An intelligent control method for building energy-saving water supply and drainage system

Abstract

This application discloses an intelligent control method for a building energy-saving water supply and drainage system. The method includes: constructing a multivariate state feature vector for each drainage pump and outputting several drainage pump groups; based on each drainage pump group, calculating the state extreme value of each drainage pump within each historical time window to form a state fluctuation sequence, and comparing it with the state deviation sequence of each other drainage pump in the same drainage pump group; if the number of historical time windows reaching a deviation threshold in a certain state deviation sequence is greater than a preset number, then it is identified as a suspected abnormal drainage pump; based on each suspected abnormal drainage pump, obtaining its multivariate deviation set and its corresponding abnormal probability value at regular intervals to generate an abnormal probability value sequence for each suspected abnormal drainage pump; generating the hit rate of the abnormal probability value sequence, and if the hit rate is greater than a hit threshold, then the suspected abnormal drainage pump is confirmed to be abnormal. This improves the reliability and accuracy of drainage pump anomaly detection while enhancing detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a control method for a building energy-saving water supply and drainage system. Background Technology

[0002] Building water supply and drainage systems are used to remove sewage from residential, public, and industrial buildings. A building's internal drainage system generally consists of water receivers for sanitary fixtures or production equipment, drainage pipes, cleaning facilities, ventilation pipes, wastewater lifting equipment, and local treatment structures. In building water supply and drainage systems, drainage pumps play an irreplaceable role. Proper control of drainage pumps in building water supply and drainage systems ensures stable operation. When a drainage pump malfunctions, it manifests as abnormal vibration.

[0003] Currently, the common method for detecting and controlling abnormal conditions in drainage pumps is to determine whether a pump is malfunctioning by monitoring the vibration data of each pump and comparing it to that of a normal drainage pump. Since drainage pumps naturally vibrate during operation, abnormal vibrations are more noticeable; in cases of severe abnormalities, simply monitoring the vibration data is sufficient to diagnose the problem. However, when a drainage pump exhibits only minor abnormalities, the abnormal vibrations are less pronounced, making it difficult to identify the malfunction solely through comparison with the vibration data of a normal drainage pump.

[0004] Chinese invention patent application number 202210818029.5 discloses an intelligent control method for building energy-saving water supply and drainage systems. Based on the multi-segment vibration sequence of each drainage pump, the common factor matrix and independent factor vector of each drainage pump are obtained, and a common scatter plot and independent scatter plot are constructed to obtain the degree of abnormality of the drainage pump. By performing an overall analysis of the working status and independent parameters of the drainage pumps working simultaneously, and combining the independent status of the drainage pumps for separate analysis, it is possible to detect minor abnormalities in the drainage pumps in a timely manner.

[0005] However, relying solely on the vibration sequence of the drainage pump for anomaly diagnosis may have certain limitations, because the vibration sequence may be affected by a variety of factors, including external factors at the drainage end and internal factors of the drainage pump itself that may cause abnormal water flow. These factors may lead to occasional vibration anomalies, but the drainage pump itself is not abnormal, thus making the diagnosis of drainage pump anomalies impossible. Summary of the Invention

[0006] This application provides an intelligent control method for building energy-saving water supply and drainage systems, which improves the reliability and accuracy of drainage pump anomaly detection while enhancing detection efficiency.

[0007] This application provides an intelligent control method for a building energy-saving water supply and drainage system, including:

[0008] S101, collect vibration characteristic data and working characteristic parameters of all drainage pumps within a preset historical time period in each historical time window, and construct a multivariate state feature vector for each drainage pump.

[0009] S102, input the multivariate state feature vectors of all drainage pumps into the pre-set clustering algorithm, and output K clusters, each cluster corresponding to a drainage pump group, and each drainage pump group including at least one drainage pump.

[0010] S103, Based on each drainage pump group, calculate the state extreme value of each drainage pump in each historical time window, form the state fluctuation sequence of the drainage pump, and generate the state deviation sequence of the drainage pump from each other drainage pump in the drainage pump group respectively.

[0011] S104, In all the state deviation sequences of each drainage pump in the drainage pump group, if the number of historical time windows that reach the deviation threshold in a certain state deviation sequence is greater than the preset number, then the corresponding drainage pump is identified as a suspected abnormal drainage pump.

