Multi-level fusion building system sensing device data anomaly attribution method and system

Abstract

The invention provides a multi-level fusion building system sensing device data anomaly attribution method and system. The method comprises the following steps: acquiring current sensing data and a preprocessed sensing data sequence of a sensing device in a building system; determining an extreme abnormal value according to whether the current sensing data exceed a reasonable range of the preprocessed sensing data sequence; data exceeding a reasonable range in the preprocessed sensing data sequence is removed, and a non-extreme sensing data sequence is obtained; calculating the deviation of the current sensing data relative to the data in the non-extreme sensing data sequence to obtain a trend abnormal value; reasoning the current sensing data through a pre-trained structural anomaly model to obtain a structural anomaly value; obtaining a rule abnormal value based on the operation rule model; and comparing each abnormal value with a preset abnormal threshold value to obtain a data abnormal reason of the sensing device. According to the invention, the data abnormity reason of the sensing device is explained, and the abnormity attribution judgment of the sensing device is credible.

Description

Technical Field

[0001] The present invention relates to the fields of intelligent building operation and maintenance and Internet of Things edge computing, and specifically to a method and system for attributing data anomalies of building system sensor devices using multi-level fusion. Background Art

[0002] With the widespread adoption of green and smart buildings, various environmental and automatic control system sensors (such as temperature, humidity, air volume, and power) provide crucial data support for building control and energy efficiency optimization. In the context of the dual energy consumption control policy, building energy consumption monitoring requires data traceability and the ability to accurately locate faults. However, traditional data collection platforms cannot meet the requirements for accurate judgment in complex scenarios. The core contradiction of this situation is that the root cause of abnormal sensor data is difficult to determine.

[0003] The root cause of sensor data anomalies is difficult to pinpoint. Sensor performance gradually degrades over time due to phenomena like aging and data drift, which in turn reduces the credibility of the data. Traditional methods obtain independent measurements (such as temperature, humidity, and power at a specific moment) from a single sensor or data source. These measurements lack correlation analysis with other related sensor data and a multi-level collaborative judgment mechanism, making it difficult to trace controller erroneous commands or logical anomalies, resulting in a high false alarm rate. Relying solely on existing AI models, due to their lack of ability to interpret engineering rules, makes anomaly detection results difficult for operations and maintenance personnel to trust, leading to a dilemma of having models but no responsiveness.

[0004] Therefore, there is an urgent need for a comprehensive judgment method that integrates multi-level analysis and engineering mechanisms to achieve reliable determination of the cause of sensor device anomalies. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides a multi-level fusion building system sensor device data anomaly attribution method and system.

[0006] In order to solve the above technical problems, the present invention is solved by the following technical solutions: A multi-level fusion building system sensor device data anomaly attribution method includes the following steps: Obtain current sensor data from different sensor devices in the building system; Acquire historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocess the historical sensor data sequence to obtain a preprocessed sensor data sequence; Determining an extreme outlier based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; Eliminate data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; Calculate the deviation of the current sensor data from the data in the non-extreme sensor data sequence to obtain the trend outlier; The current sensor data is inferred through the pre-trained structural anomaly model to obtain the current data prediction value; the current data prediction value is compared with the current sensor data to obtain the structural anomaly value; Based on the current sensor data and in combination with the operating rule model, the total value of the rule anomaly weights that meet the abnormal condition is compared with the total value of the rule anomaly weights that apply the abnormal condition to obtain the rule anomaly value; wherein the operating rule model is constructed based on the abnormal condition of the sensor device and the rule anomaly weights that meet the abnormal condition, and the abnormal condition is set according to the operating rules of the sensor device in the building system; The extreme outliers, trend outliers, structural outliers and regular outliers are compared with the preset anomaly thresholds to obtain the cause of the sensor device data anomaly.

[0007] As an implementable method, the extreme outlier value is determined based on whether the current sensor data exceeds a reasonable range, specifically as follows: When the current sensing data of the target sensing device exceeds the reasonable range of the corresponding sensing device data, the target sensing device has an extreme abnormality, and the extreme abnormality value is a first preset extreme abnormality value; when the current sensing data of the target sensing device is within the reasonable range of the corresponding sensing device data, the target sensing device does not have an extreme abnormality, and the extreme abnormality value is a second preset extreme abnormality value; The reasonable range of the sensor device data is expressed as follows:

[0008] in, Represents the first The first quartile of the sensor device data, Represents the first The third quartile of the sensor device data, Represents the first The interquartile range of the sensor data, , Respectively represent The weights of the first and third quartiles of the sensor device data.

[0009] As an implementation method, calculating the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence to obtain the trend outlier includes the following steps: Acquire the latest data of a first preset time length in the non-extreme sensor data sequence to obtain a first time length sensor data sequence, and then obtain the average value and standard deviation of the first time length sensor data sequence; Calculate the skewness of the non-extreme sensor data sequence to obtain the first sequence skewness; When the skewness of the first sequence is less than the first preset skewness threshold, the deviation reference value is the first preset deviation threshold; When the skewness of the first sequence is greater than or equal to a first preset skewness threshold, the non-extreme sensor data sequence and the current sensor data are transformed to obtain a transformed data sequence and a transformed current sensor data; the skewness of the transformed data sequence is calculated to obtain a second sequence skewness; and a deviation reference value is obtained by comparing the second sequence skewness with the first preset skewness threshold and the second preset skewness threshold; Obtaining a transformation trend deviation value based on transforming the current sensor data and the average value and standard deviation of the first time-length sensor data sequence; Based on the deviation from the baseline value and the transformed trend deviation value, the trend anomaly value is obtained; The transformation trend deviation value is expressed as follows:

[0010] The trend outliers are expressed as follows:

[0011] in, Indicates the The deviation value of the transformation trend of the sensor device data, Indicates the Transformation of sensor data Current sensor data, Indicates the current timestamp, Indicates the first preset duration, Indicates the The average value of the first time-length sensor data sequence of the sensor device data, Indicates the The standard deviation of the first time-length sensor data sequence of the sensor device data, Indicates the Trend anomalies of each sensor device, Indicates returning the smaller value among multiple values. Indicates the The deviation of the sensor device data from the baseline value.

[0012] As an implementable method, the transforming of the non-extreme sensor data sequence and the current sensor data to obtain the transformed data sequence and the transformed current sensor data respectively includes the following steps: When all data in the non-extreme sensor data sequence are positive numbers, performing maximum likelihood estimation processing on the non-extreme sensor data sequence to obtain first transformation parameters; performing first transformation processing on the non-extreme sensor data sequence and the current sensor data based on the first transformation parameters to obtain a first transformed data sequence and first transformed current sensor data; When there are negative numbers in the non-extreme sensor data sequence, performing maximum likelihood estimation processing on the non-extreme sensor data sequence to obtain second transformation parameters; performing second transformation processing on the non-extreme sensor data sequence and the current sensor data based on the second transformation parameters to obtain a second transformed data sequence and second transformed current sensor data; The first transformation process is represented as follows:

[0013] The second transformation process is expressed as follows:

[0014] in, Indicates the first transformed data sequence or the second transformed data sequence. The first sensor data sequence data, Indicates the first The first sensor data sequence data, Indicates the a first transformation parameter of the sensor device data, Indicates the A second transformation parameter of the sensor device data.

[0015] As an implementation method, obtaining the deviation reference value based on the comparison of the second sequence skewness with the first preset skewness threshold and the second preset skewness threshold includes the following steps: When the skewness of the second sequence is less than the first preset skewness threshold, the deviation reference value is the first preset deviation threshold; When the skewness of the second sequence is greater than or equal to the first preset skewness threshold and less than the second preset skewness threshold, the deviation reference value is the second preset deviation threshold; When the skewness of the second sequence is greater than or equal to the second preset skewness threshold, the deviation reference value is the third preset deviation threshold; wherein the first preset deviation threshold>the second preset deviation threshold>the third preset deviation threshold.

[0016] As an implementable method, the method uses a pre-trained structural anomaly model to infer the current sensor data to obtain a current data prediction value; and compares the current data prediction value with the current sensor data to obtain a structural anomaly value, including the following steps: Inferring the current sensor data of the target sensor device through the pre-trained target sensor device structural anomaly model to obtain the predicted value of the current data of the target sensor device; Performing square processing on the difference between the current sensing data of the target sensor device and the current data prediction value to obtain the current reconstruction error of the target sensor device; Obtaining a structural anomaly value of the target sensing device based on a preset reconstruction error threshold and a current reconstruction error of the target sensing device; Calculate the structural anomaly values ​​of other sensing devices; The structural abnormality value of the target sensor device is expressed as follows:

[0017] in, Indicates the The structural abnormal value of each sensor device is the structural abnormal value of the target sensor device. Indicates the The current timestamp of each sensor device The reconstruction error is the current reconstruction error of the target sensor device. , Indicates the The current sensing data of a sensor device, that is, the target sensor device, Indicates the The predicted value of the current data of each sensor device, Indicates the The preset reconstruction error threshold of each sensor device, Indicates that the smaller value among multiple numeric values ​​is returned.

