, and from among one or more devices that are the prescribed device type, a device in which the abnormality has occurred is determined.

Description

Technical Field

[0001] This invention relates to an anomaly detection system, an anomaly detection method for an anomaly detection system, and a recording medium for an anomaly detection system, particularly to an anomaly detection system for detecting anomalies in an air conditioning system, an anomaly detection method for an anomaly detection system, and a recording medium for an anomaly detection system. Background Technology

[0002] Detecting abnormalities and malfunctions (hereinafter referred to as abnormalities) in air conditioning systems, as well as the signs of these abnormalities (abnormality detection), is crucial for reducing the cost of air conditioning system inspection and maintenance. Furthermore, there is a recent trend towards not only providing air conditioning equipment to customers but also operating a unified service encompassing the operation and maintenance of air conditioning equipment.

[0003] Furthermore, since April 2015, the Ministry of the Environment's "Freon Emission Suppression Law" has been implemented for commercial air conditioning systems and refrigeration equipment, requiring companies to conduct "simple inspections" and "periodic inspections" of equipment using Freon gas. Therefore, understanding any abnormalities in air conditioning systems has become crucial for operational service businesses.

[0004] Furthermore, from a maintenance and repair perspective, it is important to detect abnormalities in the air conditioning system, which consists of one or more outdoor units and multiple indoor units connected by piping for refrigerant circulation, determine the cause of the abnormality, and identify which equipment is malfunctioning.

[0005] By utilizing such anomaly detection, the environmental, economic, and social values ​​shown below can be provided.

[0006] In terms of environmental value, it helps prevent the greenhouse effect by reducing leakage through early detection of refrigerant (Freon) leaks. In terms of economic value, it helps reduce life-cycle costs by mitigating business losses (production stoppages, reduced yields, food waste, etc.) caused by sudden malfunctions of air conditioning equipment, or by shifting maintenance from a time-based to a condition-based approach. Furthermore, in terms of social value, it helps ensure the stable operation of critical air conditioning equipment in medical settings and eliminates the shortage of maintenance personnel (improved efficiency of maintenance workers).

[0007] However, in the anomaly detection of air conditioning systems, the following method is known: The normal state is learned by using the capability value (characteristic quantity) of each device calculated based on sensor signals installed on the outdoor and indoor units as input, and anomaly detection is performed based on the degree of deviation from the learned normal state. As learning methods, there are individual learning using a combination of one indoor unit and one outdoor unit, and unified learning by aggregating all indoor and outdoor units.

[0008] The problem with individual learning is that the capability value (characteristic) of each device in the outdoor unit is affected by the operating status (operating / non-operating) of multiple indoor units. That is, in the case of individual learning, the characteristic distribution of the normal state of the outdoor unit changes according to the operating mode (the combination of operating / non-operating of each indoor unit), thus making it impossible to perform correct anomaly detection.

[0009] To avoid this problem, learning is required based on operating modes. However, if we consider a VRF (Variable Refrigerant Flow) air conditioning system consisting of multiple outdoor and indoor units, the number of operating modes becomes enormous, making it impractical.

[0010] Furthermore, if unified learning is used, the problem lies in learning different distributions based on the operating mode. However, in unified learning, the dimensionality increases, and the feature distribution of the normal state becomes coarser, which can lead to adverse effects such as more false alarms or a higher threshold resulting in reduced anomaly detection sensitivity.

[0011] Regarding the detection of abnormalities in the equipment constituting the system, for example, Japanese Patent Application Publication No. 2001-289492 (Patent Document 1) discloses the following system: In an engine-driven air conditioner, when the engine becomes unable to operate, it monitors in advance whether the engine has stalled, and if the engine stalls, it starts a throttle valve diagnostic mode and executes a fuel adjustment valve diagnostic mode, thereby detecting the malfunction of the internal combustion engine that drives the compressor.

[0012] Existing technical documents

[0013] Patent documents

[0014] Patent Document 1: Japanese Patent Application Publication No. 2001-289492 Summary of the Invention

[0015] The problem that the invention aims to solve

[0016] The anomaly detection system described in Patent Document 1 is as follows: In an engine-driven air conditioner, when the engine becomes unable to operate, it monitors in advance whether the engine has stalled. If the engine stalls, it starts the throttle valve diagnostic mode and executes the fuel adjustment valve diagnostic mode to detect the malfunction of the internal combustion engine that drives the compressor.

[0017] However, Patent Document 1 does not consider the application of anomaly detection in air conditioning systems consisting of multiple indoor and outdoor units. Thus, Patent Document 1 does not disclose or suggest the following: detecting anomalies in an air conditioning unit consisting of one or more outdoor units and multiple indoor units, which is connected by piping for refrigerant circulation, determining the cause of the anomaly, or identifying which unit is malfunctioning.

[0018] The purpose of this invention is to provide an anomaly detection system, an anomaly detection method for the anomaly detection system, and a recording medium for the anomaly detection system that can detect anomalies in an air conditioning unit consisting of one or more outdoor units and multiple indoor units, which is connected by piping for refrigerant circulation, determine the cause of the anomaly, and identify which unit has malfunctioned.

[0019] Methods for solving problems

[0020] This invention is an anomaly detection system, which has the following features:

[0021] An air conditioning system, which has multiple devices and sensors; and a monitoring computer, which monitors the air conditioning system.

[0022] Its features are,

[0023] Multiple devices consist of multiple types of devices.

[0024] The monitoring computer performs the following processing:

[0025] (1) Obtain multiple measurements related to multiple devices from multiple sensors.

[0026] (2) For each type of equipment

[0027] (2A) Select the measured values ​​of the sensor that are relevant to this type of equipment.

[0028] (2B) Calculate statistical values ​​based on the selected values.

[0029] (3) For each of the multiple equipment types, calculate the set of feature quantities representing the position in the feature quantity space based on the calculated statistical values.

[0030] (4) When in the process of learning

[0031] (4A) Register the feature set as normal data in the feature space.

[0032] (5) In the case of being in a post-learning monitoring state.

[0033] (5A) Determine whether there are any anomalies based on the deviation of the feature set from normal data.

[0034] (5B) In cases where an anomaly is identified

[0035] (5Ba) From multiple equipment types, the specified equipment type is identified as the equipment type that has experienced the abnormality.

[0036] (5Bb) Identify the equipment that has malfunctioned from one or more equipment that are designated as equipment types.

[0037] Furthermore, this invention provides an anomaly detection method for an anomaly detection system. The anomaly detection system comprises: an air conditioning system having multiple devices for air conditioning and multiple sensors for measuring the operational status of the multiple devices; and a monitoring computer that monitors the operational status of the air conditioning system. The monitoring computer performs the following steps:

[0038] The steps are: acquiring multiple measurements from multiple sensors related to multiple device types, selecting measurements related to the same device type (hereinafter referred to as the same type of device), and calculating statistical values ​​based on the selected measurements;

[0039] The steps for calculating the set of features representing the location in the feature space based on the statistical values ​​of at least two identical devices;

[0040] In the learning state, the step of registering the feature set as normal data in the feature space;

[0041] In the post-learning monitoring state, the steps for determining the cause of anomalies based on the deviation of the feature set from normal data; and...

[0042] Based on the cause of the anomaly, from multiple identical devices, the specified identical devices are identified as the devices that have experienced the anomaly, and then the devices that have experienced the anomaly are identified from the specified identical devices.

[0043] Furthermore, the present invention is a recording medium storing a program for an anomaly detection system, the program causing a monitoring computer used in the anomaly detection system to operate. The anomaly detection system comprises: an air conditioning system having multiple devices for air conditioning and multiple sensors for measuring the operational status of the multiple devices; and a monitoring computer that monitors the operational status of the air conditioning system. Its characteristic is that...

[0044] The recording medium contains:

[0045] The measurement acquisition procedure obtains multiple measurement values ​​related to multiple devices from multiple sensors;

[0046] The statistical value calculation procedure calculates the statistical value of the same type of equipment based on the measurements related to the same type of equipment (hereinafter referred to as the same type of equipment) among multiple devices;

[0047] A program for calculating the distribution density of statistical values, which calculates the set of statistical values ​​representing the location of statistical values ​​in a two-dimensional plane with two sets of statistical values ​​related to the same type of equipment as parameters;

[0048] The statistical value normal data storage program, in learning mode, registers the statistical value set as normal data in the statistical value two-dimensional plane;

[0049] The anomaly detection program, which executes in monitoring mode after learning mode, determines whether there is any anomaly in the same type of equipment based on the deviation between the statistical value set calculated by the statistical value distribution density calculation program and the statistical value set stored by the normal data storage program.

[0050] An anomaly cause determination procedure determines the cause of an anomaly based on the correlation of multiple anomalies detected by an anomaly detection procedure;

[0051] The program for calculating the distribution density of measured values ​​represents the set of measured values ​​whose positions are in a two-dimensional plane with two sets of measured values ​​as parameters.

