Intelligent community power consumption security data linkage processing method

Abstract

The present disclosure relates to the field of power systems and intelligent control technology, and specifically discloses a kind of wisdom community power consumption security data linkage processing method, to realize the efficient collaborative analysis and intelligent response of multi-source power consumption data and security equipment data in community. By deploying smart meters, power consumption equipment monitoring modules and security terminals, real-time acquisition and preprocessing of various data, extraction of multi-dimensional features such as time series, space and behavior, and based on multi-source semantic association embedding network, adaptive fusion of data and efficient identification of implicit association events are realized. Combined with causal reasoning graph, the system can automatically infer the occurrence path and potential risk of the event, and through the self-feedback linkage decision mechanism, dynamically adjust the response strategy, continuously optimize the identification and disposal accuracy of abnormal events, and effectively improve the intelligent security capability and emergency response efficiency of the community.

Description

Technical Field

[0001] This disclosure relates to the field of power system and intelligent control technology, specifically to a method for data linkage processing of power security in smart communities. Background Technology

[0002] With the acceleration of urbanization and the continuous advancement of smart community construction in my country, smart communities have become an important vehicle for achieving refined urban management and a high-quality life for residents. As a key component of smart community construction, community electricity monitoring and security systems have always been the core foundation for ensuring residents' daily safety and improving the community's intelligence level. However, most current community electricity monitoring and security systems still suffer from problems such as fragmented operations, data silos, and weak interoperability, making it difficult to fully leverage the comprehensive value of multi-source heterogeneous data and improve the automatic identification and handling capabilities of abnormal events. There is an urgent need to achieve multi-source data fusion and intelligent prevention and control through next-generation data processing and intelligent linkage methods.

[0003] In recent years, with the rapid evolution of big data, artificial intelligence, and the Internet of Things, how to deeply integrate heterogeneous electricity consumption and security data in communities and establish a multi-source data-driven intelligent linkage processing mechanism has become a hot and challenging issue in smart community research and engineering implementation. Multi-source data linkage models not only involve the efficient collection and preprocessing of raw data, but also need to address multiple technical challenges such as time-series alignment, missing noise repair, deep feature extraction, semantic association modeling, and causal reasoning. Currently, some preliminary studies attempt to use data warehouses, rule engines, or basic machine learning methods to achieve basic data synchronization and anomaly detection. However, due to the lack of deep semantic modeling and adaptive fusion methods for multi-source, multi-modal data, they still cannot fully solve the dilemmas of "data silos," "difficulty in discovering weak associations," and "difficulty in interpreting causal chains."

[0004] Specifically, the key bottlenecks in the linkage between community electricity consumption and security are mainly reflected in the following aspects: First, the asynchronicity and heterogeneity of multi-source data acquisition terminals make it difficult to directly align and correlate data, and how to achieve efficient fusion under a time benchmark is a fundamental challenge; Second, the raw data contains many missing values, abnormal noise, and interference signals, which can easily affect the accuracy of subsequent anomaly detection and linkage analysis; Third, the diverse types of equipment and complex behavioral characteristics make it difficult to automatically summarize highly sensitive features and model behavioral patterns under unsupervised and semi-supervised conditions; In addition, the potential linkage relationships between different data modalities are often not explicit, and common correlation coefficients and statistical characteristics are difficult to reveal weak coupling or hidden causal mechanisms in high-dimensional space, which restricts the ability to discover complex anomalies and linkage events. Finally, in the actual management and response stage, the causal links of community events are complex and variable, and how to transform the analysis results into traceable and interpretable risk warning and response decisions is also a bottleneck that the intelligent linkage system must solve. Summary of the Invention

[0005] To address the aforementioned technical issues, this disclosure provides a method for data linkage processing in smart community electricity security, comprising the following specific steps: Multi-source power consumption and security data collection and preprocessing: Collect various power consumption data and security equipment data in the community, and perform time alignment, missing value imputation and abnormal noise filtering on the raw data; Feature extraction and local semantic modeling: Based on a custom local semantic feature discovery module, temporal, spatial and behavioral features are extracted from electricity consumption and security data respectively; transformative feature modeling is adopted to automatically summarize high-sensitivity features for each type of device. Multi-source data semantic association embedding and adaptive fusion utilize the multi-source semantic association embedding network SAEN to automatically learn the potential association between electricity consumption data and security events; through dynamic embedding space, it measures the possibility of linkage between different modalities and different devices; the fusion process dynamically adjusts the weights of each source data to achieve adaptive feature recombination and high-dimensional information fusion for different scenarios; and outputs a global linkage feature vector, which significantly improves the ability to identify weakly / implicitly related events. Construct a causal reasoning graph. Based on the linkage data between abnormal power consumption and security incidents, construct an event causal reasoning graph. Nodes represent the characteristic states of different devices / events, and edges represent the temporal sequence and influence relationships. Through graph structure reasoning algorithms, discover the path and potential causal chain of event occurrence. Self-feedback linkage decision-making and response: Design a causal link self-feedback linkage decision-making algorithm CCRSD. Whenever a linkage event is detected, it automatically retrieves, infers and predicts subsequent risks and corresponding response actions.

