Energy consumption abnormity real-time detection method and system for smart energy

Abstract

The invention provides an energy consumption abnormity real-time detection method and system for smart energy, and relates to the technical field of computers, in order to improve the accuracy of abnormal energy consumption node detection, the method comprises the following steps: obtaining energy consumption data and energy flow data of each energy consumption node in a smart energy system, extracting a spatial interaction feature matrix according to the energy consumption data and the energy flow data; determining abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix to obtain an abnormal candidate node set; extracting an energy anomaly propagation feature vector corresponding to the anomaly candidate node set; and comparing the similarity between the energy anomaly propagation feature vector and an energy normal propagation feature vector corresponding to a normal energy consumption pattern library through a pre-established normal energy consumption pattern library by adopting a pattern matching algorithm, and if the similarity is lower than a third preset threshold, determining that the anomaly candidate node is a real anomaly, and generating a final anomaly detection result.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and system for real-time detection of energy consumption anomalies for smart energy. Background Art

[0002] Energy is the material basis of human activities and the most basic driving force for the development and economic growth of the entire world. There are many types of energy, but in process industries they can generally be divided into five categories: water, air, gas, electricity and steam engines. It is driven by these five energy sources that a factory can transform the input raw materials and human resources into corresponding products. As an important component of production costs, enterprises have increasingly higher requirements for various energy management. In order to gain a firm foothold and stand out in the fierce market competition, in addition to increasing revenue, they must also save costs. Therefore, the requirements for energy management are crucial, whether from the perspective of human resource protection, enterprise cost management, or improving enterprise operating efficiency.

[0003] However, current energy anomaly detection methods face significant limitations in complex systems. Traditional solutions often rely on single-dimensional statistical analysis or rule-based thresholds, making them difficult to adapt to complex scenarios involving dynamic interactions between multiple subsystems. In particular, they fail to capture the characteristics of energy flow in both time and space, resulting in low anomaly identification accuracy and high false positive rates. These methods often overlook the dynamic process of energy transfer, making it difficult to detect potential energy consumption anomalies in a timely manner. Summary of the Invention

[0004] Based on the above technical problems, the present application provides a real-time detection method and system for energy consumption anomalies for smart energy, so as to improve the accuracy of abnormal energy consumption node detection.

[0005] In the first aspect, the present application provides a real-time detection method for energy consumption anomalies of smart energy, which is applied to a smart energy system, wherein the smart energy system includes multiple energy-consuming nodes. The method includes: obtaining energy consumption data and energy flow data of each energy-consuming node in the smart energy system, and extracting a spatial interaction feature matrix based on the energy consumption data and energy flow data; the spatial interaction feature matrix is used to reflect the energy flow connectivity and energy flow intensity between each energy-consuming node; based on the spatial interaction feature matrix, abnormal candidate nodes in the smart energy system are determined to obtain an abnormal candidate node set; the energy flow connectivity of the abnormal candidate nodes is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold; energy abnormal propagation feature vectors corresponding to the abnormal candidate node set are extracted; the energy abnormal propagation feature vectors are used to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node; using a pattern matching algorithm, through a pre-established normal energy consumption pattern library, the energy abnormal propagation feature vector is compared with the energy normal propagation feature vector corresponding to the normal energy consumption pattern library for similarity; if the similarity is lower than a third preset threshold, the abnormal candidate node is determined to be a true abnormality, and a final abnormality detection result is generated.

[0006] In one possible implementation, a spatial interaction feature matrix is extracted based on energy consumption data and energy flow data, including: using a graph convolution algorithm to construct an energy transfer network between energy-consuming nodes based on the trend slope and mutation time point of the energy consumption data, and the flow intensity and flow angle in the energy flow data, and generating a weighted adjacency matrix that describes the flow connectivity and flow intensity between nodes to obtain the spatial interaction feature matrix.

[0007] In one possible implementation, based on the spatial interaction feature matrix, abnormal candidate nodes in the smart energy system are determined, including: obtaining the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, using a vector decomposition method to calculate the deviation vector, if the modulus of the deviation vector is greater than a fourth preset threshold, it is marked as a node to be optimized, and a set of nodes to be optimized is obtained; for the set of nodes to be optimized, a gradient calculation method is used to calculate the gradient amplitude and flow angle of each node, and a node feature vector containing the gradient amplitude and flow angle is obtained through vector synthesis to obtain a set of node feature vectors; based on the set of node feature vectors, a graph analysis method is used to calculate the gradient distribution uniformity and flow connectivity between nodes, and an energy balance vector containing distribution uniformity and connectivity characteristics is generated to obtain an energy balance vector set; if the connectivity feature of a vector in the energy balance vector set is less than the fifth preset threshold, the abnormal candidate nodes are screened by an anomaly detection algorithm to obtain an abnormal candidate node set.

[0008] In one possible implementation, an energy anomaly propagation feature vector corresponding to an abnormal candidate node set is extracted, including: for the abnormal candidate node set, using a gradient field modeling method, combining the volatility amplitude and periodic amplitude of energy consumption data, and the gradient amplitude and flow angle of energy flow data, calculating the local extreme value of the gradient and the flow bifurcation point of each node in the time and space dimensions, and generating a gradient vector field; using a gradient mutation detection algorithm to detect whether there is a gradient mutation point in the gradient vector field, if the gradient amplitude change rate of any area in the gradient vector field is greater than a sixth preset threshold, it is marked as an abnormal flow point, and an abnormal point position set is obtained; based on the gradient mutation points in the abnormal point position set, combined with an energy path tracing algorithm, the flow bifurcation point and flow intensity between the abnormal point and the adjacent node are analyzed, the propagation direction of the abnormal flow and the uniformity of the gradient distribution are calculated, and the energy anomaly propagation feature vector is obtained.

[0009] The technical solution provided by this application brings at least the following beneficial effects: (1) This application discloses a real-time detection method for energy consumption anomalies in smart energy. It combines energy consumption data in the time dimension with energy flow data in the space dimension to extract a spatial interaction feature matrix that reflects the connectivity and intensity of energy flow between energy-consuming nodes. By constructing a gradient vector field in the time and space dimensions and combining it with the energy balance analysis of the AC nodes, it extracts key features and tracks the energy transfer process between subsystems to accurately identify gradient mutations or abnormal flow points, thereby finding abnormal energy consumption nodes. Compared with traditional solutions that rely on single-dimensional statistical analysis or rule thresholds and are difficult to adapt to complex scenarios with dynamic interactions between multiple subsystems, this application comprehensively monitors the smart energy system in the time and space dimensions, can promptly detect abnormal flows, and improve system operation efficiency and safety.

[0010] (2) Based on the trend slope and mutation time points of energy consumption data, as well as the flow intensity and flow angle in energy flow data, this application uses a graph convolution algorithm to construct an energy transfer network between energy-consuming nodes, generate a weighted adjacency matrix describing the flow connectivity and flow intensity between nodes, and obtain a spatial interaction feature matrix. The spatial interaction feature matrix is a mathematical tool used to describe the interaction relationship between different regions or entities in geographic space. Its core lies in quantifying the interaction between spatial units, and thus can well reveal the inherent laws and dynamic characteristics of spatial structure.

[0011] (3) With the development of digital twins and urban computing technologies, the spatial interaction feature matrix will evolve towards high resolution, multimodality, and real-time. This application determines abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix, providing more accurate decision support for the screening of abnormal candidate nodes.