[0012] S105, Based on each suspected abnormal drainage pump, obtain its multivariate deviation set and its corresponding abnormal probability value every preset period, and generate an abnormal probability value sequence for each suspected abnormal drainage pump.

[0013] S106. Based on each suspected abnormal drainage pump, generate the hit rate of the abnormal probability value sequence according to the preset abnormal probability threshold and abnormal probability value sequence. If the hit rate is greater than the hit threshold, the suspected abnormal drainage pump is confirmed to be abnormal.

[0014] Preferably, the vibration characteristic data includes vibration frequency, amplitude, and vibration duration; the operating characteristic parameters include flow rate, pressure, rotational speed, and temperature; and the construction of the multivariate state feature vector for each drainage pump includes:

[0015] A1. Calculate the statistical properties of the vibration characteristic data, including the mean, variance, and kurtosis values ​​corresponding to vibration frequency, amplitude, and vibration duration, respectively.

[0016] A2. Calculate the statistical quantities of the working characteristic parameters, including the mean, variance, and kurtosis values ​​of flow rate, pressure, speed, and temperature, respectively.

[0017] A3. Combine the statistics of vibration characteristic data and the statistics of working characteristic parameters to generate a multivariate state feature vector for each drainage pump.

[0018] Preferably, the pre-set clustering algorithm is set to the K-Means clustering algorithm, and the center point of each cluster is determined as the label feature vector of the corresponding drainage pump group.

[0019] Preferably, the state fluctuation sequence of the drainage pump is represented as [ ,... The state deviation sequence is represented as [ ,... ];

[0020] in, This refers to the extreme state value within the k-th historical time window of a preset historical period. This represents the deviation of the state of this drainage pump from that of other drainage pumps in the same drainage pump group.

[0021] Preferably, the method for calculating the extreme value of the state is as follows:

[0022] Obtain the vibration characteristic data and operating characteristic parameters of each drainage pump within each historical time window, and obtain the vibration extreme value and operating extreme value corresponding to them. Then, perform a weighted summation of the vibration extreme value and the operating extreme value to obtain the state extreme value.

[0023] Preferably, obtaining its multivariate state deviation set and its corresponding anomaly probability value every preset period includes:

[0024] B1. Based on each suspected abnormal drainage pump, obtain the multivariate state feature vector of the previous cycle and the multivariate state feature vector set of all other drainage pumps in the drainage pump group at preset intervals.

[0025] B2. Calculate the Euclidean distance between the drainage pump and each multivariate state feature vector in the multivariate state feature vector set to form the multivariate state deviation set of the drainage pump. Each element in the set corresponds to an Euclidean distance value. The ratio of the number of elements in the multivariate state deviation set that are greater than a preset distance threshold to the total number of elements is determined as the anomaly probability value.

[0026] Preferably, the hit rate of the abnormal probability value sequence is set as: the number of abnormal probability values ​​greater than the abnormal probability threshold / the total number of abnormal probability values ​​in the abnormal probability value sequence.

[0027] Preferably, S106 includes:

[0028] When the hit rate of the abnormal probability value sequence is within the preset first interval, the suspected abnormal drainage pump is identified as the target drainage pump. Using the pre-set abnormal retest mechanism, the suspected abnormal drainage pump at the corresponding point is retested, and its final diagnosis result is output. The first interval is set to [30%, 90%].

[0029] Preferably, the pre-set anomaly retesting mechanism specifically includes:

[0030] S201, obtain the Euclidean distance value of the multivariate state feature vectors of the target drainage pump and other drainage pumps in the drainage pump group that are diagnosed as normal in the most recent preset period.

[0031] S202, input the Euclidean distance value into the pre-trained state anomaly prediction model, and output the diagnostic result of the target drainage pump, which includes abnormal and normal results;

[0032] Each drainage pump group corresponds to a state anomaly prediction model. The state anomaly prediction model is determined based on the drainage pump group to which the target drainage pump belongs. The specific process of obtaining the pre-trained state anomaly prediction model is as follows:

[0033] In a certain drainage pump group, a large number of multivariate state feature vectors of all drainage pumps over a long period of time are obtained. The drainage pumps that have been operating stably and normally for a long time in the drainage pump group are selected as reference objects. The Euclidean distance values ​​of the multivariate state feature vectors of each drainage pump and the corresponding reference object over all historical periods are obtained as training datasets. Each training data is labeled with the content of drainage pump state as abnormal or normal. The pre-set neural network structure is trained using the training dataset, and the model parameters are continuously optimized to obtain the final state abnormality prediction model.