[0006] In some embodiments, the multi-source power consumption and security data acquisition and preprocessing further includes: real-time acquisition of raw data from various power consumption and security devices by deploying multi-source acquisition terminals such as smart meters, power equipment monitoring modules, access control, cameras and alarm sensors, covering power consumption status, device actions, access control records, video summaries and alarm log information; The collected asynchronous data is aligned based on a unified time base, and missing values ​​are filled in using interpolation, forward or backward padding. Statistical analysis and rule filtering are combined to remove abnormal noise. Finally, all data is converted into a standardized encoding format and unified semantic labels to form a high-quality basic dataset that meets the requirements of subsequent multi-source data fusion.

[0007] In some embodiments, the feature extraction and local semantic modeling further include: extracting local statistical features, spatial neighborhood features and behavioral event features from the collected time-series data of electricity consumption and security equipment, respectively; The aforementioned multidimensional original features are then input into a trainable feature transformation network to achieve nonlinear high-dimensional mapping of the original data, ultimately obtaining a unified semantic feature vector for each device at each time step. This vector is then used for subsequent multi-source data fusion analysis and the linkage identification and modeling of complex events.

[0008] In some embodiments, the multi-source data semantic association embedding and adaptive fusion further includes: introducing a multi-source semantic association embedding network to perform cross-modal unified embedding of high-order feature representations of electrical equipment and security equipment, and dynamically measuring the potential linkage relationship between devices in different times and modalities by constructing an adaptive correlation matching mechanism. Furthermore, by employing an attention mechanism to assign weights to the globally embedded features, the adaptive fusion of multi-source features is achieved, generating a global linkage feature representation that comprehensively reflects the overall electricity consumption and security status of the community. This network can continuously adapt to changes in the environment and behavioral patterns through online updates, thereby enhancing the identification of abnormal linkage behaviors and data support capabilities.

[0009] In some embodiments, the construction of the causal reasoning graph further includes: based on the global linkage features obtained after multi-source data fusion, performing node modeling of various states and key events of electrical equipment and security equipment at different times; By abstracting the temporal sequence and correlation between devices and events through graph structures, the structural learning algorithm is used to dynamically determine the causal influence strength between nodes, and key causal chains and event propagation paths are automatically discovered through graph reasoning methods. This enables interpretable modeling and real-time reasoning of the causal mechanisms of complex events, thereby providing a scientific basis for anomaly tracing, diagnosis and intelligent response.

[0010] In some embodiments, the self-feedback linkage decision and response further includes: based on the causal reasoning graph and global linkage features, automatically identifying and activating the corresponding causal link for the detected high-risk linkage event, dynamically assessing the potential risk level by combining the link structure, causal weight and the status of each node, and generating an automated linkage response action sequence accordingly. It includes early warning, notification, power outage and security control, and collects external feedback signals after the response. Based on the feedback, it adaptively corrects the causal weights and decision parameters. Through a continuous self-feedback learning mechanism, it continuously optimizes the accuracy and response efficiency of the system's abnormal handling, and realizes intelligent closed-loop linkage prevention and control.

[0011] The main beneficial effects of this disclosure include: by introducing a causal link self-feedback linkage decision-making algorithm, it achieves deep integration of multi-source heterogeneous data on electricity consumption and security, and mines complex linkage relationships, effectively breaking through the technical bottlenecks of slow response and poor adaptability of traditional rule-driven linkage mechanisms. Based on a causal reasoning graph, the system can automatically perceive and quantify potential risks, and promptly generate targeted linkage response actions, realizing intelligent and closed-loop handling of abnormal events. Simultaneously, through a self-feedback learning mechanism, it dynamically optimizes causal weights and linkage strategies based on actual external response results, significantly improving the accuracy and timeliness of anomaly detection and linkage response, and enhancing the system's adaptive evolutionary capability in changing environments. Overall, this invention improves the automation, intelligence, and risk prevention level of community security management, providing an efficient and reliable intelligent linkage solution for smart communities, smart buildings, and other scenarios. Attached Figure Description

[0012] To more clearly illustrate the technical solutions and advantages in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 According to an embodiment of this disclosure, a flowchart illustrating a method for data linkage processing of power security in smart communities is provided. Detailed Implementation

[0014] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed herein. This disclosure can also be implemented or applied to systems through other different specific embodiments, and various details in this disclosure can also be modified or changed according to different viewpoints and application systems without departing from the spirit of this disclosure. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this disclosure can be combined with each other.

[0015] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings, so that those skilled in the art to which this disclosure pertains can readily implement it. This disclosure may be embodied in many different forms and is not limited to the embodiments described herein.

[0016] In this disclosure, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic represented in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. Furthermore, the specific features, structures, materials, or characteristics represented may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples represented in this disclosure, as well as the features of those different embodiments or examples.

[0017] Furthermore, the terms "first" and "second" are used for illustrative purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the representation of this disclosure, "a plurality of" means two or more, unless otherwise expressly and specifically defined.

