A construction site power distribution equipment remote monitoring method and system based on big data

Abstract

The application discloses a kind of based on big data's construction site power distribution equipment remote monitoring method and system, it is related to remote monitoring technical field.The steps of the method include: obtaining construction site power distribution equipment data and carrying out semantic mapping, generates power distribution equipment behavior semantic segment;Based on the behavior semantic segment and the power transmission relationship between equipment, establish cross-level behavior association chain, and form power distribution behavior evolution track;Based on graph neural network, build trend anomaly identification model, input the behavior evolution track into the model, output trend anomaly identification result;Based on the identification result, generate different intervention scheme.Different intervention scheme is realized based on big data to the remote monitoring of construction site power distribution equipment.

Description

Technical Field

[0001] This invention relates to the field of remote monitoring technology, specifically to a method and system for remote monitoring of construction site power distribution equipment based on big data. Background Technology

[0002] Temporary power supply systems at construction sites typically employ a three-tiered distribution system consisting of a main distribution box, sub-distribution boxes, and switch boxes. This system is characterized by complex load configurations, variable operating environments, and frequent equipment start-ups and shutdowns. To ensure construction safety and power continuity, effective remote status monitoring and fault early warning systems are crucial. However, construction site power distribution is not an isolated collection of equipment but a dynamic, interconnected network based on clearly defined power transmission relationships. An anomaly at any node can propagate along the power supply path, triggering cascading risks.

[0003] Traditional remote monitoring methods primarily rely on monitoring the independent parameters of individual power distribution equipment. This method involves installing sensors to collect key electrical parameters such as voltage, current, power, and temperature of each device in real time, and then uploading this data to a remote monitoring center. The monitoring center presets static safety thresholds for various parameters. When the monitored data of any device continuously exceeds its corresponding threshold, an alarm is triggered, indicating that the specific device may have problems such as overload, overheating, or insulation abnormalities. This method achieves real-time perception of the individual status of power distribution equipment and over-limit alarms.

[0004] However, the aforementioned traditional methods only focus on the instantaneous parameter status of individual equipment, and cannot perceive and analyze the chain behavior process that is triggered by actual operational activities such as equipment start-up and shutdown, load switching, etc., and propagates across levels within the three-level power distribution system. As a result, the opportunity to achieve early warning by identifying its weak and gradual cross-level transmission patterns before the abnormal chain reaction reaches the single-point threshold is lost. This leads to the fact that this traditional monitoring based on isolated threshold judgment can only passively respond to obvious faults that have already developed. When dealing with complex system risks that are latent and propagating, its timeliness and foresight in early warning are fundamentally limited. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for remote monitoring of construction site power distribution equipment based on big data, thus solving the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for remote monitoring of construction site power distribution equipment based on big data, comprising the following steps: Step S1: Obtain data on the construction site power distribution equipment, perform semantic mapping on the data, and generate semantic fragments of power distribution equipment behavior; Step S2: Based on the semantic fragments of the power distribution equipment behavior and the power transmission relationship between the power distribution equipment on the construction site, establish a cross-level behavior association chain and form a power distribution behavior evolution trajectory; Step S3: Based on the graph neural network, construct a trend anomaly identification model, input the power distribution behavior evolution trajectory into the trend anomaly identification model, and output the trend anomaly identification result; Step S4: Based on the trend anomaly identification results, generate a differentiated intervention plan.

[0007] Preferably, acquiring data from construction site power distribution equipment includes: The system collects monitoring data from all equipment within the three-tiered power distribution system of the temporary power supply system at the construction site. This three-tiered power distribution system consists of a main distribution box, distribution boxes, and switch boxes arranged from top to bottom. The main distribution box serves as the main distribution node for the incoming power supply, supplying power to multiple distribution boxes. Each distribution box supplies power to multiple switch boxes within a specific construction area. Each switch box directly controls one or a group of terminal electrical devices. The collected data on the construction site's power distribution equipment includes monitoring data synchronously collected from these three levels of equipment, forming a complete time-series data set.

[0008] Preferably, semantic mapping of the construction site power distribution equipment data to generate semantic fragments of power distribution equipment behavior includes: Perform semantic mapping operations on the time series data set D: For the main distribution box, based on the monitoring data of the main distribution box, features reflecting the power source scheduling and overall distribution logic are extracted, and the behavior is judged according to quantitative standards; For distribution boxes, based on the monitoring data of the distribution boxes, features reflecting the regional load switching, coordination and shock response logic are extracted, and the behavior is judged according to quantitative standards; For switch boxes, based on the monitoring data of the switch boxes, features reflecting the direct operation and protection precursors of terminal electrical equipment are extracted, and the behavior is determined according to quantitative standards. Time series data set Mapped to a set of structured semantic fragments of power distribution equipment behavior .

[0009] Preferably, the power supply transmission relationship between the construction site's electrical distribution equipment includes: The power supply transmission relationships between all power distribution equipment are formally predefined to form a power supply path topology diagram, which is represented by an ordered pair: ; in, This represents the set of all nodes in the graph, and this set is related to the set of device identifiers in step S1. The same, that is, the topology of each monitored power supply path. A node in the graph; E represents the set of all directed edges in the graph, used to describe direct power supply relationships.

[0010] Preferably, establishing cross-level behavioral association chains includes: This process iterates through the collection. The behavioral semantic fragments in the data, and based on spatial and temporal association rules, from the set of behavioral semantic fragments of power distribution equipment. Multiple cross-level behavioral association chains are extracted and linked together to form a new set. Each link It is an ordered list that records a series of semantic fragments of behavior that are triggered by an initial event and propagated upstream along the power supply path.

