Energy consumption hierarchical monitoring method, system, device and medium combined with flow direction prediction

Abstract

This invention relates to the field of artificial intelligence technology, specifically providing a method, system, device, and medium for energy consumption grading monitoring that combines flow direction prediction. The method includes: collecting and standardizing multi-source data from an energy system to form a standardized dataset containing flow direction data and grade data. Next, basic features and correlation features between flow direction and grade are mined from the dataset to form a feature set. This feature set is then input into a trained dual-objective prediction model, simultaneously outputting predicted energy flow rates and energy consumption grade predictions for future periods. Finally, based on the prediction results, when the probability of a target area reaching a specific energy consumption grade in the future exceeds a preset threshold, the monitoring strategy is automatically and dynamically adjusted. This invention achieves intelligent linkage between prediction and monitoring, effectively improving the timeliness of early warning and the accuracy of response.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a method, system, device, and medium for energy consumption classification monitoring that combines flow direction prediction. Background Technology

[0002] In the field of energy management, accurate prediction and tiered monitoring of energy consumption are crucial for achieving energy conservation and emission reduction. Current technologies typically treat energy flow prediction and tiered energy consumption monitoring as two separate processes: the prediction model only outputs future energy consumption figures, while the monitoring system issues static alarms based on fixed thresholds. This fragmented approach makes it difficult for prediction results to directly and automatically guide the optimization of monitoring strategies, and it cannot proactively adjust monitoring frequency and warning thresholds based on predicted risks. Simultaneously, monitoring results cannot be effectively fed back to the prediction model to correct errors, lacking an adaptive closed-loop "perception-decision" capability, resulting in problems such as delayed warnings, rigid responses, and difficulty in uncovering the deep correlation between energy dynamics and energy consumption status. Summary of the Invention

[0003] In view of the above-mentioned shortcomings of the prior art, the present invention provides an energy consumption classification monitoring method, system, device and medium that combines flow direction prediction to solve the above-mentioned technical problems.

[0004] In a first aspect, the present invention provides an energy consumption classification monitoring method combining flow direction prediction, comprising:

[0005] Collect multi-source data from the energy system and standardize the collected data to form a standardized dataset; the multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status.

[0006] Mining a feature set from the standardized dataset, the feature set including basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data;

[0007] The feature set is input into the trained bi-objective prediction model to obtain the predicted energy flow and energy consumption level for the future target period.

[0008] Based on the energy consumption level prediction results, when the probability that the target area will reach a specific energy consumption level in the future exceeds a preset threshold, the monitoring strategy will be dynamically adjusted automatically.

[0009] In one optional implementation, multi-source data from the energy system is collected, and the collected data is standardized to form a standardized dataset, including:

[0010] The flow direction data is collected in real time from IoT sensing devices deployed at various nodes of the energy system; and the grade data and its corresponding grading standards are obtained from the energy management information system.

[0011] The collected data is subjected to outlier detection and removal, as well as missing value imputation.

[0012] The cleaned numerical data is standardized to eliminate the influence of units, and all data are aggregated according to a unified time window to align the multi-source data in the time series, thereby forming the standardized dataset.

[0013] In one optional implementation, mining a feature set from the standardized dataset includes:

[0014] At least one statistical feature or state feature is extracted from the flow data and the grade data respectively; wherein, the features extracted from the flow data include flow volatility, peak duration or path concentration, and the features extracted from the grade data include grade proportion, threshold deviation or grade stability.

[0015] By mining the dynamic relationship between the flow data and the level data, correlation features are constructed;

[0016] Integrate basic features and related features into a feature set.

[0017] In an optional implementation, by mining the dynamic relationship between the flow data and the level data, a correlation feature is constructed, including:

[0018] The process of constructing dynamic matching degree features includes: within a sliding time window, obtaining the historical energy flow peak sequence and the corresponding future energy consumption level sequence; calculating the Spearman rank correlation coefficient between the two sequences as the dynamic matching degree feature value at the current moment;

[0019] The process of constructing regional sensitivity characteristics includes: for each monitoring region, obtaining its historical energy consumption change rate sequence and corresponding energy flow fluctuation rate sequence; performing linear regression analysis on the two sequences, and using the regression slope as the sensitivity characteristic of the region;

[0020] The process of constructing time-series cross samples for supervised learning includes: using the feature data of the current and historical moments as input features, and using the actual energy consumption level or energy flow value of the future target moment as the prediction label, and aligning them to form training samples.