[0018] As an implementation method, based on the current sensor data and in combination with the running rule model, the total value of the rule anomaly weights that meet the abnormal conditions is compared with the total value of the rule anomaly weights that apply the abnormal conditions to obtain the rule anomaly value, including the following steps: If the current sensor data of the target sensor device meets the abnormal condition, the corresponding rule abnormal weight is added to the target satisfaction rule abnormal weight to obtain the total value of the target satisfaction rule abnormal weight; If the current sensing data of the target sensing device is applicable to the abnormal condition, the corresponding rule abnormal weight is added to the target applicable rule abnormal weight to obtain the total target applicable rule abnormal weight; Obtaining a rule anomaly value of the target sensor device based on a ratio of a total value of the target satisfied rule anomaly weights to a total value of the target applicable rule anomaly weights; Calculate regular anomalies of other sensing devices; The regular abnormal value of the target sensor device is expressed as follows:

[0019] in, Indicates the Regular outliers for each sensor device, Indicates the The total value of the abnormal weight of the target satisfying rule of each sensor device, Indicates the The total value of the target applicable rule anomaly weight of each sensor device.

[0020] As an implementable method, the method of comparing the extreme outliers, trend outliers, structural outliers, and rule outliers with a preset outlier threshold value to obtain the cause of the sensor device data anomaly includes the following steps: Performing weighted summation on the extreme outliers, trend outliers, structural outliers, and regular outliers of the target sensor device to obtain the total outlier value of the target sensor device; When the total abnormal value of the target sensor device is greater than the preset target abnormal total threshold, the target sensor device in the building system is abnormal, and the abnormal reason is: When the rule abnormality value of the target sensor device is greater than the corresponding first preset rule abnormality threshold, the target sensor device has a physical-logical conflict abnormality; When the structural abnormality value of the target sensing device is greater than the corresponding first preset structural abnormality threshold and the rule abnormality value is less than the corresponding second preset rule abnormality threshold, the target sensing device has a structural abnormality; When the trend abnormality value of the target sensor device is greater than the corresponding first trend abnormality threshold, there is no extreme abnormality, the structural abnormality value is less than the corresponding second preset structural abnormality threshold, and the rule abnormality value is less than the corresponding third preset rule abnormality threshold, then the target sensor device has a trend abnormality; When the target sensor device has an extreme abnormality, the abnormality cause of the target sensor device is an extreme abnormality; Among them, the first preset rule abnormality threshold>the second preset rule abnormality threshold>the third preset rule abnormality threshold, and the first preset structure abnormality threshold>the second preset structure abnormality threshold.

[0021] A multi-level fusion building system sensor device data anomaly attribution system, capable of implementing the above-mentioned method, includes a current data module, a historical data module, an extreme anomaly module, a data rejection module, a trend anomaly module, a structural anomaly module, a rule anomaly module, and an anomaly attribution module; The current data module obtains current sensor data of different sensor devices in the building system; The historical data module acquires historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocesses the historical sensor data sequence to obtain a preprocessed sensor data sequence; The extreme anomaly module determines an extreme anomaly value based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; The data elimination module eliminates data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; The trend anomaly module calculates the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence to obtain a trend anomaly value; The structural anomaly module uses a pre-trained structural anomaly model to infer the current sensor data to obtain a current data prediction value; compares the current data prediction value with the current sensor data to obtain a structural anomaly value; The rule exception module compares the total value of the rule exception weights that meet the exception condition with the total value of the rule exception weights that apply the exception condition based on the current sensor data in combination with the operation rule model to obtain a rule exception value; wherein the operation rule model is constructed based on the abnormal condition of the sensor device and the rule exception weights that meet the abnormal condition, and the abnormal condition is set according to the operation rule of the sensor device in the building system; The anomaly attribution module compares extreme anomalies, trend anomalies, structural anomalies, and rule anomalies with preset anomaly thresholds to obtain the cause of the sensor device data anomaly.

[0022] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0023] A multi-level fusion building system sensor device data anomaly attribution device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the method described above is implemented.

[0024] The present invention has significant technical effects due to the adoption of the above technical solutions: The present invention provides a multi-level fusion method and system for attributing abnormal data of building system sensor devices. It is a multi-level fusion mechanism that first analyzes statistics, then structures, and finally physical logic. Based on historical data and current data of the sensor device, the extreme abnormal values, trend abnormal values, and structural abnormal values ​​of the sensor device are obtained, and the engineering rule set is executed in a linked manner to obtain rule abnormal values ​​for targeted verification. Each type of abnormal value is progressively coordinated step by step. Finally, each abnormal value is fused and analyzed to form a multi-level fusion to obtain the cause of the abnormality of the building system sensor device. The extreme abnormal values, trend abnormal values, structural abnormal values, and rule abnormal values ​​in the multi-level fusion method for attributing abnormal data of building system sensor devices proposed by the present invention have the dynamic collaborative characteristics of layer-by-layer enhancement, cross-verification, and result feedback. The final cause of the abnormality of the sensor device is explanatory, so the determination of the attribution of the abnormality of the sensor device is credible, and the operation and maintenance personnel can trust it and take corresponding responses. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 Schematic diagram of a process of an embodiment of the method of the present invention; Figure 2 Schematic diagram of the overall embodiment of the system of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described in detail below with reference to the embodiments. The following embodiments are for explanation of the present invention and the present invention is not limited to the following embodiments. In the absence of conflict, the features in the following embodiments may be combined with each other.

[0028] Example 1: A multi-level fusion method for anomaly attribution of building system sensor device data, such as Figure 1 As shown, the following steps are included: S100: Acquiring current sensor data of different sensor devices in the building system; S200: acquiring historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocessing the historical sensor data sequence to obtain a preprocessed sensor data sequence; S300: determining an extreme outlier based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; S400: Eliminate data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; S500: Calculate the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence to obtain a trend outlier; S600: Inferring the current sensor data using the pre-trained structural anomaly model to obtain a current data prediction value; comparing the current data prediction value with the current sensor data to obtain a structural anomaly value; S700: Based on the current sensor data and in combination with the operating rule model, a total value of the rule abnormality weights that satisfy the abnormal condition is compared with a total value of the rule abnormality weights that apply the abnormal condition to obtain a rule abnormal value; wherein the operating rule model is constructed based on the abnormal condition of the sensor device and the rule abnormality weights that satisfy the abnormal condition, and the abnormal condition is set according to the operating rules of the sensor device in the building system; S800: Compare the extreme outliers, trend outliers, structure outliers, and rule outliers with the preset outlier thresholds to obtain the cause of the sensor device data anomaly.

[0029] In S200, historical data of different sensor devices in the building system are acquired based on the time series to obtain a historical sensor data sequence, and the historical sensor data sequence is preprocessed to obtain a preprocessed sensor data sequence, including the following steps: S210: Acquire historical data of sensor devices in the building system based on the time series to obtain a historical sensor data sequence.

[0030] There are many sensor devices in the building system, among which temperature sensor devices, humidity sensor devices, air volume sensor devices and power sensor devices are relatively important sensor devices. Therefore, the data of the sensor devices include at least two of temperature, humidity, air volume and power.

[0031] S220: Preprocess the historical sensor data sequence to obtain a preprocessed sensor data sequence.

[0032] Invalid data of each sensor device data sequence in the historical sensor data sequence is removed and padded to obtain a preprocessed sensor device data sequence; wherein the invalid data at least includes duplicate data and missing data.

[0033] Invalid data types in sensor device data are usually caused by data quality or collection anomalies. Common invalid data types include but are not limited to the following: missing data, sensor device failure, communication interruption, storage error, etc., resulting in data points not being recorded or transmission loss, manifested as empty values, NULL, NaN, etc.; duplicate data, transmission retransmission, storage logic errors, improper sampling frequency settings, etc., resulting in exactly the same or highly similar data points being recorded multiple times; fixed value data, sensor device jamming, software crash, range exceeding, etc., resulting in data remaining unchanged for a long time; invalid format data, data format errors caused by transmission protocol errors, parsing failures, etc.; timestamp errors, clock synchronization failure, storage anomalies, etc., resulting in timestamp missing or disordered, etc.

[0034] Data filling can be done using one or more of the following methods: filling missing data with fixed values ​​(such as mean and median) or sliding window statistics; linear, polynomial, or time-weighted interpolation; leveraging the data's periodicity to fill in missing data with historical data from the same period; predicting missing values ​​using models such as ARIMA, LSTM, or random forest; or filling in data based on physical laws or business rules. The appropriate data filling method can be selected based on the characteristics of the data from different sensor devices.

[0035] If the historical data of a sensor device are all zero within a sufficiently long time window, and the physical quantity monitored by the sensor device is theoretically impossible to be zero for a long time (such as temperature, current, gas concentration, etc.), it can be preliminarily judged that it is in an extremely abnormal state, and it is necessary to check whether there is a power supply failure, sensor failure or communication interruption.