[0052] The normal data storage procedure for measured values, in learning mode, registers the set of measured values ​​as normal data onto the two-dimensional plane of measured values; and

[0053] The abnormal device identification procedure, which is executed in monitoring mode after learning mode, selects the measurement values ​​related to the abnormal cause determined by the abnormal cause identification procedure, and determines the same type of equipment related to the abnormal cause based on the deviation between the measurement value set calculated by the measurement value distribution density calculation procedure based on the selected measurement values ​​and the measurement value set stored by the measurement value normal data storage procedure, and then identifies the abnormal equipment that has malfunctioned from the same type of equipment.

[0054] Invention Effects

[0055] According to the present invention, it is possible to detect abnormalities in an air conditioning unit consisting of an outdoor unit (one or more) and an indoor unit (one or more) connected by refrigerant circulation piping, determine the cause of the abnormality, and further determine which unit has malfunctioned. Attached Figure Description

[0056] Figure 1 This is a structural diagram showing the basic structure of the anomaly detection system of the air conditioning system to which the present invention is applied.

[0057] Figure 2 This is a control block diagram illustrating the processing block of the abnormal detection system of the air conditioning system according to an embodiment of the present invention.

[0058] Figure 3 This is an explanatory diagram illustrating the structure of an air conditioning system with multiple devices.

[0059] Figure 4 It is an explanatory graph illustrating the relationship between the measured values ​​of each device and the statistical values ​​of the same type of device.

[0060] Figure 5A This is an explanatory diagram illustrating the two-dimensional distribution density of normal data under the learning mode.

[0061] Figure 5B This is a detailed explanation. Figure 5A An illustration of the two-dimensional distribution density.

[0062] Figure 6A This is to explain the... Figure 2 The diagram illustrates an example of determining the cause of anomalies caused by the two-dimensional distribution density in the overall analytical section.

[0063] Figure 6B To provide a more detailed explanation of the determination Figure 6A The diagram illustrates an example of the cause of the anomaly.

[0064] Figure 7 This is an illustrative diagram illustrating an example of determining the cause of anomalies based on combinations of statistical values.

[0065] Figure 8 This is an illustrative diagram illustrating an example of identifying anomalous devices based on a two-dimensional distribution density generated from a set of measurements of devices related to the cause of the anomaly.

[0066] Figure 9A This is a flowchart representing the first half of the learning and processing flow based on the measured and statistical values ​​of equipment in an air conditioning system under normal conditions.

[0067] Figure 9B This is a flowchart representing the latter half of the learning and processing flow based on the measured and statistical values ​​of equipment in an air conditioning system under normal conditions.

[0068] Figure 10A This is a flowchart representing the first half of the monitoring and processing flow based on the measured and statistical values ​​of the equipment in the air conditioning system during monitoring.

[0069] Figure 10B This is a flowchart representing the middle part of the monitoring and processing flow of measured and statistical values ​​of equipment in an air conditioning system during monitoring.

[0070] Figure 10C This is a flowchart representing the latter half of the monitoring and processing flow based on the measured and statistical values ​​of equipment in the air conditioning system during monitoring.

[0071] Figure 11 This is an explanatory diagram illustrating an example of the display screen of the display device constituting an anomaly detection system.

[0072] Explanation of reference numerals in the attached figures

[0073] 100…Air conditioning equipment system, Ssig sensor signals from multiple devices, 102…Sensor signal input unit, 200…Monitoring computer, 210…Overall analysis unit, 220…Detailed analysis unit, 211…Sensor signal processing unit, Ssta…Statistical values ​​of “same type of equipment”, 212…Feature vector extraction unit, 213…Anomaly measurement calculation unit, 214…Threshold calculation unit, 215…Anomaly detection unit, 216…Anomaly cause determination unit, Ssop…Measured values ​​for each device, 221…Anomaly cause feature quantity extraction unit, 222…Anomaly cause feature quantity analysis unit, 223…Anomaly cause feature quantity learning and data storage unit, 224…Anomaly location determination unit. Detailed Implementation

[0074] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to the following embodiments. Various modifications and applications are also included within the scope of the technical concept of the present invention.

[0075] Figure 1 This describes the basic structure of an anomaly detection system in an air conditioning system. The anomaly detection system in the air conditioning system consists of an air conditioning equipment system 100 and a monitoring computer 200. The air conditioning equipment system 100 has multiple devices and sensors for air conditioning, and the monitoring computer 200 monitors the operation of the air conditioning equipment system 100.

[0076] The air conditioning system in the air conditioning equipment system 100 basically consists of an outdoor unit and an indoor unit, which are connected by refrigerant piping. The outdoor unit is equipped with a compressor, heat exchanger, expansion valve, and air supply fan, while the indoor unit is equipped with a heat exchanger, expansion valve, and air supply fan. Furthermore, sensors are installed in each device to detect the operating status parameters of the device (such as temperature, pressure, and current).

[0077] An air conditioning system consists of one or more outdoor units and one or more indoor units. Therefore, the various devices that make up the outdoor and indoor units are categorized as equipment types. Here, equipment type refers to the type of equipment that performs a certain function. For example, if we consider the compressor as an equipment type, then the compressors of different outdoor units are considered "the same type of equipment." Similarly, if we consider the expansion valve and the blower fan as equipment types, then the expansion valves and blower fans of different outdoor units are considered "the same type of equipment." The same applies to the indoor units.

[0078] Action state quantities include sensor signals that can be directly measured from sensors, and measured values ​​that cannot be measured by sensors but are calculated based on sensor signals. However, the following description will generally include both types of measured values. Furthermore, in cases where they are specifically treated as sensor signals or measured values, this meaning will be explained.

[0079] Furthermore, the monitoring computer 200 has an interface with input / output functions and a processor with arithmetic functions. The processor can execute the operations of this embodiment as described below according to the control program. Additionally, as an example of a processor, a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit) can be considered, but other semiconductor devices can also be used as long as the entity performing the specified arithmetic operations is to be executed.

[0080] The monitoring computer 200 can be integrated with the air conditioning system or connected via a wired / wireless cloud system. Furthermore, since the control program has control functions, it can be understood as a control function block.

[0081] Figure 2 This describes the structure of the anomaly detection system in the air conditioning system. The air conditioning equipment system 100 is installed at multiple customer sites, transmitting sensor information to the monitoring computer 200 via a network. Each customer site corresponds to a building (such as a skyscraper), for example, to a VRF air conditioning system within the building.

[0082] The air conditioning system 100 comprises one or more air conditioning systems 101 and a sensor signal input unit 102 that inputs sensor signals (Ssig) from multiple devices installed in the air conditioning system 101 and transmits them to a monitoring computer 200. In the air conditioning system 101, sensors are installed in one or more various devices. Furthermore, the sensor signals (Ssig) from the multiple devices are, for example, signals related to temperature, current, pressure, etc.

[0083] The monitoring computer 200 consists of the following parts: an overall analysis unit (abnormal cause determination analysis unit) 210, which detects abnormalities in the air conditioning system 101 and determines the "abnormal cause"; and a detailed analysis unit (abnormal device determination analysis unit) 220, which determines the abnormal location of the air conditioning system 101 based on the "abnormal cause" obtained by the overall analysis unit 210, such as determining the "abnormal device" of multiple outdoor units or indoor units.

[0084] These overall analysis unit 210 and detailed analysis unit 220 are functions executed by the processor's control program. The processing flow executed by these control programs is... Figures 9A-9B as well as Figures 10A to 10CThe explanation is provided below.

[0085] The overall analysis unit 210 inputs sensor signals (Ssig) from multiple devices transmitted from the air conditioning equipment system 100 to the sensor signal processing unit 211. Using the statistical value (Ssta) of the "same type of device" generated by the sensor signal processing unit 211, it detects abnormalities in the air conditioning system 101 and determines the "cause of the abnormality".

[0086] The overall analysis unit 210 includes a feature vector extraction unit 212, an anomaly measurement calculation unit 213, a threshold calculation unit 214, an anomaly detection unit 215, and an anomaly cause determination unit 216.

[0087] On the other hand, the detailed analysis unit 220 inputs the sensor signals (Ssig) of multiple devices transmitted from the air conditioning equipment system 100 to the sensor signal processing unit 211, and uses the measured value (Ssop) of each device generated by the sensor signal processing unit 211 to select the measured value (Ssop) based on the "abnormal cause" obtained by the overall analysis unit 210, and determines the "abnormal device" as the abnormal location of the air conditioning system 101.

[0088] The detailed analysis unit 220 includes an anomaly cause feature quantity extraction unit 221, an anomaly cause feature quantity analysis unit 222, an anomaly cause feature quantity learning and data storage unit 223, and an anomaly location determination unit 224.

[0089] Next, the control functions of the overall analysis unit 210 and the detailed analysis unit 220 will be explained. First, the "learning mode" of the overall analysis unit 210 in the normal state will be explained.

