Water affair abnormal water use intelligent identification method, device, equipment and medium

Abstract

The application provides a water affair anomaly water intelligent identification method, device, equipment and medium, relates to the technical field of data processing, and comprises the following steps: preprocessing to-be-identified water affair data to obtain basic feature data corresponding to the to-be-identified water affair data; performing time-aware path weighting based on the basic feature data through an unsupervised anomaly detection model to obtain an unsupervised anomaly score corresponding to the to-be-identified water affair data; constructing time-embedded feature data based on the to-be-identified water affair data, and determining a supervised anomaly classification probability corresponding to the to-be-identified water affair data based on the basic feature data, the time-embedded feature data and the unsupervised anomaly score through a supervised classification model; and determining an abnormal treatment strategy corresponding to the to-be-identified water affair data according to the unsupervised anomaly score, the supervised anomaly classification probability and a preset anomaly identification rule. The application can significantly improve the accuracy and intelligent level of water affair anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method, apparatus, equipment and medium for intelligent identification of abnormal water use in water affairs. Background Technology

[0002] With the continuous expansion of urban scale and increasing pressure on water resource management, smart water systems are playing an increasingly important role in ensuring water supply security and improving operational efficiency. As a core component of smart water systems, the water revenue system undertakes key tasks such as water metering, billing, payment collection, and monitoring of abnormal water usage. With the widespread application of IoT technology, the proliferation of IoT smart water meters has significantly improved the real-time performance and accuracy of data collection, providing a richer data foundation for water usage behavior analysis and anomaly detection.

[0003] However, due to numerous challenges in the current water revenue system, such as data diversity and large differences in sampling frequency, significant differences in user behavior, complexity of tiered water pricing and billing cycles, limitations of traditional rule-based early warning methods, and lack of intelligent model support and adaptive mechanisms, the accuracy and intelligence level of existing water anomaly identification are relatively low. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for intelligent identification of abnormal water use, which can significantly improve the accuracy and intelligence level of water anomaly identification.

[0005] In a first aspect, the present invention provides a method for intelligent identification of abnormal water use, comprising: Preprocess the water data to be identified to obtain the basic feature data corresponding to the water data to be identified; By using an unsupervised anomaly detection model, time-aware path weighting is applied to basic feature data to obtain the unsupervised anomaly score corresponding to the water data to be identified. Based on the water data to be identified, time-nested feature data is constructed, and a supervised classification model is used to determine the supervised anomaly classification probability corresponding to the water data to be identified, based on the basic feature data, time-nested feature data, and unsupervised anomaly scores. Based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules, the anomaly handling strategy corresponding to the water data to be identified is determined.

[0006] In one implementation, the water resources data to be identified carries multiple type labels, and the unsupervised anomaly detection model includes an isolated forest model corresponding to combinations of different type labels, wherein the isolated forest model includes multiple isolated trees; through the unsupervised anomaly detection model, time-aware path weighting is performed based on the basic feature data to obtain the unsupervised anomaly score corresponding to the water resources data to be identified, including: Based on the type labels carried by the water data to be identified, the target isolated forest model is determined from the unsupervised anomaly detection model; The following operations are performed using any isolated tree in the target isolated forest model: the original path length of the water data to be identified on the isolated tree is determined based on the basic feature data, and the time decay factor is determined based on the reference time point and the collection time of the water data to be identified. Based on the original path length and the time decay factor, the weighted path length of the water data to be identified on the isolated tree is determined. Based on the weighted path lengths of the water data to be identified on multiple isolated trees, the unsupervised anomaly score corresponding to the water data to be identified is determined.

[0007] In one implementation, the method further includes: For any combination of label types, the following steps are performed on the isolated forest model: The tiered pricing rules are used to determine the tiered boundary points to identify the disabled segmentation zones. During the construction of internal nodes of any isolated tree in the isolated forest model, it is determined whether the candidate splitting threshold corresponding to the internal node is within the disabled splitting interval, and the candidate splitting threshold is disabled if the determination result is yes, until the isolated tree reaches the preset depth. Extract target sample data corresponding to combinations of the same type of labels from the sample data, and use the target sample data to train the isolated trees in the isolated forest model.

[0008] In one implementation, the time-nested feature data includes: the number of days from the start of the billing cycle to the current date, the number of days from the end of the billing cycle to the current date, the proportion of the current date in the billing cycle, whether the current date is at the edge of the billing cycle, and the water volume change trend of the billing cycle.

[0009] In one implementation, a supervised classification model is used to determine the supervised anomaly classification probability of the water data to be identified, based on basic feature data, time-nested feature data, and unsupervised anomaly scores, including: The type labels carried by the water data to be identified are encoded based on hierarchical division to obtain hierarchical feature codes; Hierarchical feature encoding, basic feature data, time-nested feature data, and unsupervised anomaly scores are input into a supervised classification model to determine the supervised anomaly classification probability corresponding to the water data to be identified. The supervised anomaly classification probability is used to describe the probability that the water data to be identified belongs to each anomaly type.

[0010] In one implementation, an anomaly handling strategy for the water data to be identified is determined based on unsupervised anomaly scores, supervised anomaly classification probabilities, and preset anomaly identification rules, including: The unsupervised anomaly score and the supervised anomaly classification probability are fused to obtain the first score value; And based on the water data to be identified and the preset anomaly identification rules, a second score value is determined; By using a pre-trained scoring fusion processor, a comprehensive score value corresponding to the water data to be identified is determined based on the first score value and the second score value. Based on the mapping relationship between the comprehensive score and the anomaly handling strategy, the anomaly handling strategy corresponding to the water data to be identified is determined.