[0011] Preferably, the formation of the power distribution behavior evolution trajectory includes: For a specified analysis period , and These represent the start and end times of the analysis period, and the evolution trajectory of power distribution behavior during that period. Build it in the following way: Filter out the set All time windows fall entirely within the relevant links of that time period; then, based on the start time of the initial behavior segment in each relevant link. Sort in ascending order

[0012] Preferably, the trend anomaly identification model based on graph neural networks includes: Trend Anomaly Identification Model It is a spatiotemporal graph neural network whose input is the evolution trajectory of power distribution behavior. The output is a quantitative identification result of potential abnormal trends, i.e., anomaly identification result. The model includes four core modules in sequence: data transformation, spatial graph convolution, time series modeling, and output calculation. These modules process the spatiotemporal characteristics of the power distribution behavior evolution trajectory in a serial manner.

[0013] Preferably, the trend anomaly identification result is obtained by inputting the power distribution behavior evolution trajectory into the trend anomaly identification model and outputting the trend anomaly identification result. The evolution trajectory T of the power distribution behavior to be analyzed is input into the data conversion module of the trend anomaly identification model and processed into a standard spatiotemporal graph sequence. Next, the sequence undergoes forward propagation through the spatial graph convolution module and the temporal series modeling module, sequentially completing the spatial correlation and temporal evolution analysis of behavioral features to obtain the final spatiotemporal fusion representation h of each node. And the attention weights in the process; finally, the output calculation module is based on h Calculate the anomaly score for each node. It also analyzes potential risk propagation paths by combining attention weights, thereby outputting structured trend anomaly identification results.

[0014] Preferably, based on the trend anomaly identification results, generating a differentiated intervention plan includes: Based on the node anomaly score set and the risk propagation path set, the intervention level and unique identifier of the power distribution equipment that needs intervention are determined; Intervention levels obtained from mapping And the specific device type indicated by device identifier O, retrieve and instantiate specific control instructions from a predefined, hierarchical intervention strategy knowledge base; The final differentiated intervention plan is a structured list of instructions, where each entry is a quadruple:

[0015] in, The target device identifier of the instruction is determined by the abnormal score list output in step S3 and the device node identifier determined in the risk path. The specific control command code is obtained by mapping from the predefined strategy library in step S4 according to the intervention level and device type of the target device; Params represents the command parameters, which are preset values ​​in the strategy library based on the trade-off between safety and efficiency; Priority represents the command execution priority, which is determined based on the principle of safety priority and the operation dependency of the power supply path, and is used to coordinate the execution sequence of multiple commands.

[0016] A remote monitoring system for construction site power distribution equipment based on big data includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0017] This invention provides a remote monitoring method for construction site power distribution equipment based on big data, involving machine learning and deep learning technologies, which has the following beneficial effects: (1) By adopting a graph neural network as the core model for trend anomaly identification, the traditional method overcomes the limitation of only being able to make independent threshold judgments on single-point parameters. This model can take the power distribution behavior evolution trajectory formed in step S2, which contains complex topological relationships, as the overall input and automatically learn the normal and abnormal behavior propagation patterns. This enables the method to identify the abnormal propagation trend and issue an early warning when the abnormal current, voltage disturbance, and other characteristics are still propagating and spreading between power distribution network levels and have not reached any single-point alarm threshold, thereby achieving a fundamental improvement from isolated point alarms to system-level, forward-looking risk warnings.

[0018] (2) By specifically introducing a spatial graph convolution module and a time series modeling module into the trend anomaly identification model, a unified spatiotemporal analysis framework is formed. The spatial graph convolution module explicitly utilizes the power supply topology, enabling the model to accurately model the spatial diffusion process of abnormal electrical characteristics along the power supply path; the time series modeling module can deeply capture the temporal accumulation and evolution of the behavioral characteristics of each node. The combination of the two enables the model not only to determine whether there is an anomaly, but also to deeply understand how anomalies spread in space and how they evolve in time, thereby greatly improving the accuracy of identifying risks in complex and progressive systems and the advance warning.

[0019] (3) By connecting and organizing the discrete semantic fragments of power distribution equipment behavior generated in step S1 into a continuous power distribution behavior evolution trajectory based on power transmission relationship and spatiotemporal association rules, this method elevates the underlying object of monitoring and analysis from a single data reading to a system behavior story that can completely describe when and where the behavior occurs and how it is linked across levels. This provides a unified and coherent analysis object with both temporal dynamics and spatial topology for subsequent intelligent analysis, fundamentally solving the key problem that traditional monitoring cannot understand the overall behavior logic and risk evolution process of the system due to isolated data and lack of semantic association. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for remote monitoring of construction site power distribution equipment based on big data, as proposed in this invention.

[0021] Figure 2 The present invention proposes a method for remote monitoring of construction site power distribution equipment based on big data to obtain a hierarchical diagram of the evolution trajectory of power distribution behavior.

[0022] Figure 3 This is a hierarchical diagram of the differentiated intervention scheme obtained in the remote monitoring method for construction site power distribution equipment based on big data proposed in this invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0024] Please see Figures 1-3 This invention provides a technical solution: a method for remote monitoring of construction site power distribution equipment based on big data. Specifically, it provides the following plant extraction method based on image processing; please refer to [link / reference]. Figure 1 The method includes the following steps: Step S1: Obtain the construction site power distribution equipment data, perform semantic mapping on the construction site power distribution equipment data, and generate semantic fragments of power distribution equipment behavior.

