Power load monitoring data fusion method and device, equipment and storage medium

Abstract

The invention provides a power load monitoring data fusion method, device and equipment and a storage medium, and the method comprises the steps: firstly obtaining power load multi-source data containing meteorological elements, and constructing a power physical constraint branch according to the data; a differential equation module is used for modeling a temperature response curve, a delay integral module is used for simulating a rainfall penetration process, and physical rule dynamic weights are generated and set as gating signals. And the gating fusion unit is controlled to output the weight through the signal, so that load monitoring physical rule embedding is realized. According to the method, a power physical constraint branch is constructed, a temperature-power relation and a precipitation hysteresis effect are modeled by a specific module, a dynamic weight driving gating signal is generated, a physical rule is embedded into a fusion model, and the accuracy and generalization of load monitoring are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of load monitoring, and particularly relates to a power load monitoring data fusion method, device, equipment and storage medium. BACKGROUND

[0002] In the field of power load monitoring, accurately and comprehensively grasping power demand data is of vital importance to ensuring power supply reliability, optimizing resource allocation, ensuring safe and stable operation of the power grid, and improving service quality. However, existing power load monitoring technologies face many challenges in practical application, which have prompted the proposal of a power load monitoring data fusion method.

[0003] Traditionally, power load monitoring mainly relies on data from a single source, such as smart meter data. Although these data can reflect the power consumption of users, they often ignore the profound influence of external environmental factors on power load. In particular, meteorological elements such as temperature, humidity, and precipitation have a close physical causal relationship with power load. For example, under the coal-to-electricity policy, changes in temperature directly affect the heating power load of residents, and the hysteresis effect of precipitation penetration can have a sustained impact on agricultural irrigation power. If these meteorological elements are not fully considered, the accuracy and completeness of power load monitoring will be greatly compromised. Further, existing technologies often lack a deep understanding and effective use of the unique physical characteristics of the power system when dealing with multi-source data fusion. For example, the relationship between power load and meteorological elements is not a simple linear correlation, but is strictly constrained by power physics equations. These physical equations describe how meteorological elements such as temperature and precipitation affect power generation, transmission, and consumption, and in turn have complex effects on power load.

[0004] However, in traditional data fusion methods, these physical laws are often ignored or only roughly adjusted by artificial parameters, making it difficult for the model to accurately capture the dynamic changes of power load. In addition, power data also has significant spatio-temporal coupling. Power grid failures or abnormal fluctuations can occur suddenly and spread rapidly to different parts of the network. This suddenness and propagation characteristic makes it difficult for a single model to effectively capture all dynamic changes. For example, the KNN (K-Nearest Neighbor) algorithm can utilize the similarity in spatial dimensions by finding the most similar situation in historical data, but it is difficult to handle long-term dependencies in time series; while LSTM (Long Short-Term Memory) is good at handling time series data, it may ignore important information in the spatial dimension. Therefore, how to combine the advantages of these two models to better cope with sudden fluctuations in the power grid has become a problem that needs to be solved. At the same time, the accuracy of spatial data is also one of the key factors affecting the effectiveness of power load monitoring.

[0005] In power distribution network analysis, directly using Euclidean distance to calculate the interaction between nodes often leads to inaccurate results, because the actual physical distance is not equal to the circuit distance. Circuit distance is deeply affected by the impedance of the power grid, including resistance, inductance and capacitance components. Therefore, when calculating the similarity or interaction between nodes in the power grid, the correction factor of the power grid impedance must be considered.

[0006] Finally, the reverse conduction of electricity price policy also poses new challenges to power load monitoring. Changes in electricity price policy will directly affect the electricity consumption behavior of users, and in turn cause the timing dynamic changes of power load. The traditional BiLSTM model may encounter timing conflict problems when dealing with this special timing dynamic, because its gating structure usually depends on manual parameter tuning, and it is difficult to adapt to the changes of electricity price policy in real time. SUMMARY

[0007] The purpose of the present application is to overcome the defects in the prior art described above, and to provide a power load monitoring data fusion method, device, equipment and storage medium.