[0011] In one implementation, a second score value is determined based on the water data to be identified and preset anomaly identification rules, including: From the water data to be identified, extract the target water data corresponding to the identification items contained in the preset anomaly identification rules, and assign the score corresponding to the identification item to the water data to be identified when the target water data meets the rule conditions corresponding to the identification item. A second score is determined based on the score of the water data to be identified relative to each identification item.

[0012] Secondly, the present invention also provides a smart water usage abnormality identification device, comprising: The data preprocessing module is used to preprocess the water data to be identified in order to obtain the basic feature data corresponding to the water data to be identified. The unsupervised anomaly identification module is used to obtain the unsupervised anomaly score corresponding to the water data to be identified by using an unsupervised anomaly detection model and performing time-aware path weighting based on basic feature data. The supervised anomaly identification module is used to construct time-nested feature data based on the water data to be identified, and to determine the supervised anomaly classification probability corresponding to the water data to be identified based on the basic feature data, time-nested feature data and unsupervised anomaly scores through a supervised classification model. The strategy determination module is used to determine the anomaly handling strategy corresponding to the water data to be identified based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules.

[0013] Thirdly, the present invention also provides an electronic device including a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement any of the methods provided in the first aspect.

[0014] Fourthly, the present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to implement any of the methods provided in the first aspect.

[0015] This invention provides a method, apparatus, equipment, and medium for intelligent identification of abnormal water use. First, the water data to be identified is preprocessed to obtain basic feature data. Then, an unsupervised anomaly detection model is used to perform time-aware path weighting on the basic feature data to obtain an unsupervised anomaly score. Next, time-nested feature data is constructed based on the water data to be identified, and a supervised classification model is used to determine the supervised anomaly classification probability based on the basic feature data, time-nested feature data, and unsupervised anomaly score. Finally, an anomaly handling strategy is determined based on the unsupervised anomaly score, supervised anomaly classification probability, and preset anomaly identification rules. After preprocessing the water data to be identified, this method combines an unsupervised anomaly detection model and a supervised classification model to determine the unsupervised anomaly score and supervised anomaly classification probability, respectively, and combines these with preset anomaly identification rules to determine the final anomaly handling strategy. This invention significantly improves the accuracy and intelligence level of water anomaly identification.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating an intelligent method for identifying abnormal water usage according to an embodiment of the present invention; Figure 2 This is a technical framework diagram of an intelligent water use anomaly identification method provided by an embodiment of the present invention; Figure 3 A diagram illustrating an improved Isolation Forest mechanism provided in an embodiment of the present invention; Figure 4 A layered modeling architecture diagram provided for an embodiment of the present invention; Figure 5 A schematic diagram of a new supervised model feature system provided in an embodiment of the present invention; Figure 6 A diagram illustrating a model-rule fusion scoring mechanism provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of an intelligent water use abnormality identification device provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Currently, there are many technical challenges in the water revenue system: (1) Significant differences in data diversity and sampling frequency: IoT water meters typically support high-frequency data collection, such as reading the meter every 30 minutes, while traditional mechanical water meters have a longer collection cycle, often reading the meter once a quarter. This difference in sampling frequency leads to significant differences in the manifestation of data characteristics, which brings considerable difficulty to unified modeling and anomaly detection.

[0022] (2) Significant differences in user behavior: There are huge differences in water use behavior between residential users and non-residential users (such as commercial and industrial users), with obvious differences in periodicity, water consumption, and water use patterns. For example, residential users usually have periodic patterns such as peak water usage in the morning and evening and sparse water usage at night, while non-residential users (such as industrial and commercial users) show behavior patterns with strong continuity and periodic surges. A single rule or model is difficult to cover the diverse user groups and is prone to false alarms or false alarms.

[0023] (3) Complexity of tiered water pricing and billing cycles: Most cities adopt tiered water pricing, and water consumption accumulation and tier reset are often carried out on a quarterly or other cycle basis. Traditional anomaly detection models often ignore the boundary effect of the billing cycle. If the model fails to identify the "cycle reset point", it may misjudge "natural increase within the cycle" as an anomaly, or it may ignore "abnormal surge" behavior at the beginning of the cycle, resulting in misjudgment at the tier reset point and making it difficult to accurately identify abnormal water use.

[0024] (4) Limitations of traditional rule-based early warning methods: Traditional early warning rules based on static thresholds rely on experience to set thresholds, lack adaptability to dynamic changes, and are difficult to uncover complex water usage anomaly patterns, resulting in high false alarm and missed alarm rates, affecting operational efficiency and user experience. In addition, existing models focus more on data precision and accuracy, neglecting the business personnel's need for understanding the "cause of anomalies." If the model cannot provide an "anomaly cause chain," its practical value and acceptance will be limited.

[0025] (5) Lack of intelligent model support and adaptive mechanism: Most current systems lack a mechanism to use machine learning models to assist in judgment, and even more so a closed-loop mechanism that combines rules and models for dynamic optimization, making it difficult to meet the increasingly complex needs of water use anomaly detection.

[0026] Therefore, there is an urgent need for a new type of intelligent water use anomaly early warning method that can integrate different user categories and water meter types, adapt to multiple sampling frequencies and tiered water pricing cycles, combine unsupervised and supervised machine learning models, and deeply integrate with rule engines, in order to improve the accuracy of early warning and the intelligence level of the system.

[0027] Based on this, the present invention provides a method, device, equipment, and medium for intelligent identification of abnormal water use in water affairs, which can significantly improve the accuracy and intelligence level of water affairs anomaly identification. The core objectives of the embodiments of the present invention are: (1) to construct a differentiated intelligent early warning mechanism that integrates user type and meter characteristics; (2) to overcome the limitations of data heterogeneity by utilizing the structural innovation of machine learning models (unsupervised + supervised); (3) to construct a periodic perception model by introducing knowledge in areas such as billing cycle and tiered water pricing; (4) to form a "multi-source fusion" final judgment system by combining the rule engine and model judgment results; (5) to design a model interpreter and a manual review mechanism to improve the transparency of results and business availability; and (6) to support structured alarm output and system self-learning optimization to promote intelligent upgrading. Through the above means, the accuracy, interpretability, and adaptability of abnormal water use identification in the smart water affairs system are effectively improved, providing reliable technical support for water affairs enterprise operation decision-making, leakage control, and risk early warning.