[0021] In an optional implementation, the bi-objective prediction model is a machine learning model trained to simultaneously output traffic prediction values ​​and grade prediction results, comprising:

[0022] Multimodal input layer: It includes parallel temporal feature extraction tower, static attribute encoding tower and dynamic cross feature processing tower, which are used to perform structured embedding and representation of the input real-time temporal data, static attribute data and dynamic cross features respectively, and concatenate and fuse the feature vectors output by each tower to form a fused feature vector;

[0023] The core network layer of physical perception: It constructs a spatiotemporal graph neural network based on a predefined topological graph reflecting the energy transmission relationship between regions; the spatiotemporal graph neural network is used to operate on the fused feature vector, aggregate neighbor node information through a graph attention mechanism to model spatial dependence, and model the temporal dependence of each node through a temporal neural network module, wherein the initial weights of the graph attention mechanism are initialized by the regional sensitivity coefficient or line loss parameter.

[0024] Differentiable fuzzy output layer: It includes a regression branch and a classification branch; the classification branch contains a parameterized fuzzy classifier, which is configured to: take the predicted output of the regression branch or the intermediate output of the core network layer as input, and calculate the membership probability of the predicted traffic belonging to each energy consumption level based on a preset energy consumption level threshold parameter through a differentiable fuzzy membership function, as the energy consumption level prediction result; wherein, the threshold parameter can be fine-tuned during model training.

[0025] In one optional implementation, the training of the bi-objective prediction model is optimized using a joint loss function, which is a weighted sum of the mean squared error loss for the regression branch and the Focal Loss for the classification branch, and is trained and updated by combining offline pre-training with online incremental learning.

[0026] In an optional implementation, based on the energy consumption level prediction results, when the probability that the target area will reach a specific energy consumption level in the future exceeds a preset threshold, the monitoring strategy is automatically dynamically adjusted, including:

[0027] When it is predicted that the probability of a target area reaching a specific energy consumption level in the future exceeds a preset threshold, at least one of the following operations shall be performed: increase the data acquisition frequency of the monitoring equipment in the area, dynamically adjust the energy consumption classification early warning threshold of the area, or send energy regulation suggestions to the energy dispatch system.

[0028] Secondly, the present invention provides an energy consumption classification monitoring system that incorporates flow direction prediction, comprising:

[0029] The data acquisition module is used to collect multi-source data from the energy system and to standardize the collected data to form a standardized dataset; the multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status.

[0030] The feature extraction module is used to mine a feature set from the standardized dataset. The feature set includes basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data.

[0031] The model prediction module is used to input the feature set into the trained bi-objective prediction model to obtain the predicted energy flow and energy consumption level for the future target period.

[0032] The strategy adjustment module is used to automatically perform dynamic adjustment of the monitoring strategy based on the energy consumption level prediction results. When the probability that the target area will reach a specific energy consumption level in the future exceeds a preset threshold, the module will automatically perform dynamic adjustment of the monitoring strategy.

[0033] Thirdly, a device is provided, comprising:

[0034] Memory for storing energy consumption classification monitoring programs that incorporate flow direction prediction;

[0035] A processor is configured to implement the steps of the energy consumption classification monitoring method combined with flow direction prediction as provided in the first aspect when executing the energy consumption classification monitoring program that combines flow direction prediction.

[0036] Fourthly, a computer-readable medium is provided, on which an energy consumption classification monitoring program incorporating flow direction prediction is stored, wherein when the energy consumption classification monitoring program incorporating flow direction prediction is executed by a processor, the steps of the energy consumption classification monitoring method incorporating flow direction prediction provided in the first aspect are implemented.

[0037] The beneficial effects of this invention are that the energy consumption classification monitoring method, system, equipment, and medium combined with energy flow direction prediction provided by this invention, by constructing a closed loop linking energy flow direction prediction and energy consumption classification monitoring, realize automatic and forward-looking guidance of monitoring strategies based on prediction results, effectively improving the timeliness of early warning and the accuracy of response. The proposed correlation feature engineering and dual-objective coupling model deeply explore the intrinsic correlation between energy dynamics and energy consumption status, significantly improving prediction accuracy and classification correlation. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic flowchart of a method according to an embodiment of the present invention.