[0034] Preferably, each of the drainage pump sets corresponds to a state anomaly prediction model, and each state anomaly prediction model includes several differentiated prediction models. Step S202 further includes:

[0035] S301, based on the preset first interval, is divided into three sub-intervals [30%, 50%), [50%, 70%), and [70%, 90%], with each sub-interval corresponding to a differential prediction model;

[0036] S302, obtain the hit rate corresponding to the abnormal probability value sequence of the target drainage pump, and locate the corresponding differential prediction model according to the sub-interval in which the hit rate is located.

[0037] S303, input the Euclidean distance value corresponding to the target drainage pump into the corresponding differential prediction model, and output the diagnostic result of the target drainage pump.

[0038] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0039] By introducing multivariate data related to the operating status of drainage pumps (such as operating characteristic parameters) and refined analysis methods (such as cluster analysis, state extreme value and deviation sequence calculation), abnormal drainage pumps can be identified more accurately. By calculating the abnormal probability value sequence and hit rate, more reliable basis for anomaly judgment is provided, reducing the possibility of false positives and false negatives. Cluster analysis based on multivariate state feature vectors divides drainage pumps into similar groups, reducing the workload of comprehensive self-inspection. Comparison and verification only need to be performed within the divided drainage pump groups, improving the efficiency and reliability of anomaly detection. The calculation of multivariate state feature vectors and state extreme values ​​provides comprehensive information on the changes in the operating status of drainage pumps, which helps to better understand their operating status and the causes of anomalies. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the intelligent control method for building energy-saving water supply and drainage systems according to an embodiment of the present invention. Detailed Implementation

[0041] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0042] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0044] Example 1: Figure 1 This is a flowchart illustrating the intelligent control method for building energy-saving water supply and drainage systems according to an embodiment of the present invention.

[0045] like Figure 1 As shown, an intelligent control method for a building energy-saving water supply and drainage system includes the following steps:

[0046] S101: Collect vibration characteristic data and operating characteristic parameters of all drainage pumps within a preset historical time period for each historical time window. The vibration characteristic data includes vibration frequency, amplitude, and vibration duration, while the operating characteristic parameters include flow rate, pressure, speed, and temperature. Preprocess the collected vibration characteristic data and operating characteristic parameters to construct a multivariate state feature vector for each drainage pump.

[0047] Specifically, the preset historical time period can be set to the past week, and the time window can be set to daily or 6 hours, which can be adjusted according to the actual situation.

[0048] The collected vibration characteristic data and operating characteristic parameters are preprocessed to construct a multivariate state feature vector for each drainage pump, specifically including:

[0049] A1. Calculate the statistical measures of vibration characteristic data, including the mean, variance, and kurtosis (the difference between low and high peak values) of vibration frequency, amplitude, and duration.

[0050] A2. Calculate the statistical quantities of the working characteristic parameters, including the mean, variance, and kurtosis values ​​of flow rate, pressure, speed, and temperature, respectively.

[0051] A3. Combine the statistics of vibration characteristic data and the statistics of working characteristic parameters to generate a multivariate state feature vector for each drainage pump.

[0052] S102, input the multivariate state feature vectors of all drainage pumps into a pre-set clustering algorithm, and output K clusters. Each cluster corresponds to a drainage pump group, and each drainage pump group includes at least one drainage pump.

[0053] Specifically, the pre-set clustering algorithm is the K-Means clustering algorithm, which determines the center point (center multivariate state feature vector) of each cluster as the label feature vector of the corresponding drainage pump group.

[0054] Therefore, based on the multivariate state feature vectors of all drainage pumps, drainage pump groups are divided so that the drainage pumps in each group have similar comprehensive operating states. This provides a reliable classification basis for subsequent self-inspection based on data between drainage pumps within the group, reducing the workload of comprehensive self-inspection. Only comparisons between drainage pumps within each group are needed.