[0090] In the overall analysis unit 210, sensor signals (Ssig) from multiple devices transmitted from the air conditioning equipment system 100 are converted into statistical values ​​(Ssta) for "same type of device" by the sensor signal processing unit 211. Multiple statistical values ​​(Ssta) are calculated. Furthermore, the statistical values ​​(Ssta) are defined using terms such as "total value, average value, maximum value, minimum value, standard deviation, variance, median, and histogram." These statistical values ​​(Ssta) are then input to the subsequent feature vector extraction unit 212.

[0091] The feature vector extraction unit 212 extracts feature vectors using the statistical values ​​(Ssta) of the "same type of device" and inputs the extracted feature vectors into the anomaly measurement calculation unit 213. The anomaly measurement calculation unit 213 uses the feature vectors during a pre-specified learning period and calculates the anomaly measure for each feature vector at specified time intervals (sometimes represented as each moment).

[0092] The calculated anomaly measure is input to the threshold calculation unit 214, which calculates the threshold (SL) corresponding to the anomaly measure in "learning mode". The threshold (SL) calculated by the threshold calculation unit 214 and the anomaly measure obtained by the anomaly measure calculation unit 213 are compared by the anomaly detection unit 215. Then, the comparison result is sent to the anomaly cause determination unit 216, where the "anomaly cause" is determined.

[0093] Next, the "monitoring mode" of the overall analysis unit 210 will be explained. The "monitoring mode" is executed after the "learning mode" ends. In the "monitoring mode" of the overall analysis unit 210, the sensor signals (Ssig) of multiple devices transmitted from the air conditioning equipment system 100 are converted into statistical values ​​(Ssta) of "same type of device" by the sensor signal processing unit 211, and the feature vector extraction unit 212 uses the statistical values ​​(Ssta) of "same type of device" to extract feature vectors.

[0094] The anomaly measurement calculation unit 213 uses feature vectors from a pre-specified monitoring period to calculate anomaly measurements for each feature vector at predetermined time intervals (sometimes expressed as individual moments). The anomaly detection unit 215 compares the anomaly measurements calculated in "monitoring mode" with the threshold (SL) calculated by the threshold calculation unit 214 based on the anomaly measurements in "learning mode" to determine an anomaly. The anomaly cause determination unit 216 determines the "cause of the anomaly" of the air conditioning equipment system 100 by analyzing the statistical values ​​(Ssta) of "same type of equipment" related to the anomaly, i.e., the feature vectors corresponding to the statistical values ​​(Ssta).

[0095] Next, the detailed functions of the sensor signal processing unit 211, feature vector extraction unit 212, anomaly measurement calculation unit 213, threshold calculation unit 214, anomaly detection unit 215, and anomaly cause determination unit 216 will be explained.

[0096] The sensor signal processing unit 211 extracts stable operating data (hereinafter referred to as normal data) from the sensor signals (Ssig) of multiple devices installed in the air conditioning equipment system 100, and calculates the measured value (Ssop) of each device as the measured value of the sensor related to the type of device. This measured value (Ssop) is also used in the detailed analysis unit 220.

[0097] Furthermore, the sensor signal processing unit 211 selects the measured value (Ssop) for each type of equipment, and calculates the statistical value (Ssta) for "same type of equipment" based on the selected measured value (Ssop). As mentioned above, the statistical value (Ssta) can be "total value, average value, maximum value, minimum value, standard deviation, variance, median value, histogram", etc. Of course, other statistical values ​​can also be used.

[0098] The feature vector extraction unit 212 performs standardization on the statistical value (Ssta) by setting the mean to "0" and the variance to "1". Then, for a certain time t (t = 0, 1, 2, ...), as shown in [Equation 1] below, one vector x(t) is extracted at each time step.

[0099] x(t)=(x1(t), x2(t),…,xM-1(t),xM(t)) T ...[Equation 1].

[0100] Here, xm(t) (m = 1, 2, ..., M) is the m-th statistic (Ssta) at time t after standardization.

[0101] The anomaly measurement calculation unit 213 uses the Local Subspace Classifier (LSC) method to calculate the measure. LSC is widely used as an anomaly detection technique, generating a "k-1 dimensional" subspace using k neighboring data points xi (i = 1, ..., k) of the unknown data. Furthermore, it determines whether an anomaly is based on the projected distance between the unknown data and the "k-1 dimensional" subspace. Since the anomaly measure is represented by the projected distance between the unknown data and the "k-1 dimensional" subspace, the point closest to the unknown data in the subspace can be found.

[0102] The threshold calculation unit 214 sorts the anomaly measures calculated by the anomaly measure calculation unit 213 in ascending order, and takes the value of the largest anomaly measure as the threshold (SL). According to this method, the threshold (SL) can be easily calculated. Alternatively, the threshold (SL) can also be calculated using other methods.

[0103] The anomaly detection unit 215 determines whether the anomaly measurement calculated by the anomaly measurement calculation unit 213 exceeds the threshold (SL) set by the threshold calculation unit 214. If it exceeds the threshold (SL), it is determined to be abnormal; if it does not exceed the threshold (SL), it is determined to be normal.

[0104] The anomaly cause determination unit 216 identifies continuously detected anomaly intervals. For each interval, based on a two-dimensional distribution density, it extracts the statistical value (Ssta) of the "same type of device" related to the anomaly. Therefore, in "learning mode," the two-dimensional distribution density of normal data based on the statistical values ​​(Ssta) of the two "same type of device" is calculated cyclically using these two statistical values ​​(Ssta), and this two-dimensional distribution density is saved in image format. Thus, multiple two-dimensional distribution densities calculated cyclically can be obtained. (Details are available in...) Figure 5A , Figure 5B As shown in the image.

[0105] In the patented technical solution, the two-dimensional distribution density is defined as a "feature space," which is a space representing feature vectors. Furthermore, the statistical value (Ssta) is represented by the feature vector, so it can be processed as a density image, which becomes a normal-state image obtained through a "learning mode."

[0106] On the other hand, in "monitoring mode," monitoring data based on the statistical values ​​(Ssta) of "same type of equipment" obtained from multiple measurements are plotted on a two-dimensional distribution density image, and the deviation between the statistical values ​​(Ssta) of the two types of "same type of equipment" is calculated. Then, starting from the value with the larger deviation, the statistical values ​​(Ssta) of the "same type of equipment" that are associated with the anomaly are extracted sequentially.

[0107] This means that the distribution of anomalous data will reveal statistical values ​​(Ssta) of "same type devices" that are far removed from the distribution of normal data and are related to the anomaly. After identifying the statistical values ​​(Ssta) of "same type devices" related to the anomaly, the anomaly cause is determined by referring to the combination of statistical values ​​(Ssta) of "same type devices" (see details). Figure 7 (Provide an explanation) to determine the cause of the anomaly. Typically, the cause could be, for example, refrigerant leakage, expansion valve malfunction, or compressor failure.

[0108] Next, the functions of the detailed analysis unit 220 will be explained. First, the "learning mode" of the detailed analysis unit 220 in its normal state will be explained.

[0109] In the "learning mode" of the detailed analysis unit 220, sensor signals (Ssig) from multiple devices transmitted from the air conditioning equipment system 100 are input to the sensor signal processing unit 211, which converts them into measurement values ​​(Ssop) for each device. These measurement values ​​(Ssop) are then input to the anomaly cause feature extraction unit 221, which extracts a set of measurement values ​​(Ssop) for each device related to the "anomaly cause." The set of measurement values ​​(Ssop) represents a collection of multiple measurement values ​​related to a specific device.

[0110] The anomaly cause feature quantity analysis unit 222 generates a two-dimensional distribution density of normal data using the set of measurements (Ssop) of each device related to each anomaly cause. The two-dimensional distribution density of normal data is generated using the same method as the two-dimensional distribution density of the statistics (Ssta) generated by the overall analysis unit 210. This two-dimensional distribution density of normal data is sent to the anomaly cause feature quantity learning data storage unit 223 and stored in the memory of the anomaly cause feature quantity learning data storage unit 223.

[0111] Furthermore, the two-dimensional distribution density of normal data is generated according to the number of devices, for example, the number of compressors, expansion valves, blower fans, and heat exchangers corresponding to the number of outdoor units, and the number of expansion valves, blower fans, and heat exchangers corresponding to the number of indoor units. Of course, if there are other devices, it will also be generated for those devices.

[0112] On the other hand, in the "monitoring mode" of the detailed analysis unit 220, the sensor signals (Ssig) of multiple devices transmitted from the air conditioning equipment system 100 are converted into measurement values ​​(Ssop) for each device by the sensor signal processing unit 211. In addition, the anomaly feature quantity extraction unit 221 extracts the set of measurement values ​​(Ssop) for each device related to the "cause of anomaly", and the set of measurement values ​​(Ssop) for each device is sent to the anomaly cause feature quantity analysis unit 222.