[0012] (4) This application combines the node balancing algorithm to determine abnormal candidate nodes and uses the gradient field modeling method to generate a dynamic energy flow diagram. The gradient mutation detection algorithm is used to mark abnormal flow points, and the energy path tracking algorithm is used to analyze the abnormal propagation characteristics. Finally, a pattern matching algorithm is used to compare the similarity between the abnormal propagation characteristics and the normal pattern, confirm the true anomaly, and generate the final detection result. This application realizes comprehensive monitoring of the smart energy system, can timely detect abnormal flow directions, and improve the system operation efficiency and safety.

[0013] In a second aspect, the present application provides a real-time detection system for energy consumption anomalies for smart energy, the system including an energy consumption anomaly detection device, the device including a processing unit and a determination unit; the processing unit is used to obtain energy consumption data and energy flow data of each energy consumption node in the smart energy system, and extract a spatial interaction feature matrix based on the energy consumption data and energy flow data; the spatial interaction feature matrix is used to reflect the energy flow connectivity and energy flow intensity between each energy consumption node; the determination unit is used to determine the abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix to obtain an abnormal candidate node set; the energy flow connectivity of the abnormal candidate nodes is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold; the processing unit is also used to extract an energy anomaly propagation feature vector corresponding to the abnormal candidate node set; the energy anomaly propagation feature vector is used to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node; the determination unit is also used to use a pattern matching algorithm to compare the similarity of the energy anomaly propagation feature vector with the energy normal propagation feature vector corresponding to the normal energy consumption pattern library through a pre-established normal energy consumption pattern library. If the similarity is lower than a third preset threshold, the abnormal candidate node is determined to be a true anomaly, and a final anomaly detection result is generated.

[0014] In one possible implementation, the processing unit is specifically used to: construct an energy transfer network between energy-consuming nodes using a graph convolution algorithm based on the trend slope and mutation time point of the energy consumption data, as well as the flow intensity and flow angle in the energy flow data, generate a weighted adjacency matrix describing the flow connectivity and flow intensity between nodes, and obtain a spatial interaction feature matrix.

[0015] In one possible implementation, the determination unit is specifically used to: obtain the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, use the vector decomposition method to calculate the deviation vector, if the modulus of the deviation vector is greater than the fourth preset threshold, mark it as a node to be optimized, and obtain a set of nodes to be optimized; for the set of nodes to be optimized, use the gradient calculation method to calculate the gradient amplitude and flow angle of each node, obtain the node feature vector containing the gradient amplitude and flow angle through vector synthesis, and obtain a set of node feature vectors; based on the set of node feature vectors, use the graph analysis method to calculate the gradient distribution uniformity and flow connectivity between nodes, generate an energy balance vector containing distribution uniformity and connectivity characteristics, and obtain an energy balance vector set; if the connectivity feature of a vector in the energy balance vector set is less than the fifth preset threshold, screen the abnormal candidate nodes through the anomaly detection algorithm to obtain an abnormal candidate node set.

[0016] In one possible implementation, the processing unit is specifically configured to: for a set of abnormal candidate nodes, use a gradient field modeling method, combine the volatility amplitude and periodic amplitude of energy consumption data, and the gradient amplitude and flow angle of energy flow data, calculate the local extreme value and flow bifurcation point of the gradient of each node in the time and space dimensions, and generate a gradient vector field; use a gradient mutation detection algorithm to detect whether there is a gradient mutation point in the gradient vector field; if the gradient amplitude change rate of any area in the gradient vector field is greater than a sixth preset threshold, it is marked as an abnormal flow point, and a set of abnormal point positions is obtained; based on the gradient mutation points in the abnormal point position set, combine the energy path tracing algorithm to analyze the flow bifurcation point and flow intensity between the abnormal point and the adjacent nodes, calculate the propagation direction of the abnormal flow and the uniformity of the gradient distribution, and obtain the energy anomaly propagation feature vector.

[0017] In a third aspect, the present application provides an electronic device comprising: a processor and a memory; the memory stores instructions executable by the processor; when the processor is configured to execute the instructions, the electronic device implements the method of the first aspect described above.

[0018] In a fourth aspect, the present application provides a computer program product, which, when running in an electronic device, enables the electronic device to execute the method related to the above-mentioned first aspect to implement the method of the above-mentioned first aspect.

[0019] In a fifth aspect, the present application provides a computer-readable storage medium, which includes: software instructions; when the software instructions are executed in an electronic device, the electronic device implements the method of the first aspect above.

[0020] The beneficial effects of the second to fifth aspects mentioned above can be referred to the first aspect and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A schematic diagram of the structure of a smart energy system provided in an embodiment of the present application; Figure 2 A schematic diagram of the composition of an electronic device provided in an embodiment of the present application; Figure 3 A flowchart of a method for real-time detection of abnormal energy consumption in smart energy provided in an embodiment of the present application; Figure 4 A schematic diagram of the composition of the energy consumption anomaly detection device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable ordinary people in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0024] It should be noted that the terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0025] In addition, in the description of the embodiments of this application, unless otherwise specified, " / " means or. For example, A / B can mean A or B. "And / or" in this article is simply a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of this application, "plurality" means two or more than two.

[0026] Before explaining the embodiments of the present application in detail, some relevant terms and related technologies involved in the embodiments of the present application are first introduced.

[0027] Energy management is a key pillar of sustainable development for modern industry and cities, directly impacting resource efficiency and environmental protection. Smart energy systems provide key support for achieving a low-carbon economy by monitoring and optimizing energy consumption in real time.

[0028] However, current energy anomaly detection methods face significant limitations in complex systems. Traditional solutions often rely on single-dimensional statistical analysis or rule-based thresholds, making them difficult to adapt to complex scenarios involving dynamic interactions between multiple subsystems. In particular, they fail to adequately capture energy flow characteristics in the temporal and spatial dimensions, resulting in low anomaly identification accuracy and high false positive rates. These methods often overlook the dynamic process of energy transfer, making it difficult to promptly detect potential energy consumption anomalies. The core challenge lies in effectively characterizing the dynamic changes in energy consumption in the temporal and spatial dimensions, and extracting key features from complex energy transfer networks to identify anomalies.

[0029] In the temporal dimension, fluctuations in energy consumption may vary due to load changes or equipment failures. In the spatial dimension, the energy flow between subsystems is nonlinear and heterogeneous, making it difficult to construct a unified feature description model using traditional methods.

[0030] Furthermore, energy balance analysis at AC nodes involves multivariable coupling, and existing technologies lack robustness when dealing with sudden gradient changes or abnormal flow directions, making it difficult to accurately locate anomalies. These unresolved technical issues directly restrict the anomaly detection capabilities of smart energy systems in highly dynamic environments, leading to the unique challenge of achieving efficient and accurate anomaly detection in complex systems.

[0031] In view of the above problems, an embodiment of the present application provides a real-time detection method for energy consumption anomalies for smart energy. By constructing a gradient vector field in the time and space dimensions, combined with the energy balance analysis of the AC node, key features are extracted and the energy transfer process between subsystems is tracked to accurately identify gradient mutations or abnormal flow points, and then discover abnormal energy consumption nodes.

[0032] The following describes in detail the method for real-time detection of energy consumption anomalies for smart energy provided by an embodiment of the present application in conjunction with the accompanying drawings.