[0036] In S300, an extreme abnormal value is determined based on whether the current sensor data exceeds a reasonable range, as follows: In addition to the invalid data with the aforementioned errors, abnormal sensor data also includes mutation data caused by sampling failures, bit errors, etc., which need to be detected. The reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles. The reasonable range of each sensor device data is expressed as follows:

[0037] in, Indicates the first The first quartile of the sensor device data represents the 25th percentile of the data. Indicates the first The third quartile of the sensor device data represents the 75th percentile of the data. Indicates the first The interquartile range of the data from each sensor device reflects the range of variation of the data. . Respectively represent The weights of the first and third quartiles of the sensor device data, are all positive numbers and can be set according to the characteristics of each sensor device data For example, the value of is 1.5, Also 1.5. When the current sensing data of the target sensing device exceeds the reasonable range of the corresponding sensing device data, the target sensing device has an extreme abnormality, and the extreme abnormality value is assigned to a first preset extreme abnormality value; when the current sensing data of the target sensing device is within the reasonable range of the corresponding sensing device data, the target sensing device does not have an extreme abnormality, and the extreme abnormality value is assigned to a second preset extreme abnormality value; The extreme outliers are represented as follows:

[0038] in, Indicates the Extreme outliers for each sensor device, Indicates the a first preset extreme abnormal value of a sensing device, Indicates the a second preset extreme abnormal value of a sensing device, Indicates the The current sensing data of each sensor device, Indicates the current timestamp. Typically, the value is 1 if an extreme anomaly exists, and 0 if no extreme anomaly exists. Therefore, the first preset extreme anomaly value is set to 1, and the second preset extreme anomaly value is set to 0.

[0039] The IQR value is not affected by the data distribution and is suitable for eliminating extreme values ​​of normal distribution and skewed distribution data. It is easy to detect sampling failures and bit errors in sensor devices.

[0040] In S400 , data in the pre-processed sensor data sequence that exceeds the reasonable range is eliminated to obtain a non-extreme sensor data sequence.

[0041] Mutation data is a strong interference point that affects the model training in the present invention and must be eliminated because: (1) The subsequent trend anomaly detection analysis relies on the mean and standard deviation of the pre-processed sensor data sequence. If there are extreme outliers in the pre-processed sensor data sequence, it will cause the mean to shift and the standard deviation to increase, thereby affecting the accuracy of trend anomaly detection. Therefore, it is necessary to eliminate the extreme abnormal data (data that is not within a reasonable range) in the pre-processed sensor data sequence before calculating the mean and standard deviation, so that the calculation of trend anomalies is based on more stable data. (2) The pre-processed sensor data sequence is also used to train the pre-trained artificial intelligence model in structural anomaly detection, and it is also necessary to eliminate extreme abnormal data to learn the relationship between the normal data of different sensor devices, thereby improving the accuracy of the trained model.

[0042] In S500, the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence is calculated to obtain a trend anomaly value. Sensor device abnormal data also includes drift of data that slowly deviates from the true value over time due to sensor device aging, calibration failure, and environmental cumulative effects. Data drift reflects the trend anomaly of the sensor device data. Further testing of trend anomaly data is required to detect long-term, slow-changing problems that extreme anomalies cannot capture. The following steps are included: S510: Acquire the most recent data of a first preset time period in the non-extreme sensor data sequence to obtain the first time period sensor data sequence, and then obtain the average value and standard deviation of the first time period sensor data sequence. The first preset time period can be set based on the sampling frequency of the sensor device, anomaly detection accuracy, and other factors. Selecting the most recent data as the basis for trend anomaly detection can focus on the latest data, quickly reflect the current status, thereby reducing historical noise interference, more sensitively capturing short-term anomalies, eliminating the need for long-term storage of large amounts of historical data, and reducing processing complexity and storage costs.

[0043] S520: Calculate the skewness of the non-extreme sensor data sequence using a conventional skewness calculation method, which is the first sequence skewness.

[0044] Based on the current sensor data and the average value and standard deviation of the first time-length sensor data sequence, a basic trend deviation value is obtained. The basic trend deviation value is expressed as follows:

[0045] in, Indicates the The basic trend deviation value of the sensor device data indicates the standardized degree to which the current data deviates from the historical average. Indicates the The current sensor data of each sensor device, Indicates the current timestamp, Indicates the first preset duration, Indicates the The average value of the first time-length sensor data sequence of the sensor device data, Indicates the The standard deviation of the first time-length sensor data sequence of the sensor device data.

[0046] The calculation method for the basic trend deviation value is suitable for data that follows a normal distribution. Existing technology generally uses a basic trend deviation value greater than a baseline deviation value (usually 3) as the criterion for identifying anomalies. However, when data exhibit skewness, using a baseline deviation value of 3 as the criterion can easily lead to misjudgments or missed detections. Some sensor data in building systems (such as current, voltage, and cooling load) exhibit positive skewness. This can increase the standard deviation of the sensor data, masking trend anomalies and making it impossible to detect them. It can also shift the mean of the sensor data, further distorting the basic trend deviation value. Furthermore, the probability of a basic trend deviation value of 3 in skewed data is far less than 0.13% (the standard probability of anomalies under a normal distribution), leading to false positives or missed detections.

[0047] Therefore, the present invention introduces skewness check before the existing trend anomaly judgment, and performs basic trend deviation value judgment based on the skewness or performs dynamic threshold setting after data transformation.

[0048] S530: When the first sequence skewness of a sensor device is less than a first preset skewness threshold, it indicates that the data of this sensor device follows a normal distribution. Existing techniques can be used to determine whether there is a trend anomaly. The deviation from the baseline value is set as the first preset deviation threshold. The first preset deviation threshold is preferably 3 and can be adjusted based on actual conditions.

[0049] S540: When the skewness of a first sequence of a certain sensor device is greater than or equal to a first preset skewness threshold (indicating that the data of this sensor device is positively skewed, and determining a trend anomaly based on existing techniques will result in a false positive), the non-extreme sensor data sequence and the current sensor data are transformed to obtain a transformed data sequence and transformed current sensor data, respectively; the skewness of the transformed data sequence is calculated to obtain a second sequence skewness; and a deviation reference value is obtained by comparing the second sequence skewness with the first preset skewness threshold and the second preset skewness threshold. The following steps are included: (1) When all the data in the non-extreme sensor data sequence are positive numbers, the non-extreme sensor data sequence is subjected to maximum likelihood estimation processing to obtain first transformation parameters; based on the first transformation parameters, the non-extreme sensor data sequence and the current sensor data are subjected to first transformation processing respectively to obtain first transformation data sequence and first transformation current sensor data. When the data in the non-extreme sensor data sequence include negative numbers, the non-extreme sensor data sequence is subjected to maximum likelihood estimation processing to obtain second transformation parameters; based on the second transformation parameters, the non-extreme sensor data sequence and the current sensor data are subjected to second transformation processing respectively to obtain second transformation data sequence and second transformation current sensor data. The first transformation processing is expressed as follows:

[0050] The second transformation process is expressed as follows:

[0051] in, Indicates the first transformed data sequence or the second transformed data sequence. The first sensor data sequence data, Indicates the first The first sensor data sequence data, Indicates the a first transformation parameter of the sensor device data, Indicates the The second transformation parameter of the sensor device data. Current sensor data of each sensor device The first transformation processing or the second transformation processing is performed to obtain the first transformed current sensing data or the second transformed current sensing data. The above formulas for the first transformation processing and the second transformation processing can be used and will not be repeated here.

[0052] The first transformation process can perform a power transformation on positive numbers, thereby reducing the skewness of non-extreme sensor data sequences. The second transformation process can transform negative and zero values, thereby reducing the skewness of non-extreme sensor data sequences. Through either the first or second transformation process, the skewness of non-extreme sensor data sequences can be reduced, resulting in a normal distribution and improving the accuracy of trend anomaly determination. Non-extreme sensor data sequences typically do not contain all zeros. Even if this occurs, indicating an extreme anomaly in the sensor device, the extreme anomaly can still be identified, thus not affecting the anomaly attribution results.

[0053] In the maximum likelihood estimation process, both the first and second transformation parameters are obtained by performing maximum likelihood estimation on the non-extreme sensor data sequence. Statistical software, such as Python's scipy.stats.boxcox or R's car::powerTransform(), can be used to automatically calculate the optimal first and second transformation parameters. It should be noted that the first and second transformation parameters are not identical because the underlying non-extreme sensor data sequences used to obtain them are different.

[0054] (2) The skewness of the transformed data sequence is calculated using the conventional skewness calculation method, which is the second sequence skewness.