[0025] First, monitoring data was collected covering all equipment within the three-tiered power distribution system of the construction site's temporary power supply. This three-tiered system consists of a main distribution box, sub-distribution boxes, and switch boxes, arranged from top to bottom. The main distribution box serves as the central distribution node for incoming power, supplying power to multiple sub-distribution boxes; each sub-distribution box supplies power to multiple switch boxes within a specific construction area; and each switch box directly controls one or a group of terminal electrical devices. The collected data on the construction site's power distribution equipment, including monitoring data synchronously collected from these three levels of equipment, forms a complete time-series data set, denoted as [data missing]. This collection contains each power distribution device. At continuous sampling time Monitoring data vector ,Right now: ; in, This represents the set of identifiers for all monitored power distribution equipment. For discretized time indexing; This represents the total length of the data acquisition time window. The purpose of this formula is to formally represent the raw monitoring dataset, encompassing multiple types of parameters, synchronously acquired from the three-level power distribution system at the construction site.

[0026] It should be noted that the identifier set May include, for example Represents the No. 1 main distribution box, Represents belonging to A C3 represents elements such as switch box No. 3 belonging to B2, and its identifier itself encodes the hierarchy and affiliation of the equipment; Monitoring data vector Includes equipment At any moment The following specific monitoring data: three-phase phase voltage Three-phase phase current Total active power Total reactive power Power factor Zero-sequence current used for leakage current monitoring Equipment point temperature and the on / off state of the switch. .

[0027] Then, a semantic mapping operation is performed on the time-series data set D. The purpose of this operation is to transform discrete, context-free instantaneous parameter readings into behavioral description units with clear operational semantics and well-defined evaluation criteria, based on the actual functional logic of equipment at various levels in a construction site temporary power supply scenario. The mapping process follows differentiated semantic extraction and judgment rules for different levels of power distribution equipment: For the main distribution box, based on the monitoring data of the main distribution box, features reflecting the power source scheduling and overall distribution logic are extracted, and the behavior is determined according to quantitative standards: Calculate the current imbalance of each phase based on the bus load distribution characteristics. The evaluation criteria are: if If the current imbalance persists for more than 15% for 10 seconds, it is determined that a three-phase imbalance of the bus load has occurred. The current imbalance threshold of 15% and the duration threshold of 10 seconds are set according to the allowable values ​​and common monitoring intervals for short-term imbalance in low-voltage systems in the national standard GB / T15543-2008 "Power Quality Three-Phase Voltage Imbalance".

[0028] For main circuit switching characteristics: monitor the status of the main incoming line switch. Status of standby incoming line switch The change event. The evaluation criteria are: if From 1 to 0 and If the value changes from 0 to 1 within 500 milliseconds, it is determined that a main / backup power supply circuit switching behavior has occurred. The 500-millisecond time window is set according to the typical operating time value of commonly used dual-power automatic transfer switches.

[0029] To assess the active power supply characteristics of branch lines, the active power of each output branch is statistically analyzed. Exceeding its rated power within a continuous time window The cumulative time percentage. The evaluation criteria are: if If the cumulative percentage of the load in a given state exceeds 30% within a continuous 5-minute observation window, the branch is considered to be experiencing active overload behavior. The overload coefficient of 1.2, the 5-minute observation window, and the percentage threshold of 30% were determined based on statistical analysis of the fluctuation characteristics of temporary power loads at construction sites, and are used to identify abnormal branches that are in a state of high load rate for a long period of time.

[0030] For distribution boxes, based on monitoring data from the distribution boxes, features reflecting the regional load switching, coordination, and impact response logic are extracted, and behavior is determined according to quantitative standards: Based on the branch circuit connection sequence characteristics, identify the status of each branch circuit switch. Calculate the transition time from 0 to 1 and the time difference between any two branch connections. The evaluation criteria are: if two or more branch switches are detected closing successively within 60 seconds, and the time difference is... If both intervals are less than 2 seconds, a sequential power-on behavior is determined to have occurred. The 60-second observation window and 2-second interval threshold are based on observations of typical operational rhythms of multiple devices being started collaboratively by construction work teams. For the regional load combination switching characteristics, monitor the total current. The step change is judged by the following criteria: if If the increase exceeds 50% within 1 second, and at least two branch switches are simultaneously detected to change to state 1, then a combined load startup event is determined to have occurred. Record the steady-state value of the total current during combined startup. With power factor A threshold of 50% increase within 1 second is used to distinguish between the startup of large combined loads and the switching of small devices; For the response characteristics of impact loads, short-time current surges are captured. The evaluation criterion is: if the total current increases sharply within 0.5 seconds, the peak value... Reaching the steady-state value before the impact It was more than 3.5 times higher, and then dropped back to 1.5 times within 2 seconds. The following criteria determine whether an impact load response has occurred. The thresholds of 3.5 times, 0.5 seconds, 1.5 times, and 2 seconds are derived from the measured current waveform characteristics of typical impact loads such as tower cranes and welding machines.

[0031] For switch boxes, based on monitoring data from the switch boxes, features reflecting the direct operation and protection precursors of terminal electrical equipment are extracted, and the behavior is determined according to quantitative standards: For the transient characteristics of motor startup, the starting current waveform is captured. The evaluation criterion is: if the current in any phase jumps from near zero to exceeding the rated current within 0.2 seconds. If the starting current is 5 times the rated value and gradually stabilizes near the rated value within the following 2 to 10 seconds, then the motor is considered to have undergone full-voltage starting behavior. Record the starting current multiple. and transition time A starting current multiple of 5 and a rise time threshold of 0.2 seconds conform to the typical electrical characteristics of direct starting of an asynchronous motor.