[0028] To facilitate understanding of this embodiment, a detailed description of the intelligent water use identification method disclosed in this embodiment of the invention will be provided first. (See [link to relevant documentation]). Figure 1 The diagram shows a flowchart of an intelligent method for identifying abnormal water use. The method mainly includes the following steps S102 to S108: Step S102: Preprocess the water data to be identified to obtain the basic feature data corresponding to the water data to be identified.

[0029] The water data to be identified carries multiple type tags, such as resident type, water meter type, meter reading cycle, tiered billing cycle, and the identification of the corresponding pipe network zone. Different preprocessing methods are used for the water data to be identified for different water meter types to obtain corresponding basic feature data. The basic feature data corresponding to the water data to be identified collected by IoT water meters includes average daily water consumption, nighttime water consumption ratio, volatility, and peak frequency, etc. The basic feature data corresponding to the water data to be identified collected by mechanical water meters includes quarterly fluctuation trends, comparison with the average water consumption of similar users, and cumulative rate, etc.

[0030] Step S104: Using an unsupervised anomaly detection model, time-aware path weighting is performed based on basic feature data to obtain the unsupervised anomaly score corresponding to the water data to be identified.

[0031] In one example, the unsupervised anomaly detection model includes isolated forest models corresponding to combinations of different type labels. Each isolated forest model comprises multiple isolated trees. The combinations of type labels can be categorized as residential-IoT water meters, residential-mechanical water meters, non-residential-IoT water meters, and non-residential-mechanical water meters, with each combination corresponding to an isolated forest model. In the implementation, the target isolated forest model corresponding to the combination of type labels carried by the water data to be identified is first invoked. This target isolated forest model is then used to perform time-aware path weighting based on the feature data to obtain the unsupervised anomaly score for the water data to be identified.

[0032] Furthermore, in the process of training the isolated forest model, this embodiment of the invention introduces a ladder-sensitive pruning mechanism to avoid splitting at ladder boundaries when constructing isolated trees, thereby enhancing the tolerance of isolated trees to water jumps caused by ladder triggering.

[0033] Step S106: Construct time-nested feature data based on the water data to be identified, and determine the supervised anomaly classification probability corresponding to the water data to be identified through a supervised classification model based on the basic feature data, time-nested feature data, and unsupervised anomaly scores.

[0034] The time-nested feature data includes: the number of days from the start of the billing cycle to the current date, the number of days from the end of the billing cycle to the current date, the proportion of the current date within the billing cycle, whether the current date is at the edge of the billing cycle, and the water consumption trend within the billing cycle. In one example, the type label carried by the water data to be identified can be encoded using a hierarchical partitioning method to obtain hierarchical feature codes. These codes are then merged with the basic feature data, time-nested feature data, and unsupervised anomalies, and input into a supervised classification model to obtain supervised anomaly classification probabilities. These probabilities describe the probability that the water data to be identified belongs to each anomaly type. For example, anomaly types could include high nighttime water consumption or a sudden increase in water consumption at the end of the cycle.

[0035] Step S108: Determine the anomaly handling strategy corresponding to the water data to be identified based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules.

[0036] The preset anomaly identification rules can include multiple pre-defined identification rules corresponding to different anomaly types. For example, the preset anomaly identification rules can include "nighttime water consumption exceeds 50% of the daily average", "water consumption in this cycle is more than twice that of the previous cycle", "single-day water consumption exceeds the historical high value", etc. In one example, a rule-based score can be obtained based on the water data to be identified and the preset anomaly identification rules. The unsupervised anomaly score and the supervised anomaly classification probability are then fused to obtain a comprehensive score. Based on the mapping relationship between the comprehensive score and the anomaly handling strategy, the anomaly handling strategy corresponding to the water data to be identified is determined. The anomaly handling strategy includes "immediate alarm", "weak anomaly, enter the composite zone", "normal, no processing required", etc.

[0037] The intelligent water use anomaly identification method provided in this invention preprocesses the tax data to be identified, and then combines an unsupervised anomaly detection model and a supervised classification model to determine the unsupervised anomaly score and supervised anomaly classification probability, respectively. Finally, it combines preset anomaly identification rules to determine the anomaly handling strategy. This invention can significantly improve the accuracy and intelligence level of water anomaly identification.

[0038] For ease of understanding, this invention provides a specific implementation of a method for intelligent identification of abnormal water usage. This method is an intelligent water management early warning technology solution that integrates water meter type, user type, meter reading cycle, and tiered billing mechanism, combining unsupervised anomaly detection and supervised classification prediction. See [link to relevant documentation]. Figure 2The diagram shows a technical framework for an intelligent water usage anomaly identification method, including: data input; user water meter tagging: high-frequency data processing is performed on water usage data to be identified collected by IoT water meters, and low-frequency data processing is performed on water usage data to be identified collected by mechanical water meters; hierarchical unsupervised monitoring; supervised classification model; model and rule fusion engine, and outputs the following anomaly handling strategies: strong anomaly alarm (structured output) and weak anomaly manual review (feedback to optimize model / rules).

[0039] The specific implementation process is as follows: (1) User water meter labeling: Based on user profiles and meter registration information in the water system, the following tags are automatically extracted for model training and decision logic control: Table 1. Type labels and their corresponding functions

[0040] (2) Data acquisition and periodic processing module: Different data preprocessing methods are adopted according to different types of water meters, so that high-frequency and low-frequency data can be uniformly entered into the model system.