[0033] The method for real-time detection of abnormal energy consumption for smart energy provided in the embodiment of the present application can be applied to smart energy systems. Figure 1 A structural diagram of the smart energy system is shown in FIG. Figure 1As shown, the smart energy system 10 includes an energy consumption anomaly detection device 11 and multiple energy consumption nodes 12. The energy consumption nodes 12 are connected to each other via wired or wireless communication. The energy consumption anomaly detection device 11 is connected to the multiple energy consumption nodes 12 via wired or wireless communication. Specifically, the energy consumption anomaly detection device 11 can be connected to multiple energy consumption nodes individually or to a single energy consumption node, although this is not a limitation in the present embodiment.

[0034] The energy consumption anomaly detection device 11 can be any electronic device with data processing capabilities. For example, the energy consumption anomaly detection device 11 can be a server, a computer, or a server cluster consisting of multiple servers. In some embodiments, the server cluster can also be a distributed cluster. Optionally, the server can be a central server, and the server can also be implemented on a cloud platform. For example, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud, and a multi-cloud, or any combination thereof. This embodiment of the present application is not limited to this.

[0035] The execution entity of the real-time energy consumption anomaly detection method for smart energy provided in the embodiments of the present application can be the aforementioned energy consumption anomaly detection device 11. As described above, the energy consumption anomaly detection device 11 can be an electronic device with data processing capabilities, such as a computer or server. Alternatively, the energy consumption anomaly detection device 11 can be a processor (e.g., a central processing unit (CPU)) in the aforementioned electronic device; or an application (APP) installed in the aforementioned electronic device with model training capabilities; or further, the energy consumption anomaly detection device 11 can be a functional module in the aforementioned electronic device with model training capabilities, etc. This embodiment of the present application is not limited to this.

[0036] For simplicity of description, the energy consumption anomaly detection device 11 is taken as an electronic device as an example for introduction.

[0037] Figure 2 This is a schematic diagram of the composition of the electronic device provided in the embodiment of the present application. Figure 2 As shown, the electronic device may include: a processor 20 , a memory 21 , a communication line 22 , a communication interface 23 , and an input / output interface 24 .

[0038] The processor 20 , the memory 21 , the communication interface 23 and the input / output interface 24 may be connected via a communication line 22 .

[0039] The processor 20 is used to execute the instructions stored in the memory 21 to implement the fault analysis method provided in the following embodiments of the present application. The processor 20 can be a CPU, a general-purpose processor network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller (MCU), a programmable logic device (PLD), or any combination thereof. The processor 20 can also be any other device with processing functions, such as a circuit, a device, or a software module, which is not limited in the embodiments of the present application. In one example, the processor 20 can include one or more CPUs, such as Figure 2 As an optional implementation, the electronic device may include multiple processors, for example, in addition to the processor 20, it may also include a processor 25 ( Figure 2 The dashed line is shown as an example).

[0040] Memory 21 is used to store instructions. For example, the instructions can be computer programs. Optionally, the memory 21 can be a read-only memory (ROM) or other type of static storage device that can store static information and / or instructions, or a random access memory (RAM) or other type of dynamic storage device that can store information and / or instructions. It can also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium, or other magnetic storage device, etc., and the embodiments of the present application are not limited to this.

[0041] It should be noted that the memory 21 may exist independently of the processor 20 or may be integrated with the processor 20. The memory 21 may be located inside the electronic device or outside the electronic device, which is not limited in the embodiment of the present application.

[0042] The communication line 22 is used to transmit information between the various components included in the electronic device.

[0043] Communication interface 23 is used to communicate with other devices or other communication networks. Such other communication networks may include Ethernet, radio access networks (RAN), wireless local area networks (WLAN), etc. Communication interface 23 may be a module, circuit, transceiver, or any other device capable of communication.

[0044] The input / output interface 24 is used to implement human-computer interaction between a user and the electronic device, for example, to implement action interaction or information interaction between the user and the electronic device.

[0045] For example, the input / output interface 24 may be a mouse, keyboard, display screen, or touch screen screen, etc. Action interaction or information interaction between a user and the electronic device may be achieved through the mouse, keyboard, display screen, or touch screen screen, etc.

[0046] It should be noted that Figure 2 The structure shown in the figure does not constitute a limitation on the electronic device, except Figure 2 In addition to the components shown, the electronic device may include more or fewer components than shown, or a combination of certain components, or a different arrangement of components.

[0047] The following introduces the real-time detection method for energy consumption anomalies for smart energy provided in an embodiment of the present application.

[0048] Figure 3 The flowchart of the method for real-time detection of abnormal energy consumption of smart energy provided in the embodiment of the present application is shown in FIG. Figure 2 The electronic device of the hardware structure shown is executed as Figure 3 As shown, the method includes S301 to S304.

[0049] S301. Obtain energy consumption data and energy flow data of each energy-consuming node in the smart energy system, and extract a spatial interaction feature matrix based on the energy consumption data and energy flow data.

[0050] Among them, the spatial interaction feature matrix is used to reflect the energy flow connectivity and energy flow intensity between energy-consuming nodes.

[0051] As a possible implementation method, electronic devices can obtain real-time energy consumption data (also called energy consumption data) and energy flow data from each energy-consuming node in the smart energy system through a sensor network.

[0052] Specifically, in smart energy systems, real-time energy consumption and energy flow data are obtained through sensor networks, and the real-time and accuracy of data collection must be ensured.

[0053] For example, sensors deployed at substations and user terminals in smart grids monitor parameters like voltage and current in real time, collecting data every second and generating datasets containing timestamps and consumption figures. Time series databases like InfluxDB can efficiently store this high-frequency data.

[0054] Preferably, query efficiency is optimized through time indexing to ensure low latency for subsequent analysis. When processing energy flow data, the space vector analysis tool can vectorize the energy flow of each node in the power grid into a vector form.

[0055] For example, the flow intensity represents the power transmitted per unit time, and the flow angle reflects the geometric distribution of energy flow. A second dataset containing these features is generated. For the initial time series dataset, Fourier transform is used to analyze the periodic fluctuations of energy consumption.

[0056] Specifically, a Fourier transform of daily electricity consumption data from an industrial park reveals 24-hour periodic fluctuations with an amplitude of 200kW. If the amplitude exceeds a preset threshold of 150kW, a peak detection algorithm is used to identify daily peaks in electricity consumption, such as the peak time at 2 pm. This generates a time series feature set containing the amplitude of the fluctuation, the periodic frequency, and the peak time.

[0057] It can be understood that this feature set can be used to predict peak electricity consumption, optimize energy scheduling, and reduce peak load pressure. In the processing of spatially distributed data sets, the vector decomposition algorithm decomposes the flow intensity and angle into orthogonal components.

[0058] For example, if the energy flow intensity at a node is 500kW and the angle is 45 degrees, the decomposition can clearly show the energy distribution in the north-south and east-west directions. If the intensity falls below the threshold of 100kW, an interpolation algorithm, such as Kriging, is used to estimate the missing data based on the intensity values of neighboring nodes to generate a complete set of spatial distribution features.

[0059] It should be noted that this method can improve data integrity and ensure the reliability of subsequent analysis. The matrix generation algorithm aligns the dimensions of the time series feature set and the spatial distribution feature set.

[0060] Preferably, dimensionality reduction is performed through principal component analysis to retain the main feature dimensions.

[0061] For example, the fluctuation amplitude and periodic frequency of the time series are aligned with the flow intensity and angle of the spatial distribution to generate a unified dimensional matrix. Data fusion algorithms such as weighted average fusion integrate all features to generate the initial energy state matrix.

[0062] In one embodiment, the matrix can reflect the overall operating status of a regional power grid at a specific time, such as the peak energy flow is concentrated in the industrial area, with an intensity of 800kW and a fluctuation amplitude of 250kW.