[0055] S103, Based on each drainage pump group, calculate the extreme state value of each drainage pump within each historical time window, forming the state fluctuation sequence of that drainage pump. ,... ], and generate the state deviation sequence of the drainage pump from each of the other drainage pumps in the same drainage pump group. ,... ].

[0056] in, This refers to the extreme state value within the k-th historical time window of a preset historical period. This represents the deviation of the state of this drainage pump from that of other drainage pumps in the same drainage pump group.

[0057] Specifically, the method for calculating the state extreme value is as follows: obtain the vibration characteristic data and operating characteristic parameters of each drainage pump in each historical time window, respectively, and obtain the vibration extreme value (set as: the average or weighted sum of the differences between the maximum and minimum values ​​of all vibration characteristic data, used to reflect the degree of influence of different vibration characteristic data) and the operating extreme value (set as: the average or weighted sum of the differences between the maximum and minimum values ​​of all operating characteristic parameters, used to reflect the degree of influence of different operating characteristic parameters). The state extreme value is obtained by weighted summing of the vibration extreme value and the operating extreme value. The weight values ​​of the vibration extreme value and the operating extreme value are set according to the actual situation to represent the importance of vibration data and operating parameters to the state assessment of drainage pumps, and the sum of the weight values ​​of the two is 1.

[0058] S104, in all state deviation sequences of each drainage pump in the drainage pump group, if the number of historical time windows that reach the deviation threshold in a certain state deviation sequence is greater than the preset number, then the corresponding drainage pump is identified as a suspected abnormal drainage pump.

[0059] The deviation threshold is preset according to the actual situation and is used to measure the degree of deviation between the states of different drainage pumps in the same drainage pump group. The preset number is set according to the k value, which can be set to half of the k value. This indicates that when the number is greater than half, the drainage pump may be abnormal. This step performs a large-scale preliminary screening to screen out drainage pumps that may be abnormal and identify them as suspected abnormal drainage pumps.

[0060] S105: Based on each suspected abnormal drainage pump, obtain its multivariate deviation set and its corresponding abnormal probability value every preset period until a preset monitoring duration is reached, generating an abnormal probability value sequence for each suspected abnormal drainage pump. The monitoring duration is longer than the preset period and is an integer multiple of the preset period; for example, the integer multiple can be set to 2.

[0061] Specifically, at preset intervals, its multivariate state deviation set and its corresponding anomaly probability value are obtained, including:

[0062] B1. Based on each suspected abnormal drainage pump, obtain the multivariate state feature vector of the previous cycle and the multivariate state feature vector set of all other drainage pumps in the same drainage pump group every preset period.

[0063] B2. Calculate the Euclidean distance between the drainage pump and each multivariate state feature vector in the multivariate state feature vector set, forming a multivariate state deviation set for the drainage pump. Each element in the set corresponds to an Euclidean distance value. The ratio of the number of elements in the multivariate state deviation set that are greater than a preset distance threshold to the total number of elements is determined as the anomaly probability value. The preset distance threshold is set according to actual conditions and is used to measure the degree of deviation between the multivariate state feature vectors of the two drainage pumps.

[0064] S106. Based on each suspected abnormal drainage pump, generate the hit rate of the abnormal probability value sequence according to the preset abnormal probability threshold and abnormal probability value sequence. If the hit rate is greater than the hit threshold, the suspected abnormal drainage pump is confirmed to be abnormal, and the drainage control terminal is notified to carry out maintenance.

[0065] The hit rate of the anomaly probability value sequence is set as: the number of anomaly probability values ​​greater than the anomaly probability threshold / the total number of anomaly probability values ​​in the anomaly probability value sequence. The hit threshold is set according to the actual situation, for example, 90%.

[0066] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0067] By introducing multivariate data related to the working status of drainage pumps (such as working characteristic parameters) and refined analysis methods (such as cluster analysis, state extreme value and deviation sequence calculation), abnormal drainage pumps can be identified more accurately.

[0068] By calculating the anomaly probability value sequence and hit rate, a more reliable basis for anomaly judgment is provided, reducing the possibility of false positives and false negatives.