[0040] Figure 2This is a schematic block diagram of a system according to an embodiment of the present invention.

[0041] Figure 3 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0042] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0044] The energy consumption classification monitoring method combined with flow direction prediction provided in this embodiment of the invention is executed by a computer device, and correspondingly, the energy consumption classification monitoring system combined with flow direction prediction runs in the computer device.

[0045] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention. Wherein, Figure 1 The implementing entity can be an energy consumption grading monitoring system that incorporates flow direction prediction. Depending on different requirements, the order of steps in this flowchart can be changed, and some steps can be omitted.

[0046] like Figure 1 As shown, the method includes:

[0047] S1. Collect multi-source data from the energy system and standardize the collected data to form a standardized dataset; the multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status;

[0048] S2. Mining a feature set from the standardized dataset, the feature set including basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data;

[0049] S3. Input the feature set into the trained bi-objective prediction model to obtain the predicted energy flow rate and energy consumption level for the future target period;

[0050] S4. Based on the energy consumption level prediction results, when the probability that the target area will reach a specific energy consumption level in the future exceeds a preset threshold, the monitoring strategy is automatically dynamically adjusted.

[0051] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0052] S101. Collect the flow direction data in real time from IoT sensing devices deployed at various nodes of the energy system; and obtain the grade data and its corresponding grading standards from the energy management information system.

[0053] Real-time energy flow data, including time-series signals such as current, flow rate, pressure, and temperature, is collected from IoT sensing devices (such as SCADA systems, smart meters, and flow meters) deployed at key nodes in energy networks, workshops, and buildings, and timestamps and device identifiers are recorded simultaneously. Simultaneously, predefined energy consumption grading standards (e.g., threshold ranges for high, medium, and low energy consumption) and historical energy consumption records and attribute information (such as area and equipment list) for each area are obtained from the Energy Management Information System (EMS) or management platform via API interfaces or direct database connections.

[0054] S102. Perform outlier detection and removal, as well as missing value imputation on the collected data.

[0055] The collected raw data is cleaned. The 3σ principle or the isolated forest algorithm is used to detect and remove abnormal data points such as instantaneous spikes and zero-value drift caused by sensor malfunctions. For randomly missing values, time series interpolation methods (such as linear interpolation or spline interpolation) are used to fill them in; for continuous large segments of missing data, they are filled in or marked according to business rules and are not included in subsequent model training.

[0056] S103. Standardize the cleaned numerical data to eliminate the influence of units, and aggregate all data according to a unified time window to align the multi-source data in the time series, thereby forming the standardized dataset.

[0057] The cleaned numerical features are standardized. To eliminate the influence of dimensions, Z-Score standardization or Min-Max standardization is used to transform various numerical values ​​to similar scales. Subsequently, a uniform time aggregation window (e.g., 15 minutes) is set, and aggregation operations (such as averaging and cumulative summing) are performed on all data streams within this window to ensure that energy flow data, energy consumption level data, and auxiliary data such as meteorological and production plans obtained from external sources are accurately aligned in time series. The final output is a high-quality, consistent standardized dataset for subsequent feature engineering.

[0058] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0059] S201. Extract at least one statistical feature or state feature from the flow data and the grade data respectively; wherein, the features extracted from the flow data include flow volatility, peak duration or path concentration, and the features extracted from the grade data include grade proportion, threshold deviation or grade stability.

[0060] For standardized flow data, within a set aggregation time window (e.g., 15 minutes), key statistical characteristics are calculated: flow volatility ((maximum value - minimum value) / average value), peak duration (number of consecutive time points where flow exceeds 110% of average flow), and path concentration (the proportion of flow from the main energy supply path to the total flow). For graded data, the following status characteristics are calculated: grade proportion (the proportion of high, medium, and low energy consumption areas to the total number of areas), threshold deviation (the difference between the current actual energy consumption of an area and the median threshold of its grade), and graded stability (the duration for which the area continuously maintains the same energy consumption level).