[0113] Next, the anomaly cause feature quantity analysis unit 222 uses the input set of measured values ​​(Ssop) for each device related to the "anomaly cause" to generate an image of the two-dimensional distribution density of the monitoring data. Then, the anomaly cause feature quantity learning data storage unit 223 selects the two-dimensional distribution density of normal data based on the set of measured values ​​(Ssop) for each device related to the "anomaly cause" determined by the anomaly cause determination unit 216 of the overall analysis unit 210.

[0114] Furthermore, the two-dimensional distribution density of the monitoring data based on the measurement set (Ssop) of each device is compared with the two-dimensional distribution density of the normal data to identify "abnormal devices". In addition, the two-dimensional distribution density of the monitoring data is generated according to the number of devices, for example, according to the number of compressors, expansion valves, blower fans, and heat exchangers corresponding to the number of outdoor units, and the number of expansion valves, blower fans, and heat exchangers corresponding to the number of indoor units.

[0115] Next, the functions of the abnormal cause feature extraction unit 221, the abnormal cause feature parsing unit 222, the abnormal cause feature learning data storage unit 223, and the abnormal location determination unit 224 will be explained.

[0116] The anomaly cause feature extraction unit 221 extracts the set of measurement values ​​(Ssop) of each device related to the "anomaly cause" from the measurement values ​​(Ssop) of each device obtained by the sensor signal processing unit 211.

[0117] The anomaly cause feature analysis unit 222 takes the two types of measurement values ​​(Ssop) related to the anomaly cause from the extracted measurement value (Ssop) set of each device as input, and generates a two-dimensional distribution density of normal data and monitoring data based on cyclic combination according to each device (e.g., each of the compressor, expansion valve, and air supply fan corresponding to the number of outdoor and indoor units).

[0118] The abnormal cause feature quantity learning data storage unit 223 stores the two-dimensional distribution density of normal data obtained by the abnormal cause feature quantity analysis unit 222 in the memory for each device (e.g., each of the compressor, expansion valve, and blower fan corresponding to the number of outdoor and indoor units).

[0119] The anomaly location determination unit 224, based on the "anomaly cause" determined by the anomaly cause determination unit 216 of the overall analysis unit 210, selects a two-dimensional distribution density of normal data based on the measurement value (Ssop) set of each device related to the "anomaly cause" from the anomaly cause feature quantity learning data storage unit 223, and compares it with the two-dimensional distribution density of monitoring data based on the measurement value (Ssop) set of each device related to the "anomaly cause". Based on this comparison result, the "abnormal device" can be determined. Furthermore, the two-dimensional distribution density of the monitoring data is generated according to each device (e.g., compressors, expansion valves, and blower fans corresponding to the number of outdoor and indoor units).

[0120] Next, the structure of an air conditioning system with multiple devices will be explained. Figure 3 The air conditioning system 101 shown consists of, for example, multiple outdoor units 110, 120, 130 and multiple indoor units 140, 150, 160, 170. The number of these outdoor and indoor units is not limited and can be increased or decreased.

[0121] In addition, outdoor units 110, 120, and 130 may have the same type of equipment 111, 121, 131, as well as the same type of equipment 112, 122, and 132. For example, the same type of equipment 111 to 131 may be heat exchangers, and the same type of equipment 112 to 132 may be compressors. There may be several of these same type of equipment.

[0122] Similarly, indoor units 140, 150, 160, and 170 also have the same type of equipment 141, 151, 161, 171, as well as the same type of equipment 142, 152, 162, and 172. For example, the same type of equipment 141 to 171 is an expansion valve, and the same type of equipment 142 to 172 is a heat exchanger. There can also be several of these same types of equipment.

[0123] The air conditioning system 101, consisting of outdoor units 110-130 and indoor units 140-170, is connected by a refrigerant circulation pipe 201. Therefore, the sensors installed on the outdoor units 110-130 are affected by the operating status (operation / daily operation) of the indoor units 140-170.

[0124] On the other hand, the sensor signal (Ssig) of the sensor that detects the operating state of multiple devices, the measured value (Ssop) of each device obtained from the sensor signal (Ssig), and the statistical value (Ssta) of the "same type of device" are regarded as values ​​that reflect the operating state of the device itself. Figure 4 This represents one example.

[0125] Figure 4 The system displays the project number, the name of the characteristic quantity, the measured value (Ssop) for each device, and the statistical value (Ssta) for "same type of device". For devices (compressors, heat exchangers, etc.) that cannot be directly measured by sensors, the capability values ​​(characteristic quantities) can be calculated as simulated measured values ​​using formulas based on the operational status quantities of existing sensors in the actual setup. Using these values, abnormal states can be easily identified.

[0126] Figure 4 The first column represents the project number, the second column represents the name of the characteristic quantity, the third column represents the measured value (Ssop) of each device obtained by using an arithmetic expression (e.g., four arithmetic operations) based on the action state quantities of existing sensors, and the fourth column represents the statistical value (Ssta) of "same type of device" obtained by summarizing the measured values ​​(Ssop) of each device for each same model.

[0127] In the second column of features, for example, F1 represents the control of the expansion valve, F2 represents the state of the refrigerant, and F3 represents the efficiency of the compressor. The same applies to F4 through F7. Furthermore, many more features can be set.

[0128] The measured values ​​(Ssop) for each device in the third column are calculated by device type. For example, when the characteristic is associated with the outdoor unit, the measured values ​​(Ssop) for the devices corresponding to the number of outdoor units, such as compressors, heat exchangers, expansion valves, and blower fans, are calculated. Similarly, when the characteristic is associated with the indoor unit, the measured values ​​(Ssop) for the devices corresponding to the number of indoor units, such as heat exchangers, expansion valves, and blower fans, are calculated. Of course, if the number of devices increases, the number of measured values ​​(Ssop) for each device also increases.

[0129] The statistical value (Ssta) for "Same Type of Equipment" in the fourth column is calculated as a summary for each type of equipment. For example, if there are multiple measured values ​​(Ssop) for each of the same indoor unit's characteristic quantities, these measured values ​​(Ssop) are summarized into a statistical value (Ssta), so the statistical value (Ssta) for "Same Type of Equipment" is 1. The same applies to characteristic quantities associated with the outdoor unit and those associated with the compressor; the statistical value (Ssta) for "Same Type of Equipment" is 1.

[0130] Furthermore, the statistical value (Ssta) corresponds to the eigenvector; therefore, the statistical value (Ssta) will sometimes be explained in the context of the eigenvector.

[0131] Statistical values ​​(Ssta) are defined, for example, as "total value, mean, maximum value, minimum value, standard deviation, variance, median, histogram," etc. For instance, characteristic quantity F1 is defined as the "mean" and statistic (Ssta), characteristic quantity F2 is defined as the "total value" and statistic (Ssta), and characteristic quantity F3 is also defined as the "mean" and statistic (Ssta).

[0132] For example, characteristic quantities that represent the capacity of equipment and are related to heat or control quantities are calculated by summing the measured values ​​(Ssop) of each type of equipment, while characteristic quantities that represent the state, efficiency, or index of equipment are calculated by averaging the measured values ​​(Ssop) of each type of equipment.

[0133] In this way, the statistical value (Ssta) in the measured value (Ssop) of each type of device can be calculated according to the type and state of the characteristic quantity. Furthermore, when calculating the statistical value (Ssta) of "same type of device", in the case of thermal shutdown (outdoor unit stopped), it is preferable to set the measured value (Ssop) of each device that will be the thermal shutdown condition to "0". This is because the signal output from the thermally shut-off device becomes noise.

[0134] In this way, multiple devices have at least operating and non-operating states, and the calculation of statistical values ​​can exclude the measured values ​​related to the stopped devices from being considered, or treat them as predetermined values ​​for calculation.

[0135] Figure 5A , Figure 5B This is an example of an image showing the two-dimensional distribution density of normal data and monitoring data generated by the anomaly cause determination unit 216 of the overall analysis unit 210. The same applies to the anomaly cause characteristic quantity analysis unit 222 of the detailed analysis unit 220. Furthermore, the following... Figure 4 Based on the records, the statistical values ​​are labeled "Fall".

[0136] Figure 5A The horizontal axis in the middle is Figure 4 The figure shows the statistical values ​​(Fall1) of a characteristic quantity (F1) of a certain "same type of equipment". The vertical axis is related to the horizontal axis of "same type of equipment". Figure 4 The statistical value (Fall2) of a characteristic quantity (F2) of a certain "same type of equipment" is shown.

[0137] Figure 5AThis is a two-dimensional density distribution chart (G2dg) in image format, representing the "feature space" as a feature vector. It's a chart that represents pixel values ​​of two-dimensional density as varying shades: "0" is white, "maximum" is black, and intermediate values ​​are gray. The chart shows that the monitored data (Derr) deviates from the normal data (Dnor), indicating an anomaly.