[0069] Clustering analysis based on multivariate state feature vectors divides drainage pumps into similar groups, reducing the workload of comprehensive self-inspection. It only requires comparison and verification within the divided drainage pump groups, improving the efficiency and reliability of anomaly detection.

[0070] The calculation of multivariate state feature vectors and state extrema provides comprehensive information on the changes in the operating state of the drainage pump, which helps to better understand its working state and the causes of abnormalities.

[0071] Example 2: In Example 1, although the hit rate of the anomaly probability value sequence is used to determine whether the drainage pump is abnormal, the complex and variable working environment of the drainage pump may lead to false positives or false negatives due to the hit rate being close to the threshold. In particular, under certain special circumstances, such as when the drainage pump is in a transitional state or is subject to brief interference, its hit rate may fall near the judgment threshold, making it difficult to directly determine whether it is abnormal.

[0072] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0073] In some embodiments, step S106 includes:

[0074] When the hit rate of the abnormal probability value sequence is within the preset first interval, the suspected abnormal drainage pump is identified as the target drainage pump. Using a pre-set abnormal retest mechanism, the suspected abnormal drainage pump at the corresponding point is retested, and its final diagnostic result is output. The first interval is set according to the actual situation, for example, it is set to: [30%, 90%].

[0075] In some embodiments, the pre-configured anomaly retesting mechanism specifically includes:

[0076] S201, obtain the Euclidean distance value of the multivariate state feature vectors of the target drainage pump and other drainage pumps in the drainage pump group that are diagnosed as normal in the most recent preset period.

[0077] S202, input the Euclidean distance value into the pre-trained state anomaly prediction model, and output the diagnostic result of the target drainage pump, which includes abnormal and normal results.

[0078] Each drainage pump group corresponds to a state anomaly prediction model. The state anomaly prediction model is determined based on the drainage pump group to which the target drainage pump belongs. The specific process of obtaining the pre-trained state anomaly prediction model is as follows:

[0079] In a certain drainage pump group, a large number of multivariate state feature vectors of all drainage pumps over a long period of time are obtained. The drainage pumps that have been operating stably and normally for a long time (as determined by the management personnel) are selected as reference objects. The Euclidean distance values ​​of the multivariate state feature vectors of each drainage pump and the corresponding reference object (the model reference object within the same drainage pump group) over all historical periods are obtained as training datasets. Each training data is labeled (whether the drainage pump is actually normal or abnormal). The pre-set neural network structure is trained using the training dataset, and the model parameters are continuously optimized to obtain the final state abnormality prediction model.

[0080] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0081] By setting a first interval to make a more detailed judgment on the hit rate of the abnormal probability value sequence, when the hit rate falls within this interval, it is not directly determined whether the drainage pump is abnormal, but further abnormal re-inspection is carried out, which avoids misjudgment or missed judgment due to the hit rate being close to the threshold, and improves the accuracy of abnormal judgment.

[0082] An anomaly retesting mechanism is proposed. When the hit rate of the target drainage pump is within the first interval, the mechanism is used for retesting. The retesting process compares the Euclidean distance values ​​of the multivariate state feature vectors of the target drainage pump and other normal drainage pumps in the group, and inputs them into the state anomaly prediction model for judgment, outputting more reliable diagnostic results.

[0083] An anomaly prediction model was trained for each drainage pump group. The model was trained based on historical data of the drainage pumps in the group. The model can learn the normal operating status of the drainage pumps in the group and judge whether the target drainage pump is abnormal, thus improving the pertinence of anomaly identification.

[0084] Because the working environment of drainage pumps is complex and variable, training a state anomaly prediction model specific to drainage pump sets enhances the method's adaptability to different working environments. This approach enables the method to better adapt to the operating characteristics of different drainage pump sets and improves the accuracy of anomaly identification.

[0085] Example 3: In Example 2, although the accuracy of anomaly identification was improved by introducing a state anomaly prediction model to re-inspect the drainage pump, the possibility of false detection still exists due to the complex and variable working environment of the drainage pump and the diversity of anomaly characteristics. Especially when the hit rate is close to the sub-interval boundary, a single state anomaly prediction model may not be able to accurately determine whether the drainage pump is abnormal.