[0061] S202. Construct association features by mining the dynamic relationship between the flow data and the level data;

[0062] Constructing a dynamic matching feature: A sliding window of 24 hours is defined. Within this window, the peak energy flow sequences for the current and past 23 hours are acquired. A simple predictor (such as LightGBM) is then used to generate a corresponding energy consumption level prediction sequence for the next hour based on historical energy consumption data (the levels are quantified into numerical values, such as low=1, medium=2, high=3). The Spearman rank correlation coefficient between these two sequences is calculated; this coefficient serves as the matching feature characterizing the strength of the dynamic correlation between "flow direction" and "level" at the current moment.

[0063] Constructing regional sensitivity characteristics: For each independent monitoring area, collect its historical hourly data for the past 3 months. Calculate the energy consumption change rate (Energy Consumption Change Rate) at each time point. t -Energy consumption t-1 Energy consumption t-1 ) and energy flow volatility (inflow) t -Inflow t-1 ) / Inflow t-1 Using the rate of change in flow direction as the independent variable (X) and the rate of change in energy consumption as the dependent variable (Y), a univariate linear regression analysis was performed. The slope k obtained from the regression was used as the flow direction sensitivity coefficient for the region, which characterizes the inherent responsiveness of the region's energy consumption to inflow fluctuations.

[0064] Constructing temporal cross-samples: Organizing data using supervised learning. For any time t, combine all basic features and associated features (including dynamic matching degree and sensitivity coefficient) from this time and several previous times into an input feature vector X. t The actual energy consumption level (classification label) or actual energy flow value (regression label) at the future time t+1 is defined as the prediction target y. t . (X) t ,y t Pair them up to form a training sample. Iterate through the timeline to generate all samples.

[0065] S203. Integrate basic features and related features into a feature set.

[0066] The basic features extracted in the above steps (such as volatility, tiered proportion, etc.) are combined and integrated with the constructed related features (dynamic matching degree, regional sensitivity coefficient) to form a unified, multi-dimensional feature set (i.e., feature vector), which serves as the direct input for the subsequent bi-objective prediction model.

[0067] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0068] The integrated feature set is input into a bi-objective prediction model pre-trained offline. This model is a hybrid machine learning model specifically designed for energy systems, and its core architecture consists of three layers:

[0069] A multimodal input layer is designed to efficiently process heterogeneous input features. It comprises three parallel feature processing towers:

[0070] Suppose there are N regions (nodes), and each node has corresponding characteristics at time t.

[0071] Temporal Feature Extraction Tower: Receives strong temporal signals such as energy flow and volatility. First, a one-dimensional convolutional layer is used to extract local temporal patterns and short-term abrupt changes. Then, the output is fed into a bidirectional long short-term memory network (BiLSTM) to capture long-term temporal dependencies.

[0072] Recent time-series window data for each node i , where T is the time step, This represents the temporal feature dimension.

[0073] Operation: First, use one-dimensional convolution to capture local patterns, and then use BiLSTM to extract temporal representations.

[0074] Local feature extraction formula:

[0075] Forward and backward LSTM formulas:

[0076] Final timing representation formula: ,in Indicates splicing.

[0077] Static attribute encoding tower: Receives static attributes such as region type and device density. Encodes them using an embedding layer or a fully connected layer, transforming them into dense vectors.

[0078] Static attribute vector of each node i

[0079]

[0080] and These are learnable parameters.

[0081] Dynamic cross-feature processing tower: Receives derived correlation features such as dynamic matching degree and region sensitivity coefficient. Dimensionality reduction and nonlinear transformation are performed directly through fully connected layers.

[0082] Dynamic cross-feature vector of each node i .

[0083]

[0084] in, and These are learnable parameters.

[0085] The output vectors of the three towers are then concatenated to form a unified, multimodal fusion feature vector, which serves as the input to the core network.

[0086]

[0087] The final fusion feature of node i will be sent to the core network layer.

[0088] The core network layer of physical perception is a spatiotemporal graph neural network designed to model the spatial topological dependencies between regions and the temporal evolution of each region itself.

[0089] Graph Construction: First, based on the physical connections (such as power grid and pipeline topology) or logical transmission relationships of the energy system, a regional relationship graph is constructed. Each monitoring region serves as a node in the graph, and the edges between nodes represent energy transmission paths. The initial weights of the edges can be set based on historical average flow or line transmission capacity.