[0063] For example, this matrix can be used to monitor the state of the power grid in real time, predict potential failures, and improve energy efficiency.

[0064] In one possible implementation, the integrated application of the above technical solutions can significantly improve the operating efficiency of the smart energy system.

[0065] For example, peak load can be predicted through time series feature sets, and power generation plans can be adjusted in advance; energy distribution can be optimized through spatial distribution feature sets to reduce transmission losses.

[0066] Understandably, these approaches collectively support efficient management of energy systems, reduce operating costs, and improve system stability.

[0067] Specifically, in smart energy systems, extracting time series data from the initial energy state matrix and performing feature analysis is an important step in optimizing energy management.

[0068] For example, the initial energy state matrix contains voltage, current, and power data for an industrial park's power grid, with a time resolution of once per minute. The wavelet transform algorithm decomposes time series data by breaking the signal into components of different frequencies, preserving low-frequency trends and high-frequency fluctuations.

[0069] Specifically, the low-frequency trend component reflects the overall trend of daily electricity consumption, such as the gradual increase in electricity consumption on weekdays; the high-frequency fluctuation component captures instantaneous changes, such as power spikes caused by equipment startup.

[0070] In one embodiment, power data from a substation is subjected to a wavelet transform to extract the low-frequency component of the daily average power variation, showing that power is stable at around 500 kW from 8:00 AM to 5:00 PM. The high-frequency component reveals hourly power fluctuations of approximately 50 kW, forming a first feature set. For this first feature set, an autoregressive moving average algorithm is used to assess the stationarity of the time window.

[0071] It should be noted that the stationarity index measures whether the data has a stable mean and variance. If the index is lower than the threshold of 0.8, the data may be missing or abnormal.

[0072] For example, due to a sensor failure at a certain node, the power data for some time periods is missing, and the stability index drops to 0.6.

[0073] Preferably, a Lagrange interpolation algorithm is used to estimate missing data based on the power values at the preceding and following time points, such as interpolating the missing 10 power values to 480 kW, to generate an optimized second feature set. Based on the second feature set, a fast Fourier transform is used to analyze the periodic frequency and amplitude.

[0074] As can be understood, the algorithm converts time series data into the frequency domain and identifies the main periodic patterns.

[0075] In one possible implementation, a user's electricity usage data, after fast Fourier transformation, shows a 24-hour periodic amplitude of 150 kW and a 12-hour periodic amplitude of 80 kW. If the amplitude exceeds a threshold of 100 kW, a peak detection algorithm is used to identify peak times. For example, if the power peak reaches 650 kW at 3 pm each day, a third feature set is generated, including the periodic frequency, amplitude, and peak time. A matrix generation algorithm then aligns the dimensions of the third feature set, integrating the fluctuation amplitude, periodic frequency, and peak time.

[0076] For example, align the 24-hour period amplitude of 150kW, the frequency of 1 / 24 hours, and the peak time of 15:00 with other features to generate a time dimension feature vector.

[0077] Specifically, this vector can be used to describe the operating characteristics of the power grid in a specific time window, such as the high load state on weekday afternoons.

[0078] It should be noted that dimension alignment ensures that the feature vector format is uniform, which is convenient for subsequent analysis.

[0079] In one embodiment, the above method generates high-precision feature vectors by decomposing and optimizing time series data, which helps to accurately identify power usage patterns.

[0080] For example, the eigenvector can be used to predict peak loads, adjust energy distribution in advance, and reduce grid pressure.

[0081] It is understandable that these methods jointly improve the reliability of data analysis and provide technical support for the efficient operation of smart energy systems.

[0082] As a possible implementation method, electronic devices can use graph convolution algorithms to construct an energy transfer network between energy-consuming nodes based on the trend slope and mutation time point of energy consumption data, as well as the flow intensity and flow angle in energy flow data, and generate a weighted adjacency matrix that describes the flow connectivity and flow intensity between nodes to obtain a spatial interaction feature matrix.

[0083] Specifically, in smart energy systems, extracting the time trend slope and mutation time point from the time dimension feature vector is the key to analyzing dynamic changes in energy.

[0084] For example, the time trend slope reflects the rate of change of power in an industrial park's power grid over a period of time, such as a 10 kW increase in power per hour. The time series segmentation method divides the data into windows by hour or day and calculates the slope of each window.

[0085] Preferably, if the slope exceeds a preset threshold of 15 kW / hour, the starting point of the window is marked as a sudden change time point, such as a sudden power increase at 9 am due to concentrated startup of equipment.

[0086] It should be noted that mutation time points are often associated with production scheduling or failures. After labeling, they form a time-dimensional feature set containing slope and mutation time information to facilitate subsequent analysis. Based on this time-dimensional feature set, flow intensity and flow angle are extracted from the spatial-dimensional data to capture the energy transfer characteristics between subsystems.

[0087] In one possible implementation, the flow intensity represents the amount of power transferred from a substation to a load node, such as 100 kW; the flow angle reflects the direction of transfer, such as a current phase angle of 30 degrees. Vector decomposition methods split the flow vector into intensity and angle components and analyze their dynamics.

[0088] For example, if the flow intensity at a node increases to 150kW during peak hours and the angle shifts to 45 degrees, a spatial flow feature matrix is generated to record the flow characteristics of each node. Using a graph convolution algorithm, the spatial flow feature matrix and node connectivity are combined to construct an energy transfer network between subsystems.

[0089] Specifically, graph convolution calculates the influence of energy transfer through the connection weights between nodes.

[0090] For example, a substation has strong connectivity with multiple load nodes, and its graph convolution weight is high, generating a weighted adjacency matrix that reflects the intensity of energy interaction between nodes.

[0091] In one embodiment, a weight value of 0.8 indicates that a node has a significant power impact on neighboring nodes, and the matrix clearly shows the network topology. By combining the weighted adjacency matrix with node energy distribution and subsystem interaction data, a spatial interaction feature matrix is generated.

[0092] It can be understood that the node energy distribution describes the power allocation of each node, such as a certain node accounts for 20% of the total power; the subsystem interaction data records the frequency of energy exchange between nodes.

[0093] For example, if the energy transfer efficiency between nodes is higher than 80%, the interaction is considered stable, and the final spatial interaction characteristic matrix is determined. This matrix integrates flow direction, weight, and interaction information to fully describe the spatial dynamic characteristics of the power grid.

[0094] For example, during weekday peak hours in an industrial park, the aforementioned method identified a sudden change at 10:00 AM, when the flow intensity increased to 200 kW. The graph convolution weights revealed a significant impact of the core substation on the load, and the spatial interaction feature matrix further revealed efficient energy distribution patterns. These features provided data support for optimized scheduling.

[0095] S302: Based on the spatial interaction feature matrix, determine abnormal candidate nodes in the smart energy system to obtain an abnormal candidate node set.

[0096] Among them, the energy flow connectivity of the abnormal candidate node is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold.

[0097] As a possible implementation method, the electronic device can obtain the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, and use the vector decomposition method to calculate the deviation vector. If the modulus of the deviation vector is greater than the fourth preset threshold, it is marked as a node to be optimized, and a set of nodes to be optimized is obtained. Furthermore, for the set of nodes to be optimized, the gradient calculation method is used to calculate the gradient amplitude and flow angle of each node, and the node feature vector containing the gradient amplitude and flow angle is obtained through vector synthesis to obtain a set of node feature vectors. Based on the set of node feature vectors, the electronic device can use the graph analysis method to calculate the gradient distribution uniformity and flow connectivity between nodes, generate an energy balance vector containing distribution uniformity and connectivity features, and obtain an energy balance vector set; if the connectivity feature of a vector in the energy balance vector set is less than the fifth preset threshold, the abnormal candidate node is screened by the anomaly detection algorithm to obtain an abnormal candidate node set.