[0138] Furthermore, the method of image creation is not limited to the above methods. For example, it may not be a simple frequency distribution, but rather a Gaussian distribution or other weighted filter may be assigned to a data point and the data points may be overlapped.

[0139] Alternatively, a maximum value filter of a specified size, or an average filter or other weighted filters can be applied to the image obtained by the above method. Furthermore, it does not necessarily need to be saved in image format; the two-dimensional array can also be saved in text format. Moreover, the pixel values ​​do not necessarily have to be shades of gray; instead, they can be saved as a text format of a binary two-dimensional array where pixels with distributions are set to 1 and pixels without distributions are set to 0.

[0140] Figure 5B As an example, a two-dimensional distribution density chart (G2dg) related to two interrelated "identical devices" is shown in air conditioning system 101. The statistical value (Fall1) of the first "identical device" on the horizontal axis is, for example, a characteristic quantity representing the control of the expansion valve, and the statistical value (Fall2) of the second "identical device" on the vertical axis is, for example, a characteristic quantity representing the refrigerant state.

[0141] exist Figure 5B In the diagram, the minimum (MIN) to the maximum (MAX) of the statistical values ​​(Fall1) of the first "same type of device" after scaling is set as the horizontal axis, and the minimum (MIN) to the maximum (MAX) of the statistical values ​​(Fall2) of the second "same type of device" after similar scaling is set as the vertical axis.

[0142] Furthermore, the refrigerant state characteristic (p1) in the characteristic quantity (t1) representing the control of the expansion valve, which is the first parameter, is plotted as (X1, Y1), and the refrigerant state characteristic (p2) in the characteristic quantity (t2) representing the control of the expansion valve, which is the second parameter, is plotted as (X2, Y2).

[0143] In the generated two-dimensional distribution density chart (G2dg) in image format, identification information "S0001 A0001" is assigned to identify customer sites and air conditioners in the air conditioning system (here, S0001 is the customer site identification number, and A0001 is the air conditioner identification number). Thus, based on the statistical values ​​(Fall) of all "same type of equipment" associated with the air conditioning system 101, a combination of the statistical values ​​(Fall) of the two types of "same type of equipment" is generated as follows: Figure 5A , Figure 5B The resulting two-dimensional distribution density chart (G2dg) is visualized and associated with the identification number.

[0144] The two-dimensional distribution density chart (G2dg) generated in the above processing can be considered a chart representing the correlation between the statistical values ​​(Fall) of two "identical devices". Furthermore, the two-dimensional distribution density chart (G2dg) can be applied not only to the statistical values ​​(Fall) of "identical devices" but also to the measured values ​​(Ssop) of each device. This applies to the detailed analysis section 220 described later.

[0145] Figure 6A An example is shown of determining the "cause of anomaly" using a two-dimensional distribution density chart (G2dg) generated by summarizing statistical values ​​(Fall) for each device type in the monitoring mode of the overall analysis unit 210. In particular, the outputs of the anomaly measurement calculation unit 213, the threshold calculation unit 214, the anomaly detection unit 215, and the anomaly cause determination unit 216 are shown.

[0146] exist Figure 6A In the anomaly detection graph (Grp), the horizontal axis represents the passage of time (date), "AM" represents the anomaly measurement generated by the anomaly measurement calculation unit 213, "SL" represents the threshold (SL) generated by the threshold calculation unit 214, and "AD" represents whether the anomaly detection unit 215 has detected anomalies. Then, it is shown that when the anomaly measurement (AM) exceeds the threshold (SL), the anomaly detection (AD) is output.

[0147] exist Figure 6A For example, anomaly detection was performed continuously within anomaly intervals (Wd1) and (Wd2). Then, taking the anomaly intervals where anomalies were continuously detected as objects, for each anomaly interval, based on a two-dimensional distribution density chart (G2dg), the correlation between the anomaly detection results and the anomaly detection results was extracted. Figure 5A The abnormal correlation features (statistics) corresponding to the monitoring data (Derr).

[0148] Figure 6B This demonstrates the anomaly intervals (Wd1) and (Wd2) detected in monitoring mode, flexibly utilizing statistical values ​​(Fall) summarized by each device type, and... Figure 5A Two-dimensional distribution density charts (G2dg-1), (G2dg-2) and contribution charts of outlier correlation features (statistics) based on the two-dimensional distribution density chart (G2dg) (i.e., Gcnt-1), (Gcnt-2).

[0149] Here, the two-dimensional distribution density chart (G2dg-1) and contribution chart (Gcnt-1) represent the case of the outlier interval (Wd1), and the two-dimensional distribution density chart (G2dg-2) and contribution chart (Gcnt-2) represent the case of the outlier interval (Wd2).

[0150] In "learning mode", Figure 5A The two-dimensional distribution density chart (G2dg) of the normal data shown is calculated cyclically for the statistical values ​​(Fall) of two "same devices" and saved as an image. Therefore, multiple two-dimensional distribution density charts (G2dg) calculated cyclically can be obtained. Hereinafter, this image will be referred to as a "distribution density image".

[0151] Then, during anomaly detection, according to Figure 5A The outlier data (Derr) shown is plotted on the distribution density map to calculate the contribution of features, and features with high contribution are extracted sequentially as outlier-related features. This means that features that deviate from the distribution of normal data will be found in the distribution of outlier data.

[0152] exist Figure 6B In the abnormal interval (Wd1), as in the contribution chart (Gcnt-1), the statistical values ​​are extracted in the order of (Fall7), (Fall2), (Fall1)... Similarly, in the abnormal interval (Wd2), as in the contribution chart (Gcnt-2), the statistical values ​​are extracted in the order of (Fall7), (Fall4), (Fall1)...

[0153] Figure 7 This example illustrates a mapping for determining the cause of anomalies based on a combination of aggregated statistics (Fall) for each equipment type. The horizontal axis represents the aggregated statistics (Fall) for each equipment type, i.e., the characteristic quantities (e.g., cooling / heating capacity, refrigerant status, compressor efficiency, etc.), and the vertical axis represents the cause of the anomaly (e.g., refrigerant leakage, expansion valve failure, compressor failure, etc.). The "low / high" values ​​within the mapping represent the state (consistency) of the associated characteristic quantities (abnormal statistics) compared to the state of the characteristic quantities (normal statistics) in "learning mode." Here, "high" indicates close to the normal statistics.

[0154] Then, as Figure 7As shown, the correlation between the detected feature set (a set of multiple features from Fall1 to Fall7) and each anomaly cause is compared to determine the "anomaly cause" with high correlation. For example, in the case where the detected feature set is a combination of (Fall1), (Fall2), and (Fall7), the "anomaly cause" is AC5. That is, multiple statistical values ​​(features) and multiple anomaly causes are registered in a mapping form, and the anomaly cause is extracted from the mapping based on the correlation of the combinations of anomaly states of each statistical value (feature).

[0155] In this way, it is possible to evaluate whether there are any anomalies in a specified equipment type from multiple equipment types based on specified statistical values ​​associated with the specified equipment type, and to determine the "cause of the anomaly".

[0156] Next, the detailed analysis unit 220, which identifies the "abnormal device" from the "abnormal cause," will be explained. Here, the case where there are 3 indoor units is shown. Figure 8 It shows from Figure 2 The overall analysis unit 210 inputs a two-dimensional distribution density chart generated from the set of measured values ​​of the equipment related to the "cause of the anomaly" to the detailed analysis unit 220 to determine examples of "abnormal equipment". The idea behind the two-dimensional distribution density chart is the same as that for the statistical values ​​described above.

[0157] For example, for in Figure 6A In the "monitoring mode", the abnormal intervals (Wd1) and (Wd2) detected by the anomaly detection system are used flexibly. Figure 7 The abnormal cause is determined by mapping, assuming the "abnormal cause" is AC5, and the "abnormal cause" is an abnormality associated with the indoor unit (same type of equipment).

[0158] Next, an example of a two-dimensional distribution density image of normal data and monitoring data generated by the abnormal cause characteristic quantity analysis unit 222 of the detailed analysis unit 220 will be explained.

[0159] Figure 8 Indicates and Figure 5A Two-dimensional distribution density charts (G2sdg-1) and (G2sdg-2) of the same form, and charts of the number of device anomalies based on the two-dimensional distribution density chart (G2sdg) (Gscnt-1) and (Gscnt-2).

[0160] The two-dimensional distribution density chart (G2sdg-1) and the equipment anomaly occurrence count chart (Gscnt-1) represent the anomaly interval (Wd1), while the two-dimensional distribution density chart (G2sdg-2) and the equipment anomaly occurrence count chart (Gscnt-2) represent the anomaly interval (Wd2).

[0161] In "learning mode," the two-dimensional distribution density chart (G2sdg) calculates the measurements (Ssop) for two "identical devices" in a cyclic manner and saves them as an image. Therefore, multiple two-dimensional distribution densities (G2sdg) calculated cyclically can be obtained. This is the same as the case for statistical values.