[0090] The goal of this layer is to capture both spatial (between nodes) and temporal (within the nodes themselves) dependencies. Define a graph G=(V,E), where V is the set of nodes (regions) and E is the set of edges (energy connections).

[0091] Spatial Modeling: A graph attention network is employed, enabling each node to aggregate feature information from its neighboring nodes. The key innovation lies in the fact that the initial weights of this attention mechanism are initialized and guided by physical knowledge such as region sensitivity coefficients or known line loss parameters, giving the model physical awareness from the early stages of training and allowing it to focus more on physically strong connections.

[0092] For each node i, aggregate the information of its neighbors j∈N(i).

[0093]

[0094] It is the raw attention score between nodes i and j. It is a shared linear transformation weight matrix. It is a learnable parameter vector of the attention mechanism.

[0095] At the start of training, The calculations can incorporate prior physical knowledge, such as line loss. , where λ is a hyperparameter, so that the initial attention of high-loss connections will be lower.

[0096]

[0097] These are the attention weights after softmax normalization.

[0098]

[0099] σ is the output of node i after aggregating spatial neighbor information. σ is the activation function, such as ELU.

[0100] Temporal modeling: At each node (region), the aggregated spatial features and the node's own historical temporal features (from the fused feature vector) are input into a temporal neural network module (such as BiLSTM or temporal convolutional network) to capture its dynamic change patterns.

[0101] The sequence after aggregating spatial information for each node is processed.

[0102] Each node i The sequence consisting of this feature from its past few time steps.

[0103] Formula (using GRU or TCN):

[0104]

[0105] The final high-level feature of node i after spatiotemporal modeling will be sent to the output layer.

[0106] Differentiable fuzzy output layer, which contains two parallel branches, each corresponding to a dual prediction target:

[0107] This layer receives the output from the core layer. This generates two prediction targets.

[0108] Regression branch: This is usually a fully connected layer that directly outputs a predicted value (a specific numerical value) of energy flow for a future target time period (such as the next hour).

[0109] Regression branch (predicting energy flow values):

[0110]

[0111] It is the predicted future energy flow value of node i.

[0112] The classification branch's core is an innovative parametric fuzzy classifier. This classifier takes the output of the previous regression branch (or intermediate features from the core network layer) as input. Internally, it pre-sets learnable energy consumption level threshold parameters (e.g., critical values ​​for high, medium, and low energy consumption, initially set to industry standards). Using a differentiable sigmoid membership function, it calculates the membership probabilities (summing up to 1) of the predicted flow value belonging to the "high," "medium," and "low" energy consumption levels, respectively. These smoothed probability values ​​are the final energy consumption level prediction results. This design not only directly embeds domain knowledge (thresholds) into the model, but its differentiability also supports end-to-end training and elegantly handles uncertainties in critical states.

[0113] The classification branch classifies directly based on high-level features:

[0114]

[0115] It is a probability distribution obtained by direct classification.

[0116] Using regression predictions Fuzzy reasoning is performed using learnable thresholds.

[0117] Define the learnable threshold parameter: , , representing the fuzzy dividing points between medium / low energy consumption and high / medium energy consumption, respectively.

[0118] Constructing fuzzy membership degrees using a sigmoid function:

[0119] "High energy consumption" level classification:

[0120]

[0121] "Low energy consumption" level membership:

[0122]

[0123] "Medium energy consumption" level membership degree (the sum of the three must be 1):

[0124]

[0125] It is the sigmoid function, where s is the slope parameter that controls the width of the fuzzy interval (it can be learned or fixed).

[0126] Final classification probability: Normalize the membership vector to obtain the output of the fuzzy classifier. .

[0127] Final classification result: The results of direct classification and fuzzy classification can be combined, for example, by taking an average or weighted average, to obtain the final probability distribution. .

[0128] During model training, a joint loss function (such as a weighted sum of the mean squared error for regression and the Focal Loss for classification) is used for optimization. A strategy combining offline pre-training and online incremental learning after deployment is employed to maintain model performance. In application, the model receives a feature set, processes it sequentially through the aforementioned three-layer structure, and ultimately outputs both accurate energy flow predictions and probabilistic energy consumption level predictions, providing direct and structured decision-making support for subsequent dynamic linkage control.