[0032] For the characteristics of leakage disturbance, continuously monitor the zero-sequence current. The evaluation criteria are: if The effective value of the 1-minute moving average shows a monotonically increasing trend over 5 consecutive minutes, with a cumulative increase exceeding 30mA, or its minute-level variance. If the current continues to increase, it is determined that there is a gradual change in leakage current. The cumulative increase threshold of 30mA is based on the typical rated non-operating current of the residual current device, and a 5-minute window is used to identify the gradual change process.

[0033] For the shutdown release characteristic, record the current flow after the equipment stops. The evaluation criterion is: if the switching state... If, after the circuit transitions from 1 to 0, the current decays to below 0.5A within 3 seconds, a circuit shutdown release is considered to have occurred. The 3-second time threshold and the 0.5A current threshold are used to confirm that the load has been completely disconnected.

[0034] By applying the rules described above, which are based on specific monitoring data and quantitative evaluation criteria, the time series data set... Mapped to a set of structured semantic fragments of power distribution equipment behavior Each behavioral semantic fragment It is a tuple, which can be formally represented as: ; in, A unique identifier representing the power distribution equipment that generated this behavior fragment. ; This indicates the start and end time window of the observed behavior, identified according to the above evaluation criteria; Indicates the semantic type of the behavior of this segment; It is a feature vector used to quantify the specific characteristics of this behavioral segment, and its dimension is related to the specific... Correspondingly. The purpose of this formula is to define a standardized data structure for semantic fragments of power distribution equipment behavior.

[0035] It should be noted that behavioral semantic types Its value comes from a predefined dictionary that corresponds to the electricity consumption scenario on construction sites and contains the aforementioned explicit evaluation criteria. ,For example Includes {busbar negative three-phase imbalance, main and backup power supply circuit switching, branch overload activity, branch sequential power-on, area combined load start-up, impact load response, motor full-voltage start-up, leakage current gradual change, circuit shutdown release, ...}.

[0036] This step transforms and elevates the monitored underlying objects from simple parameter readings into standardized behavioral semantic fragments that carry contextual information about the construction site operations and have clear, quantifiable definitions. This processing provides a unified and unambiguous semantic input foundation for subsequent analysis of behavioral relationships between different levels.

[0037] Step S2: Based on the semantic fragments of the power distribution equipment behavior and the power transmission relationship between the power distribution equipment on the construction site, establish a cross-level behavior association chain and form a power distribution behavior evolution trajectory.

[0038] Inherited from the set of semantic fragments of power distribution equipment behavior obtained in step S1 This step aims to associate and integrate discrete behavioral semantic fragments, which are isolated units of equipment, based on the inherent power transmission path of the three-level power distribution system on the construction site, so as to construct a coherent view that can depict the cross-level propagation and time evolution of behavior in the power distribution network.

[0039] First, the power supply transmission relationships between all power distribution equipment are formally predefined, forming a power supply path topology diagram. This topology is represented by an ordered pair: ; in, This represents the set of all nodes in the graph, and this set is related to the set of device identifiers in step S1. The same, that is, the topology of each monitored power supply path. In the graph, E represents a node; E represents the set of all directed edges in the graph, used to describe direct power supply relationships. For any two distinct device nodes... and If device node The electrical energy required for operation comes directly from the device nodes. That is, in the three-level power distribution system of the construction site, yes The upstream distribution box has a line from point to A directed edge is denoted as The purpose of this formula is to abstract the physical connections between the main distribution box, the sub-distribution boxes, and the switch boxes in the construction site power distribution system into a clear graph data structure, providing a spatial topological basis for subsequent behavioral associations based on the graph structure.

[0040] Next, based on the power supply path topology diagram and a collection of semantic fragments of the behavior of power distribution equipment Establish cross-level behavioral association chains. This process iterates through the set. The behavioral semantic fragments are chained together based on spatial and temporal association rules: For spatial association rules, the device nodes to which multiple behavioral semantic fragments to be concatenated must be in the graph. The components are connected by a directed path. This path reflects the likelihood of abnormal behavior or events propagating along the actual power supply path.

[0041] For time-related rules, the behaviors of different nodes belonging to the same power supply path are considered as segments, and their time windows are defined as follows: Temporal proximity should be satisfied to reflect the causal transmission of behaviors. A maximum permissible association interval should be set. In a specific embodiment, this interval can be set to This value is determined based on the electromagnetic transient processes and typical response times of protection devices in the construction site's power distribution system. For two directly adjacent equipment nodes on the path... and ,like The time window of the segment on the device is ,but The time window of the segment associated with it on the device. Must meet Or the two windows may overlap. This rule aims to capture the causal timing relationship between a lower-level device's action and the response triggered by its direct parent device within a reasonable delay.

[0042] Applying the two rules mentioned above, from the set of semantic fragments of power distribution equipment behavior Multiple cross-level behavioral association chains are extracted and linked together. The specific algorithm flow is as follows:

[0043] First, initialization is performed by creating an empty set of association chains C and assigning an unused state marker to each behavioral semantic fragment in set F. Then, set F is traversed sequentially. If the behavioral semantic type of a fragment belongs to a predefined set of initial event types (e.g., full-voltage start of a motor in a switch box) and its state is unused, a new association chain is created starting from this fragment and marked as used. Next, the iterative chaining phase begins. Let the end fragment of the currently constructed association chain be the current fragment. Candidate subsequent fragments in set F that satisfy the following conditions are searched: their corresponding device is the direct superior power supply node of the current fragment's device in the power supply path topology G (this is the spatial association rule); the interval between the start time of their time window and the end time of the current fragment's time window is less than the preset maximum allowable association interval δt (this is the temporal association rule); and their state is unused. During linking, if a unique candidate segment that meets the criteria is found, it is linked to the end of the current association chain, its state is updated, and the search continues upstream. If multiple candidate segments are found, a pre-defined conflict resolution strategy is used, such as selecting the segment with the highest overlap in the time window for linking. If no candidate segments are found, the current association chain is complete. Finally, the completed association chain is added to set C, and the traversal steps are returned until all segments in set F have been processed, forming a new set. Each link It is an ordered list that records a series of semantic fragments of behavior that are triggered by an initial event and propagated upstream along the power supply path. For example, a chain might be represented as... ,in It is a switch chain The full-voltage start-up segment of the motor, It is a distribution box The subsequent impact load response segment, It is the main distribution box This describes a segment of three-phase unbalanced bus load. This chain fully describes the hierarchical propagation and impact of a specific operational event in the distribution network.