[0041] The data collection cycle of the IoT water meter is 30 minutes. The processing method is to construct a sliding window (7 days, 30 days) and extract the following features based on the constructed sliding window: average daily water consumption, nighttime water consumption ratio, fluctuation, peak frequency, etc. The data collection period for mechanical water meters is quarterly. The processing method is year-on-year and month-on-month enhancement (compared with the previous 1 quarter and the previous 2 quarters) and regression prediction residuals. Based on this, the following features are extracted: quarterly fluctuation trend, comparison with the average water consumption of similar users, and cumulative speed.

[0042] (3) An unsupervised anomaly detection model based on improved Isolation Forest. The unsupervised anomaly detection model includes an isolated forest model corresponding to combinations of different types of labels. The isolated forest model includes multiple isolated trees.

[0043] Considering that standard Isolation Forest performs poorly in identifying temporally continuous anomalies (such as water leaks) and is prone to misjudgment due to sparse data structures or unreasonable partitioning, this invention improves Isolation Forest, proposing an unsupervised anomaly detection model based on the improved Isolation Forest. For example... Figure 3 The diagram illustrates an improved IsolationForest mechanism, comprising three parts: time-aware path weighting, ladder-sensitive pruning, and hierarchical sub-model training. Specifically: (a) Time-aware path weighting: When constructing the splitting path, nodes whose time is "closest to the current time point" are weighted to enhance the model's perception of recent persistent anomalies. The code for time-aware path weighting is shown below: deftime_weighted_path_length(node,sample_time,current_time): path_length=0 while not node.is_leaf(): path_length += 1 feature,threshold=node.split_feature,node.threshold ifsample_time <threshold: node=node.left else: node=node.right #Weightedness: The closer to the current time and the shorter the path, the higher the anomaly score. time_decay=1.0 / (1+abs(current_time-sample_time).days) returnpath_length time_decay (ii) Step-sensitive pruning: Avoid splitting at step boundaries during tree construction (e.g., when water volume is low at the beginning of a quarter), enhancing the model's tolerance to "water usage jumps caused by step triggers." The code for step-sensitive pruning is shown below: efsplit_node(data,feature_candidates): forfeatureinfeature_candidates: iffeature=='cumulative_usage': #Skip the staircase breakpoint to avoid splitting near the breakpoint. ifis_near_step_boundary(data[feature]): continue threshold=random_split(data[feature]) left,right=data[data[feature]<=threshold],data[data[feature]>threshold] if len(left) > 0 and len(right) > 0: returnfeature,threshold returnNone,None defis_near_step_boundary(values): forstepin[10,15,20]: # Assume a step breakpoint ifany(abs(v-step)<0.5forvinvalues): returnTrue returnFalse (III) Hierarchical Sub-model Training: Isolation forest models are trained independently for different combinations. After scoring, unified normalization and weighted fusion are performed, such as... Figure 4 The diagram illustrates a hierarchical modeling architecture, including isolated forest models corresponding to different combinations such as residential-IoT water meters, residential-mechanical water meters, non-residential-IoT water meters, and non-residential-mechanical water meters. The code for training the hierarchical sub-models is shown below: classLayeredIsolationForest: def__init__(self): self.models={} deffit(self, data): grouped_data=data.groupby(['user_type','meter_type']) for(utype,mtype),groupingrouped_data: model=IsolationForest(n_estimators=100) model.fit(group.features) self.models[(utype,mtype)]=model defscore(self,sample): key=(sample['user_type'],sample['meter_type']) model = self.models.get(key) returnmodel.score(sample.features) Based on this, the embodiments of the present invention first provide a specific process for obtaining the unsupervised anomaly score corresponding to the water affairs data to be identified by using an unsupervised anomaly detection model and performing time-aware path weighting based on basic feature data, including: Step 1: Based on the type labels carried by the water data to be identified, determine the target isolated forest model from the unsupervised anomaly detection model. That is, based on the user type and water meter type carried by the water data to be identified, determine the corresponding combination, and then use the isolated forest model corresponding to the combination as the target isolated forest model.

[0044] Step 2: Perform the following operations using any isolated tree in the target isolated forest model: determine the original path length of the water data to be identified on the isolated tree based on the basic feature data, and determine the time decay factor based on the reference time point and the collection time of the water data to be identified. Based on the original path length and the time decay factor, determine the weighted path length of the water data to be identified on the isolated tree.

[0045] In one example, the process for determining the original path length is as follows: Starting from the root node: Input the basic feature data of the water data to be identified into the root node of the isolated tree.

[0046] Traversing the decision path: Based on the fundamental feature values ​​of the water data to be identified at each internal node, determine whether to move to the left or right subtree. For example, the segmentation feature of a node is night_usage_ratio (nighttime water usage ratio), and the segmentation threshold is 0.3. If the night_usage_ratio of the water data to be identified is 0.4, then the water data to be identified will be directed to the right subtree.

[0047] Record path length: Continue traversing until the water data to be identified reaches a leaf node. Record the number of edges traversed from the root node to the leaf node; this number is the original path length (path_length) of the water data to be identified on this isolated tree.

[0048] In one example, the process of determining the time decay factor is as follows: Obtaining timestamps: For water data to be identified, it is necessary to know its collection time (sample_time). Simultaneously, a reference time point (current_time) is needed, typically the current time for model inference or scoring.

[0049] Calculate the time difference: Calculate the difference between the collection time of the water data to be identified and the reference time point (e.g., the number of days). The formula is as follows: time_diff = abs(current_time - sample_time).days; where time_diff is the time difference, current_time is the reference time point, and sample_time is the collection time.