[0018] For the purpose of clarity, devices unrelated to the description are omitted, and the same or similar components throughout the specification are given the same reference numerals.

[0019] Although the terms first, second, etc., are used in some instances herein to refer to various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, first interface and second interface, etc., are used. Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to also include the plural forms unless the context indicates otherwise. It should be further understood that the terms “comprising,” “including,” indicate the presence of features, steps, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, steps, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” Exceptions to this definition will only occur if the combination of elements, functions, steps, or operations is inherently mutually exclusive in some way.

[0020] The technical terms used herein are for reference only to specific embodiments and are not intended to limit the scope of this disclosure. The singular form used herein includes the plural form unless the statement explicitly indicates otherwise. The word "comprising" as used in this specification means to specify a particular characteristic, region, integer, step, operation, element, and / or component, and does not exclude the presence or addition of other characteristics, regions, integers, steps, operations, elements, and / or components.

[0021] Although not explicitly defined, all terms, including technical and scientific terms used herein, shall have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains. Terms defined in commonly used dictionaries shall be further interpreted as having a meaning consistent with the relevant technical literature and the content of this present disclosure, and shall not be over-interpreted as having an ideal or overly formulaic meaning unless otherwise defined.

[0022] To address the problems raised in the background art, this disclosure provides a method for data linkage processing in smart community electricity security. Specifically, in some embodiments of this disclosure, Figure 1 A flowchart illustrating a community electricity safety early warning method based on the Internet of Things (IoT) is shown. Figure 1 As shown, the specific steps may include the following: Step 1: Multi-source electricity consumption and security data collection and preprocessing. Specifically, this involves collecting various types of electricity consumption data (smart meters, electrical device status) and security device data (access control, cameras, alarm / crime logs, etc.) within the community. The raw data undergoes time alignment, missing value imputation, and abnormal noise filtering. A unified data encoding format is established to prepare for subsequent data fusion.

[0023] In the multi-source data collection and preprocessing stage of electricity and security data in smart communities, the first step is to deploy various types of data collection devices in a reasonable distribution throughout the community. This includes smart meters distributed in residential buildings, power distribution rooms, and public areas to collect detailed power parameters such as electricity consumption, power, current, and voltage for each household and public area; it also includes monitoring modules configured for key electrical equipment (such as elevators, air conditioning, and lighting systems) to record equipment operating status, switching actions, fault alarms, and other information. In terms of security, access control systems need to be installed to collect entry and exit records of residents and visitors; video surveillance cameras covering key passages, unit doors, and public spaces need to be installed to obtain video streams and generate brief summaries through intelligent analysis (such as moving target detection and abnormal behavior recognition); at the same time, various security sensors such as infrared, smoke detectors, and vibration sensors, as well as alarm recording terminals that integrate alarm information, need to be deployed to achieve real-time reporting and categorized tracking of security incidents.

[0024] These multi-source devices, based on their design requirements, can collect data periodically at fixed time intervals (e.g., every minute, every 5 seconds) or generate and report data based on event-triggered patterns (e.g., abnormal actions, access control card swipes, alarm events). Therefore, the sampling times of the data from different sources are highly asynchronous, resulting in significant temporal heterogeneity and uneven distribution of the raw data. To address this issue, a unified time reference (e.g., NTP clock synchronization) needs to be introduced after data aggregation to align all data at granularities such as milliseconds, seconds, or minutes. This can be achieved by using a sliding window aggregation method to aggregate and normalize data from different devices within the same time window based on timestamps, or by using event-driven methods to correlate related data segments over time. This ensures that in subsequent analysis and feature extraction stages, accurate temporal relationships and causal connections can be established for all types of data.

[0025] During data collection, missing and outlier values ​​are unavoidable. For example, due to network communication interruptions, equipment failures, and signal interference, some devices may miss or fail to report data at certain times. To address this, diverse data completion methods are needed: for quantitative data, linear interpolation, sample mean imputation, and forward / backward value imputation can be used; for categorical or event-based data, predictive completion can be performed by combining business rules and historical behavior patterns to ensure that important features are not overlooked. For abnormal noise such as bit errors, extreme value interference, and false alarms in various device signals, reasonable threshold judgments, statistical distribution-based anomaly detection methods, and rule engine filtering are required to remove, correct, or mark data points exceeding the normal range, ensuring data quality.

[0026] After data cleaning, noise reduction, and completion, all collected raw data needs to be standardized and semantically annotated. This mainly includes: parsing and structuring the encoding and data formats (such as JSON, binary, and custom protocols) of various device messages, uniformly converting them into standardized data tables or object structures; adding clear semantic tags to each field; mapping and normalizing fields for data from different devices that have similar functions but different names or values ​​(such as "switch status," "operation flag," and "alarm event type"); and standardizing the data value range and units to ensure seamless integration of data from different manufacturers, models, and batches during subsequent fusion analysis. These detailed and systematic data preprocessing operations not only effectively improve the consistency, integrity, and usability of multi-source data but also provide a solid and reliable foundation of data for subsequent deep semantic fusion, causal reasoning analysis, and intelligent linkage decision-making of electricity and security data.