[0129] Regression Loss: Mean Squared Error (MSE)

[0130]

[0131] Classification loss: Focal loss, which addresses class imbalance.

[0132]

[0133] 'c' represents the category (low, medium, high). It is a prediction probability. These are the true labels (one-hot). α and γ are the hyperparameters of the focus loss.

[0134] Adaptive weighted total loss:

[0135]

[0136] and This is the homoscedastic uncertainty parameter that the model automatically learns for regression and classification tasks, representing the noise level for each task. This formula allows the model to automatically assign smaller weights (smaller 1 / σ²) to tasks with high uncertainty (more difficult), while the regularization term... To prevent them from growing indefinitely.

[0137] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0138] The system deploys a lightweight rules engine that continuously receives real-time output from a dual-objective prediction model, which represents the predicted probability of each target area reaching different energy consumption levels (high, medium, and low) in the future target time period (e.g., the next 2 hours). When the engine detects that the prediction for a specific area (e.g., "Workshop A") meets preset triggering conditions, it automatically executes the corresponding strategy adjustment instructions.

[0139] The specific triggering conditions and execution operations are as follows:

[0140] Triggering condition: The rule engine sets a preset probability threshold (e.g., 80%). When the model predicts that the probability of a certain area entering a high-energy consumption level within the next 2 hours exceeds this threshold, the adaptation action is triggered.

[0141] Operations to be performed (can be performed individually or in combination):

[0142] Increase data collection frequency: The rule engine automatically sends instructions to the corresponding IoT sensing devices (such as smart meters) in the area to temporarily increase their data sampling and reporting frequency from the usual 5 minutes / time to 1 minute / time, so as to achieve more intensive and real-time monitoring of the energy status of the area and make it easier to capture rapidly changing details.

[0143] Dynamically adjust energy consumption grading and early warning thresholds: The system temporarily and specifically adjusts the virtual monitoring thresholds for the area. For example, the threshold for triggering a "high energy consumption" alarm can be lowered by 5% from the original standard (e.g., 1000 kWh) (i.e., adjusted to 950 kWh). This makes the monitoring system more sensitive during that period and able to issue early warning signals earlier.

[0144] Sending Energy Regulation Recommendations: The rules engine automatically generates a structured early warning recommendation message and pushes it to the energy dispatch system or the administrator's console via a standard interface (such as an API). This recommendation message includes a region identifier, the predicted high energy consumption risk, the risk time window, and recommended actions (such as "Recommend reducing unnecessary loads in workshop A in advance" or "Please check the status of power supply lines B"), providing forward-looking information for manual or automated dispatch decisions.

[0145] This positive adaptation process achieves an automatic closed loop from "prediction" to "action," transforming the monitoring strategy from static, passive response to dynamic, proactive prevention, significantly improving the agility and intelligence of the energy management system.

[0146] In some embodiments, the energy consumption classification monitoring system incorporating flow direction prediction may include multiple functional modules composed of computer program segments. The computer programs for each program segment in the energy consumption classification monitoring system incorporating flow direction prediction may be stored in the memory of a computer device and executed by at least one processor to perform (see details). Figure 1 (Description) Functionality of energy consumption classification monitoring combined with flow direction prediction.

[0147] In this embodiment, the energy consumption classification monitoring system combined with flow direction prediction can be divided into multiple functional modules according to the functions it performs, such as... Figure 2 As shown. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, and is stored in memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0148] The data acquisition module is used to collect multi-source data from the energy system and to standardize the collected data to form a standardized dataset; the multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status.

[0149] The feature extraction module is used to mine a feature set from the standardized dataset. The feature set includes basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data.

[0150] The model prediction module is used to input the feature set into the trained bi-objective prediction model to obtain the predicted energy flow and energy consumption level for the future target period.

[0151] The strategy adjustment module is used to automatically perform dynamic adjustment of the monitoring strategy based on the energy consumption level prediction results. When the probability that the target area will reach a specific energy consumption level in the future exceeds a preset threshold, the module will automatically perform dynamic adjustment of the monitoring strategy.