[0050] Calculate the Time Decay Factor: This is a weighted factor between 0 and 1 calculated based on the time difference. The further back in time the time, the closer the factor is to 0; the closer the time, the closer the factor is to 1. The formula is as follows: time_decay = 1.0 / (1 + time_diff); where time_decay is the time decay factor and time_diff is the time difference.

[0051] In one example, the process of determining the weighted path length is as follows: the product of the original path length and the time decay factor is used as the weighted path length.

[0052] Step 3: Based on the weighted path lengths of the water data to be identified across multiple isolated trees, determine the unsupervised anomaly score for the water data to be identified. In one example, the average of the weighted path lengths corresponding to each isolated tree can be used as the unsupervised anomaly score for the water data to be identified.

[0053] This invention also provides a training process for an isolated forest model, wherein the isolated forest model corresponding to any combination of label types performs the following steps: Step 1: Determine the tiered billing boundary points based on the preset tiered billing rules to identify the prohibited segmentation intervals. In practical applications, clearly define the tiered billing rules in the water system and extract key boundary values ​​as the prohibited segmentation intervals. For example, assuming the tiered breakpoints are 10, 15, and 20 (actually, this can be configured according to business needs, such as 15 tons for the first tier of residential water use and 25 tons for the second tier), the characteristic corresponding to these breakpoints is the cumulative water usage, which is the core feature of the pruning mechanism.

[0054] Step 2: During the construction of internal nodes of any isolated tree in the isolated forest model, determine whether the candidate splitting threshold corresponding to the internal node is within the disabled splitting interval, and disable the candidate splitting threshold if the determination result is yes, until the isolated tree reaches the preset depth.

[0055] In practical applications, during the construction phase of each internal node of the isolation tree, a pruning check is triggered when the cumulative water consumption is selected as the splitting feature. Specifically, the normal process is as follows: the isolation tree randomly selects a feature (such as cumulative water consumption) and a threshold (such as 12 tons), dividing the data into two groups: "≤12 tons" and ">12 tons". The pruning check checks the following: if a candidate splitting threshold is within the prohibited splitting interval, that threshold is discarded, and the feature is not used for splitting. If all candidate thresholds for that feature are within the prohibited splitting interval, the cumulative water consumption feature is skipped, and other features (such as nighttime water consumption percentage or average daily water consumption) are randomly selected for splitting.

[0056] Step 3: Extract target sample data corresponding to combinations of the same type of labels from the sample data, and use the target sample data to train the isolated trees in the isolated forest model. For example, for the isolated forest model corresponding to residential IoT water meters, the sample data corresponding to residential IoT water meters will be used to train the isolated forest model.

[0057] (4) Supervised classification model based on LightGBM (structural enhancement): Considering that existing models cannot perceive periodic patterns and lack interpretability, leading to a lack of trust in business applications, this embodiment of the invention provides a new feature system for supervised models, such as... Figure 5 The diagram illustrates a novel supervised model feature system. It inputs temporal features, business features, and explanatory features into a supervised classification model to output supervised anomaly classification probabilities. Temporal features can include basic feature data and time-nested feature data; business features are hierarchical feature codes obtained based on a hierarchical splitting mechanism of classification variables; and explanatory features can be understood as an anomaly explanation chain generation module. Specifically: (i) Time-nested feature engineering: Constructing features such as the number of days from the start of the billing cycle to the current date, the number of days from the end of the billing cycle to the current date, the proportion of the current date in the billing cycle, and whether the current date is on the edge of the billing cycle, and incorporating the water consumption change trend of the billing cycle (cumulative growth rate within the cycle, average daily rate). The code is shown below: defgenerate_time_features(row,billing_cycle_start,billing_cycle_end): row['days_from_cycle_start']=(row['timestamp']-billing_cycle_start).days row['days_to_cycle_end']=(billing_cycle_end-row['timestamp']).days row['cycle_progress']=row['days_from_cycle_start'] / (billing_cycle_end-billing_cycle_start).days row['weekly_avg_usage']=row['usage'].rolling(7).mean() row['cycle_usage_rate']=row['cumulative_usage'] / max(1,row['days_from_cycle_start']) returnrow (ii) Categorical variable hierarchical splitting mechanism: such as user category, water usage type, pipeline area, etc., are split hierarchically to avoid collinearity between categories, which could cause difficulties in model fitting. The code is shown below: defhierarchical_encoding(data,category_field): hierarchy={ 'user_type':['residential','commercial','industrial'], 'pipe_zone_id':['zone_A','zone_B','zone_C'] } forfieldinhierarchy: forlevelinhierarchy[field]: data[f'{field}_{level}']=(data[field]==level).astype(int) returndata (III) Anomaly Explanation Chain Generation Module (Feature Path Tracker): After each prediction, this module outputs the feature contribution path, indicating the main reason for the "anomaly" judgment (such as "high water consumption at night" or "sudden increase in the later stage of the cycle"), facilitating manual review and feedback optimization. Its code is shown below: defexplain_lightgbm_prediction(model,sample): shap_values=shap.TreeExplainer(model).shap_values(sample) explanation=[] fori,valinumerate(shap_values[0]): ifabs(val)>0.1: # Set the minimum contribution threshold explanation.append((sample.columns[i],val)) explanation.sort(key=lambdax:abs(x[1]),reverse=True) returnexplanation[:5] # Returns the top 5 main features Based on this, embodiments of the present invention provide a specific implementation method for determining the supervised anomaly classification probability of water affairs data to be identified by using a supervised classification model based on basic feature data, time-nested feature data, and unsupervised anomaly scores, including: Step 1: The type label carried by the water data to be identified is encoded based on hierarchical division to obtain hierarchical feature codes.