[0027] Step 2: Feature extraction and local semantic modeling. Specifically, based on a custom local semantic feature discovery module, temporal, spatial, and behavioral features can be extracted from electricity consumption and security data respectively. Transformative feature modeling is employed, automatically summarizing highly sensitive features for each type of device, rather than using traditional manually defined raw features. The extraction results are represented as semantic vectors, preparing for subsequent deep-level association.

[0028] In the feature extraction and local semantic modeling stage, the local semantic feature discovery module is first used to extract structured features from different types of data (electricity consumption data and security data). Given N electrical devices and M security devices, the original data sequence of any device i can be represented as... ,in This represents the observation value of device i at time t. For example, the smart meter... It can be a real-time power value, for access control. It can be used to count entry and exit events.

[0029] For time series characteristics, the moving average of each sequence is first calculated. (in (for the window length), and the sliding standard deviation. This is used to characterize local volatility. Furthermore, the autocorrelation function is utilized. To reveal periodicity and behavioral inertia.

[0030] For spatial characteristics, the location information of the equipment is further refined. Beyond simply calculating the Euclidean distance between devices, spatial relationship modeling techniques such as spatial clustering and thermal zoning can be introduced based on the actual layout of the community. For example, based on the affiliation of equipment with building units and functional areas (such as elevator shafts, public passageways, and machine rooms), the consistency of joint power consumption behavior and the co-occurrence probability of security events of multiple devices in different spatial domains can be statistically analyzed, thereby extracting spatially sensitive collective characteristics. Furthermore, the radius r of the neighborhood definition can be dynamically adjusted, or a density adaptive algorithm can be used to fully reflect the heterogeneity of power consumption / security behavior in different areas.

[0031] Assuming the device has location information The Euclidean distance between devices can be calculated. Furthermore, it allows for the statistical analysis of collective behavioral characteristics within a specific spatial neighborhood, such as neighborhood characteristics. Average electricity consumption within .

[0032] When modeling behavioral features, the event types and their spatiotemporal patterns are further refined. For electrical equipment, this can be subdivided into active operational behaviors (such as power on / off, power surges) and passive abnormal behaviors (such as overload alarms, prolonged operation without shutdown, etc.). For security equipment, it distinguishes between regular entry / exit, suspicious intrusion, and emergency alarms, extracting statistics such as operation duration, interval period, number of event mutations, and spatiotemporal diffusion characteristics of sudden events from the event sequence. Simultaneously, for scenarios such as behavioral state transitions and abnormal behavior chains, conditional probability statistical analysis is used to extract composite sequence features such as "sudden high power consumption → abnormal entry / exit in the area → security alarm". This defines the behavioral event sequence. ,in This indicates whether a specific operation (such as switching on / off or alarming) occurred at time t, and the frequency of the behavior can be further statistically analyzed. And the distribution of the duration of the behavior.

[0033] Unlike traditional fixed features, this approach employs transformative feature modeling. It not only performs multi-scale fusion of preliminary statistical features but also flexibly expands the feature space according to actual business needs. For example, auxiliary attributes such as equipment type codes, operating status indicators, and equipment health scores can be introduced to further enhance feature discrimination. All original and statistical features are concatenated into an original feature vector. This vector is then input into a nonlinear transformation network (such as a multilayer perceptron, convolutional neural network, or temporal self-attention mechanism) and, through an end-to-end trainable process, automatically mines hidden high-order combined features and behavioral representations. The final output high-dimensional semantic vector not only contains rich spatiotemporal behavioral semantics but also possesses strong discriminative and generalization abilities, laying a solid foundation for subsequent downstream tasks such as causal inference, cross-modal data fusion, and intelligent linkage modeling.

[0034] For the original feature vector Input a nonlinear transform network, a multilayer perceptron (MLP), to achieve high-dimensional feature mapping: in It is a trainable feature transformation function. For parameter set.

[0035] All features are ultimately encoded into high-dimensional semantic vectors. These semantic vectors serve as a deep semantic representation of device i at time t. They provide a solid data foundation for the heterogeneous linkage, cross-modal representation, and subsequent deep correlation modeling of electricity consumption and security data.

[0036] Step 3: Multi-source data semantic association embedding and adaptive fusion Leveraging the innovative multi-source semantic association embedding network (SAEN), this system automatically learns the potential correlations between electricity consumption data and security events. Through a dynamic embedding space, it measures the likelihood of interaction between different modalities and devices. The fusion process dynamically adjusts the weights of each source data, achieving adaptive feature recombination and high-dimensional information fusion for different scenarios. It outputs a global linkage feature vector, significantly improving the ability to identify weakly / implicitly correlated events.