[0044] Finally, based on the relationship set This forms the evolutionary trajectory of power distribution behavior. This trajectory is a macroscopic sequence of system behavior over time. For a given analysis period... , and These represent the start and end times of the analysis period, and the evolution trajectory of power distribution behavior during that period. It is constructed in the following way: First, the set is selected. All time windows fall entirely within the relevant links of that time period; then, based on the starting behavior segment in each relevant link, that is, the start time of the earliest segment representing the lowest-level device in the chain. Sort in ascending order.

[0045] ; in, Indicates the connection link The total time span covered is the union of the time windows of all segments in the related chain; operations Indicates the start time of the initial segment. Sort in ascending order. The purpose of this formula is to organize discrete, event-triggered connections into a global behavioral evolution sequence arranged chronologically, thereby providing a structured, temporal view that can be used to analyze how the macroscopic state of the system changes over time.

[0046] It should be noted that the relevant links Depend on It consists of several behavioral segments, each with a time window of 1. , The mathematical expression is: ; in, It is a minimum value function. It is used here to find all... The earliest start time of each behavioral segment is used as the starting point of the total time span. ; It's a maximum value function. Here it's used to find all... The latest end time among the behavioral segments As the end point of the total time span .

[0047] This step shifts the focus of analysis from the isolated behavioral semantics of individual devices to the collaborative behavioral evolution across devices and levels. The constructed power distribution behavior evolution trajectory... It organically integrates the spatial propagation path and temporal progression information of behavior, providing a core analysis object with both topological structure and temporal dynamics for identifying potential and progressive abnormal risk trends in subsequent steps.

[0048] Step S3: Based on the graph neural network, construct a trend anomaly identification model, input the power distribution behavior evolution trajectory into the trend anomaly identification model, and output the trend anomaly identification result.

[0049] The power distribution behavior evolution trajectory generated in step S2 This step aims to construct and apply a deep analysis model based on graph neural networks, namely a trend anomaly identification model. This model can automatically learn dynamic patterns under normal operating conditions from trajectory data representing the spatiotemporal evolution of behavior, and thereby identify potential abnormal development trends, thus achieving a leap from describing phenomena to assessing risks.

[0050] Trend Anomaly Identification Model It is a spatiotemporal graph neural network whose input is the evolution trajectory of power distribution behavior. The output is a quantitative identification result of potential abnormal trends, i.e., the trend anomaly identification result. The model consists of four core modules in sequence: data transformation, spatial graph convolution, time series modeling, and output calculation, which process the spatiotemporal characteristics of the power distribution behavior evolution trajectory in a serial manner.

[0051] For the data transformation module, this module will convert the behavioral evolution trajectory. Convert the sequence into a spatiotemporal graph sequence that the model can process. Divide the continuous time into... A length of Continuous time slices, the first Each slice is denoted as For each time slice According to the power supply relationship diagram defined in step S2 and fall Construct a spatiotemporal graph snapshot from the behavioral semantic fragments within. : ; in, Represents a fixed set of nodes. Sum of edges The power supply topology diagram is as follows; It is a time slice The corresponding node feature matrix. The OK It is a device node exist The eigenvectors within. Through the Internal occurrence in equipment All behavioral semantic fragments on The aggregation yields: First, for each fragment The original feature vector and its behavioral semantic types Perform concatenation; then, average pooling is applied along the time dimension to all weighted concatenation results, and finally mapped through a trainable fully connected layer. dimensional vector If the equipment In the absence of behavioral fragments, then The zero vector is the vector. The purpose of this formula is to normalize non-uniform, event-driven behavioral trajectories into a series within a fixed topological graph. Dynamic graph sequence of node features evolving over time steps This provides standard input for spatiotemporal modeling.

[0052] For the spatial graph convolutional module, this module aggregates node information and its neighbors in the power supply topology within each time slice using a graph neural network to model the spatial propagation of behavioral features. A graph attention network layer is employed, with its l-th layer focusing on nodes... In the The operation at each time step is represented as follows: ; in, It is a node In the The hidden state of the layer during initialization ; It is a node In the figure The set of first-order neighbor nodes in the; It is the first The learnable weight matrix of the layer; It is the ReLU nonlinear activation function. These are the normalized attention weights, calculated using the following formula: ; ; in, Normalized attention weights; It is the unnormalized attention score between node v and its neighbor u; It is an iterable variable that represents all the neighbors of node v; ( () is a non-linear activation function; It is a learnable parameter vector used to map the concatenated high-order features to a scalar score; the superscript T refers to the transpose sign. It is a distance metric between nodes v and u. In this power distribution monitoring scenario, it refers to electrical distance, such as the normalized line impedance value in the power supply path, which reflects the tightness of the electrical connection. Hyperparameters used to control the strength of physical constraint terms; It is a very small positive number, added to the denominator to prevent division by zero when the electrical distance is zero; This module introduces a bias term based on electrical distance into the graph attention mechanism, enabling the model to not only rely on data-driven feature similarity when aggregating information, but also follow the physical law that electrical influence decays with distance. This allows for a more accurate modeling of the spatial diffusion patterns of anomalous behavior in power distribution networks. go through After spatial convolution, a representation of each node rich in spatial context is obtained. .