[0058] In one implementation, for each category at each level, an independent binary feature (0 or 1) is generated to achieve "one feature per category". The specific operation is as follows: First-level hierarchical coding: A feature is created for each first-level category. A sample belonging to this category is assigned a value of 1, otherwise it is assigned a value of 0. Example: The first-level coding of user type generates "user_type_resident" and "user_type_non-resident". The coding result for a resident user is (1, 0).

[0059] Second-level coding: Based on the first-level coding, subdivided features are created for each second-level category, also using binary assignment. Example: Based on the first-level "resident", "user_type_resident_family", "user_type_resident_dormitory", and "user_type_resident_apartment" are generated. The coding result for a family user is (1, 0, 0).

[0060] Step 2 involves inputting the hierarchical feature encoding, basic feature data, temporally nested feature data, and unsupervised anomaly scores into a supervised classification model. This model determines the supervised anomaly classification probability for the water resources data to be identified, describing the probability that the data belongs to each anomaly type. In one example, the hierarchical feature encoding, basic feature data, temporally nested feature data, and unsupervised anomaly scores can be merged. The merged data is then input into the LightGBM supervised classification model to obtain the probability that the water resources data belongs to each anomaly type.

[0061] (5) Model and Rule Fusion Engine: To ensure compatibility with existing business logic and expert knowledge while enhancing automated recognition capabilities, a dual-channel fusion strategy of "model and rule" is introduced, such as... Figure 6 The diagram illustrates a model-rule fusion scoring mechanism. Unsupervised anomaly scores, supervised anomaly classifications, and rule trigger scores (i.e., the second score) are input into the fusion engine, and the final anomaly determination is obtained through logistic regression weighting, such as strong anomaly alerts or weak anomaly reviews.

[0062] Based on this, the anomaly handling strategy for the water data to be identified is determined, including: 1) The unsupervised anomaly score and the supervised anomaly classification probability are fused to obtain the first score value.

[0063] The model and rules are evaluated independently and scored separately. Isolation Forest outputs an unsupervised anomaly score score_if, ranging from [0, 1]. LightGBM outputs the supervised anomaly classification probability prob_abnormal, ranging from [0, 1]. The weighted combination can be used to form the model score (i.e., the first score value) model_score: model_score = α × score_if + (1 - α) × prob_abnormal, where α is a configurable parameter.

[0064] 2) Based on the water data to be identified and the preset anomaly identification rules, determine the second score value, including: extracting the target water data corresponding to the identification items contained in the preset anomaly identification rules from the water data to be identified, and assigning the score corresponding to the identification item to the water data to be identified when the target water data meets the rule conditions corresponding to the identification item; and determining the second score value based on the score of the water data to be identified relative to each identification item.

[0065] Each rule match generates a recognition item, for example: Table 2. Rule number, rule content, and score upon triggering for each identification item.

[0066] The total score for a rule is the sum of all triggered scores, denoted as the second score value rule_score, which can be normalized to the interval [0, 1].

[0067] 3) Using a pre-trained scoring fusion unit, a comprehensive score value is determined based on the first and second score values ​​to identify the water data to be identified.

[0068] Specifically, construct a scoring fusion engine (logistic regression fusion): Using model_score and rule_score as input features, train a logistic regression model P(y=1 |X), and output the final anomaly probability, which is the comprehensive score final_score: final_score = sigmoid(β0 + β1 model_score + β2 rule_score) 4) Based on the mapping relationship between the comprehensive score and the anomaly handling strategy, determine the anomaly handling strategy corresponding to the water data to be identified, as shown in Table 3: Table 3. Mapping Relationship between Overall Score and Anomaly Handling Strategy

[0069] (6) Alarm generation and structured feedback module: After the system makes a comprehensive judgment based on the model and rules, it outputs the abnormal behavior as a structured message, including: User ID, water meter number, anomaly type, confidence level, anomaly time, and main cause of anomaly; The main cause of the anomaly is automatically generated based on the explanatory chain (e.g., "a surge in water volume during the cycle + higher nighttime water consumption"). Supports manual review and feedback, which can be used as a basis for model retraining or rule optimization; All alarms are stored in the database and can be used for subsequent statistics, reconciliation, and water loss analysis.

[0070] Furthermore, this embodiment of the invention provides an application example of a smart water usage identification method, which has been actually deployed in a local smart water revenue system, covering both residential and non-residential users, and involving two meter types: IoT water meters and mechanical water meters.

[0071] The system first performs tiered modeling based on user type and meter type, and then uses the unsupervised Isolation Forest model and the supervised LightGBM model to model and predict user water usage behavior.

[0072] During operation: For IoT water meter users, the system analyzes water usage data from the past 7 days in real time, identifies typical abnormal behaviors such as continuous high flow rates at night, and scores them using a rule engine (e.g., "nighttime water usage exceeds 50% of the daily average"). If the combined score exceeds a threshold, an automatic alarm is triggered. For mechanical water meter users, the system collects water usage data quarterly, compares it with historical data and data from other users in the same period to determine if there is any abnormal growth, and triggers a "weak anomaly alert" or directs the system to manual review based on model prediction scores.

[0073] Deployment results show that the model + rule fusion mechanism of this invention significantly reduces false positives and false negatives, improves the anomaly identification accuracy by about 40%, and has good interpretability and business adaptability.

[0074] In summary, the embodiments of the present invention have at least the following characteristics: 1) Intelligent early warning architecture based on hierarchical modeling of "user type + meter type": Based on the different water usage behavior characteristics and data frequencies of users (residential / non-residential) and meters (IoT / mechanical), differentiated modeling paths and feature systems are constructed to achieve refined anomaly identification under a unified system. This strategy solves the problem that traditional models cannot uniformly process multi-source data and is the core architectural innovation of this invention.