[0086] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0087] In some embodiments, each drainage pump group corresponds to a state anomaly prediction model, and each state anomaly prediction model includes several differentiated prediction models. Step S202 further includes:

[0088] S301, based on the preset first interval, is divided into three sub-intervals [30%, 50%), [50%, 70%), and [70%, 90%], with each sub-interval corresponding to a differentiated prediction model.

[0089] S302, obtain the hit rate corresponding to the abnormal probability value sequence of the target drainage pump, and locate the corresponding differential prediction model according to the sub-interval in which the hit rate is located.

[0090] S303, input the Euclidean distance value corresponding to the target drainage pump into the corresponding differential prediction model, and output the diagnostic result of the target drainage pump.

[0091] In some embodiments, the training method for the differential prediction model corresponding to each sub-interval specifically includes:

[0092] S401, In the acquired training dataset, obtain the hit rate corresponding to the sequence of abnormal probability values ​​obtained by each training data pump within its respective drainage pump group.

[0093] S402, based on the hit rate of each training data point, divide it into corresponding sub-intervals to obtain the sub-training set corresponding to each sub-interval.

[0094] S403 uses a sub-training set to train and optimize the parameters of a pre-defined neural network structure, resulting in a differentiated prediction model for each sub-interval.

[0095] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0096] The state anomaly prediction model corresponding to each drainage pump group is refined into several differentiated prediction models, with each sub-interval corresponding to a differentiated prediction model. This enables the model to handle drainage pumps with different hit rates more specifically, improving the model's adaptability and accuracy.

[0097] The preset first interval [30%, 90%] is divided into three sub-intervals [30%, 50%), [50%, 70%), and [70%, 90%]. Each sub-interval corresponds to a differential prediction model. Through this detailed division, the hit rate of the abnormal probability value sequence of the drainage pump can be more accurately located, so as to select a more suitable differential prediction model for re-examination.

[0098] Since drainage pumps with different hit rates may have different abnormal characteristics, a differentiated prediction model is used to re-examine the drainage pumps in each sub-interval, which improves the accuracy and reliability of the re-examination. This approach reduces false detections caused by insufficient model generalization ability and improves the accuracy of anomaly identification.

[0099] During model training, the training data is divided into corresponding sub-intervals based on the hit rate, resulting in a sub-training set for each sub-interval. This sub-training set is then used to train and optimize the parameters of a pre-defined neural network structure, yielding a differentiated prediction model for each sub-interval. This approach enhances the relevance of model training and improves model performance.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart control method for a building energy-saving water supply and drainage system, characterized in that, include: S101, collect vibration characteristic data and working characteristic parameters of all drainage pumps within a preset historical time period in each historical time window, and construct a multivariate state feature vector for each drainage pump. S102, input the multivariate state feature vectors of all drainage pumps into the pre-set clustering algorithm, and output K clusters, each cluster corresponding to a drainage pump group, and each drainage pump group including at least one drainage pump. S103, based on each drainage pump group, calculate the state extreme value of each drainage pump in each historical time window, form the state fluctuation sequence of the drainage pump, and generate the state deviation sequence of the drainage pump from each other drainage pump in the drainage pump group; the calculation method of the state extreme value is: obtain the vibration extreme value and working extreme value corresponding to the vibration characteristic data and working characteristic parameters of each drainage pump in each historical time window, and then perform a weighted summation of the vibration extreme value and working extreme value to obtain the state extreme value; S104, In all the state deviation sequences of each drainage pump in the drainage pump group, if the number of historical time windows that reach the deviation threshold in a certain state deviation sequence is greater than the preset number, then the corresponding drainage pump is identified as a suspected abnormal drainage pump. S105, Based on each suspected abnormal drainage pump, obtain its multivariate deviation set and its corresponding abnormal probability value every preset period, and generate an abnormal probability value sequence for each suspected abnormal drainage pump. The process of obtaining the multivariate state deviation set and its corresponding abnormal probability value every preset period includes: B1. Based on each suspected abnormal drainage pump, obtaining the multivariate state feature vector of the previous period and the multivariate state feature vector set of all other drainage pumps in the drainage pump group every preset period. B2. Calculate the Euclidean distance between the drainage pump and each multivariate state feature vector in the multivariate state feature vector set to form the multivariate state deviation set of the drainage pump. Each element in the set corresponds to an Euclidean distance value. The ratio of the number of elements in the multivariate state deviation set that are greater than a preset distance threshold to the total number of elements is determined as the anomaly probability value. S106. Based on each suspected abnormal drainage pump, generate the hit rate of the abnormal probability value sequence according to the preset abnormal probability threshold and abnormal probability value sequence. If the hit rate is greater than the hit threshold, the suspected abnormal drainage pump is confirmed to be abnormal.

2. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 1, characterized in that, The vibration characteristic data includes vibration frequency, amplitude, and vibration duration; the operating characteristic parameters include flow rate, pressure, rotational speed, and temperature; and the construction of the multivariate state feature vector for each drainage pump includes: A1. Calculate the statistical properties of the vibration characteristic data, including the mean, variance, and kurtosis values ​​corresponding to vibration frequency, amplitude, and vibration duration, respectively. A2. Calculate the statistical quantities of the working characteristic parameters, including the mean, variance, and kurtosis values ​​of flow rate, pressure, speed, and temperature, respectively. A3. Combine the statistics of vibration characteristic data and the statistics of working characteristic parameters to generate a multivariate state feature vector for each drainage pump.

3. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 1, characterized in that, The pre-set clustering algorithm is the K-Means clustering algorithm, which determines the center point of each cluster as the label feature vector of the corresponding drainage pump group.

4. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 1, characterized in that, The state fluctuation sequence of the drainage pump is represented as [ ,... The state deviation sequence is represented as [ ,... ]; in, This refers to the extreme state value within the k-th historical time window of a preset historical period. This represents the deviation of the state of this drainage pump from that of other drainage pumps in the same drainage pump group.

5. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 1, characterized in that, The hit rate of the abnormal probability value sequence is set as: the number of abnormal probability values ​​greater than the abnormal probability threshold / the total number of abnormal probability values ​​in the abnormal probability value sequence.

6. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 5, characterized in that, S106 includes: When the hit rate of the abnormal probability value sequence is within the preset first interval, the suspected abnormal drainage pump is identified as the target drainage pump. Using the pre-set abnormal retest mechanism, the suspected abnormal drainage pump at the corresponding point is retested, and its final diagnosis result is output. The first interval is set to [30%, 90%].

7. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 6, characterized in that, The pre-set anomaly retesting mechanism specifically includes: S201, obtain the Euclidean distance value of the multivariate state feature vectors of the target drainage pump and other drainage pumps in the drainage pump group that are diagnosed as normal in the most recent preset period. S202, input the Euclidean distance value into the pre-trained state anomaly prediction model, and output the diagnostic result of the target drainage pump, which includes abnormal and normal results; Each drainage pump group corresponds to a state anomaly prediction model. The state anomaly prediction model is determined based on the drainage pump group to which the target drainage pump belongs. The specific process of obtaining the pre-trained state anomaly prediction model is as follows: In a certain drainage pump group, a large number of multivariate state feature vectors of all drainage pumps over a long period of time are obtained. The drainage pumps that have been operating stably and normally for a long time in the drainage pump group are selected as reference objects. The Euclidean distance values ​​of the multivariate state feature vectors of each drainage pump and the corresponding reference object over all historical periods are obtained as training datasets. Each training data is labeled with the content of drainage pump state as abnormal or normal. The pre-set neural network structure is trained using the training dataset, and the model parameters are continuously optimized to obtain the final state abnormality prediction model.

8. The intelligent control method for building energy-saving water supply and drainage systems as described in claim 7, characterized in that, Each of the drainage pump sets corresponds to a state anomaly prediction model, and each state anomaly prediction model includes several differentiated prediction models. S202 further includes: S301, based on the preset first interval, is divided into three sub-intervals [30%, 50%), [50%, 70%), and [70%, 90%], with each sub-interval corresponding to a differential prediction model; S302, obtain the hit rate corresponding to the abnormal probability value sequence of the target drainage pump, and locate the corresponding differential prediction model according to the sub-interval in which the hit rate is located. S303, input the Euclidean distance value corresponding to the target drainage pump into the corresponding differential prediction model, and output the diagnostic result of the target drainage pump.

Images