[0037] In the multi-source data semantic association embedding and adaptive fusion stage, a multi-source semantic association embedding network (SAEN) is introduced to automatically discover the potential linkage between electricity consumption data and security events. Specifically, SAEN first uses the electricity consumption feature vector obtained in the previous step... Security feature vector Mapped to the same dynamic semantic embedding space ,in This represents the higher-order feature representation of the i-th electrical device at time t. This represents the feature representation of the j-th security device at the same time. SAEN uses a set of nonlinear embedding functions with shared parameters. This unifies the features of data from different modalities into a cross-modal vector representation, in the form of... , ,in For their respective network parameters.

[0038] Building upon this foundation, the SAEN network constructs an adaptive linkage correlation matching function to measure the potential association between electricity consumption and security features. First, SAEN receives multi-dimensional representation vectors of electricity consumption and security features at its input, generated through high-order feature extraction and local semantic modeling. To fully eliminate modal differences and uneven feature distribution, SAEN employs nonlinear mapping with embedding layers that are structurally similar but have independent or partially shared parameters for both types of features. Representation capabilities can be further enhanced through structures such as deep multilayer perceptrons, cross-modal normalization, or residual connections. Furthermore, the feature normalization strategy can be adaptively adjusted based on historical data statistics, ensuring that the final electricity consumption embedding and security embedding are comparable and geometrically proximate within the same semantic space. This embedding space not only maps individual device information but also contains multi-dimensional composite semantics such as device historical behavior, spatial attributes, and clustering relationships, laying a solid foundation for subsequent efficient linkage.

[0039] Specifically, for any pair of devices Define the correlation metric as ,in Represents the vector dot product. For learnable bias terms, For the Sigmoid activation function, such that The natural mapping is a correlation strength score between 0 and 1. This can be used to dynamically generate a time-device-modal three-dimensional correlation matrix. This reflects the potential interconnectedness among devices across the entire domain.

[0040] Furthermore, SAEN employs an attention mechanism for adaptive feature fusion of multi-source data. The SAEN network measures the association between each pair of devices through a parameterized correlation matching module. Its construction can be extended to be spatiotemporal context-aware, incorporating auxiliary information such as device category, physical distance, historical linkage behavior, spatial layout, and time period patterns into the correlation calculation. For example, the correlation between security points and high-power electrical devices physically close to each other dynamically increases over a specific time period. The generation of the correlation matrix not only reflects the potential "synergy" between individual pairs of devices in real time but also utilizes sparsity constraints, graph regularization terms, or knowledge graph priors to effectively mine "abnormal hotspot areas" and "high-risk linkage links," providing a data foundation for subsequent anomaly warnings and causal chain inference. At each time point, dynamic weights are assigned to the embedded features of electricity consumption and security. For example, through softmax normalization:

[0041] in These are trainable parameters. Ultimately, the globally linked feature vector is obtained through weighted aggregation:

[0042] The vector By fully integrating spatiotemporal-behavioral semantic information of all types of electricity and security devices in the community, the ability to capture and identify weakly correlated and implicitly correlated events (such as the potential linkage between abnormal electricity use and security warnings) can be significantly improved.

[0043] SAEN's self-evolving adaptability is reflected in its continuous model increment / online learning mechanism. The system can flexibly set a data sliding window, periodically / in real-time collect the latest electricity consumption and security linkage samples in the community, and dynamically fine-tune the embedding network and attention parameters through unsupervised representation learning, semi-supervised label guidance, or self-supervised loss functions, effectively adapting to data distribution drift caused by changes in residents' lives, additions or subtractions of equipment, and environmental disturbances. For newly emerging equipment types and patterns, the system can introduce advanced paradigms such as transfer learning and meta-learning to quickly expand embedding capabilities and shorten the adaptation cycle.

[0044] SAEN not only achieves effective alignment of multi-source heterogeneous data in a unified high-dimensional semantic space, but also possesses highly sensitive perception and flexible adaptive capabilities for various explicit / implicit linkages, providing a persistent and dynamically optimized semantic foundation and technical guarantee for the deep integration of smart community electricity consumption and security, early warning of abnormal risks, and intelligent response decision-making.

[0045] Step 4: Construct a causal reasoning graph Based on the linkage data between abnormal power consumption and security incidents, a causal reasoning graph is constructed. Nodes represent the characteristic states of different devices / events, and edges represent temporal sequences and influence relationships. Through graph structure reasoning algorithms, possible paths and potential causal chains of event occurrence are discovered.

[0046] In the causal reasoning graph construction stage, the global linkage feature vector output by the aforementioned SAEN model is used. Based on this, a spatial-temporal distribution and correlation model of abnormal power consumption and security incidents is performed. Specifically, the characteristic states of all electrical equipment, security equipment, and their potential key events at each moment are first abstracted into nodes in a graph. Formally, the graph... , where the set of nodes This represents the state vector of the k-th device or event at time t, typically or This refers to the aforementioned high-dimensional semantic features. The edge set E is used to express the temporal causal or influence relationships between different nodes. Common edge types include: state temporal transition edges of the same device. Connections between different devices / events ,in The time delay that reflects the potential causal event.