[0053] For the time series modeling module, this module is used to capture the dynamic trend of the state characteristics of each device node evolving over time. Each node... Spatial feature sequence Input a gated recurrent unit network: ; in, It is the GRU at the time step The hidden state encodes the node. The temporal evolution information up to the current moment; This represents all learnable parameters within the GRU unit. The purpose of this formula is to learn the normal variation pattern of each power distribution device's behavior over time, thereby enabling sensitive identification of abnormal timing patterns that deviate from this pattern, such as continuous gradual changes in characteristics, periodic disruptions, or sudden disappearances of correlations.

[0054] For the output computation module, this module is based on the final node representation obtained from spatiotemporal modeling. Calculate the results of trend anomaly identification.

[0055] Node-level anomaly scoring: For each power distribution equipment node Calculate its abnormal score: ; in, Represents a node Abnormal scores, with a value range of The closer the value is to 1, the higher the probability that the node is involved in the evolution of anomalous behavior during the observation period; and These are learnable weight vectors and bias scalars.

[0056] For risk propagation path assessment, the attention weight matrix output by the spatial graph convolutional module at each time step is analyzed. The identification process shows that the attention weights are consistently and significantly higher during abnormal evolution, such as edges whose weights are more than two standard deviations above the average weight for more than 80% of the time steps. The ordered sequence of these edges constitutes a potential risk propagation path, indicating the main direction of diffusion of anomalous features.

[0057] The training of the trend anomaly identification model is as follows: Trend Anomaly Identification Model parameter set The data is obtained through a training process; for the training data, the power distribution behavior evolution trajectory dataset, collected under historical normal operating conditions and processed through steps S1 and S2, is used. Training was conducted, among which This represents the number of training samples.

[0058] For the training objective and loss function, an unsupervised graph sequence reconstruction strategy is adopted. The trend anomaly identification model is extended to an encoder-decoder structure, with the front part serving as the encoder and the decoder consisting of symmetric spatiotemporal layers designed to reconstruct input features from the encoder's hidden states. The training objective is to minimize the reconstruction error. The loss function L uses the mean squared error: ; in, The model is for the first training samples, nodes At time step The feature reconstruction value; These correspond to the actual input features. The purpose of this formula is to force the encoder to learn and compress typical spatiotemporal patterns of normal behavioral evolution trajectories, so that the model has low reconstruction error for normal data, but high reconstruction error for data containing abnormal trends, thereby revealing anomalies.

[0059] After the trend anomaly identification model has been trained, it can be applied to online monitoring and analysis. The specific application process is as follows: First, the power distribution behavior evolution trajectory T to be analyzed is input into the data conversion module of the trend anomaly identification model and processed into a standard spatiotemporal graph sequence. Next, the sequence undergoes forward propagation through the spatial graph convolution module and the temporal series modeling module, sequentially completing the spatial correlation and temporal evolution analysis of behavioral features to obtain the final spatiotemporal fusion representation h of each node. And the attention weights in the process; finally, the output calculation module is based on h Calculate the anomaly score for each node. By combining attention weight analysis, potential risk propagation paths are identified, resulting in a structured trend anomaly identification result. This result includes a list of node-level anomaly scores and risk propagation path information, providing a direct basis for subsequent precise intervention.

[0060] It should be noted that the trend anomaly identification result can be formally defined as an ordered pair R: ; in, This represents the set of node anomaly scores, defined as follows: This collection contains a power supply topology diagram. Each device node and their corresponding anomaly scores ; This represents the set of risk transmission paths, defined as follows: Each of these paths It is a sequence of directed edges, for example The sequence consists of attention weights output by the spatial graph convolution module. The analysis revealed the main diffusion direction of the anomalous features in the distribution network.

[0061] Step S4: Based on the trend anomaly identification results, generate a differentiated intervention plan.

[0062] The formalized result output from step S3, namely the trend anomaly identification result. This step aims to automatically generate and issue tiered and object-oriented remote intervention commands based on the severity of the anomaly score and its propagation topology in the power distribution network, forming a decision-making closed loop from intelligent early warning to precise control.

[0063] First, based on the node anomaly score set and risk transmission path set To determine the intervention level and unique identifier for power distribution equipment that requires intervention.

[0064] For intervention level determination, continuous abnormal scores are used. Mapped to discrete severity levels, with two decision thresholds set. and The specific values ​​are derived from statistical analysis of a historical anomaly event sample database. In specific embodiments, for example, the optimal segmentation point for classification performance can be determined by using the receiver operation characteristic curve. It is a sensitivity threshold designed to capture the vast majority of potential anomalies and avoid omissions, but it will include some suspicious or minor anomalies. This is a specific threshold designed to identify severe anomalies with high certainty, ensuring that events triggering high-level interventions have a high degree of confidence. Device Node Intervention level Determined by the following formula: ; in, Represents a node The intervention level takes the value of a discrete integer; and The threshold values ​​are preset for each level of anomaly. Level 1 corresponds to the concern level, indicating that the anomaly is in its nascent stage; Level 2 corresponds to the control level, indicating that the anomaly has begun to spread and needs to be actively suppressed; Level 3 corresponds to the isolation level, indicating that the anomaly is approaching the critical boundary and needs to be immediately blocked. The purpose of this formula is to quantify the model's anomaly score output into a limited and clear action priority.