[0008] The present application provides a power load monitoring data fusion method, comprising:

[0009] Obtaining multi-source data of a power load monitoring system, wherein the multi-source data includes meteorological element data;

[0010] Constructing a power physical constraint branch according to the meteorological element data;

[0011] Modeling a temperature response curve through a differential equation module in the power physical constraint branch, wherein the temperature response curve represents the relationship between temperature and power;

[0012] Simulating a precipitation infiltration process through a delay integral module in the power physical constraint branch, wherein the precipitation infiltration process represents the precipitation infiltration hysteresis effect;

[0013] Generating a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process;

[0014] Setting the physical law dynamic weight as a physical law driven gating signal;

[0015] Controlling the output weight of a gating fusion unit through the physical law driven gating signal to realize the embedding of physical law in load monitoring.

[0016] Optionally, modeling a temperature response curve through a differential equation module in the power physical constraint branch comprises:

[0017] Converting the meteorological element data into a feature vector;

[0018] Solving a temperature response differential equation according to the feature vector, generating an output gradient parameter, the gradient parameter being used to generate a physical law dynamic weight and representing a temperature-power relationship.

[0019] Optionally, simulating a precipitation infiltration process by a delay integral module in the power physical constraint branch, comprising:

[0020] Processing the meteorological element data to obtain a precipitation feature vector;

[0021] Time delay integration is performed on the precipitation feature vector to obtain a time integration result;

[0022] The time integration result is input into a nonlinear function to generate a precipitation weight value, the precipitation weight value being used to generate a physical law dynamic weight.

[0023] Optionally, generating a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process, comprising:

[0024] Extracting a line load rate from grid operation data;

[0025] Superimposing the gradient parameter and the precipitation weight value to generate an initial dynamic weight;

[0026] When the line load rate exceeds a preset threshold, the upper limit of the precipitation weight component in the initial dynamic weight is constrained by a weight truncation function, and the constrained initial dynamic weight is output as a physical law dynamic weight.

[0027] Optionally, setting the physical law dynamic weight as a physical law driven gating signal, comprising:

[0028] Constructing a dynamic weight matrix according to the physical law dynamic weight, the dynamic weight matrix mapping an initial gating signal;

[0029] Extracting a current real-time electricity price rule from an electricity price rule library;

[0030] Generating a timing adjustment parameter based on the current real-time electricity price rule;

[0031] Integrating the timing adjustment parameter into the dynamic weight matrix by a gating remapping function, and outputting the reconstructed dynamic weight matrix as a physical law driven gating signal.

[0032] Optionally, controlling the output weight of the gating fusion unit by the physical law driven gating signal to realize physical law embedding of load monitoring, comprising:

[0033] Obtaining an actual output weight of the gating fusion unit;

[0034] Extracting a reference weight corresponding to the power line property from a weight parameter table;

[0035] comparing the actual output weight with the reference weight by a difference calculation function, and adjusting modeling parameters of the differential equation module and the delay integral module when the difference exceeds an allowed threshold value;

[0036] feeding the adjusted modeling parameters back to the differential equation module and the delay integral module.

[0037] Optionally, the construction of the power physical constraint branch satisfies:

[0038] activating the temperature response curve modeling only when there is a coal-to-electricity load;

[0039] activating the precipitation infiltration process simulation only when there is an irrigation load.