[0098] Specifically, in smart energy systems, extracting node flow intensity and energy inflow / outflow deviation from the spatial interaction feature matrix is key to analyzing the dynamic balance of the power grid. The spatial interaction feature matrix records the energy transfer characteristics between nodes. For example, if the flow intensity of a node is 150kW, the energy inflow / outflow deviation reflects the difference in energy received and output by the node.

[0099] For example, if a substation has 200kW of power flowing in and 180kW of power flowing out, the deviation is 20kW. The vector decomposition method decomposes the deviation into a modulus and a direction. The modulus represents the magnitude of the deviation, while the direction reflects the energy flow trend.

[0100] Specifically, nodes whose module values exceed a preset threshold, such as 30kW, are marked as nodes to be optimized, and a set of nodes to be optimized is generated.

[0101] For example, in an industrial park power grid, a core substation was flagged as a node for optimization due to peak load fluctuations, with a deviation modulus of 40kW. This method helps identify nodes with uneven energy distribution. For the set of nodes to be optimized, a gradient calculation method is used to calculate the gradient magnitude and flow angle for each node. The gradient magnitude reflects the severity of the node's energy fluctuations, while the flow angle indicates the direction of energy transfer.

[0102] In one possible implementation, the gradient amplitude of a node is 25 kW / hour and the flow angle is 60 degrees. A node feature vector is generated by vector synthesis, which includes amplitude and angle information to form a node feature vector set.

[0103] It can be understood that the node feature vector set provides a structured data basis for subsequent analysis.

[0104] For example, a sudden increase in gradient amplitude at a load node due to device startup indicates a significant impact on grid stability. Based on the set of node feature vectors, graph analysis methods are used to calculate the uniformity of gradient distribution and flow connectivity between nodes. Gradient uniformity measures the degree of balance in energy changes between nodes, while flow connectivity reflects the closeness of energy transfer between nodes.

[0105] Preferably, if the gradient distribution uniformity in a certain area is high, it means that the energy distribution is stable.

[0106] In one embodiment, the flow connectivity between the core substation and surrounding load nodes reaches 0.9, indicating efficient energy transfer. This generates an energy balance vector, which includes distribution uniformity and connectivity characteristics, forming a set of energy balance vectors. These vectors provide a basis for assessing the overall stability of the power grid. If the connectivity characteristic of a vector in the set of energy balance vectors falls below a preset threshold, such as 0.7, an anomaly detection algorithm is used to identify candidate nodes with abnormal connectivity. The anomaly detection algorithm uses statistical analysis or machine learning to identify nodes with abnormal connectivity.

[0107] It should be noted that low connectivity may be caused by equipment failure or line aging.

[0108] For example, a remote load node has an aging line and its connectivity is only 0.5, so it is marked as an abnormal candidate node and an abnormal candidate node set is generated.

[0109] Specifically, the set of abnormal candidate nodes provides precise targets for subsequent maintenance.

[0110] For example, after detecting an anomaly at a node, the line condition can be checked first. This approach improves the efficiency of power grid anomaly detection.

[0111] In one example, during peak weekday hours, an industrial park identified a substation with an excessive deviation modulus. After calculating its gradient amplitude and flow angle, it was determined that its connectivity was below a threshold. Anomaly detection further confirmed the node as a candidate for anomaly, prompting maintenance personnel to inspect the equipment. These steps form a closed-loop analysis, ensuring the dynamic balance and operational efficiency of the power grid.

[0112] S303: Extract the energy anomaly propagation feature vector corresponding to the abnormal candidate node set.

[0113] Among them, the energy anomaly propagation feature vector is used to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node.

[0114] As a possible implementation, for a set of abnormal candidate nodes, the electronic device can use a gradient field modeling method, combining the volatility amplitude and periodic amplitude of energy consumption data with the gradient amplitude and flow angle of energy flow data, to calculate the local gradient extrema and flow bifurcation points of each node in the time and space dimensions, thereby generating a gradient vector field. Furthermore, the electronic device can use a gradient mutation detection algorithm to detect whether there are gradient mutation points in the gradient vector field. If the gradient amplitude change rate of any region in the gradient vector field exceeds a sixth preset threshold, it is marked as an abnormal flow point, thereby obtaining a set of abnormal point locations. Based on the gradient mutation points in the set of abnormal point locations, the electronic device combines an energy path tracing algorithm to analyze the flow bifurcation points and flow intensity between the abnormal point and adjacent nodes, calculate the propagation direction of the abnormal flow and the uniformity of the gradient distribution, and obtain an energy anomaly propagation feature vector.

[0115] Specifically, in smart energy systems, the analysis of time dimension eigenvectors is an important means to study the dynamic characteristics of power grids. Time fluctuations reflect the changing trend of node energy over time.

[0116] For example, by collecting power data from a substation, such as recording power values every minute, a time series is formed. This time series is decomposed into periodic components of different frequencies using Fourier transform to obtain a periodic amplitude series.

[0117] For example, a substation's periodic amplitude series shows a 24-hour periodic amplitude of 50kW, indicating significant daily load fluctuations. If the amplitude exceeds a preset threshold, such as 30kW, local extreme points in the time series are extracted to generate a local extreme point set.

[0118] It's important to note that these extreme points may correspond to peak or valley periods, such as the power surge in industrial parks when equipment is started up on weekday mornings. The gradient operator convolves the set of local extreme points with the spatial interaction feature matrix to generate a gradient magnitude distribution. The spatial interaction feature matrix records the energy transfer relationship between nodes, such as the power difference between a node and its neighbors. The gradient magnitude distribution reflects the spatial characteristics of energy variation.

[0119] In one possible implementation, the gradient amplitude of a core substation is 20kW / hour, indicating that its energy changes are relatively drastic. If the amplitude exceeds a threshold, such as 15kW / hour, it is marked as a candidate bifurcation point.

[0120] Specifically, bifurcation points can be caused by sudden load changes or line switching, such as abnormal energy flow at a node due to device startup. Using candidate bifurcation points and flow angles, vector field interpolation methods calculate the change in flow angle and construct a vector field structure. The flow angle represents the direction of energy transfer. For example, if the flow angle at a node suddenly changes from 45 degrees to 90 degrees, a rate of change of 50%, the bifurcation point is confirmed if the rate of change exceeds a threshold, such as 30%.

[0121] Preferably, the bifurcation point location may indicate a potential discontinuity point in the energy flow in the power grid.

[0122] For example, a substation's load fluctuations caused rapid changes in flow angle, identifying it as a bifurcation point and providing a basis for subsequent optimization. Flow characteristics were extracted from the vector field structure, and a streamline tracing algorithm was used to plot the streamline distribution, creating a dynamic graph. Flow characteristics describe the energy transfer path between nodes.

[0123] In one embodiment, streamline tracing reveals dense streamlines around a core substation in an industrial park, indicating concentrated energy output. Dynamic graphs visually depict energy flow paths, such as streamlines diverging from the substation toward load nodes.

[0124] It can be understood that the dynamic graph provides a visual tool for analyzing the operation status of the power grid and helps to identify bottleneck nodes.