[0055] (3) According to the comparison between the skewness of the second sequence and the first preset skewness threshold and the second preset skewness threshold, a deviation reference value is obtained, which specifically includes: When the skewness of the second sequence is less than the first preset skewness threshold, it indicates that the transformed data sequence is approximately normally distributed, and the existing technology can be used to determine whether there is a trend anomaly. The deviation reference value is the first preset deviation threshold; When the skewness of the second sequence is greater than or equal to the first preset skewness threshold and less than the second preset skewness threshold, it indicates that the transformed data sequence is moderately skewed, and the deviation from the baseline value needs to be reduced before the existing technology can be used to determine whether there is a trend anomaly, that is, the deviation from the baseline value is the second preset deviation threshold; When the skewness of the second sequence is greater than or equal to the second preset skewness threshold, it indicates that the transformed data sequence is obviously skewed or has a long tail, and the deviation from the baseline value needs to be further reduced before the existing technology can be used to determine whether there is a trend anomaly. The deviation from the baseline value is the third preset deviation threshold; Therefore, the first preset deviation threshold>the second preset deviation threshold>the third preset deviation threshold.

[0056] Wherein, the deviation from the reference value is expressed as follows;

[0057] in, Indicates the The deviation from the reference value of each sensor device, Indicates the a first preset deviation threshold value of a sensing device, Indicates the a second preset deviation threshold value of a sensing device, Indicates the a third predetermined deviation threshold value of each sensor device, and , Indicates the The second sequence skewness of the sensor device data sequence, Indicates the a first preset skewness threshold of a sensing device, Indicates the A second preset skewness threshold of each sensing device.

[0058] S550: Obtain a transformation trend deviation value based on the transformation of the current sensing data and the average value and standard deviation of the first time-length sensing data sequence; the transformation trend deviation value is expressed as follows:

[0059] in, Indicates the Transformation of sensor data Current sensor data, Indicates the The transformation trend deviation value of the data of each sensor device.

[0060] S560: Obtain a trend abnormal value based on the deviation from the baseline value and the transformed trend deviation value; the trend abnormal value is represented as follows:

[0061] in, Indicates the Trend anomalies of each sensor device, Indicates returning the smaller value among multiple values. Indicates the The deviation of the sensor device data from the baseline value.

[0062] Trend outliers are mainly used to capture the statistical deviation behavior of a single variable within a certain time window. They are particularly suitable for discovering slowly rising or falling drift-type anomalies, periodic deviations or control instability, and slight errors caused by sensor aging or position offset.

[0063] Trend anomalies can trigger subsequent verification of physical operating mechanism rules. Trend anomalies can also trigger coordinated verification of system anomalies: This can determine whether a trend anomaly in a particular sensor device causes anomalies in other system sensors (i.e., structural anomalies), thereby distinguishing whether the issue is a drift problem with a single sensor device, or a problem with associated devices or the entire system. This embodiment supports dynamic analysis with a sliding window: the first preset duration can be set to 1 to 3 hours, making it suitable for real-time edge deployment. Furthermore, trend anomalies can form a complementary and collaborative mechanism with other anomalies. When a trend anomaly is confirmed by a structural anomaly or a rule anomaly, the building system anomaly can be upgraded to a medium / high-level warning; otherwise, it can be downgraded to observation status to avoid false alarms.

[0064] The trend anomaly detection method provided by the present invention has the following advantages: (1) It has the ability to detect trend anomalies / offsets with high sensitivity and can effectively identify anomalies that slowly accumulate and gradually offset in sensor device data. It is suitable for early detection of drift-type faults, aging errors, and control system response lag problems, and is an important component of the "early warning" mechanism. (2) It can adapt to the adaptive threshold mechanism of different distributions. By combining skewness calculation with the first transformation / second transformation, it can realize the normalization of skewed data, and then dynamically adjust the deviation baseline value according to the degree of skewness to avoid false alarms caused by non-normal distribution. (3) It has low computational cost and is easy to deploy at the edge. Compared with neural networks or complex graphical models, the trend anomaly detection method provided by the present invention only relies on basic statistics such as mean and standard deviation, and consumes extremely low computing resources. It is suitable for real-time anomaly prediction in low-power devices such as AI Edge Box. (4) Data is standardized and output for easy integration analysis. The output trend anomaly value is a continuous normalized score, which can be seamlessly integrated into the subsequent total anomaly value model (weighted summation with mutation anomalies, structural anomalies, and rule anomalies). It can also be used as a "pre-screening module" for other algorithms. (5) Supports sliding window and trend analysis extensions. The first preset time length (preset value according to needs) can be expanded to support analysis methods such as moving average, volatility judgment, and segmented regression, to achieve further modeling and tracking of complex trends such as periodic offsets and control instability. (6) It can be linked with physical rules to locate the root cause. After the building system triggers an anomaly, the trend anomaly value can drive the directional rule screening mechanism, and only verify the engineering rules involved in the relevant variables, greatly improving the accuracy and interpretability of anomaly judgment.

[0065] The complementarity between mutation anomalies and trend anomalies can be illustrated by the following examples: (1) Continuous data of the air supply temperature of the terminal air conditioners in the building system were collected to form a time series sample of 30 points. The first 20 points were basically stable at around 22°C, and the last 10 points slowly rose to 27°C, simulating the trend of abnormal scenarios where the sensor device drifts or the control accuracy decreases.

[0066] The sequence of preprocessed sensor data is as follows: 22.0, 22.1, 22.2, 22.0, 22.3, 22.2, 22.1, 22.4, 22.2, 22.0, 22.1, 22.0, 22.3, 22.1, 22.0, 22.2, 22.4, 22.3, 22.2, 22.1, 22.6, 23.1, 23.8, 24.2, 24.8, 25.3, 25.9, 26.4, 26.7, 27.0.

[0067] (2) Determination of abnormal mutation: The calculated values ​​were Q1 = 22.1, Q3 = 24.35, and IQR = 2.25. The upper limit of the reasonable range was Q3 + 1.5 × IQR = 27.725. Since the maximum value of 27.0 < 27.725, no abnormal mutation was identified.

[0068] (3) Eliminate the data that exceeds the reasonable range in the preprocessed sensor data sequence to obtain a non-extreme sensor data sequence.

[0069] (4) Skewness analysis: The calculated skewness of the first sequence of the non-extreme sensor data sequence is about 1.01, which is significantly right-leaning, triggering the first transformation process.

[0070] (5) Perform maximum likelihood estimation on the non-extreme sensor data sequence and obtain the first transformation parameter to be approximately 0.2.

[0071] (6) The first transformation parameter performs the first transformation processing on the non-extreme sensor data sequence and the current sensor data, respectively, to obtain the transformed data sequence and the transformed current sensor data. The skewness of the transformed data sequence (the second sequence skewness) is about 0.65, which is moderately skewed / close to normal. The deviation from the baseline value is the second preset deviation threshold (2.5).

[0072] (7) Trend anomaly determination: The (second) trend deviation value calculated based on the transformation of the current sensor data is approximately 2.76, based on which the trend anomaly value is calculated. Moreover, since the trend deviation value is greater than the deviation baseline value, the building system successfully identifies the presence of a trend anomaly in the sensor device and triggers an alarm.

[0073] This example demonstrates how, when a sudden anomaly in a building system fails to detect a trend anomaly, a trend anomaly can be successfully detected through further processing. This demonstrates the complete mechanism chain of "statistics → transformation → standardization → judgment" in the system.

[0074] In S600, the current sensor data is inferred by the pre-trained structural anomaly model to obtain a current data prediction value; the current data prediction value is compared with the current sensor data to obtain a structural anomaly value, including the following steps: The pre-trained structural anomaly model can be an existing trained model, or a model that is self-built and trained based on specific building system characteristics. In this embodiment, a method for self-building and training is provided, which is as follows: (1) Based on the time series, the data of different sensor devices in the building system during normal operation are obtained to obtain the original normal sensor data sequence; based on the unified timestamp, the data sequences of different sensor devices in the original normal sensor data sequence are aligned to obtain the aligned sensor data sequence; the aligned sensor data sequence is normalized to obtain the normalized aligned sensor data sequence; the target sensor device data is set, and combined with the normalized aligned sensor data sequence to obtain the target training sample data set.

[0075] Time alignment can unify the time bases of different sensor devices, address differences in sampling frequency and latency, ensure data synchronization (e.g., matching camera and radar data), avoid timing errors, and meet the AI ​​model's requirement for regular time-series input. Normalization eliminates dimensional differences between sensor devices (e.g., temperature and voltage) and scales all features to a similar range (e.g., [0, 1]), preventing large features from dominating model training while accelerating convergence and improving generalization. The resulting normalized, aligned sensor data sequence, obtained through time alignment and normalization, is high-quality, temporally and spatially consistent. Using this as training data enables pre-trained models to more accurately learn correlated features across sensor devices, improving robustness. Ignoring these steps can significantly degrade model performance.