[0162] Then, during anomaly detection in "monitoring mode", according to Figure 8 The abnormal data (Dserr) shown is plotted on multiple two-dimensional distribution density charts (G2sdg) calculated in a cyclic manner to detect where the anomaly occurs. In this case, it can be seen that the indoor unit 3 has an anomaly in the abnormal intervals (Wd1) and (Wd2).

[0163] Furthermore, the number of anomalies in the two-dimensional distribution density chart (G2sdg) of the measured values ​​(Ssop) of the two types of "same equipment" calculated in a cyclic manner is counted. If the anomaly occurrence count charts (Gscnt-1) and (Gscnt-2) are in the same manner, a specific "abnormal equipment" with a high number of anomalies is considered to be highly likely to have an abnormality in the indoor unit 3.

[0164] That is, using two-dimensional distribution density charts (G2sdg-1) and (G2sdg-2), the measured values ​​of individual devices related to the "cause of abnormality" are compared between normal data in "learning mode" and monitoring data in "monitoring mode" under normal conditions. The number of times the abnormality occurs for each indoor unit is counted, and the higher the count value, the higher the probability of the abnormality. In this case, indoor unit 3 is considered to have malfunctioned. Furthermore, while the "abnormal device" is captured as the indoor unit, it is also possible to include the constituent devices of the lower-level indoor unit as the object.

[0165] In this way, it is possible to evaluate whether there are any abnormalities in the specified equipment from the specified equipment types detected by anomaly detection, based on the specified measurement values ​​associated with the specified equipment, and to determine the location of the anomaly (abnormal equipment).

[0166] According to the above-described embodiment,

[0167] (1) By importing the statistical values ​​summarized by each type of equipment, even if some of the equipment stops or the operating status of some equipment changes, the statistical values ​​can still be obtained, so that monitoring can continue;

[0168] (2) By importing statistical values ​​summarized by each equipment type, anomalies can be detected with less operational data, thus shortening the initial learning period;

[0169] (3) Compared to generating the feature space using the measurement set of individual devices, the feature space is less dimensional and therefore not sparse. Therefore, it can make normal data dense enough to enable high-precision anomaly detection.

[0170] Next, the processing flow of the "learning mode" executed by the processor of the monitoring computer 200 will be explained. Figure 9 shows the learning processing flow based on the measured values ​​of individual devices in a normal air conditioning system and the statistical values ​​summarized by each device type (same type of device).

[0171] Step S901

[0172] In step S901, sensor signals (Ssig) from multiple devices transmitted from the air conditioning system 101 are first input into the sensor signal input unit 102. When a sensor signal (Ssig) is input, the processing in step S902 is executed.

[0173] Step S902

[0174] In step S902, during any period of the "learning mode", loop processing 1 is started, and the following processing is repeatedly executed until the arbitrary period ends. When loop processing 1 starts, the processing in step S903 is executed.

[0175] Step S903

[0176] In step S903, when loop processing 1 begins in step S902, loop processing 2 begins. Data is taken from each sensor signal when it is in normal condition, and the following processing is repeated until loop processing 2 ends. When loop processing 2 begins, the processing in step S904 is executed.

[0177] Step S904

[0178] In step S904, the measured value for each device is calculated based on the acquired data. For example, measured values ​​are calculated at 1-minute intervals, 1-hour intervals, and 1-day intervals. The measured values ​​are obtained by the sensor signal processing unit 211 converting the sensor signal into the measured value for each device, and the set of measured values ​​for each device related to the "cause of the anomaly" is input to the anomaly feature extraction unit 221. When this process is performed, the process in step S905 is executed.

[0179] Step S905

[0180] In step S905, the detailed analysis unit 220 performs loop processing 3 according to each number of abnormal causes, repeatedly executing the following processing until loop processing 3 ends. When this processing is performed, the processing in step S906 is executed.

[0181] Step S906

[0182] In step S906, the detailed analysis unit 220 further performs cyclic processing 4 by the number of each device (e.g., by the number of each indoor unit, by the number of each outdoor unit, by the number of each compressor), and repeatedly executes the following processing until cyclic processing 4 ends. When this processing is performed, the processing in step S907 is executed.

[0183] Step S907

[0184] In step S907, the anomaly feature extraction unit 222 extracts the measurement set of each device associated with the "anomaly cause," and the anomaly cause feature analysis unit 222 uses the measurement set of each device associated with each anomaly cause to generate a two-dimensional distribution density of normal data. When this process is performed, the process in step S908 is executed.

[0185] Step S908

[0186] In step S908, the two-dimensional distribution density of normal data is stored in the abnormal cause feature quantity learning data storage unit 223. This storage process completes loop process 4. When this process is executed, the process in step S909 is performed.

[0187] Step S909

[0188] In step S909, the sensor signal processing unit 211 calculates the statistical value of "same type of device" based on the measured value of each device, and inputs the statistical value set of "same type of device" into the anomaly cause determination unit 216. When this process is performed, the process of step S910 is executed.

[0189] Step S910

[0190] In step S910, the anomaly cause determination unit 216 extracts a set of statistical values ​​for "same type of equipment" associated with the "anomaly cause," and uses the set of statistical values ​​for the "same type of equipment" associated with each anomaly cause to generate a two-dimensional distribution density of normal data. When this process is performed, the process in step S911 is executed.

[0191] Step S911

[0192] In step S911, the two-dimensional distribution density of normal data is stored in the anomaly cause determination unit 217. This storage process ends loop process 3. When this process is executed, the process in step S912 is performed.

[0193] Step S912

[0194] In step S912, the statistical value set of "same type of device" is input to the feature vector extraction unit 212 through the overall analysis unit 210. When this process is performed, the process of step S913 is executed.

[0195] Step S913

[0196] In step S913, the feature vector extraction unit 212 extracts feature vectors using statistical values ​​from "same type of device". When this process is performed, the process in step S914 is executed.

[0197] Step S914

[0198] In step S914, the anomaly measurement calculation unit 213 calculates the anomaly measure for each feature vector at predetermined intervals (sometimes represented as each moment) using the feature vectors from a pre-specified learning period. When this process is executed, the process in step S915 is performed.

[0199] Step S915

[0200] In step S915, the threshold (SL) corresponding to the anomaly measure is calculated by the threshold calculation unit 214. When this process is performed, the process of step S916 is executed.

[0201] Step S916

[0202] In step S916, loop processing 2 ends. When this processing is executed, the processing in step S917 is executed.

[0203] Step S917

[0204] In step S917, loop processing 1 ends. When this processing is executed, the "learning mode" ends.

[0205] Next, the processing flow of the "monitoring mode" executed by the processor of the monitoring computer 200 will be explained. Figures 10A to 10C This describes the monitoring process based on the measured values ​​of individual devices in a monitored air conditioning system after learning, as well as the statistical values ​​summarized by each device type. First, according to... Figure 10A Please provide an explanation.

[0206] Step S1001

[0207] In step S1001, sensor signals (Ssig) from multiple devices transmitted from the air conditioning system 101 are first input into the sensor signal input unit 102. When a sensor signal (Ssig) is input, the processing in step S1002 is executed.

[0208] Step S1002

[0209] In step S1002, during any period of the monitoring mode, loop processing 1 begins and the following processing is repeatedly executed until the arbitrary period ends. When loop processing 1 begins, the processing in step S1003 is executed.

[0210] Step S1003

[0211] In step S1003, when loop processing 1 begins in step S1002, loop processing 2 begins, and the data monitored for each sensor signal is acquired and the following processing is repeatedly executed. When loop processing 2 begins, the processing in step S1004 is executed.

[0212] Step S1004

[0213] In step S1004, the measured value for each device is calculated. For example, measured values ​​at 1-minute intervals, 1-hour intervals, and 1-day intervals are calculated. The measured values ​​are obtained by the sensor signal processing unit 211 converting the sensor signal into the measured value for each device, and the set of measured values ​​for each device associated with the "abnormal cause" is input to the abnormal feature quantity extraction unit 221. When this process is performed, the process in step S1005 is executed.

[0214] Step S1005

[0215] In step S1005, the sensor signal processing unit 211 calculates the statistical value of "same type of device" based on the measured value of each device, and inputs the statistical value set of "same type of device" to the feature vector extraction unit 212. When this processing is performed, the processing of step S1006 is executed.

[0216] Step S1006

[0217] In step S1006, the feature vector extraction unit 212 extracts the feature vectors of the statistical values ​​used for "same type of device". When this process is performed, the process in step S1007 is executed.

[0218] Step S1007

[0219] In step S1007, the anomaly measurement calculation unit 213 uses the feature vectors from a pre-specified monitoring period to calculate the anomaly degree for each feature vector at predetermined intervals (sometimes represented as individual moments). This corresponds to... Figure 6A The anomaly measure (AM) is shown. When this process is performed, the process of step S1008 is executed.