[0125] For example, areas with sparse streamlines may indicate low transmission efficiency, prompting operators to optimize line layout. This method combines the characteristics of time and space to comprehensively characterize the dynamic characteristics of the power grid and provide data support for operational optimization.

[0126] Specifically, in smart energy systems, gradient vector field analysis of dynamic energy flow diagrams is an important method for studying the dynamic characteristics of power grids. The amplitude distribution of the gradient vector field reflects the intensity of node energy changes.

[0127] For example, power data from a substation is collected every five minutes to form a time series. The gradient amplitude at each node is then calculated using a differential operator. This calculation calculates the rate of change by dividing the power difference between adjacent time points by the time interval.

[0128] In a possible implementation, the amplitude change rate of a substation is 25kW / minute, and the preset threshold is 15kW / minute. If the threshold is exceeded, it is marked as a candidate mutation point.

[0129] It should be noted that candidate mutation points may correspond to sudden load changes or equipment switching in the power grid, such as the rapid increase in power caused by the startup of large equipment in an industrial park. For this set of candidate mutation points, a vector field interpolation method is used to calculate the rate of change of the flow angle. The flow angle represents the direction of energy transfer, and the interpolation method estimates the angle change using vector field data from neighboring nodes.

[0130] Specifically, the flow angle of a candidate mutation point changes from 30 degrees to 80 degrees, with a change rate of 55%. The preset threshold is 30%. If the threshold is exceeded, it is confirmed as a mutation point.

[0131] In one embodiment, a mutation point may be caused by a line switch or fault, such as a short circuit at a node causing abnormal energy flow. The set of mutation points provides key location information for subsequent analysis. The gradient vector field is gridded using a region partitioning algorithm, dividing the vector field into regular grids and assigning a unique identifier to each grid.

[0132] Preferably, a certain mutation point is located in an area with a grid ID of G-23, and an area annotation set is generated.

[0133] As you can see, gridding helps locate the spatial distribution of mutation points, such as identifying high-load areas within an industrial park. The regional annotation set provides the basis for subsequent clustering. The coordinates of the mutation points are extracted from the regional annotation set. The K-means clustering algorithm is used to cluster the mutation points, generating a set of outlier locations.

[0134] For example, the clustering results show that three mutation points in a certain industrial park are concentrated in the same area, with coordinates of (100, 200), (110, 210), and (105, 205).

[0135] In one embodiment, these abnormal points may correspond to high-voltage line load concentration areas near the power station. Cluster analysis helps to find abnormal energy flow concentration areas in the power grid, providing a basis for optimizing scheduling.

[0136] It is understandable that the set of abnormal point locations intuitively reflects the potential operational risk points in the power grid, such as areas with high incidence of line overload or equipment failure, providing data support for operation and maintenance decisions.

[0137] Specifically, in smart energy systems, processing of abnormal point location sets is an important step in optimizing grid operation.

[0138] For example, coordinate information is extracted from a set of outlier locations. For example, suppose an outlier is detected near a substation with coordinates (150, 300) and (160, 310). Using a grid partitioning algorithm, the gradient vector field is divided into regular grids, each 50×50 meters in size. This generates a set of grids, each assigned a unique identifier, such as G-45 or G-46. The outlier (150, 300) is mapped to grid G-45, generating a set of mappings between outliers and region identifiers.

[0139] It can be understood that this mapping facilitates the subsequent analysis of the spatial distribution of outliers.

[0140] Specifically, based on the mapping set, an energy path tracing algorithm is used to calculate the energy flow trajectory of the subsystem where the outlier point is located. For example, if the outlier point (150, 300) is located in an industrial park subsystem, the tracing algorithm analyzes the power data and plots the energy flow trajectory from that point. It finds that the trajectory extends to the intersection of adjacent subsystems, forming a flow bifurcation point, such as the coordinates (180, 320).

[0141] It should be noted that the bifurcation point reflects the redistribution of energy flow, which may be caused by load changes.

[0142] Preferably, if the flow intensity at the bifurcation point is 30kW, which exceeds a preset threshold of 20kW, it is confirmed as a critical bifurcation point.

[0143] In one embodiment, a vector field interpolation method is used to calculate the propagation direction and gradient amplitude distribution at key bifurcation points. Assuming the flow angle at the bifurcation point (180, 320) changes from 45 degrees to 60 degrees, the interpolation method estimates the propagation direction using data from neighboring nodes, generating a propagation direction set. Simultaneously, the gradient amplitude distribution shows an increase in amplitude at this point from 20 kW / m to 35 kW / m, forming an amplitude distribution set.

[0144] Understandably, this reflects a localized concentration of energy flow and may indicate abnormal equipment operation.

[0145] For example, a K-means clustering algorithm can be used to extract features from the propagation direction and amplitude distribution set to generate a set of abnormal propagation feature vectors. Suppose cluster analysis finds that the propagation directions of multiple bifurcation points in a subsystem are concentrated between 60 and 70 degrees, and the gradient amplitude is highly uniform, indicating that the energy flow anomaly is consistent.

[0146] Preferably, this set of feature vectors can be used to identify potential high-risk areas in the power grid and provide a basis for operation and maintenance decisions.

[0147] In one possible implementation, the above method forms a complete analysis chain from outlier location to feature extraction through multi-dimensional analysis.

[0148] It should be noted that the analysis of bifurcation points and propagation characteristics helps to detect hidden dangers in power grid operation in advance, such as line overload or equipment failure, and improve system stability.

[0149] S304. Using a pattern matching algorithm, the normal energy consumption pattern library is pre-established to compare the similarity between the energy abnormal propagation feature vector and the normal energy propagation feature vector corresponding to the normal energy consumption pattern library. If the similarity is lower than the third preset threshold, the abnormal candidate node is determined to be a real abnormality, and the final abnormality detection result is generated.

[0150] Specifically, in smart energy systems, extracting flow intensity values and gradient distribution data from the set of abnormal propagation feature vectors is the core link in analyzing the abnormal propagation laws of power grids.

[0151] For example, the flow intensity value reflects the power of the energy flow at the outlier point, while the gradient distribution data describes the rate of change of the energy flow in space. For example, suppose a substation detects an outlier point with a flow intensity of 25 kW. The gradient distribution in the adjacent area shows a gradual change from 15 kW / m to 30 kW / m. Vector decomposition algorithms are used to process eigenvectors, aiming to break down complex eigenvectors into independent components for easier analysis.

[0152] Specifically, the vector decomposition algorithm separates the flow intensity and gradient distribution, generating a set of flow intensity values such as {20kW, 25kW, 30kW} and a set of gradient distribution values such as {15kW / m, 22kW / m, 30kW / m}. This decomposition method facilitates capturing the local characteristics of anomaly propagation.

[0153] Preferably, based on the above set, a fast Fourier transform algorithm is used to analyze the periodic frequency of abnormal propagation. Fast Fourier transform identifies the periodic pattern of energy flow changes by converting time domain signals into frequency domain.

[0154] For example, after analyzing the flow intensity set, it is found that there is a power fluctuation every 10 seconds near the abnormal point, and the corresponding periodic frequency is 0.1 Hz, generating a periodic frequency set {0.1 Hz, 0.15 Hz}.

[0155] It should be noted that if the similarity between this frequency set and the typical frequencies in the normal pattern library, such as {0.2 Hz, 0.25 Hz}, is lower than the preset threshold of 0.8, it indicates that the abnormal propagation has atypical periodicity, which may be caused by equipment failure or load mutation.