[0047] To automatically identify potential causal relationships between devices and events, a structural learning algorithm (attention-based graph neural network, GNN) is introduced. This algorithm performs structural learning on the spatiotemporal dependencies and sequential influences between device and event nodes. It not only dynamically updates node states through weighted aggregation of neighbor node features (incorporating mechanisms such as multi-order neighbors, multi-hop path aggregation, and multi-scale spatiotemporal windows), but also utilizes strategies such as trainable edge weights, node attribute gating, and spatial-modal multi-head attention to make the weight scores of each edge more granular and interpretable. For example, by monitoring changes in edge weights over time, it automatically identifies changes in the strength of causal coupling under different time, space, and behavioral patterns, enabling a sensitive response to periodic, sudden, and abnormal behaviors in community life.

[0048] Node state is updated through neighbor-weighted aggregation:

[0049] in, For the j-th node in delay The attention weight assigned to the k-th node reflects the causal strength of the event. This is a node aggregation function. If, after a power anomaly is detected, an anomaly is subsequently observed in an adjacent access control area, then a higher-level access control can be established. The system automatically learns its potential causal pathways. Simultaneously, by traversing possible directed paths in the graph, combining edge weights and node attributes, and applying graph reasoning algorithms (such as shortest path, maximum flow, and probabilistic propagation methods), it can efficiently discover causal chains and main propagation paths from a given initial event to a final event. For example, the weights of causal paths can be defined as follows: (P represents a certain path) to score and rank all candidate causal chains, and select the event chain with the most significant causal relationship to provide a scientific basis for event tracing, anomaly diagnosis and emergency response.

[0050] Ultimately, the constructed event causal reasoning graph not only clearly defines the complex spatiotemporal coupling between different devices and events in terms of structure, but also can continuously adapt to new community data streams through dynamic graph reasoning, thereby achieving interpretable modeling and real-time reasoning of the evolution mechanism of abnormal events.

[0051] By constructing a highly dynamic, semantically strong, and interpretable causal reasoning event graph, it not only accurately depicts the spatiotemporal coupling relationship between community electricity consumption and security data in terms of structure, but also assists in community risk early warning, emergency response and behavior governance with adaptive and intelligent reasoning capabilities, greatly improving the scientific nature, initiative and security resilience of smart community management.

[0052] Step 5: Self-feedback linkage decision-making and response Design a causal link self-feedback linkage decision algorithm (CCRSD) to implement the following process: whenever a linkage event is detected, automatically retrieve, infer, and predict subsequent risks and corresponding response actions. A self-feedback mechanism is adopted: the response effect (such as whether the alarm is effective, whether the power outage is timely) is fed back to the causal inference model, and the weights and decision parameters are adjusted in real time. Linkage responses may include: automatic alarm, notification of security personnel, cutting off part of the circuit, controlling the camera's direction, etc.

[0053] A Causal Chain-based Responsive Self-adaptive Decision (CCRSD) algorithm is introduced to achieve an intelligent, dynamic, and continuously optimized closed-loop anomaly response. Based on a causal reasoning graph and global semantic linkage features, this algorithm achieves collaborative prevention and dynamic evolution of electricity consumption and security incidents in smart communities through six stages: detection, reasoning, decision-making, response, feedback, and adaptive optimization.

[0054] In the self-feedback linkage decision-making and response phase, a causal chain-based responsive self-adaptive decision algorithm (CCRSD) is introduced to achieve intelligent, closed-loop anomaly linkage response. When the multi-source semantic association embedding network (SAEN) and the causal reasoning graph module jointly detect potential linkage events within the community (such as frequent abnormal power consumption accompanied by access control anomalies, camera obstruction, etc.), CCRSD automatically retrieves and activates the causal chain related to the event. Each node in the causal chain represents a key device or event state, and the edge weights quantify the influence and transmission probability between events. The system comprehensively analyzes the current node state, historical data, and chain structure to form a targeted risk state sequence.

[0055] Specifically, when SAEN and the causal inference graph module detect potential linkage events—that is, when a node (such as abnormal electricity consumption) is found to have a high-confidence association with certain security events (such as abnormal access control or abnormal camera) after path inference and weight scoring in the causal inference graph—the CCRSD algorithm first automatically retrieves and activates the corresponding causal chain. Each node in the chain Its edge weights represent the event sequence and transmission strength.

[0056] Based on the current link structure and causal weights Combined global linkage features Based on the node time sequence state, a risk scoring function is used.

[0057] Quantitative predictions of potential risks are made, including Let be the semantic anomaly strength of the k-th node at time t. If the risk score exceeds a preset threshold... Generate a sequence of linked response actions in real time Each response action Automatic matching based on event type and scenario, such as automatic alarm ( ), notify security ( ), power outage ( (e.g., camera pre-positioning).

[0058] The core of CCRSD lies in its self-feedback learning mechanism: after each linked response, it collects external feedback signals. Examples include whether alarms are handled promptly (1 / 0), the delay in power outage operations, and the response time of security personnel. Feedback signals serve as reward or punishment signals, guiding causal reasoning and adaptive adjustments to decision parameters. Specifically, a loss function is defined.