[0075] 2) Time-aware path weighting and tiered sensitive pruning mechanism of Isolation Forest: To address data jumps caused by temporal continuity anomalies (such as nighttime leaks) and tiered water pricing systems, this invention introduces the following into the unsupervised model: Path weighting mechanism: Path length is weighted based on time distance; Pruning mechanism: Avoid splitting nodes at the boundaries of water price tiers.

[0076] These two structural modifications improve the accuracy and adaptability of the anomaly detection model, providing key support for the application of the embodiments of this invention in the industry.

[0077] 3) Model-rule fusion scoring mechanism and interpretable early warning output: This invention constructs a fusion engine for model scoring and rule scoring, and uses weighted and logistic regression methods to fuse decisions. It also supports structured output of "explanation paths for abnormal causes" (such as "abnormal growth rate within a period and higher water consumption at night"), realizing a verifiable and feedback-enabled business closed loop. This reflects the intelligence and business compatibility of this invention.

[0078] Based on the foregoing embodiments, this invention provides a smart water usage abnormality identification device, see [link to previous embodiment]. Figure 7 The diagram shows a structural schematic of a smart water usage abnormality identification device. The device mainly includes the following components: The data preprocessing module 702 is used to preprocess the water data to be identified in order to obtain the basic feature data corresponding to the water data to be identified. The unsupervised anomaly identification module 704 is used to obtain the unsupervised anomaly score corresponding to the water data to be identified by performing time-aware path weighting based on basic feature data through an unsupervised anomaly detection model. The supervised anomaly identification module 706 is used to construct time-nested feature data based on the water affairs data to be identified, and to determine the supervised anomaly classification probability corresponding to the water affairs data to be identified based on the basic feature data, time-nested feature data and unsupervised anomaly scores through a supervised classification model. The strategy determination module 708 is used to determine the anomaly handling strategy corresponding to the water data to be identified based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules.

[0079] The intelligent water use anomaly identification method provided in this invention preprocesses the tax data to be identified, and then combines an unsupervised anomaly detection model and a supervised classification model to determine the unsupervised anomaly score and supervised anomaly classification probability, respectively. Finally, it combines preset anomaly identification rules to determine the anomaly handling strategy. This invention can significantly improve the accuracy and intelligence level of water anomaly identification.

[0080] In one implementation, the water resources data to be identified carries multiple types of labels, and the unsupervised anomaly detection model includes an isolated forest model corresponding to combinations of different types of labels, wherein the isolated forest model includes multiple isolated trees; the unsupervised anomaly identification module 704 is specifically used for: Based on the type labels carried by the water data to be identified, the target isolated forest model is determined from the unsupervised anomaly detection model; The following operations are performed using any isolated tree in the target isolated forest model: the original path length of the water data to be identified on the isolated tree is determined based on the basic feature data, and the time decay factor is determined based on the reference time point and the collection time of the water data to be identified. Based on the original path length and the time decay factor, the weighted path length of the water data to be identified on the isolated tree is determined. Based on the weighted path lengths of the water data to be identified on multiple isolated trees, the unsupervised anomaly score corresponding to the water data to be identified is determined.

[0081] In one implementation, the unsupervised anomaly detection module 704 is further configured to: For any combination of label types, the following steps are performed on the isolated forest model: The tiered pricing rules are used to determine the tiered boundary points to identify the disabled segmentation zones. During the construction of internal nodes of any isolated tree in the isolated forest model, it is determined whether the candidate splitting threshold corresponding to the internal node is within the disabled splitting interval, and the candidate splitting threshold is disabled if the determination result is yes, until the isolated tree reaches the preset depth. Extract target sample data corresponding to combinations of the same type of labels from the sample data, and use the target sample data to train the isolated trees in the isolated forest model.

[0082] In one implementation, the time-nested feature data includes: the number of days from the start of the billing cycle to the current date, the number of days from the end of the billing cycle to the current date, the proportion of the current date in the billing cycle, whether the current date is at the edge of the billing cycle, and the water volume change trend of the billing cycle.

[0083] In one implementation, the supervisory anomaly identification module 706 is specifically used for: The type labels carried by the water data to be identified are encoded based on hierarchical division to obtain hierarchical feature codes; Hierarchical feature encoding, basic feature data, time-nested feature data, and unsupervised anomaly scores are input into a supervised classification model to determine the supervised anomaly classification probability corresponding to the water data to be identified. The supervised anomaly classification probability is used to describe the probability that the water data to be identified belongs to each anomaly type.

[0084] In one implementation, the strategy determination module 708 is specifically used for: The unsupervised anomaly score and the supervised anomaly classification probability are fused to obtain the first score value; And based on the water data to be identified and the preset anomaly identification rules, a second score value is determined; By using a pre-trained scoring fusion processor, a comprehensive score value corresponding to the water data to be identified is determined based on the first score value and the second score value. Based on the mapping relationship between the comprehensive score and the anomaly handling strategy, the anomaly handling strategy corresponding to the water data to be identified is determined.

[0085] In one implementation, the strategy determination module 708 is specifically used for: From the water data to be identified, extract the target water data corresponding to the identification items contained in the preset anomaly identification rules, and assign the score corresponding to the identification item to the water data to be identified when the target water data meets the rule conditions corresponding to the identification item. A second score is determined based on the score of the water data to be identified relative to each identification item.

[0086] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0087] This invention provides an electronic device, specifically, the electronic device includes a processor and a memory; the memory stores a computer program, which, when run by the processor, executes the method described in any of the above embodiments.

[0088] Figure 8The present invention provides a schematic diagram of the structure of an electronic device 100, which includes a processor 80, a memory 81, a bus 82 and a communication interface 83. The processor 80, the communication interface 83 and the memory 81 are connected through the bus 82. The processor 80 is used to execute executable modules, such as computer programs, stored in the memory 81.