[0220] Step S1008

[0221] In step S1008, the anomaly detection unit 215 compares the anomaly measure calculated in step S1007 with the threshold (SL) obtained based on the anomaly measure under the "learning mode," and determines whether an anomaly exists based on their magnitude. This corresponds to... Figure 6A The anomaly measure (AM) is compared with the threshold (SL). If an anomaly is determined, an anomaly detection (AD) is generated. When this process is performed, the process in step S1009 is executed.

[0222] Step S1009

[0223] In step S1009, if the anomaly detection unit 216 determines that the anomaly measure obtained in step S1007 is normal, the process returns to step S1003, and the processing of S1003 to S1008 is repeated for the next anomaly measure. On the other hand, if the anomaly measure obtained in step S1007 is determined to be anomaly, the processing of step S1010 is executed. The processing after step S1010 is as follows... Figure 10B As shown.

[0224] Step S1010

[0225] In step S1010, the overall analysis unit 210 performs cyclic processing 3 based on the number of abnormal causes, repeatedly executing the following processing. When this processing is executed, the processing in step S1011 is performed.

[0226] Step S1011

[0227] In step S1011, the anomaly cause determination unit 216 extracts a set of statistical values ​​for "same type of equipment" associated with the "anomaly cause". Then, using the set of statistical values ​​for the "same type of equipment" associated with each anomaly cause, a two-dimensional distribution density of the monitoring data is generated. Furthermore, "anomaly causes" include, for example, refrigerant leakage, expansion valve failure, and compressor failure. When this process is performed, the process in step S1012 is executed.

[0228] Step S1012

[0229] In step S1012, loop processing 3 ends. When this processing is executed, the processing in step S1013 is executed.

[0230] Step S1013

[0231] In step S1013, the anomaly cause determination unit 216 compares the normal data and the two-dimensional distribution density of the monitoring data using the statistical value set of "same type of equipment" associated with each anomaly cause. When this process is performed, the process in step S1014 is executed.

[0232] Step S1014

[0233] In step S1014, based on the comparison in step S1013, the statistical values ​​(hereinafter referred to as abnormal characteristic quantities) of the "same type of equipment" that are abnormal are counted according to the abnormal characteristic quantities. This corresponds to Figure 6B The contribution graph (Gcnt) of the abnormal feature quantities (statistics) in the data. When this process is performed, the processing in step S1015 is executed.

[0234] Step S1015

[0235] In step S1015, the abnormal features are sorted according to the order in which they were counted in step S1014. This is also consistent with... Figure 6B The contribution chart (Gcnt) of the abnormal feature quantities (statistics) in the data corresponds to, for example, sorted in the manner of "Fall7", "Fall2", ... When this process is performed, the processing in step S1016 is executed.

[0236] Step S1016

[0237] In step S1016, the specified upper N-bit abnormal feature values ​​are extracted from the sorted abnormal feature values. This is also consistent with... Figure 6B The contribution chart (Gcnt) of the abnormal feature quantities (statistics) corresponds to this, for example, extracting "Fall7", "Fall2", "Fall1", and "Fall4". The upper N bits are determined based on the number of main feature quantities that can distinguish the "cause of the anomaly". When this process is performed, the processing in step S1017 is executed.

[0238] Step S1017

[0239] In step S1017, the degree of consistency with the main characteristic quantities of each abnormal cause is calculated. This is consistent with... Figure 7 Each feature is assigned a "low / high" correspondence. When this process is performed, the process in step S1018 is executed.

[0240] Step S1018

[0241] In step S1018, the proportion (correlation) of each abnormal cause is calculated based on the consistency of the characteristic quantities. This is for... Figure 7 The consistency score is calculated by evaluating the set shown. When this process is performed, step S1018 is executed.

[0242] Step S1019

[0243] In step S1019, abnormal causes (ACs) are assigned a ranking order starting from those with the highest occurrence rates. For example, they are ranked in the order of "AC5", "AC4", etc. When this process is executed, the process in step S1020 is performed. The processing after step S1020 is as follows... Figure 10C As shown.

[0244] Step S1020

[0245] In step S1020, the first position of the occurrence rate of "abnormal cause" in the ranking obtained in step S1019 is determined as "abnormal cause".

[0246] Step S1021

[0247] In step S1021, the anomaly cause feature quantity extraction unit 222 of the detailed analysis unit 220 extracts the set of measurement values ​​for each device related to the "anomaly cause" from the measurement values ​​of each device stored in the sensor signal processing unit 211, based on the "anomaly cause" determined by the anomaly cause determination unit 216. When this process is performed, the process of step S1021 is executed.

[0248] Step S1022

[0249] In step S1021, the following process is repeatedly executed, starting with the number of devices associated with each "abnormal cause". While this process is being executed, the process in step S1023 is performed.

[0250] Step S1023

[0251] In step S1023, the anomaly cause feature analysis unit 222 takes the two types of data related to each anomaly cause from the extracted measurement set of each device as input, and generates a two-dimensional distribution density of the monitoring data based on cyclic combinations for each device. This corresponds to... Figure 8 The two-dimensional distribution density chart (G2sdg) is generated. When this process is performed, the process of step S1024 is executed.

[0252] Step S1024

[0253] In step S1024, loop processing 4 ends. When this processing is executed, the processing in step S1025 is executed.

[0254] Step S1025

[0255] In step S1025, the anomaly location determination unit 224, based on the "anomaly cause" determined by the anomaly cause determination unit 216, selects the two-dimensional distribution density of normal data generated by the measurement value set of each device related to the "anomaly cause" from the anomaly cause feature quantity learning data storage unit 223, and compares it with the two-dimensional distribution density of monitoring data generated by the measurement value set of each device in "monitoring mode". When this process is performed, the process of step S1026 is executed.

[0256] Step S1026

[0257] In step S1026, based on the comparison in step S1025, the number of faulty devices is counted for each faulty device. This corresponds to... Figure 8 The chart showing the number of device anomalies (Gscnt) is used. The device with the highest count is identified as the "abnormal device". When this process is executed, step S1027 is performed.

[0258] Step S1027

[0259] In step S1027, loop processing 2 ends. When this processing is executed, the processing in step S1028 is executed.

[0260] Step S1028

[0261] In step S1028, loop processing 1 ends. When this processing is executed, the monitoring mode ends.

[0262] Next, the screen structure of the display screen connected to the monitoring computer 200 will be explained. Figure 11 An example of a screen showing abnormality warnings, the cause of the abnormality, and the abnormal device (abnormal location) is shown.

[0263] exist Figure 11 In this context, the display screen (DSPY) is shown on the screen of the display device connected to the monitoring computer 200. The display screen (DSPY) shows the abnormality warning detection display box (Dpd), the abnormality cause determination box (Dic), and the abnormal device determination box (Ddi).

[0264] The anomaly detection display box (Dpd) displays the time sequence. Figure 6A The anomaly detection graph (Grp) shown is used to read the anomaly detection over time, thus enabling the identification of early warning signs of anomalies.

[0265] Displayed in the exception cause determination box (Dic) Figure 6BThe diagram shows a two-dimensional distribution density chart (G2dg) and a contribution chart of anomaly-related features (Gcnt). By visualizing the basis for determining the "cause of the anomaly," the user's understanding is enhanced. Furthermore, the specific text display box (SC) for the cause of the anomaly clearly indicates the cause of the failure.

[0266] Displayed in the Abnormal Device Identification Box (Ddi) Figure 8 The diagram shows a two-dimensional distribution density chart (G2sdg) and a chart showing the number of times the device malfunctions (Gscnt). This visualizes the basis for identifying "malfunctioning devices," thus enhancing user understanding. Furthermore, the specific text display box (SE) for the malfunctioning device is also shown, allowing for a clear understanding of the "malfunctioning device."

[0267] The invention is characterized by acquiring multiple measurement values ​​related to multiple device types from multiple sensors, selecting measurement values ​​from sensors related to the device type, calculating statistical values ​​based on the selected values, calculating a set of feature quantities representing the position in the feature quantity space for each of the multiple device types based on the calculated statistical values, registering the feature quantity set as normal data in the feature quantity space when in a learning state, determining whether there is an anomaly based on the deviation of the feature quantity set from the normal data when in a monitoring state after learning, and determining whether there is an anomaly when it is determined to be an anomaly, identifying a specified device type as the device type that has an anomaly from the multiple device types, and identifying the device that has an anomaly from one or more devices that are specified as device types.

[0268] Therefore, it is possible to detect abnormalities in air conditioning equipment consisting of one or more outdoor units and one or more indoor units, which are connected by piping for refrigerant circulation, determine the cause of the abnormality, and identify which equipment has malfunctioned.

[0269] Furthermore, the present invention is not limited to the above-described embodiments, but includes various modifications. The embodiments described in detail above are for the purpose of readily understanding and illustrating the present invention, and are not limited to possessing all the described structures. Additionally, a portion of the structure of one embodiment can be replaced with the structure of another embodiment, and structures of other embodiments can be added to the structure of one embodiment. For the structures of each embodiment, other structures can also be added, deleted, or replaced.