[0156] In one embodiment, if the periodic frequency is abnormal, the eigenvector is further processed using a linear regression algorithm to calculate a set of trend slopes of the abnormal propagation. The linear regression algorithm estimates the long-term trend of the abnormal propagation by fitting the time series data of the eigenvector.

[0157] For example, by analyzing the change of flow intensity over time, it is found that the intensity value increases from 20kW to 30kW within 1 hour, with a slope of about 0.028kW / min, generating a trend slope set {0.028kW / min, 0.032kW / min}.

[0158] Understandably, this trend analysis helps determine if the anomaly is continuing to worsen.

[0159] For example, a pattern matching algorithm compares a trend slope set with slope data in a normal pattern library, such as {0.01kW / min, 0.015kW / min}. If the similarity falls below a threshold of 0.7, the anomaly is confirmed to be a true anomaly. The pattern matching algorithm quantifies the difference between anomalies and normal patterns by calculating the Euclidean distance or cosine similarity of the slope set.

[0160] Specifically, let's assume that the slope of an outlier point in a subsystem of an industrial park differs significantly from the normal pattern. A set of detection results is generated, and the outlier point is marked as a "true anomaly" with an identifier such as A-001. This labeling makes it easier for subsequent operations and maintenance personnel to quickly locate the problem area.

[0161] In one possible implementation, the above analysis forms a complete analysis chain from feature vector decomposition to anomaly detection.

[0162] Preferably, the combined analysis of periodic frequency and trend slope can verify the authenticity of the anomaly from multiple dimensions.

[0163] For example, an abnormal point exhibits both an atypical frequency and an abnormal trend slope, indicating that it may be caused by line overload rather than temporary interference.

[0164] It should be noted that this multi-faceted analysis improves the accuracy of anomaly detection and provides a reliable basis for optimized power grid operation.

[0165] The technical solution provided by the embodiment of the present application brings at least the following beneficial effects: combining energy consumption data in the time dimension with energy flow data in the space dimension, extracting a spatial interaction feature matrix that reflects the connectivity and intensity of energy flow between energy-consuming nodes, and by constructing a gradient vector field in the time and space dimensions, combined with the energy balance analysis of the AC nodes, extracting key features and tracking the energy transfer process between subsystems, so as to accurately identify gradient mutations or abnormal flow points, and then find abnormal energy-consuming nodes. Compared with traditional solutions that rely on single-dimensional statistical analysis or rule thresholds and are difficult to adapt to complex scenarios of dynamic interactions between multiple subsystems, the present application comprehensively monitors the smart energy system in the time and space dimensions, can promptly detect abnormal flows, and improve system operation efficiency and safety.

[0166] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to realize the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. It should be easy to realize that the technical goals in this field are combined with the units and algorithm steps of each example described in the embodiments disclosed herein, and the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional technical goals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0167] In an exemplary embodiment, the present application also provides an energy consumption anomaly detection device. Figure 4 This is a schematic diagram of the composition of the energy consumption anomaly detection device provided in the embodiment of the present application. Figure 4 As shown, the energy consumption anomaly detection device includes: a processing unit 401 and a determination unit 402.

[0168] The processing unit 401 is used to obtain energy consumption data and energy flow data of each energy-consuming node in the smart energy system, and extract a spatial interaction feature matrix based on the energy consumption data and energy flow data; the spatial interaction feature matrix is used to reflect the energy flow connectivity and energy flow intensity between each energy-consuming node; the determination unit 402 is used to determine the abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix to obtain an abnormal candidate node set; the energy flow connectivity of the abnormal candidate nodes is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold; the processing unit 401 is also used to extract the energy abnormal propagation feature vector corresponding to the abnormal candidate node set; the energy abnormal propagation feature vector is used to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node; the determination unit 402 is also used to use a pattern matching algorithm to compare the similarity between the energy abnormal propagation feature vector and the energy normal propagation feature vector corresponding to the normal energy consumption pattern library through a pre-established normal energy consumption pattern library. If the similarity is lower than a third preset threshold, the abnormal candidate node is determined to be a true abnormality, and a final abnormality detection result is generated.

[0169] In one possible implementation, the processing unit 401 is specifically used to: construct an energy transfer network between each energy-consuming node using a graph convolution algorithm based on the trend slope and mutation time point of the energy consumption data, as well as the flow intensity and flow angle in the energy flow data, generate a weighted adjacency matrix describing the flow connectivity and flow intensity between nodes, and obtain a spatial interaction feature matrix.

[0170] In one possible implementation, the determination unit 402 is specifically used to: obtain the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, use the vector decomposition method to calculate the deviation vector, if the modulus of the deviation vector is greater than the fourth preset threshold, mark it as a node to be optimized, and obtain a set of nodes to be optimized; for the set of nodes to be optimized, use the gradient calculation method to calculate the gradient amplitude and flow angle of each node, obtain the node feature vector containing the gradient amplitude and flow angle through vector synthesis, and obtain a set of node feature vectors; based on the set of node feature vectors, use the graph analysis method to calculate the gradient distribution uniformity and flow connectivity between nodes, generate an energy balance vector containing distribution uniformity and connectivity characteristics, and obtain an energy balance vector set; if the connectivity feature of a vector in the energy balance vector set is less than the fifth preset threshold, filter the abnormal candidate nodes through the anomaly detection algorithm to obtain an abnormal candidate node set.

[0171] In one possible implementation, the processing unit 401 is specifically configured to: for a set of abnormal candidate nodes, employ a gradient field modeling method, combine the volatility amplitude and periodic amplitude of energy consumption data, and the gradient amplitude and flow angle of energy flow data, calculate the local extreme value and flow bifurcation point of the gradient of each node in the time and space dimensions, and generate a gradient vector field; employ a gradient mutation detection algorithm to detect whether there are gradient mutation points in the gradient vector field; if the gradient amplitude change rate of any region in the gradient vector field is greater than a sixth preset threshold, mark it as an abnormal flow point, and obtain an abnormal point position set; and analyze the flow bifurcation points and flow intensity between the abnormal point and adjacent nodes in combination with an energy path tracing algorithm based on the gradient mutation points in the abnormal point position set, calculate the propagation direction of the abnormal flow and the uniformity of the gradient distribution, and obtain an energy anomaly propagation feature vector.

[0172] It should be noted that Figure 4 The module division described is illustrative and represents only one logical functional division. Actual implementations may employ different divisions. For example, two or more functions may be integrated into a single processing module. These integrated modules may be implemented as either hardware or software functional units.

[0173] In an exemplary embodiment, the present application also provides a computer-readable storage medium including software instructions, which, when executed on an electronic device, enables the electronic device to execute any one of the methods provided in the above embodiments.

[0174] In an exemplary embodiment, the present application also provides a computer program product including computer-executable instructions, which, when executed on an electronic device, enables the electronic device to execute any one of the methods provided in the above embodiments.

[0175] The above embodiments can be implemented in whole or in part through software, hardware, firmware, or any combination thereof. When implemented using a software program, they can be implemented in whole or in part in the form of a computer program product. This computer program product includes one or more computer-executable instructions. When the computer-executable instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are fully or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer-executable instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer-executable instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. Available media can include magnetic media (e.g., floppy disks, hard disks, tapes), optical media (e.g., DVDs), or solid-state drives (SSDs).

[0176] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0177] Although the present application has been described with reference to specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely illustrative of the present application as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, the present application is intended to include such modifications and variations as fall within the scope of the claims of the present application and their equivalents.