[0152] Figure 3 The energy consumption classification monitoring method combined with flow direction prediction provided in the embodiments of this application can be applied to devices. Those skilled in the art will understand that the device structure involved in the embodiments of this invention does not constitute a limitation on the device; the device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the device includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0153] The device 300 may include a processor 310, a memory 320, and a communication unit 330. These components communicate via one or more buses. Those skilled in the art will understand that the server structure shown in the figure does not constitute a limitation of the present invention. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0154] The memory 320 can be used to store execution instructions of the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 is able to perform some or all of the steps in the above method embodiments.

[0155] The processor 310 serves as the control center of the storage device, connecting various parts of the electronic device via various interfaces and lines. It executes software programs and / or modules stored in the memory 320, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 310 may consist only of a central processing unit (CPU). In this embodiment of the invention, the CPU may have a single processing core or include multiple processing cores.

[0156] The communication unit 330 is used to establish a communication channel, enabling the storage device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0157] The present invention also provides a computer medium, wherein the computer medium may store a program, which, when executed, may include some or all of the steps provided in the embodiments of the present invention. The medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0158] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a medium such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer device (which may be a personal computer, a server, or a second device, network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0159] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0160] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or modules may be electrical, mechanical, or other forms.

[0161] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0162] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0163] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. A method for energy consumption classification monitoring that combines flow direction prediction, characterized in that, include: Collect multi-source data from the energy system and standardize the collected data to form a standardized dataset; the multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status. Mining a feature set from the standardized dataset, the feature set including basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data; The feature set is input into the trained bi-objective prediction model to obtain the predicted energy flow and energy consumption level for the future target period. Based on the energy consumption level prediction results, when the probability of the target area reaching a specific energy consumption level in the future exceeds a preset threshold, the monitoring strategy is automatically dynamically adjusted. The dual-objective prediction model is a machine learning model trained to simultaneously output traffic flow predictions and grade prediction results, which includes: Multimodal input layer: It includes parallel temporal feature extraction tower, static attribute encoding tower and dynamic cross feature processing tower, which are used to perform structured embedding and representation of the input real-time temporal data, static attribute data and dynamic cross features respectively, and concatenate and fuse the feature vectors output by each tower to form a fused feature vector; The core network layer of physical perception: It constructs a spatiotemporal graph neural network based on a predefined topological graph reflecting the energy transmission relationship between regions; the spatiotemporal graph neural network is used to operate on the fused feature vector, aggregate neighbor node information through a graph attention mechanism to model spatial dependence, and model the temporal dependence of each node through a temporal neural network module, wherein the initial weights of the graph attention mechanism are initialized by the regional sensitivity coefficient or line loss parameter. Differentiable fuzzy output layer: It includes a regression branch and a classification branch; the classification branch contains a parameterized fuzzy classifier, which is configured to: take the predicted output of the regression branch or the intermediate output of the core network layer as input, and calculate the membership probability of the predicted traffic belonging to each energy consumption level based on a preset energy consumption level threshold parameter through a differentiable fuzzy membership function, as the energy consumption level prediction result; wherein, the threshold parameter can be fine-tuned during model training.

2. The method according to claim 1, characterized in that, Collect multi-source data from the energy system and standardize the collected data to form a standardized dataset, including: The flow direction data is collected in real time from IoT sensing devices deployed at various nodes of the energy system; and the grade data and its corresponding grading standards are obtained from the energy management information system. The collected data is subjected to outlier detection and removal, as well as missing value imputation. The cleaned numerical data is standardized to eliminate the influence of units, and all data are aggregated according to a unified time window to align the multi-source data in the time series, thereby forming the standardized dataset.

3. The method according to claim 1, characterized in that, Mining a feature set from the standardized dataset includes: At least one statistical feature or state feature is extracted from the flow data and the grade data respectively; wherein, the features extracted from the flow data include flow volatility, peak duration or path concentration, and the features extracted from the grade data include grade proportion, threshold deviation or grade stability. By mining the dynamic relationship between the flow data and the level data, correlation features are constructed; Integrate basic features and related features into a feature set.