[0076] In real physical systems, data from multiple sensor devices often exhibit selective correlations, meaning that a structural anomaly in a target sensor device is only correlated with data from a subset of sensors. Traditional methods require manual screening of associated sensors, but the present invention automatically learns these correlations through a pre-trained artificial intelligence model. The model autonomously identifies sensor data associated with the target sensor device and dynamically assigns weights, assigning higher weights to sensor data with high correlations and reducing or zeroing weights to sensor data with low correlations. Simply specifying the target sensor device allows the trained model to automatically weight and process multi-source sensor data based on the learned correlation patterns, eliminating the need for manual intervention in determining correlations.

[0077] (2) Based on the characteristics of specific building systems, a structural anomaly pre-training model is constructed based on the artificial intelligence model.

[0078] In this embodiment, a structural anomaly pre-training model is constructed based on AutoEncoder (a neural network model for unsupervised learning) to illustrate how to construct a structural anomaly pre-training model. However, the present invention is not limited to the structural anomaly pre-training model in this embodiment. In practice, different types of structural anomaly pre-training models can be constructed as needed.

[0079] The structural anomaly pre-training model in this embodiment consists of two parts: an encoder and a decoder. The encoder includes an input layer, several hidden layers, and a bottleneck layer, while the decoder includes a bottleneck layer, several hidden layers, and an output layer. In the encoder, the input layer receives input data and converts it into a format that a neural network can process (such as a vector or tensor). The hidden layers gradually extract high-order features of the input data and reduce the data's dimensionality through nonlinear transformations (such as ReLU and Sigmoid). The number of neurons in each hidden layer typically decreases layer by layer, forming a "compressed" structure. The bottleneck layer forces the network to compress the input data into a low-dimensional latent space with a much lower dimensionality than the input layer, generating a compact representation (i.e., encoding) of the data. The encoder's output also serves as the decoder's input. In the decoder, the bottleneck layer is shared with the encoder and receives the latent code generated by the encoder, serving as the decoder's starting input. The hidden layers gradually reconstruct the low-dimensional latent code into a high-dimensional representation of the original data, restoring data details through operations such as upsampling or deconvolution. The number of neurons in each layer typically increases layer by layer, forming an "expanded" structure. The output layer generates reconstructed data with the same dimensionality as the input layer.

[0080] There is a weight matrix between each layer of the encoder and decoder. The goal of the structural anomaly pre-training model is to minimize the reconstruction error by continuously adjusting each weight matrix, that is, to minimize the error between the output value and the input value of the model. The reconstruction error model is expressed as follows:

[0081] in, Indicates the The reconstruction error of the structural anomaly pre-trained model for each sensor device, Indicates the The first sensor device data, Indicates the The structural anomaly pre-training model of the sensor device is used to The first of the sensor devices The predicted value of data, Indicates the The total amount of data from each sensor device used for model training.

[0082] (3) The structural anomaly pre-training model is trained and tested using the target training sample data set to obtain a structural anomaly model.

[0083] The task of model training is to determine the corresponding weight matrices based on the principle of minimizing reconstruction error. After training, testing and verification, a trained model, namely the structural anomaly model, is obtained. The structural anomaly model can be used to: predict the target sensor device data based on the input data of multiple sensor devices (excluding the data of the target sensor device) to obtain the predicted value of the target sensor device data. This predicted value is used to compare with the actual data of the target sensor device to determine whether there is a structural abnormality in the sensor device.

[0084] S610: Inferring the current sensing data of the target sensing device using a pre-trained structural anomaly model of the target sensing device to obtain a predicted value of the current data of the target sensing device.

[0085] Because the structural anomaly model is trained based on data from properly functioning sensor devices, the predicted values ​​for the target sensor device based on the structural anomaly model are reasonable. Reconstruction errors for properly functioning sensor devices are small, while those for abnormal conditions are larger. If the error between the actual value and the predicted (reasonable) value for the target sensor device exceeds a preset limit, it indicates a structural anomaly in the target sensor device.

[0086] S620: The current sensor data of the target sensing device and the current data prediction value are subjected to difference and square processing to obtain the current reconstruction error of the target sensing device, the current reconstruction error of the target sensing device is expressed as follows;

[0087] in, Indicates the The current reconstruction error of each sensor device, Indicates the The current sensing data of each sensor device, Indicates the The structural anomaly model of the sensor device predicts the The predicted value of the current data of each sensor device.

[0088] S630: Obtaining a structural anomaly value of the target sensor device based on a preset reconstruction error threshold and a current reconstruction error of the target sensor device; the structural anomaly value of the target sensor device is expressed as follows:

[0089] in, Indicates the The structural abnormal value of each sensor device is the structural abnormal value of the target sensor device. Indicates the The current timestamp of each sensor device The reconstruction error is the current reconstruction error of the target sensor device. Indicates the The preset reconstruction error threshold of each sensor device can be set based on the maximum or minimum reconstruction error generated during the training process of the structural abnormality pre-training model multiplied by a preset coefficient. The coefficient corresponding to the maximum reconstruction error should be less than 1, and the coefficient corresponding to the minimum reconstruction error should be greater than 1. Indicates that the smaller value among multiple numeric values ​​is returned.

[0090] S640: The structural anomaly values ​​of other sensor devices are obtained by the same method as above, and it is only necessary to set the other sensor devices in the normalized aligned sensor data sequence as target sensor devices one by one.

[0091] The structural anomaly value provided by the present invention is suitable for identifying implicit anomalies of imbalanced relationships between multiple sensing devices, and plays the role of system-level consistency judgment in the overall architecture. The structural anomaly detection method provided by the present invention does not require manual setting of rules, has strong adaptability, can handle complex collaborative relationship anomalies between sensing devices, can be embedded in the edge computing gateway for operation, and can be jointly inferred with anomalies of other sensing devices to enhance the system anomaly recognition capability. In the subsequent anomaly attribution method, the structural anomaly value can be linked with the trend anomaly value to determine whether it is a systematic imbalance or a local drift. The structural anomaly value can be jointly verified with the rule anomaly value to determine whether the structural anomaly violates the equipment operation logic.

[0092] The linkage between structural anomalies and other anomalies can be illustrated by the following example: During the noon period, the supply air temperature in a certain office area was maintained at 22.0°C, and the return air temperature was 26.0°C, but the space temperature failed to drop. The supply air temperature was judged to be normal based on the trend anomaly value, and the mutation anomaly value and the rule anomaly value did not trigger an anomaly. However, it was detected that the air volume during this period was only 180CMH, far below the 300CMH required for normal cooling. At the same time, the host power was normal and the set temperature did not fluctuate. The system calculated that the current multivariate structural reconstruction error was 1.8, and the preset reconstruction error threshold was 1.2, resulting in a structural anomaly value of 1.0. The building system judged the anomaly as "abnormal cooling efficiency due to insufficient supply air flow" and prompted it as a medium-level structural anomaly.

[0093] In S700, based on the current sensor data and in combination with the running rule model, the total abnormal weight value of the rule that meets the abnormal condition is compared with the total abnormal weight value of the rule that applies the abnormal condition to obtain the rule abnormal value, including the following steps: The operational rule model is constructed based on abnormal conditions of sensor devices and the abnormal weights of rules that meet these abnormal conditions. These abnormal conditions are defined according to the operational rules of the sensor devices in the building system. The abnormal conditions are defined based on the operational rule mechanisms for a single sensor device or across different sensor devices in the building system. The rule weights corresponding to each abnormal condition are determined based on the importance of the abnormal condition within the operational rule mechanism. Table 1 shows an example operational rule model. Those skilled in the art can adjust it as needed, including but not limited to adding or removing abnormal conditions and adjusting the weights of rules corresponding to abnormal conditions.

[0094] Table 1: Example of a running rule model

[0095] Among them, the first preset rule abnormal weight (Corresponding to strong rules) > Second preset rule abnormal weight (Corresponding to the middle rule) > The third preset rule abnormal weight (corresponding to weak rules), among which exceptions 1, 3, and 11 are key exceptions.

[0096] S710: If the current sensor data of the target sensor device meets the abnormal condition, the corresponding rule abnormality weight is added to the target rule abnormality weight, thereby obtaining the total target rule abnormality weight. The abnormal condition is considered met when the sensor data meets both of the following conditions: the device and sensor device match the device and sensor device mentioned in the abnormal condition; and the sensor data value meets the threshold range or specification of the abnormal condition. The abnormal conditions are traversed. If the current sensor data meets the abnormal condition, the rule abnormality weight corresponding to the abnormal condition is added to the target rule abnormality weight (initial value is 0). The resulting target rule abnormality weight is the total target rule abnormality weight.

[0097] S720: If the current sensor data of the target sensor device meets the abnormal condition, the corresponding rule abnormality weight is added to the target applicable rule abnormality weight, thereby obtaining the total target applicable rule abnormality weight. When the sensor data only meets the matching requirements of the device and sensor device, but the value does not meet the abnormal condition, the abnormal condition is applicable (but not satisfied). The abnormal conditions are traversed. If the current sensor data meets (but does not meet) the abnormal condition, the rule abnormality weight corresponding to the abnormal condition is added to the target applicable rule abnormality weight (initial value is 0). The resulting target applicable rule abnormality weight is the total target applicable rule abnormality weight.