[0178] The above is only a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for real-time detection of energy consumption anomalies for smart energy, characterized in that: Applied to a smart energy system, the smart energy system includes multiple energy-consuming nodes, and the method includes: Obtaining energy consumption data and energy flow data of each energy-consuming node in the smart energy system, and extracting a spatial interaction feature matrix based on the energy consumption data and the energy flow data; the spatial interaction feature matrix is used to reflect the energy flow connectivity and energy flow intensity between each energy-consuming node; Based on the spatial interaction feature matrix, determining abnormal candidate nodes in the smart energy system to obtain an abnormal candidate node set; the energy flow connectivity of the abnormal candidate nodes is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold; Extracting an energy anomaly propagation feature vector corresponding to the abnormal candidate node set; the energy anomaly propagation feature vector is used to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node; A pattern matching algorithm is used to compare the similarity between the energy abnormality propagation feature vector and the energy normal propagation feature vector corresponding to the normal energy consumption pattern library through a pre-established normal energy consumption pattern library. If the similarity is lower than the third preset threshold, the abnormal candidate node is determined to be a real abnormality, and the final abnormality detection result is generated.

2. The method according to claim 1, characterized in that The extracting a spatial interaction feature matrix according to the energy consumption data and the energy flow data includes: Based on the trend slope and mutation time point of the energy consumption data, as well as the flow intensity and flow angle in the energy flow data, a graph convolution algorithm is used to construct an energy transfer network between each energy-consuming node, and a weighted adjacency matrix describing the flow connectivity and flow intensity between nodes is generated to obtain the spatial interaction feature matrix.

3. The method according to claim 1, characterized in that The determining of abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix includes: Obtaining the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, calculating the deviation vector using a vector decomposition method, and marking the node as a node to be optimized if the modulus of the deviation vector is greater than a fourth preset threshold, thereby obtaining a set of nodes to be optimized; For the set of nodes to be optimized, a gradient calculation method is used to calculate the gradient amplitude and flow angle of each node, and a node feature vector including the gradient amplitude and the flow angle is obtained through vector synthesis to obtain a set of node feature vectors; According to the node feature vector set, a graph analysis method is used to calculate the gradient distribution uniformity and flow connectivity between nodes, and an energy balance vector including the distribution uniformity and the connectivity characteristics is generated to obtain an energy balance vector set; If the connectivity feature of a certain vector in the energy balance vector set is less than a fifth preset threshold, abnormal candidate nodes are screened by an anomaly detection algorithm to obtain the abnormal candidate node set.

4. The method according to claim 1, wherein The extracting the energy anomaly propagation feature vector corresponding to the abnormal candidate node set includes: For the set of abnormal candidate nodes, a gradient field modeling method is used. Combining the volatility amplitude and period amplitude of the energy consumption data, and the gradient amplitude and flow angle of the energy flow data, the local extreme value and flow bifurcation point of the gradient of each node in the time and space dimensions are calculated to generate a gradient vector field. Using a gradient mutation detection algorithm to detect whether there is a gradient mutation point in the gradient vector field, if the gradient amplitude change rate of any area in the gradient vector field is greater than a sixth preset threshold, it is marked as an abnormal flow direction point, and a set of abnormal point positions is obtained; According to the gradient mutation points in the abnormal point location set, combined with the energy path tracing algorithm, the flow bifurcation points and flow intensity between the abnormal point and the adjacent nodes are analyzed, the propagation direction and gradient distribution uniformity of the abnormal flow are calculated, and the energy anomaly propagation feature vector is obtained.

5. A real-time detection system for abnormal energy consumption of smart energy, characterized in that: The system includes an energy consumption anomaly detection device, which includes a processing unit and a determination unit; The processing unit is configured to obtain energy consumption data and energy flow data of each energy-consuming node in the smart energy system, and extract a spatial interaction feature matrix based on the energy consumption data and the energy flow data; the spatial interaction feature matrix is configured to reflect the energy flow connectivity and energy flow intensity between each energy-consuming node; The determining unit is configured to determine abnormal candidate nodes in the smart energy system based on the spatial interaction feature matrix to obtain an abnormal candidate node set; the energy flow connectivity of the abnormal candidate nodes is less than a first preset threshold, and the energy flow intensity is less than a second preset threshold; The processing unit is further configured to extract an energy anomaly propagation feature vector corresponding to the abnormal candidate node set; the energy anomaly propagation feature vector is configured to reflect the energy flow intensity and energy gradient distribution uniformity of each abnormal candidate node; The determination unit is also used to adopt a pattern matching algorithm to compare the similarity between the energy abnormality propagation feature vector and the energy normal propagation feature vector corresponding to the normal energy consumption pattern library through a pre-established normal energy consumption pattern library. If the similarity is lower than a third preset threshold, the abnormal candidate node is determined to be a real abnormality, and the final abnormality detection result is generated.

6. The system according to claim 5, characterized in that The processing unit is specifically configured to: Based on the trend slope and mutation time point of the energy consumption data, as well as the flow intensity and flow angle in the energy flow data, a graph convolution algorithm is used to construct an energy transfer network between each energy-consuming node, and a weighted adjacency matrix describing the flow connectivity and flow intensity between nodes is generated to obtain the spatial interaction feature matrix.

7. The system according to claim 5, characterized in that The determining unit is specifically configured to: Obtaining the flow intensity and energy inflow and outflow deviation of each node from the spatial interaction feature matrix, calculating the deviation vector using a vector decomposition method, and marking the node as a node to be optimized if the modulus of the deviation vector is greater than a fourth preset threshold, thereby obtaining a set of nodes to be optimized; For the set of nodes to be optimized, a gradient calculation method is used to calculate the gradient amplitude and flow angle of each node, and a node feature vector including the gradient amplitude and the flow angle is obtained through vector synthesis to obtain a set of node feature vectors; According to the node feature vector set, a graph analysis method is used to calculate the gradient distribution uniformity and flow connectivity between nodes, and an energy balance vector including the distribution uniformity and the connectivity characteristics is generated to obtain an energy balance vector set; If the connectivity feature of a certain vector in the energy balance vector set is less than a fifth preset threshold, abnormal candidate nodes are screened by an anomaly detection algorithm to obtain the abnormal candidate node set.

8. The system according to claim 5, wherein: The processing unit is specifically configured to: For the set of abnormal candidate nodes, a gradient field modeling method is used. Combining the volatility amplitude and period amplitude of the energy consumption data, and the gradient amplitude and flow angle of the energy flow data, the local extreme value and flow bifurcation point of the gradient of each node in the time and space dimensions are calculated to generate a gradient vector field. Using a gradient mutation detection algorithm to detect whether there is a gradient mutation point in the gradient vector field, if the gradient amplitude change rate of any area in the gradient vector field is greater than a sixth preset threshold, it is marked as an abnormal flow direction point, and a set of abnormal point positions is obtained; According to the gradient mutation points in the abnormal point location set, combined with the energy path tracing algorithm, the flow bifurcation points and flow intensity between the abnormal point and the adjacent nodes are analyzed, the propagation direction and gradient distribution uniformity of the abnormal flow are calculated, and the energy anomaly propagation feature vector is obtained.

9. An electronic device, characterized in that: include: processor and memory; The memory stores instructions executable by the processor; When the processor is configured to execute the instructions, the electronic device implements the method according to any one of claims 1 to 4.

10. A computer-readable storage medium, characterized in that The readable storage medium includes: software instructions; When the software instructions are executed in an electronic device, the electronic device is enabled to implement the method according to any one of claims 1 to 4.

Images