[0059] in , For weight hyperparameters, To correct the target causal weights based on the latest feedback, optimization strategies such as gradient descent are used to adjust the edge weights in the causal inference graph and the decision parameters in the linkage response strategy in real time, ensuring that the model can continuously evolve in response to new risk patterns and feedback results.

[0060] Feedback signals are recorded as reward / penalty factors in the system, enabling real-time adjustments to edge weights, node state awareness parameters, and response decision strategies in the causal inference graph. The system defines a loss function to quantify the gap between the current response and the ideal target. By combining gradient descent, reinforcement learning, or other adaptive optimization methods, the system continuously adjusts the inference graph weights and linkage decision parameters, allowing the model to self-evolve in response to new risks and feedback results. Over long periods of operation, the system can autonomously learn new abnormal linkage patterns within the community, shortening response time and improving the ability to correct false positives and false negatives. Ultimately, CCRSD achieves a closed-loop process encompassing "detection-reasoning-decision-response-feedback-adaptation." Each response and its effect are used to strengthen or correct the causal reasoning model, continuously improving the accuracy, timeliness, and relevance of coordinated responses, and significantly enhancing the community's overall intelligent security coordination capabilities and risk prevention and control level.

[0061] Example: Smart Community Abnormal Electricity Consumption Security Monitoring and Closed-Loop Response System Based on Data Linkage I. Community Scenarios and Equipment Deployment A smart community, Area A, comprises two residential buildings, each 10 stories high, with 4 households per floor, totaling 80 households. To achieve integrated monitoring of electricity usage and security, the following data acquisition terminals are deployed: Smart meters: 80 sets (1 set per household) Electrical equipment monitoring modules (including high-power home appliance monitoring): 160 sets (2 sets per household, one for air conditioner and one for water heater). Access control system: 4 sets (2 sets per building, distributed at the main entrance and underground passage) Video surveillance cameras: 20 units (covering all corridors, elevator lobbies, and main entrances) Alarm sensors: 10 units (for key areas such as smoke, gas leaks, and intrusion detection) II. Data Acquisition and Preprocessing Original data sample

[0062] Data preprocessing workflow 1. Time Alignment: All device data is aligned to a unified time based on the community gateway server, using a 5-minute collection window.

[0063] 2. Missing value imputation: If the monitoring module is missing data due to a fault, forward imputation is used to complete the data.

[0064] 3. Abnormal noise filtering: Records of power surges, inconsistencies between equipment status and behavior, etc., are automatically filtered out using median filtering and set rules.

[0065] III. Feature Extraction and Local Semantic Modeling For each type of device, the following feature modeling is used:

[0066] IV. Semantic Association and Adaptive Fusion of Multi-Source Data SAEN (Multi-Source Semantic Association Embedding Network) was used to model the associations of multimodal data. During the period from 13:00 to 13:10 on a certain day, "Household 0301" exhibited the following linked characteristics:

[0067] Fusion Analysis Explanation Feature vector fusion: Based on the transformative high-sensitivity features, the system assigns higher weights to "power surge + water heater on + smoke alarm + no one entering or exiting the access control system" and dynamically synthesizes a global event feature vector.

[0068] Hidden linkage detection: No personnel entered or exited, but the power suddenly increased and triggered the smoke alarm, which was initially determined to be a high-risk smoke alarm event caused by electricity use.

[0069] V. Construction of Causal Relationship Reasoning Graph Using the aforementioned linked data, a causal reasoning graph is constructed, where nodes represent device characteristic states and edges represent relationships across time / devices: [Water heater starts (13:05)] --(Sudden power surge)-->[Abnormal power meter reading] [Smoke Generation (13:06)] --(Camera Detection)-->[Abnormal Video Summary] [Alarm triggered (13:07)] [No entry or exit through access control (13:05-13:08)] The diagram clearly illustrates the following: abnormal start-up of the water heater → sudden increase in power → smoke generation → smoke alarm → failure of personnel to evacuate in time, which may be a fire hazard caused by an electrical accident.

[0070] VI. Self-feedback linkage decision-making and response process Intelligent closed-loop processing 1. High-risk event identification: The system determines that the abnormal start-up of the water heater causes abnormal smoke, and synthesizes a linkage event, with a risk level of high.

[0071] 2. Linked response sequence: Automatically executes three steps: "Power outage (household 0301) → synchronized early warning to residents and property management (SMS / APP push) → dispatching security personnel for on-site inspection".

[0072] 3. External feedback collection: After the security personnel handled the situation, they reported through the APP that "the site has been investigated and it has been confirmed that the aging of the water heater caused a local arc short circuit, and there were no casualties."

[0073] 4. Causal weight correction: Based on feedback, the system lowers the weight of the water heater-power surge-smoke alarm path for this household and optimizes the subsequent response trigger threshold.

[0074] 5. Continuous self-learning optimization: If similar events occur in the future, the system will be able to identify and respond more quickly, greatly improving the accuracy and efficiency of event handling.