[0065] Regarding the identification of intervention targets, intervention targets The location is determined by the identifier of the power distribution equipment requiring the operation and its role in the risk context. The location rules include: node self-scoring. Devices triggering level 2 or 3; located within the risk propagation path set. any path Key nodes on the topology, especially the starting or ending nodes of a path; upstream or downstream devices adjacent to these key nodes, used for isolation or load transfer. Object Identifier Taken directly from the device identifier set .

[0066] Intervention levels obtained from mapping The system retrieves and instantiates specific control commands from a predefined, hierarchical intervention strategy knowledge base, based on the specific equipment type indicated by the equipment identifier O (i.e., main distribution box, sub-distribution box, switch box). This knowledge base defines the sets of safe, executable operations and their parameters for various types of equipment at different levels.

[0067] make This indicates that it is for level 1 Object identifier is The set of intervention actions generated by the power distribution equipment. Their generation follows the principles of hierarchy and object-specificity: when Attention Level: Intervention actions generated at this level The main purpose is to enhance monitoring and early warning without changing the system's operating status.

[0068] For example: for the main distribution box or sub-distribution box of the object ,action To increase the sampling frequency of its monitoring data from 1 time / second to 4 times / second for 10 minutes; for the object switch box ,action To lower the residual current protection warning value of its circuit by 20%, for example, from 300mA to 240mA; to the object The associated construction area administrator's mobile terminal pushes a text alert message with the template: Attention Alert: Equipment [O] is showing an abnormal trend, please observe.

[0069] when Control level: Intervention actions generated at this level This involves proactive adjustments to the load or operating mode.

[0070] For example: For object distribution boxes ,action To remotely set its maximum permissible output power to 85% of the current value for 30 minutes; if the object is a switch box Control non-critical loads, actions To remotely issue a delayed start command, postpone the next planned start time by 15 minutes; under the condition of having dual-circuit power supply, the main distribution box of the object... Specific branch, action To automatically switch the load to the backup power supply circuit.

[0071] when Isolation level: Intervention actions generated at this level The aim is to quickly and accurately isolate sources of risk.

[0072] For example: for the object switch box ,action To issue remote tripping commands to its intelligent circuit breakers; to distribute electrical boxes to the target devices. Specific high-risk routes, actions To remotely analyze this branch circuit breaker; in the isolated object At the same time, if the topology allows it and it has passed security checks, the action... To close the preset contact switch, the important load it carries is transferred to the adjacent healthy circuit for power supply.

[0073] The final differentiated intervention plan is a structured list of instructions. Each entry in this list is a quadruple: ; in, The target device identifier of the instruction is determined by the abnormal score list output in step S3 and the device node identifier determined in the risk path. This formula represents specific control command codes, such as power limiting, circuit switching, and circuit breaker tripping. Their values ​​are mapped from a predefined strategy library in step S4 based on the intervention level and equipment type of the target device. Params represent command parameters, which are preset values ​​in the strategy library based on a trade-off between safety and efficiency, such as a power limit of 85% and a delay time of 15 minutes. Priority represents the command execution priority, whose value is determined based on the principle of safety priority and the operational dependencies of the power supply path, and is used to coordinate the execution sequence of multiple commands. The purpose of this formula is to encode the intervention strategy into a clear, unambiguous data structure that can be directly parsed and executed by the on-site intelligent control unit. This structured scheme will be transmitted in real-time to the corresponding power distribution equipment on the construction site via a wireless communication network.

[0074] This step transforms the anomaly identification results into layered, progressive control actions that precisely match the risk level and optimally balance construction continuity and safety. By achieving a closed loop of early warning, analysis, and control, this method represents a leap from the traditional, crude approach of tripping the circuit breaker when monitoring exceeds a certain threshold to an intelligent, proactive management and control model characterized by precise early warning, tiered intervention, and minimal impact.

[0075] This invention proposes an innovative remote monitoring method for construction site power distribution equipment based on big data. The core of this method lies in changing the traditional approach of monitoring only isolated equipment parameters. Through a progressive intelligent analysis framework, it achieves proactive early warning and precise intervention for behavioral-level risks in the power distribution system. Its technical solution comprises four key steps: The raw monitoring data from the main distribution box, sub-distribution boxes, and switch boxes are transformed into meaningful behavioral semantic fragments based on the construction site's power consumption logic. Based on the power supply relationships between devices, these behavioral fragments are concatenated according to spatiotemporal rules to form a power distribution behavior evolution trajectory describing how anomalies propagate across levels. A spatiotemporal graph neural network model is constructed, and the behavioral evolution trajectory is input into this model. The model learns the diffusion patterns of behavior in the power distribution network through a spatial graph convolution module and its evolution trends through a time series modeling module, thereby identifying potential anomaly risks and their propagation paths. Based on the identified anomaly level and risk path, immediate and accurate differentiated remote control commands are automatically matched and generated.

[0076] The beneficial effect of this method is that it elevates the monitoring object from isolated parameters to system behavior, thereby achieving proactive early warning of complex hidden risks; and through tiered intervention, it minimizes interference with construction while ensuring safety.

[0077] The present invention also protects a remote monitoring system for construction site power distribution equipment based on big data, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "including a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. A method for remote monitoring of construction site power distribution equipment based on big data, characterized in that, Includes the following steps: Step S1: Obtain data on the construction site power distribution equipment, perform semantic mapping on the data, and generate semantic fragments of power distribution equipment behavior; Step S2: Based on the semantic fragments of the power distribution equipment behavior and the power transmission relationship between the power distribution equipment on the construction site, establish a cross-level behavior association chain and form a power distribution behavior evolution trajectory; Step S3: Based on the graph neural network, construct a trend anomaly identification model, input the power distribution behavior evolution trajectory into the trend anomaly identification model, and output the trend anomaly identification result; Step S4: Based on the trend anomaly identification results, generate a differentiated intervention plan.