[0040] The application further provides an electric power load monitoring data fusion device, comprising:

[0041] an acquisition module that acquires multi-source data of an electric power load monitoring system, the multi-source data including meteorological element data;

[0042] a construction module that constructs a power physical constraint branch according to the meteorological element data;

[0043] a temperature module that models a temperature response curve through a differential equation module in the power physical constraint branch, the temperature response curve representing a temperature-power relationship;

[0044] an infiltration module that simulates a precipitation infiltration process through a delay integral module in the power physical constraint branch, the precipitation infiltration process representing a precipitation infiltration hysteresis effect;

[0045] a weight module that generates a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process;

[0046] a signal module that sets the physical law dynamic weight as a physical law driven gating signal;

[0047] an output module that controls an output weight of a gating fusion unit through the physical law driven gating signal, and realizes physical law embedding for load monitoring.

[0048] The application further provides an electronic device comprising a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the above method.

[0049] The application further provides a computer readable storage medium storing a computer program, and the computer program, when executed in a computer, causes the computer to execute the above method.

[0050] The beneficial effects of the present application are:

[0051] The present application provides a power load monitoring data fusion method, comprising: acquiring multi-source data of a power load monitoring system, the multi-source data containing meteorological element data; constructing a power physical constraint branch according to the meteorological element data; modeling a temperature response curve through a differential equation module in the power physical constraint branch, the temperature response curve representing a temperature-power relationship; simulating a precipitation infiltration process through a delay integral module in the power physical constraint branch, the precipitation infiltration process representing a precipitation infiltration hysteresis effect; generating a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process; setting the physical law dynamic weight as a physical law driven gating signal; controlling an output weight of a gating fusion unit through the physical law driven gating signal, realizing physical law embedding of load monitoring. The present application constructs a power physical constraint branch, models temperature-power relationship and precipitation hysteresis effect through a differential equation and a delay integral module, generates a dynamic weight driven gating signal, realizes physical law embedding fusion model, and improves load monitoring accuracy and generalization. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a schematic diagram of the power load monitoring data fusion process in the present application. DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it is understood that various forms implement the present disclosure and should not be limited by the embodiments set forth herein. On the contrary, the embodiments are provided to enable a more thorough understanding of the present disclosure and to convey the scope of the present disclosure to those skilled in the art.

[0054] Please refer to Figure 1 The present application provides a power load monitoring data fusion method, comprising:

[0055] S101, acquiring multi-source data of a power load monitoring system, the multi-source data containing meteorological element data.

[0056] The process of collecting data from multiple sources of a power load monitoring system. It involves actively accessing and extracting data sources to ensure that data is fully input into the system.

[0057] Specifically, it includes connecting intelligent electric meter system, meteorological data interface and other hardware or software modules to read data streams in real time or in batches. For example, power load data recorded by intelligent electric meters and element data such as temperature and precipitation provided by meteorological stations are obtained through API calls or database queries. This step ensures that subsequent processing is based on comprehensive and real-time input, avoiding the problem of data island.

[0058] “Multi-source data” refers to heterogeneous datasets from different systems, including smart meter data and meteorological element data. This feature emphasizes the diversity and complementarity of data: smart meter data provides real-time power information of the substation load, while meteorological element data includes temperature, humidity, precipitation, and other environmental factors that affect the load. For example, in the clean heating scenario, temperature data is directly related to the change in coal-to-electricity load; in the irrigation scenario, precipitation data is related to the hysteresis effect. Format unification and preprocessing are required before data fusion to ensure consistency.

[0059] By collecting data from multiple sources, including smart meters, meteorological data, etc., comprehensive load-related information is obtained. Obtaining multi-source data of the power load monitoring system is the starting point of the monitoring method, which solves the technical problem of multi-source data fusion: in traditional methods, data dispersion leads to incomplete load pattern recognition, while the present application integrates power and meteorological systems to provide a basis for subsequent physical constraint embedding. For example, in substation load monitoring, smart meter data captures electricity fluctuations, and meteorological data provides external influencing factors, which, when combined, make load prediction more accurate.

[0060] S102, constructing a power physical constraint branch according to the meteorological element data.

[0061] “Constructing according to meteorological element data” refers to creating a special data processing branch based on meteorological input. It involves designing a network structure, taking meteorological elements (such as temperature, precipitation) as input, and outputting physical law-related features.