4. The method according to claim 3, characterized in that, By mining the dynamic relationship between the flow data and the level data, correlation features are constructed, including: The process of constructing dynamic matching degree features includes: within a sliding time window, obtaining the historical energy flow peak sequence and the corresponding future energy consumption level sequence; calculating the Spearman rank correlation coefficient between the two sequences as the dynamic matching degree feature value at the current moment; The process of constructing regional sensitivity characteristics includes: for each monitoring region, obtaining its historical energy consumption change rate sequence and corresponding energy flow fluctuation rate sequence; performing linear regression analysis on the two sequences, and using the regression slope as the sensitivity characteristic of the region; The process of constructing time-series cross samples for supervised learning includes: using the feature data of the current and historical moments as input features, and using the actual energy consumption level or energy flow value of the future target moment as the prediction label, and aligning them to form training samples.

5. The method according to claim 1, characterized in that, The training of the bi-objective prediction model is optimized using a joint loss function, which is a weighted sum of the mean squared error loss for the regression branch and the Focal Loss for the classification branch. The model is trained and updated by combining offline pre-training with online incremental learning.

6. The method according to claim 1, characterized in that, Based on the energy consumption level prediction results, when the probability of the target area reaching a specific energy consumption level in the future exceeds a preset threshold, the monitoring strategy is automatically dynamically adjusted, including: When it is predicted that the probability of a target area reaching a specific energy consumption level in the future exceeds a preset threshold, at least one of the following operations shall be performed: increase the data acquisition frequency of the monitoring equipment in the area, dynamically adjust the energy consumption classification early warning threshold of the area, or send energy regulation suggestions to the energy dispatch system.

7. An energy consumption classification monitoring system combining flow direction prediction, characterized in that, include: The data acquisition module is used to collect multi-source data from the energy system and to standardize the collected data to form a standardized dataset. The multi-source data includes at least flow data reflecting energy dynamics and level data reflecting energy consumption status; The feature extraction module is used to mine a feature set from the standardized dataset. The feature set includes basic features extracted from the flow data and the grade data, and correlation features between the flow data and the grade data. The model prediction module is used to input the feature set into the trained bi-objective prediction model to obtain the predicted energy flow and energy consumption level for the future target period. The strategy adjustment module is used to automatically perform dynamic adjustment of the monitoring strategy based on the energy consumption level prediction results, when the probability of the target area reaching a specific energy consumption level in the future exceeds a preset threshold. The dual-objective prediction model is a machine learning model trained to simultaneously output traffic flow predictions and grade prediction results, which includes: Multimodal input layer: It includes parallel temporal feature extraction tower, static attribute encoding tower and dynamic cross feature processing tower, which are used to perform structured embedding and representation of the input real-time temporal data, static attribute data and dynamic cross features respectively, and concatenate and fuse the feature vectors output by each tower to form a fused feature vector; The core network layer of physical perception: It constructs a spatiotemporal graph neural network based on a predefined topological graph reflecting the energy transmission relationship between regions; the spatiotemporal graph neural network is used to operate on the fused feature vector, aggregate neighbor node information through a graph attention mechanism to model spatial dependence, and model the temporal dependence of each node through a temporal neural network module, wherein the initial weights of the graph attention mechanism are initialized by the regional sensitivity coefficient or line loss parameter. Differentiable fuzzy output layer: It includes a regression branch and a classification branch; the classification branch contains a parameterized fuzzy classifier, which is configured to: take the predicted output of the regression branch or the intermediate output of the core network layer as input, and calculate the membership probability of the predicted traffic belonging to each energy consumption level based on a preset energy consumption level threshold parameter through a differentiable fuzzy membership function, as the energy consumption level prediction result; wherein, the threshold parameter can be fine-tuned during model training.

8. An energy consumption classification monitoring device that combines flow direction prediction, characterized in that, include: Memory for storing energy consumption classification monitoring programs that incorporate flow direction prediction; A processor, configured to implement the steps of the energy consumption classification monitoring method according to any one of claims 1-6 when executing the energy consumption classification monitoring program that incorporates flow direction prediction.

9. A computer-readable medium storing a computer program, characterized in that, The readable medium stores an energy consumption classification monitoring program that incorporates flow direction prediction. When the energy consumption classification monitoring program that incorporates flow direction prediction is executed by a processor, it implements the steps of the energy consumption classification monitoring method that incorporates flow direction prediction as described in any one of claims 1-6.

Images