2. The method for remote monitoring of construction site power distribution equipment based on big data according to claim 1, characterized in that, Obtain data on the construction site's power distribution equipment, including: The system collects monitoring data from all equipment within the three-tiered power distribution system of the temporary power supply system at the construction site. This three-tiered power distribution system consists of a main distribution box, distribution boxes, and switch boxes arranged from top to bottom. The main distribution box serves as the main distribution node for the incoming power supply, supplying power to multiple distribution boxes. Each distribution box supplies power to multiple switch boxes within a specific construction area. Each switch box directly controls one or a group of terminal electrical devices. The collected data on the construction site's power distribution equipment includes monitoring data synchronously collected from these three levels of equipment, forming a complete time-series data set.

3. The method for remote monitoring of construction site power distribution equipment based on big data according to claim 2, characterized in that, Semantic mapping is performed on the data of the construction site power distribution equipment to generate semantic fragments of power distribution equipment behavior, including: Perform semantic mapping operations on the time series data set D: For the main distribution box, based on the monitoring data of the main distribution box, features reflecting the power source scheduling and overall distribution logic are extracted, and the behavior is judged according to quantitative standards; For distribution boxes, based on the monitoring data of the distribution boxes, features reflecting the regional load switching, coordination and shock response logic are extracted, and the behavior is judged according to quantitative standards; For switch boxes, based on the monitoring data of the switch boxes, features reflecting the direct operation and protection precursors of terminal electrical equipment are extracted, and the behavior is determined according to quantitative standards. Time series data set Mapped to a set of structured semantic fragments of power distribution equipment behavior .

4. The method for remote monitoring of construction site power distribution equipment based on big data according to claim 3, characterized in that, The power transmission relationships between electrical equipment on the construction site include: The power supply transmission relationships between all power distribution equipment are formally predefined to form a power supply path topology diagram. This topology is represented by an ordered pair: ； in, This represents the set of all nodes in the graph, and this set is related to the set of device identifiers in step S1. The same, that is, the topology of each monitored power supply path. A node in the graph; E represents the set of all directed edges in the graph.

5. A method for remote monitoring of construction site power distribution equipment based on big data according to claim 4, characterized in that, Establish cross-level behavioral association chains, including: This process iterates through the collection. The behavioral semantic fragments in the data, and based on spatial and temporal association rules, from the set of behavioral semantic fragments of power distribution equipment. Multiple cross-level behavioral association chains are extracted and linked together to form a new set. Each link It is an ordered list that records a series of semantic fragments of behavior that are triggered by an initial event and propagated upstream along the power supply path.

6. The method for remote monitoring of construction site power distribution equipment based on big data according to claim 5, characterized in that, The evolution trajectory of power distribution behavior includes: For a specified analysis period , and These represent the start and end times of the analysis period, and the evolution trajectory of power distribution behavior during that period. Build it in the following way: Filter out the set All time windows fall entirely within the relevant links of that time period; then, based on the start time of the initial behavior segment in each relevant link. Sort in ascending order.

7. A method for remote monitoring of construction site power distribution equipment based on big data according to claim 6, characterized in that, A trend anomaly identification model is constructed based on graph neural networks, including: Trend Anomaly Identification Model It is a spatiotemporal graph neural network whose input is the evolution trajectory of power distribution behavior. The output is the quantitative identification result of potential abnormal trends, namely the trend anomaly identification result. The model includes four core modules in sequence according to the processing flow: data transformation, spatial graph convolution, time series modeling and output calculation. These modules process the spatiotemporal characteristics of the power distribution behavior evolution trajectory in a serial manner.

8. A method for remote monitoring of construction site power distribution equipment based on big data according to claim 7, characterized in that, The power distribution behavior evolution trajectory is input into the trend anomaly identification model, and the trend anomaly identification results are output, including: The evolution trajectory T of the power distribution behavior to be analyzed is input into the data conversion module of the trend anomaly identification model and processed into a standard spatiotemporal graph sequence. Next, the sequence undergoes forward propagation through the spatial graph convolution module and the temporal series modeling module, sequentially completing the spatial correlation and temporal evolution analysis of behavioral features to obtain the final spatiotemporal fusion representation h of each node. And the attention weights in the process; finally, the output calculation module is based on h Calculate the anomaly score for each node. It also analyzes potential risk propagation paths by combining attention weights, thereby outputting structured trend anomaly identification results.

9. A method for remote monitoring of construction site power distribution equipment based on big data according to claim 8, characterized in that, Based on the trend anomaly identification results, a differentiated intervention plan is generated, including: Based on the node anomaly score set and the risk propagation path set, the intervention level and unique identifier of the power distribution equipment that needs intervention are determined; Intervention levels obtained from mapping And the specific device type indicated by device identifier O, retrieve and instantiate specific control instructions from a predefined, hierarchical intervention strategy knowledge base; The final differentiated intervention plan is a structured list of instructions, where each entry is a quadruple: ； in, The target device identifier of the instruction is determined by the abnormal score list output in step S3 and the device node identifier determined in the risk path. The specific control command code is obtained by mapping from the predefined strategy library in step S4 according to the intervention level and device type of the target device; Params represents the command parameters, which are preset values ​​in the strategy library based on the trade-off between safety and efficiency; Priority represents the command execution priority, which is determined based on the principle of safety priority and the operation dependency of the power supply path, and is used to coordinate the execution sequence of multiple commands.

10. A remote monitoring system for construction site power distribution equipment based on big data, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-9.

Images