[0098] S730: Based on the ratio of the total value of the target satisfied rule anomaly weight to the total value of the target applicable rule anomaly weight, a rule anomaly value of the target sensor device is obtained. The rule anomaly value of the target sensor device is expressed as follows:

[0099] in, Indicates the The regular abnormal value of each sensor device is the regular abnormal value of the target sensor device. Indicates the The total value of the abnormal weight of the target satisfying rule of each sensor device, Indicates the The total value of the target applicable rule anomaly weight of each sensor device.

[0100] In the multi-level fusion sensor data anomaly attribution method proposed in this paper, rule-based outliers are a type of outlier that must be consistently evaluated. They are used to continuously monitor sensor data for violations of engineering logic or physical constraints. During rule-based outlier calculation, the system defaults to performing the calculation at a fixed interval (e.g., every minute), verifying the engineering consistency of all sensor devices point by point.

[0101] However, considering the practical application requirements of large-scale field sensor data or limited edge computing resources, the calculation frequency of rule outliers can be flexibly configured to a lower period (e.g., once every 5 minutes), i.e., a low-frequency rule scanning strategy, to balance resource consumption and system load. In this low-frequency operation mode, the calculation of trend outliers and structural outliers is still performed every minute. However, when the trend outlier value of the target sensor device is within the preset weak trend anomaly range or the structural outlier value of the target sensor device is within the preset weak structural anomaly range, even if it is not currently within the fixed operation cycle of the rule outlier value, the physical rule set corresponding to the sensor device will be triggered to perform a directional verification and calculate the rule outlier value to assist in identifying potential engineering logic errors and avoid missed judgments. The preset weak trend anomaly range is [second trend anomaly threshold, first trend anomaly threshold], and the preset weak structural anomaly range is [second preset structural anomaly threshold, first preset structural anomaly threshold]. For example, the preset weak trend anomaly range is [0.4, 0.7], and the preset weak structural anomaly range is [0.3, 0.8].

[0102] The method of the present invention always retains the periodic rule scanning mechanism, and performs a complete physical rule verification on all sensor devices within the set period to improve the abnormal coverage and the integrity of the logical consistency judgment. If the system operation mode prioritizes economy and energy efficiency, a low-frequency rule scanning strategy can be adopted, and the rule judgment is only locally linked in the event of weak anomalies; if the system attaches importance to the real-time and comprehensiveness of anomaly identification, the execution frequency of the physical rule module can be kept consistent with the trend / structure module, thereby achieving full real-time engineering logic verification. The specific method is not limited in this application, but the low-frequency rule scanning strategy is used for illustration in the embodiment.

[0103] S740: Using the above method, calculate the regular abnormal values ​​of other sensor devices.

[0104] The rule-based exception mechanism embodies the present invention's multi-level, integrated judgment path, from data-driven to physical-logical constraints. It supports flexible configuration of computational intensity based on scenario requirements, ensuring accuracy and interpretability while achieving dynamic balancing and regulation of system resource consumption. The primary function of rule-based exceptions is to coordinate the execution of the engineering rule set corresponding to a sensor device for targeted verification. This provides physical-level constraints to supplement anomaly detection for the entire building system, and is particularly valuable in scenarios where sensor devices fail to detect anomalies but system logic errors exist. Furthermore, a periodic rule-based scanning mechanism can be implemented within the low-frequency rule-based scanning strategy to ensure that even sensor data that is not significantly flagged is fully covered by the physical-logical rules / constraints. The anomaly level and corresponding response behavior can also be determined by comparing the rule-based exception value with a corresponding preset threshold. Within the low-frequency rule-based scanning strategy, when a sensor device's trend anomaly is weak (not obvious), the engineering rules associated with that sensor device are checked for targeted rule screening, avoiding computational redundancy and false positives caused by global judgment.

[0105] The linkage between trend anomalies and rule anomalies can be illustrated by the following example: When the supply air temperature is marked as a weak anomaly, the system will link Rule 1 (cooling supply air temperature is lower than return air temperature), Rule 6 (the temperature difference between set temperature and feedback temperature is lower than the preset temperature difference threshold), Rule 9 (temperature trend is consistent with air volume) and other related rules for verification. Once a rule is found to be triggered, the original weak anomaly can be upgraded to an effective alarm, realizing the closed-loop logic of trend anomaly → engineering rule confirmation → graded output, thereby improving diagnostic efficiency and reducing building system resource consumption.

[0106] When a weak structural anomaly is detected, the building anomaly detection system locates the likely abnormal sensor device based on the sensor device with the largest reconstruction error in the anomaly data and performs a directional logic check on the associated engineering physics rules. The following example illustrates the linkage between structural and rule anomalies: a structural anomaly is detected in a cooling water machine room (reconstruction error of 2.2, with a preset reconstruction error threshold of 1.5), where the flow rate, return water temperature, and main engine power variables have the largest errors. The system triggers Rule 11 (water system energy balance) and Rule 8 (main engine operating power exceeds the minimum operating power). Finding that the power is normal but the return water temperature remains unchanged, the system identifies insufficient water flow as a risk of equipment waterway blockage, escalates the anomaly to a critical level, and triggers an alarm. This mechanism forms a closed-loop linkage chain: structural anomaly → identification of the abnormal primary variable → linkage to physical rules → engineering confirmation → graded output.

[0107] The calculation method for rule anomaly values ​​provided by this invention is derived from the actual physical logic and operating mechanisms of building systems. All judgments can be traced back to clear equipment status and operating rules, making it easier for operations and maintenance personnel to understand, verify, and track them. Different rule bases and weighting systems can be configured according to different building system scenarios, compatible with a variety of air conditioning system structures (such as fan coil units + fresh air systems, water-cooled central air conditioning, VRV systems, etc.). In scenarios where sudden anomalies, trend anomalies, and structural anomalies are difficult to identify or easily misjudged (such as skewed changes and soft logic errors), rule anomalies can provide physical constraints as a last line of defense. Certain strong rules can also be set as independent alarm channels, ensuring immediate response even if the rule anomaly value does not exceed the threshold.

[0108] In S800, the extreme outliers, trend outliers, structure outliers, and rule outliers are compared with the preset outlier thresholds to obtain the cause of the sensor device data anomaly, including the following steps: S810: Perform weighted summation on the extreme outliers, trend outliers, structure outliers, and rule outliers of the target sensor device to obtain a total outlier value of the target sensor device. The total outlier value of the target sensor device is expressed as follows:

[0109] in, Indicates the The total abnormal value of the sensor devices, Indicates the Extreme outliers for each sensor device, Indicates the Trend anomalies of each sensor device, Indicates the The structural abnormal value of each sensor device is the structural abnormal value of the target sensor device. Indicates the Regular outliers for each sensor device, Respectively represent The weights of extreme outliers, trend outliers, structural outliers, and regular outliers for each sensor device, and . Extreme outliers represent extreme value jump anomalies caused by sampling errors and other reasons. Extreme outliers have two fixed values. When there is an extreme anomaly, it is the first preset extreme outlier value (1), and when there is no extreme anomaly, it is the second preset extreme outlier value (0). Trend outliers represent trend-type drift anomalies caused by reasons such as aging of the sensor device. Structural outliers represent anomalies caused by imbalances in the relationship between different sensor devices. Rule outliers represent anomalies that violate engineering physics rules between different sensor devices under weak trend anomalies or weak structural anomalies. Trend outliers, structural outliers, and rule outliers have been normalized during calculation and are continuous values ​​between [0,1] (including 0 and 1). This ensures comparability between different outliers.

[0110] The weights for extreme outliers, trend outliers, structural outliers, and regular outliers in sensor devices can be learned and dynamically adjusted based on labeled data to adapt to the distribution of abnormal patterns across different sensor devices, buildings, seasons, and control strategies. Table 2 provides examples of weight setting strategies. However, the weight settings for different strategy types in Table 2 only illustrate numerical trends; specific values ​​are not specified. Model-driven strategies are suitable for mature systems with sufficient samples and sufficient model training. In the early stages, balanced or empirical strategies are recommended.

[0111] Table 2 Outlier weight settings for different strategy types

[0112] If an outlier is missing, its weight is automatically reset to zero, and the weights of the remaining outliers are normalized.

[0113] S820: When the total abnormal value of the target sensor device is greater than the preset target abnormal total threshold, the target sensor device in the building system is abnormal. The order of determining the cause of the abnormality is as follows: (1) First, determine whether there is a rule anomaly. When the rule anomaly value of the target sensor device is greater than the corresponding first preset rule anomaly threshold (for example, 0.8), that is, when there is a clear rule anomaly, the target sensor device has a physical-logical conflict anomaly. The rule anomaly value is calculated when it is impossible to determine whether there is a trend anomaly or a structural anomaly. Therefore, if there is a rule anomaly, there will not be both a trend anomaly and a structural anomaly.