[0075] VII. Results and Effects Analysis

[0076] This embodiment combines automatic collection of multi-source power consumption and security data, semantic feature extraction, adaptive linkage fusion, causal reasoning, and self-feedback learning to achieve intelligent linkage monitoring and efficient response to abnormal power consumption and security incidents in the community, greatly improving community safety and incident handling efficiency.

[0077] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0078] The various embodiments in this disclosure are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for data linkage processing of power security in smart communities, characterized in that, The method includes: Multi-source power consumption and security data collection and preprocessing: Collect various power consumption data and security equipment data in the community, and perform time alignment, missing value imputation and abnormal noise filtering on the raw data; Feature extraction and local semantic modeling: Based on a custom local semantic feature discovery module, temporal, spatial and behavioral features are extracted from electricity consumption and security data respectively; transformative feature modeling is adopted to automatically summarize high-sensitivity features for each type of device. Multi-source data semantic association embedding and adaptive fusion utilize the multi-source semantic association embedding network SAEN to automatically learn the potential association between electricity consumption data and security events; through dynamic embedding space, it measures the possibility of linkage between different modalities and different devices; the fusion process dynamically adjusts the weights of each source data to achieve adaptive feature recombination and high-dimensional information fusion for different scenarios; and outputs a global linkage feature vector, which significantly improves the ability to identify weakly / implicitly related events. Construct a causal reasoning graph. Based on the linkage data between abnormal power consumption and security incidents, construct an event causal reasoning graph. Nodes represent the characteristic states of different devices / events, and edges represent the temporal sequence and influence relationships. Through graph structure reasoning algorithms, discover the path and potential causal chain of event occurrence. Self-feedback linkage decision-making and response: Design the causal link self-feedback linkage decision-making algorithm CCRSD. Whenever a linkage event is detected, it automatically retrieves, infers and predicts subsequent risks and corresponding response actions. The aforementioned multi-source data semantic association embedding and adaptive fusion includes: introducing a multi-source semantic association embedding network to perform cross-modal unified embedding of high-order feature representations of electrical equipment and security equipment; and dynamically measuring the potential linkage relationships between devices in different times and modalities by constructing an adaptive correlation matching mechanism. Furthermore, by employing an attention mechanism to assign weights to the globally embedded features, the adaptive fusion of multi-source features is achieved, generating a global linkage feature representation that comprehensively reflects the overall electricity consumption and security status of the community. This network can continuously adapt to changes in the environment and behavioral patterns through online updates, thereby enhancing the identification of abnormal linkage behaviors and data support capabilities. The self-feedback linkage decision and response includes: based on the causal reasoning graph and global linkage features, automatically identifying and activating the corresponding causal links for detected high-risk linkage events, dynamically assessing the potential risk level by combining the link structure, causal weight and the status of each node, and generating an automated linkage response action sequence accordingly. It includes early warning, notification, power outage and security control, and collects external feedback signals after the response. Based on the feedback, it adaptively corrects the causal weights and decision parameters. Through a continuous self-feedback learning mechanism, it continuously optimizes the accuracy and response efficiency of the system's abnormal handling, and realizes intelligent closed-loop linkage prevention and control.

2. The smart community electricity security data linkage processing method according to claim 1, characterized in that, The aforementioned multi-source power consumption and security data acquisition and preprocessing includes: By deploying smart meters, electrical equipment monitoring modules, access control systems, cameras, and alarm sensors, multi-source acquisition terminals are used to collect raw data from various electrical and security devices in real time, covering power status, device actions, access control records, video summaries, and alarm log information. The collected asynchronous data is aligned based on a unified time base, and missing values ​​are filled in using interpolation, forward or backward padding. Statistical analysis and rule filtering are combined to remove abnormal noise. Finally, all data is converted into a standardized encoding format and unified semantic labels to form a high-quality basic dataset that meets the requirements of subsequent multi-source data fusion.

3. The smart community electricity security data linkage processing method according to claim 1, characterized in that, The feature extraction and local semantic modeling include: extracting local statistical features, spatial neighborhood features and behavioral event features from the collected time-series data of electricity consumption and security equipment, respectively. The multidimensional original features are then input into a trainable feature transformation network to achieve nonlinear high-dimensional mapping of the original data, ultimately obtaining a unified semantic feature vector for each device at each time step. This vector is then used for subsequent fusion analysis of multi-source data and linkage recognition and modeling of complex events.

4. The smart community electricity security data linkage processing method according to claim 1, characterized in that, The construction of the causal relationship reasoning graph includes: based on the global linkage features obtained after multi-source data fusion, performing node modeling on various states and key events of electrical equipment and security equipment at different times; By abstracting the temporal sequence and correlation between devices and events through graph structures, the structural learning algorithm is used to dynamically determine the causal influence strength between nodes, and key causal chains and event propagation paths are automatically discovered through graph reasoning methods. This enables interpretable modeling and real-time reasoning of the causal mechanisms of complex events, thereby providing a scientific basis for anomaly tracing, diagnosis and intelligent response.

Images