[0062] Specifically, it includes initializing branch parameters, configuring sub-network layers (such as MLP), and activating branch processing. For example, a small neural network is used to process temperature data when constructing, generating temperature response features; or a convolutional layer is used to process precipitation data, capturing time delay effects. This action is only performed when a specific load type exists, ensuring efficient use of resources.

[0063] “Power physical constraint branch” refers to a special sub-network added in the fusion layer, which is used to embed the physical laws specific to the power system. This feature is designed for temperature-power physical causality (such as coal-to-electricity load) and precipitation penetration hysteresis effect (such as irrigation load), and is not effective in other fields (such as transportation). It specifically includes components such as differential equation modules and delay integral modules, and the branch output directly adjusts the fusion weight. For example, in a distribution network, this branch guides model learning through physical rules (such as the linear relationship of coal-to-electricity or the hysteresis of irrigation), improving generalization.

[0064] Further, in the above-mentioned “power physical constraint branch refers to a special sub-network added in the fusion layer”, it further includes: a smart meter data branch: a Transformer time series encoder; a meteorological data branch: a CNN-LSTM.

[0065] This application constructs a power physics constraint branch based on meteorological element data as part of the fusion layer. Logically, this step solves the problem of missing physical laws in load pattern recognition: traditional fusion models rely on manual parameter tuning, while this branch, through structured modeling of physical rules, makes the model more consistent with the reality of the power system. For example, in the coal-to-electricity conversion scenario, the branch activates temperature response modeling; in the irrigation scenario, it activates precipitation infiltration simulation. The steps rely on load type detection: temperature response curve modeling is activated only when coal-to-electricity load exists; precipitation infiltration process simulation is activated only when irrigation load exists.

[0066] S103. Model the temperature response curve using the differential equation module in the electrical physics constraint branch. The temperature response curve characterizes the temperature-power relationship.

[0067] Modeling via the differential equation module refers to using mathematical differential equations to process data and generate response curves. This involves transforming input data, solving equations, and outputting parameters. Specifically, it includes converting meteorological element data (such as temperature values) into feature vectors, inputting them into the differential equation solver, and generating output gradient parameters. The gradient parameters characterize the linear rate of change of the temperature-power relationship; for example, the weight increases by 0.05 for every 1°C decrease in temperature. This action is only activated in coal-to-electricity load scenarios to ensure model relevance.

[0068] The temperature response curve is a mathematical model characterizing the relationship between temperature and power, with its shape defined by a differential equation. This feature captures the physical causality of coal-to-electricity load changes: temperature variations directly affect power demand. For example, the curve shows a linear increase in power as temperature decreases, enhancing load pattern recognition. The feature outputs gradient parameters for dynamic weight calculation, ensuring that the weights reflect physical laws rather than manual settings.

[0069] This module models the temperature response curve using a differential equation module within the power physics constraint branch to characterize the temperature-power relationship. Traditional methods neglect the causality of temperature-power, while this module solves differential equations (e.g., gradient calculation based on eigenvectors) and outputs gradient parameters, which are directly used to generate dynamic weights for physical laws. For example, in areas undergoing coal-to-electricity conversion during winter, as the temperature drops from 5°C to 0°C, the module generates gradient parameters (e.g., +0.05 per°C) to drive an increase in weights, accurately reflecting load growth. The steps include converting meteorological data into eigenvectors, solving the temperature response differential equation based on the eigenvectors, and generating output gradient parameters, which are used to generate dynamic weights for physical laws and characterize the temperature-power relationship.

[0070] S104. The precipitation infiltration process is simulated by the delay integration module in the electrical physics constraint branch, and the precipitation infiltration process characterizes the precipitation infiltration hysteresis effect.