[0114] (2) Secondly, determine whether there are structural or trend anomalies: When the structural abnormality value of the target sensor device is greater than the corresponding first preset structural abnormality threshold (for example, 0.8) and the rule abnormality value is less than the corresponding second preset rule abnormality threshold (for example, 0.5), that is, there is an obvious structural abnormality but the rule abnormality is not obvious, then the target sensor device has a structural abnormality; When the trend abnormality value of the target sensor device is greater than the corresponding first trend abnormality threshold (for example, 0.7), the extreme abnormality value is the second preset extreme abnormality value (0, no abnormality), the structural abnormality value is less than the corresponding second preset structural abnormality threshold (lower value), and the rule abnormality value is less than the corresponding third preset rule abnormality threshold (lower value), that is, there is an obvious trend abnormality, and there is no extreme abnormality, structural abnormality, or rule abnormality, then the target sensor device has a trend abnormality.

[0115] Since the rule anomaly value is calculated when it is impossible to determine whether there is a trend anomaly or a structural anomaly, when a rule anomaly exists, there may be either a trend anomaly or a structural anomaly, or there may be no trend anomaly or structural anomaly.

[0116] (3) Then, it is determined whether there is an extreme abnormality. When the extreme abnormality value of the target sensor device is a first preset extreme abnormality value (1, abnormality exists), the target sensor device has an extreme abnormality. (4) Finally, when the extreme outliers, trend outliers, structural outliers, and rule outliers of the target sensor device do not meet any of the above conditions, the anomaly of the target sensor device is an unattributed anomaly. For unattributed anomalies, the one or two outliers with the highest values ​​are listed as reference attribution prompts, which can be: "Suspected trend anomaly + structural disturbance" or "Low-level logical conflict". Unattributed anomalies will also be recorded in the background manual review or training sample queue for further optimization of thresholds or judgment rules.

[0117] Among them, the first preset rule abnormality threshold > the second preset rule abnormality threshold > the third preset rule abnormality threshold, the first preset structure abnormality threshold > the second preset structure abnormality threshold, and the values ​​of the second preset structure abnormality threshold and the third preset rule abnormality threshold are both relatively low and can be set to 0.3 or lower.

[0118] The total abnormality value can also be used to set the abnormality level and corresponding response strategy for the building system. For example, if the total abnormality value is 0.00 ≤ < 0.30, it is normal and can be ignored, without any response; if the total abnormality value is 0.30 ≤ < 0.60, it is a mild abnormality, and recording and observation are recommended; if the total abnormality value is 0.60 ≤ < 0.85, it is a moderate abnormality, and a backend warning is recommended; if the total abnormality value is ≥ 0.85, it is a severe abnormality, and an immediate alarm and manual review are recommended. These ranges are set based on the abnormality distribution debugging of the building system and engineering experience, and can be further adaptively optimized based on historical annotation data.

[0119] Furthermore, the response strategy can be further refined based on the cause of sensor device data anomaly, anomaly level, anomaly duration, etc. As shown in Table 3, the correspondence between the cause of sensor device data anomaly, anomaly level, anomaly duration and response strategy is only for example.

[0120] Table 3 Correspondence between sensor device data anomaly causes, anomaly levels, anomaly duration, and response strategies

[0121] Furthermore, the total abnormal values ​​of different sensor devices can be weighted and summed to obtain the overall abnormal situation of the sensor devices in the building system. When the overall abnormal value exceeds the set overall abnormal threshold, it is considered that there is an overall abnormality in the sensor devices in the building system, and the specific sensor device with the abnormality and the cause of the abnormality of the abnormal sensor device can be determined according to the above method.

[0122] Different from the traditional single algorithm judgment, the multi-level fusion method for attributing abnormal data of building system sensor devices proposed in the present invention is a multi-level fusion mechanism that first considers statistics, then structures, and finally physical logic, and has the dynamic collaborative characteristics of layer-by-layer enhancement, cross-verification, and result feedback. On the basis of historical data, the extreme abnormal values, trend abnormal values, and structural abnormal values ​​of the current sensor data are obtained, and the engineering rule set is executed in conjunction to obtain the rule abnormal values ​​for directional verification. After obtaining the abnormal values ​​corresponding to each type of abnormality, these abnormal values ​​are uniformly scored, and the abnormality level, type, and response suggestions are output accordingly. The multi-level fusion method for attributing abnormal data of building system sensor devices proposed in the present invention is fault-tolerant, interpretable, and practical, and is suitable for the abnormal warning needs of complex building environments.

[0123] Example 2: A multi-level fusion building system sensor device data anomaly attribution system, such as Figure 2 As shown, it includes a current data module 100, a historical data module 200, an extreme anomaly module 300, a data elimination module 400, a trend anomaly module 500, a structure anomaly module 600, a rule anomaly module 700 and an anomaly attribution module 800; The current data module 100 obtains current sensor data of different sensor devices in the building system; The historical data module 200 acquires historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocesses the historical sensor data sequence to obtain a preprocessed sensor data sequence; The extreme anomaly module 300 determines an extreme anomaly value based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; The data elimination module 400 eliminates data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; The trend anomaly module 500 calculates the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence to obtain a trend anomaly value; The structural anomaly module 600 uses a pre-trained structural anomaly model to infer the current sensor data to obtain a current data prediction value; compares the current data prediction value with the current sensor data to obtain a structural anomaly value; The rule exception module 700 compares the total value of the rule exception weights that meet the exception condition with the total value of the rule exception weights that apply the exception condition based on the current sensor data and in combination with the operating rule model to obtain a rule exception value. The operating rule model is constructed based on the abnormal condition of the sensor device and the rule exception weights that meet the abnormal condition, and the abnormal condition is set according to the operating rules of the sensor device in the building system. The anomaly attribution module 800 compares extreme anomalies, trend anomalies, structural anomalies, and rule anomalies with preset anomaly thresholds to obtain the cause of the sensor device data anomaly.

[0124] Various changes and modifications can be made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also fall within the scope of the present invention.

[0125] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referenced to each other.

[0126] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0127] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0130] It should be noted that: References in this specification to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention. Therefore, appearances of the phrases "one embodiment" or "an embodiment" in various places throughout this specification do not necessarily refer to the same embodiment.

[0131] Furthermore, it should be noted that the specific embodiments described in this specification may vary in the shapes and names of their components. Any equivalent or simple variations based on the structure, features, and principles described in the patented concept of this invention are included within the scope of protection of this patent. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments, and these modifications, as long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, shall fall within the scope of protection of this invention.

Claims

1. A multi-level fusion building system sensor device data anomaly attribution method, characterized by: The following steps are involved: Obtain current sensor data from different sensor devices in the building system; Acquire historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocess the historical sensor data sequence to obtain a preprocessed sensor data sequence; Determining an extreme outlier based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; Eliminate data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; Calculate the deviation of the current sensor data from the data in the non-extreme sensor data sequence to obtain the trend outlier; The current sensor data is inferred through the pre-trained structural anomaly model to obtain the current data prediction value; the current data prediction value is compared with the current sensor data to obtain the structural anomaly value; Based on the current sensor data and in combination with the operating rule model, the total value of the rule anomaly weights that meet the abnormal condition is compared with the total value of the rule anomaly weights that apply the abnormal condition to obtain the rule anomaly value; wherein the operating rule model is constructed based on the abnormal condition of the sensor device and the rule anomaly weights that meet the abnormal condition, and the abnormal condition is set according to the operating rules of the sensor device in the building system; The extreme outliers, trend outliers, structural outliers and regular outliers are compared with the preset anomaly thresholds to obtain the cause of the sensor device data anomaly.

2. The multi-level fusion building system sensor device data anomaly attribution method according to claim 1 is characterized in that: The extreme outlier value is determined based on whether the current sensor data exceeds the reasonable range, as follows: When the current sensing data of the target sensor device exceeds the reasonable range of the corresponding sensor device data, the target sensor device has an extreme abnormality, and the extreme abnormality value is a first preset extreme abnormality value; when the current sensing data of the target sensor device is within the reasonable range of the corresponding sensor device data, the target sensor device does not have an extreme abnormality, and the extreme abnormality value is a second preset extreme abnormality value; The reasonable range of the sensor device data is expressed as follows: in, Indicates the first The first quartile of the sensor device data, Indicates the first The third quartile of the sensor device data, Indicates the first The interquartile range of the sensor data, , Respectively represent The weights of the first and third quartiles of the sensor device data.