[0089] The memory 81 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 83 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0090] Bus 82 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0091] The memory 81 is used to store programs. After receiving an execution instruction, the processor 80 executes the program. The method executed by the device for defining the flow process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 80 or implemented by the processor 80.

[0092] The processor 80 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 80 or by instructions in software form. The processor 80 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 81. The processor 80 reads the information in memory 81 and, in conjunction with its hardware, completes the steps of the above method.

[0093] The computer program product of the readable storage medium provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For specific implementation, please refer to the foregoing method embodiments, which will not be repeated here.

[0094] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent identification of abnormal water use, characterized in that, include: The water data to be identified is preprocessed to obtain the basic feature data corresponding to the water data to be identified; By using an unsupervised anomaly detection model, time-aware path weighting is performed based on the basic feature data to obtain the unsupervised anomaly score corresponding to the water data to be identified; Based on the water affairs data to be identified, time-nested feature data is constructed, and through a supervised classification model, the supervised anomaly classification probability corresponding to the water affairs data to be identified is determined based on the basic feature data, the time-nested feature data, and the unsupervised anomaly score. Based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules, the anomaly handling strategy corresponding to the water data to be identified is determined.

2. The intelligent water use anomaly identification method according to claim 1, characterized in that, The water data to be identified carries multiple types of labels, and the unsupervised anomaly detection model includes an isolated forest model corresponding to combinations of different types of labels. The isolated forest model includes multiple isolated trees. By using an unsupervised anomaly detection model, time-aware path weighting is applied to the basic feature data to obtain the unsupervised anomaly score corresponding to the water affairs data to be identified, including: Based on the type label carried by the water data to be identified, the target isolated forest model is determined from the unsupervised anomaly detection model; The following operations are performed using any of the isolated trees in the target isolated forest model: the original path length of the water data to be identified on the isolated tree is determined based on the basic feature data, and the time decay factor is determined based on the reference time point and the collection time of the water data to be identified. Based on the original path length and the time decay factor, the weighted path length of the water data to be identified on the isolated tree is determined. Based on the weighted path lengths corresponding to the water data to be identified on multiple isolated trees, the unsupervised anomaly score corresponding to the water data to be identified is determined.

3. The intelligent water use anomaly identification method according to claim 2, characterized in that, The method further includes: For any combination of the aforementioned type labels, the following steps are performed on the isolated forest model: The tiered pricing rules are used to determine the tiered boundary points to identify the disabled segmentation zones. During the construction of any internal node of the isolated tree in the isolated forest model, it is determined whether the candidate splitting threshold corresponding to the internal node is within the disabled splitting interval, and the candidate splitting threshold is disabled if the determination result is yes, until the isolated tree reaches the preset depth. Extract target sample data corresponding to combinations of labels of the same type from the sample data, and use the target sample data to train the isolated trees in the isolated forest model.

4. The intelligent water use anomaly identification method according to claim 1, characterized in that, The time-nested feature data includes: the number of days from the start of the billing cycle to the current date, the number of days from the end of the billing cycle to the current date, the proportion of the current date in the billing cycle, whether the current date is at the edge of the billing cycle, and the water volume change trend of the billing cycle.

5. The intelligent water use anomaly identification method according to claim 1, characterized in that, Using a supervised classification model, based on the basic feature data, the time-nested feature data, and the unsupervised anomaly score, the supervised anomaly classification probability corresponding to the water affairs data to be identified is determined, including: The type label carried by the water data to be identified is encoded based on hierarchical division to obtain hierarchical feature codes; The hierarchical feature encoding, the basic feature data, the time-nested feature data, and the unsupervised anomaly score are input into the supervised classification model to determine the supervised anomaly classification probability corresponding to the water data to be identified. The supervised anomaly classification probability is used to describe the probability that the water data to be identified belongs to each anomaly type.

6. The intelligent water use anomaly identification method according to claim 1, characterized in that, Based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules, an anomaly handling strategy corresponding to the water data to be identified is determined, including: The unsupervised anomaly score and the supervised anomaly classification probability are fused to obtain a first score value; And based on the water data to be identified and the preset anomaly identification rules, a second score value is determined; A comprehensive score value corresponding to the water data to be identified is determined based on the first score value and the second score value using a pre-trained score fusion processor. Based on the mapping relationship between the comprehensive score and the anomaly handling strategy, the anomaly handling strategy corresponding to the water data to be identified is determined.

7. The intelligent water use anomaly identification method according to claim 6, characterized in that, Based on the water data to be identified and the preset anomaly identification rules, a second score value is determined, including: From the water data to be identified, extract the target water data corresponding to the identification items contained in the preset anomaly identification rules, and when the target water data meets the rule conditions corresponding to the identification item, assign the score corresponding to the identification item to the water data to be identified; A second score is determined based on the score of the water data to be identified relative to each of the identification items.

8. A smart water usage abnormality identification device, characterized in that, include: The data preprocessing module is used to preprocess the water data to be identified in order to obtain the basic feature data corresponding to the water data to be identified. The unsupervised anomaly identification module is used to obtain the unsupervised anomaly score corresponding to the water data to be identified by performing time-aware path weighting based on the basic feature data through the unsupervised anomaly detection model. The supervised anomaly identification module is used to construct time-nested feature data based on the water affairs data to be identified, and to determine the supervised anomaly classification probability corresponding to the water affairs data to be identified based on the basic feature data, the time-nested feature data and the unsupervised anomaly score through a supervised classification model. The strategy determination module is used to determine the anomaly handling strategy corresponding to the water data to be identified based on the unsupervised anomaly score, the supervised anomaly classification probability, and the preset anomaly identification rules.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing computer-executable instructions executable by the processor, the processor executing the computer-executable instructions to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images