[0071] The time-delay integration module simulates data processing using time integration methods to model lag processes. This action involves time-delayed accumulation and integration of precipitation data. Specifically, it includes processing meteorological element data (such as precipitation amount) to obtain a precipitation feature vector, performing time-delay integration (such as accumulating 12 hours of precipitation), and generating the time integration result. Then, a nonlinear function (such as the hyperbolic tangent function) is input to generate precipitation weight values. This action is only activated in irrigation load scenarios to ensure model efficiency.

[0072] The precipitation infiltration process refers to the lagged impact model of precipitation events on power load, characterizing the infiltration effect in irrigation scenarios. This feature captures the cumulative impact 12 hours after precipitation, such as load changes caused by delayed start-up of irrigation equipment. The feature outputs precipitation weight values ​​for dynamic weight calculation, ensuring that the weights reflect physical lags rather than instantaneous responses. Nonlinear functions (such as tanh) smooth the integral results to avoid abrupt changes.

[0073] This module simulates the precipitation infiltration process using a time-delay integration module within the electrical physics constraint branch to characterize the precipitation infiltration lag effect. It addresses the problem of irrigation load identification: traditional models ignore precipitation lag, while this module generates precipitation weight values ​​through time-delay integration (e.g., cumulative precipitation in mm) and nonlinear transformation. These weight values ​​are used to generate dynamic weights for physical laws. For example, in agricultural irrigation areas, the precipitation feature vector (e.g., 10 mm of precipitation) is integrated and processed using the tanh function to generate a weight value: weight = tanh(precipitation / 10), where precipitation is in millimeters, and tanh is a hyperbolic tangent function with an output value between 0 and 1, representing the intensity of the infiltration effect. The process includes processing meteorological data to obtain a precipitation feature vector, performing time-delay integration on the precipitation feature vector, obtaining the time integration result, and inputting the time integration result into a nonlinear function to generate precipitation weight values, which are then used to generate dynamic weights for physical laws.

[0074] S105. Generate dynamic weights based on the physical laws according to the temperature response curve and the precipitation infiltration process.

[0075] The dynamic weights are calculated by combining the outputs of the two modules based on the temperature response curve and the precipitation infiltration process. This process involves superimposing gradient parameters and precipitation weight values, and applying constraint rules.

[0076] Specifically, this involves extracting line load rates (e.g., real-time percentages) from grid operation data, overlaying gradient parameters (from temperature response) and precipitation weight values ​​(from precipitation infiltration) to generate initial dynamic weights. If the line load rate exceeds a preset threshold (e.g., 85%), a weight truncation function (e.g., a hard constraint function) is used to constrain the upper limit of the precipitation weight component in the initial dynamic weights (e.g., setting the meteorological weight ≤ 0.3). The constrained initial dynamic weights are then output as the physical law dynamic weights. This process ensures that the weights conform to the grid operation status.

[0077] The dynamic weighting based on physical laws refers to the fused weight values ​​driven by physical rules, used to adjust the gating signal. This feature is based on weight parameter table rules; for example, the temperature coefficient for coal-to-electricity conversion is in the range of 0.6-0.8, and the weight increases by 0.05 for every 1°C decrease; the weight for irrigation precipitation is calculated using the tanh function; and the line load rate affects the upper limit of the weight. This feature ensures dynamic adjustment of the weights, reflecting real-time physical conditions and improving monitoring accuracy.

[0078] Furthermore, the aforementioned "dynamic weight of physical laws refers to the fusion weight value driven by physical rules" also includes:

[0079] Weighting parameter table

[0080] Parameter category Typical value / range Adjustment rule Coal-to-power temperature coefficient 0.6-0.8 Weight per 1 °C drop +0.05 Irrigation precipitation lag 12-hour cumulative amount Weight = tanh(precipitation mm / 10) Line load factor Weather weight ≤0.3 when >85% Hard constraints