3. The multi-level fusion building system sensor device data anomaly attribution method according to claim 1 is characterized in that: The step of calculating the deviation of the current sensor data from the data in the non-extreme sensor data sequence to obtain the trend outlier includes the following steps: Acquire the latest data of a first preset time length in the non-extreme sensor data sequence to obtain a first time length sensor data sequence, and then obtain the average value and standard deviation of the first time length sensor data sequence; Calculate the skewness of the non-extreme sensor data sequence to obtain the first sequence skewness; When the skewness of the first sequence is less than the first preset skewness threshold, the deviation reference value is the first preset deviation threshold; When the skewness of the first sequence is greater than or equal to a first preset skewness threshold, the non-extreme sensor data sequence and the current sensor data are transformed to obtain a transformed data sequence and a transformed current sensor data; the skewness of the transformed data sequence is calculated to obtain a second sequence skewness; and a deviation reference value is obtained by comparing the second sequence skewness with the first preset skewness threshold and the second preset skewness threshold; Obtaining a transformation trend deviation value based on transforming the current sensor data and the average value and standard deviation of the first time-length sensor data sequence; Based on the deviation from the baseline value and the transformed trend deviation value, the trend anomaly value is obtained; The transformation trend deviation value is expressed as follows: The trend outliers are expressed as follows: in, Indicates the The deviation value of the transformation trend of the sensor device data, Indicates the Transformation of sensor data Current sensor data, Indicates the current timestamp, Indicates the first preset duration, Indicates the The average value of the first time-length sensor data sequence of the sensor device data, Indicates the The standard deviation of the first time-length sensor data sequence of the sensor device data, Indicates the Trend anomalies of each sensor device, Indicates returning the smaller value among multiple values. Indicates the The deviation of the sensor device data from the baseline value.

4. The multi-level fusion building system sensor device data anomaly attribution method according to claim 3 is characterized in that: The step of transforming the non-extreme sensor data sequence and the current sensor data to obtain the transformed data sequence and the transformed current sensor data comprises the following steps: When all data in the non-extreme sensor data sequence are positive numbers, performing maximum likelihood estimation processing on the non-extreme sensor data sequence to obtain first transformation parameters; performing first transformation processing on the non-extreme sensor data sequence and the current sensor data based on the first transformation parameters to obtain a first transformed data sequence and first transformed current sensor data; When there are negative numbers in the non-extreme sensor data sequence, performing maximum likelihood estimation processing on the non-extreme sensor data sequence to obtain second transformation parameters; performing second transformation processing on the non-extreme sensor data sequence and the current sensor data based on the second transformation parameters to obtain a second transformed data sequence and second transformed current sensor data; The first transformation process is represented as follows: The second transformation process is expressed as follows: in, Indicates the first transformed data sequence or the second transformed data sequence. The first sensor data sequence data, Indicates the first The first sensor data sequence data, Indicates the a first transformation parameter of the sensor device data, Indicates the A second transformation parameter of the sensor device data.

5. The multi-level fusion building system sensor device data anomaly attribution method according to claim 3 is characterized in that: The step of obtaining the deviation reference value based on the comparison of the second sequence skewness with the first preset skewness threshold and the second preset skewness threshold comprises the following steps: When the skewness of the second sequence is less than the first preset skewness threshold, the deviation reference value is the first preset deviation threshold; When the skewness of the second sequence is greater than or equal to the first preset skewness threshold and less than the second preset skewness threshold, the deviation reference value is the second preset deviation threshold; When the skewness of the second sequence is greater than or equal to the second preset skewness threshold, the deviation reference value is the third preset deviation threshold; wherein the first preset deviation threshold>the second preset deviation threshold>the third preset deviation threshold.

6. The multi-level fusion building system sensor device data anomaly attribution method according to claim 1 is characterized in that: The method of inferring the current sensor data using the pre-trained structural anomaly model to obtain a current data prediction value and comparing the current data prediction value with the current sensor data to obtain a structural anomaly value includes the following steps: Inferring the current sensor data of the target sensor device through the pre-trained target sensor device structural anomaly model to obtain the predicted value of the current data of the target sensor device; Performing square processing on the difference between the current sensing data of the target sensor device and the current data prediction value to obtain the current reconstruction error of the target sensor device; Obtaining a structural anomaly value of the target sensing device based on a preset reconstruction error threshold and a current reconstruction error of the target sensing device; Calculate the structural anomaly values ​​of other sensing devices; The structural abnormality value of the target sensor device is expressed as follows: in, Indicates the The structural abnormal value of each sensor device is the structural abnormal value of the target sensor device. Indicates the The current timestamp of each sensor device The reconstruction error is the current reconstruction error of the target sensor device. , Indicates the The current sensing data of a sensor device, that is, the target sensor device, Indicates the The predicted value of the current data of each sensor device, Indicates the The preset reconstruction error threshold of each sensor device, Indicates that the smaller value among multiple numeric values ​​is returned.

7. The multi-level fusion building system sensor device data anomaly attribution method according to claim 1 is characterized in that: The method of comparing the total abnormal weight of the rules that meet the abnormal conditions with the total abnormal weight of the rules that apply the abnormal conditions based on the current sensor data in combination with the running rule model to obtain the rule abnormal value includes the following steps: If the current sensor data of the target sensor device meets the abnormal condition, the corresponding rule abnormal weight is added to the target satisfaction rule abnormal weight to obtain the total value of the target satisfaction rule abnormal weight; If the current sensing data of the target sensing device is applicable to the abnormal condition, the corresponding rule abnormal weight is added to the target applicable rule abnormal weight to obtain the total target applicable rule abnormal weight; Obtaining a rule anomaly value of the target sensor device based on a ratio of a total value of the target satisfied rule anomaly weights to a total value of the target applicable rule anomaly weights; Calculate regular anomalies of other sensor devices; The regular abnormal value of the target sensor device is expressed as follows: in, Indicates the Regular outliers for each sensor device, Indicates the The total value of the abnormal weight of the target satisfying rule of each sensor device, Indicates the The total value of the target applicable rule anomaly weight of each sensor device.

8. The multi-level fusion building system sensor device data anomaly attribution method according to claim 1 is characterized in that: The process of comparing the extreme outliers, trend outliers, structural outliers, and rule outliers with the preset outlier thresholds to obtain the cause of the sensor device data anomaly includes the following steps: Performing weighted summation on the extreme outliers, trend outliers, structural outliers, and regular outliers of the target sensor device to obtain the total outlier value of the target sensor device; When the total abnormal value of the target sensor device is greater than the preset target abnormal total threshold, the target sensor device in the building system is abnormal, and the abnormal reason is: When the rule abnormality value of the target sensor device is greater than the corresponding first preset rule abnormality threshold, the target sensor device has a physical-logical conflict abnormality; When the structural abnormality value of the target sensing device is greater than the corresponding first preset structural abnormality threshold and the rule abnormality value is less than the corresponding second preset rule abnormality threshold, the target sensing device has a structural abnormality; When the trend abnormality value of the target sensor device is greater than the corresponding first trend abnormality threshold, there is no extreme abnormality, the structural abnormality value is less than the corresponding second preset structural abnormality threshold, and the rule abnormality value is less than the corresponding third preset rule abnormality threshold, then the target sensor device has a trend abnormality; When the target sensor device has an extreme abnormality, the abnormality cause of the target sensor device is an extreme abnormality; Among them, the first preset rule abnormality threshold>the second preset rule abnormality threshold>the third preset rule abnormality threshold, and the first preset structure abnormality threshold>the second preset structure abnormality threshold.

9. A multi-level integrated building system sensor device data anomaly attribution system, characterized by: A method capable of implementing any one of claims 1 to 8, comprising a current data module, a historical data module, an extreme anomaly module, a data elimination module, a trend anomaly module, a structural anomaly module, a rule anomaly module, and an anomaly attribution module; The current data module obtains current sensor data of different sensor devices in the building system; The historical data module acquires historical data of different sensor devices in the building system based on the time series to obtain a historical sensor data sequence, and preprocesses the historical sensor data sequence to obtain a preprocessed sensor data sequence; The extreme anomaly module determines an extreme anomaly value based on whether the current sensor data exceeds a reasonable range; wherein the reasonable range is obtained by dividing the pre-processed sensor data sequence based on quartiles; The data elimination module eliminates data that exceeds the reasonable range in the pre-processed sensor data sequence to obtain a non-extreme sensor data sequence; The trend anomaly module calculates the deviation of the current sensor data relative to the data in the non-extreme sensor data sequence to obtain a trend anomaly value; The structural anomaly module uses a pre-trained structural anomaly model to infer the current sensor data to obtain a current data prediction value; compares the current data prediction value with the current sensor data to obtain a structural anomaly value; The rule exception module compares the total value of the rule exception weights that meet the exception condition with the total value of the rule exception weights that apply the exception condition based on the current sensor data in combination with the operation rule model to obtain a rule exception value; wherein the operation rule model is constructed based on the abnormal condition of the sensor device and the rule exception weights that meet the abnormal condition, and the abnormal condition is set according to the operation rule of the sensor device in the building system; The anomaly attribution module compares extreme anomalies, trend anomalies, structural anomalies, and rule anomalies with preset anomaly thresholds to obtain the cause of the sensor device data anomaly.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

11. A multi-level fusion building system sensor device data anomaly attribution device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Images