Claims

1. An anomaly detection system comprising: an air conditioning system having multiple devices and sensors; and a monitoring computer that monitors the air conditioning system. Its features are, The multiple devices consist of multiple types of devices. The monitoring computer performs the following processing: (1) Obtain multiple measurements related to the multiple devices from the multiple sensors. (2) For the aforementioned multiple equipment types: (2A) Select the sensor measurement value that is relevant to the type of device. (2B) Calculate statistical values ​​based on the selected measured values. (3) For each of the plurality of device types, calculate a set of feature quantities representing the position in the feature quantity space based on the calculated plurality of statistical values. (4) When in the process of learning: (4A) Register the set of features as normal data into the feature space. (5) In the case of monitoring after learning: (5A) Based on the deviation of the feature set from normal data, determine whether there is an anomaly. (5B) In cases where an anomaly is identified: (5Ba) From the plurality of equipment types, the specified equipment type is determined as the equipment type that has experienced the abnormality. (5Bb) Identify the device that has malfunctioned from one or more devices that are the specified device types.

2. The anomaly detection system according to claim 1, characterized in that, Step (5Ba) involves evaluating whether there are any abnormalities in the specified equipment types from among the multiple equipment types, based on the specified statistical values ​​associated with the specified equipment types, and determining the cause of the abnormalities.

3. The anomaly detection system according to claim 2, characterized in that, The step (5Bb) is to evaluate whether the specified equipment has any abnormalities from the specified equipment types detected by the anomaly detection, based on the specified measurement values ​​associated with the specified equipment, and to determine the location of the anomaly.

4. The anomaly detection system according to claim 3, characterized in that, The plurality of devices have at least two states: one is operational and the other is not operational. The calculation of the statistical value in step (2B) treats the measured values ​​related to the stopped equipment as outside the scope of consideration, or as predetermined values ​​for the calculation.

5. A monitoring computer having a processor and an interface for monitoring an air conditioning system, characterized in that, The air conditioning system has multiple devices and sensors. The multiple devices consist of multiple types of devices. The processor performs the following processing: (1) Measurement values ​​are obtained from the plurality of said sensors via said interface. (2) For each type of equipment constituting the air conditioning system: (2A) Select the sensor measurement value relevant to the type of equipment. (2B) Statistical values ​​are calculated based on the selected measurements. (3) For each of the multiple device types, based on the calculated multiple statistical values, calculate the set of feature quantities representing the position in the feature quantity space. (4) When in the process of learning: (4A) Register the set of features as normal data into the feature space. (5) In the case of monitoring after learning: (5A) Based on the deviation of the feature set from normal data, determine whether there is an anomaly. (5B) In cases where an anomaly is identified: (5Ba) From multiple equipment types, the specified equipment type is identified as the equipment type that has experienced the abnormality. (5Bb) Identify the device that has malfunctioned from one or more devices that are the specified device types.

6. An anomaly detection method for an anomaly detection system, the anomaly detection system comprising: an air conditioning system having multiple devices for air conditioning and multiple sensors for measuring the operational status of the multiple devices; and a monitoring computer for monitoring the operational status of the air conditioning system. Its features are, The monitoring computer performs the following steps: The steps are: acquiring multiple measurement values ​​related to multiple device types from multiple sensors, selecting measurement values ​​related to the same type of device, and calculating statistical values ​​based on the selected measurement values; The step of calculating a set of features representing a position in the feature space based on statistical values ​​of at least two of the same devices; In the learning state, the step of registering the feature set as normal data into the feature space; In the case of monitoring after learning, the step of determining the cause of the anomaly based on the degree of deviation between the feature set and the normal data; Based on the stated cause of the anomaly, the step of identifying the specified device from among a plurality of such devices as the device that has malfunctioned, and then identifying the device that has malfunctioned from among the specified such devices.

7. An anomaly detection system comprising: an air conditioning system having a plurality of devices for performing air conditioning and a plurality of sensors for measuring the operational status of the plurality of devices; and a monitoring computer for monitoring the operational status of the air conditioning system. Its features are, The monitoring computer has an overall analysis unit and a detailed analysis unit. The overall analysis unit has the following features: The measurement acquisition unit acquires multiple measurement values ​​related to the multiple devices from the multiple sensors; The statistical value calculation unit calculates the statistical value of the same type of equipment based on the measured values ​​related to the same type of equipment among the plurality of equipment; The statistical value distribution density calculation unit calculates a set of statistical values ​​representing the positions of the statistical values ​​in a two-dimensional plane of statistical values, with the statistical values ​​of the two sets of the same type of equipment as parameters; The statistical value normal data storage unit, in learning mode, registers the statistical value set as normal data in the statistical value two-dimensional plane; The anomaly detection unit, in the monitoring mode executed after the learning mode, determines whether the same type of device has any anomalies based on the deviation between the statistical value set calculated by the statistical value distribution density calculation unit and the statistical value set stored in the statistical value normal data storage unit. as well as The anomaly cause determination unit determines the cause of the anomaly based on the correlation of multiple anomalies detected by the anomaly detection unit. The detailed analysis section includes: The measured value distribution density calculation unit calculates a set of measured values ​​representing the positions of the measured values ​​in a two-dimensional plane of measured values, with the two sets of measured values ​​as parameters; The normal data storage unit for measured values ​​registers the set of measured values ​​as normal data in the two-dimensional plane of the measured values ​​in the learning mode. as well as The abnormal device determination unit, in a monitoring mode executed after the learning mode, selects the measurement value related to the cause of the abnormality from the abnormality cause determination unit, and determines the same type of device related to the cause of the abnormality based on the deviation state between the measurement value set calculated by the measurement value distribution density calculation unit based on the selected measurement value and the measurement value set stored in the measurement value normal data storage unit, and then determines the abnormal device that has malfunctioned from the same type of device.

8. The anomaly detection system according to claim 7, characterized in that, The two-dimensional plane of statistical values ​​obtained by the statistical value distribution density calculation unit is only calculated by iteratively calculating the number of two groups of statistical values ​​out of a plurality of statistical values. The two-dimensional plane of the measured values ​​obtained by the measured value distribution density calculation unit is only calculated in a cyclic manner for two groups of the measured values.

9. The anomaly detection system according to claim 7, characterized in that, The statistical value distribution density calculation unit converts the statistical values ​​into feature vectors and registers them in the form of an image on the two-dimensional plane of the statistical values. The measured value distribution density calculation unit converts the measured value into a feature vector and registers it in the form of an image on the two-dimensional plane of the measured value.

10. The anomaly detection system according to claim 7, characterized in that, The anomaly cause determination unit registers the abnormal states of multiple statistical values ​​and the multiple anomaly causes in a mapping form, and extracts the anomaly causes from the mapping based on the correlation of the combinations of abnormal states of each statistical value.

11. A recording medium storing a program for an anomaly detection system, the program causing a monitoring computer used in the anomaly detection system to perform actions. The anomaly detection system comprises: an air conditioning system having multiple devices for air conditioning and multiple sensors for measuring the operational status of the multiple devices; and a monitoring computer for monitoring the operational status of the air conditioning system. Its features are, The recording medium includes: A measurement acquisition procedure that acquires multiple measurement values ​​related to the multiple devices from the multiple sensors; A statistical value calculation program that calculates a statistical value for the same type of equipment based on the measured values ​​associated with the same type of equipment among the plurality of equipment; A statistical value distribution density calculation program, which calculates a set of statistical values ​​representing the positions of the statistical values ​​in a two-dimensional plane of statistical values, with two sets of statistical values ​​associated with the same type of equipment as parameters; The statistical value normal data storage program, in learning mode, registers the statistical value set as normal data in the statistical value two-dimensional plane; An anomaly detection program, which executes in monitoring mode after learning mode, determines whether the same type of device has any anomalies based on the deviation between the statistical value set calculated by the statistical value distribution density calculation program and the statistical value set stored by the normal data storage program. An anomaly cause determination procedure determines the cause of an anomaly based on the correlation of multiple anomalies detected by the anomaly detection procedure. A program for calculating the distribution density of measured values, which calculates a set of measured values ​​representing the positions of the measured values ​​in a two-dimensional plane with two sets of measured values ​​as parameters; The normal data storage procedure for measured values, in learning mode, registers the set of measured values ​​as normal data in the two-dimensional plane of the measured values; as well as An abnormal device identification procedure, which is executed in monitoring mode after learning mode, selects the measurement value related to the cause of the abnormality determined by the abnormality cause identification procedure, and determines the same type of device related to the cause of the abnormality based on the deviation state of the measurement value set calculated by the measurement value distribution density calculation procedure based on the selected measurement value and the measurement value set stored by the measurement value normal data storage procedure, and then identifies the abnormal device that has malfunctioned from the same type of device.

Images