[0081] Dynamic weights based on physical laws are generated according to temperature response curves and precipitation infiltration processes. Traditional weight settings are static, while this application outputs adaptive weights by superimposing gradient parameters and precipitation weight values, combined with line load rate constraints. For example, in areas where coal-to-electricity conversion and irrigation coexist, gradient parameters (e.g., 0.7) and precipitation weight values ​​(e.g., tanh(5 / 10) = 0.46) are superimposed to generate initial weights; if the line load rate reaches 90%, a weight truncation function constrains the precipitation weight component to 0.3, outputting the final dynamic weights. The steps include extracting the line load rate from grid operation data, superimposing gradient parameters and precipitation weight values ​​to generate initial dynamic weights, and when the line load rate exceeds a preset threshold, constraining the upper limit of the precipitation weight component in the initial dynamic weights through a weight truncation function, outputting the constrained initial dynamic weights as the dynamic weights based on physical laws.

[0082] S106. Set the dynamic weight of the physical law as the physical law driving gating signal.

[0083] Setting the dynamic weights of physical laws as the transformation weight values ​​refers to the gating control signal. This action involves constructing a weight matrix and mapping it to the gating system.

[0084] Specifically, this involves constructing a dynamic weight matrix based on physical laws, which maps to the initial gating signal (such as the input to the gating fusion unit). The matrix elements are filled with weight values, ensuring the signal is directly driven by physical laws. This process converts weights into control signals, providing input to the fusion unit.

[0085] Physically driven gating signals refer to the gating input signals used to control the fusion unit, dynamically generated by physical rules. This feature emphasizes the non-artificial nature of the signal: it is directly driven by the physical equations of the power grid, rather than parameter tuning. The signal is expressed through a dynamic weight matrix, adjusting the fusion process to ensure that the output weights conform to the temperature-power relationship and the precipitation hysteresis effect.

[0086] This invention sets dynamic weights based on physical laws to drive the gating signal. Traditional gating signals rely on manual settings, while this application constructs a dynamic weight matrix and maps it to the initial gating signal, allowing the signal to be driven by physical laws (such as coal-to-electricity conversion or irrigation rules). For example, in the fusion layer, the dynamic weight matrix (e.g., a matrix based on temperature coefficients and precipitation weights) serves as the gating signal input, directly controlling the output of the gating fusion unit.

[0087] S107. The output weight of the gate control fusion unit is controlled by the gate control signal driven by the physical law, so as to realize the embedding of the physical law of load monitoring.

[0088] The gating signal control, driven by physical laws, refers to the use of gating signals to adjust the output of the fusion unit.

[0089] This process involves signal input, weight comparison, and feedback adjustment. Specifically, it includes obtaining the actual output weights of the gating fusion unit and extracting reference weights (e.g., coal-to-electricity temperature coefficient of 0.6-0.8) from the weight parameter table corresponding to the power line properties (e.g., load, feeder impedance). The actual output weights are compared with the reference weights using a difference calculation function (e.g., mean square error function). If the difference exceeds an allowable threshold (e.g., 10%), the modeling parameters (e.g., gradient parameters or delay time) of the differential equation module and delay integration module are adjusted. The adjusted modeling parameters are then fed back to the modules for iterative optimization. This process ensures that the output weights are always aligned with the physical constraints.

[0090] The output weight of the gating fusion unit refers to the final weight value of the fusion layer, used to weight multi-source data (such as meter and meteorological data branches). This feature is controlled by a gating signal, and the unit structure can be an attention mechanism or a gating unit. This feature embeds the physical laws of load monitoring, meaning the weights reflect real power rules (such as temperature-power linearity or precipitation lag), improving monitoring accuracy and interpretability.

[0091] By driving gating signals based on physical laws to control the output weights of the gating fusion unit, the physical laws of load monitoring are embedded. Logically, this step completes a closed loop from physical rules to monitoring output: traditional fusion ignores physical interpretability, while this application controls the output weights through signals and adjusts the modeling parameters based on the degree of difference to ensure effective embedding. For example, in transformer area load monitoring, the actual output weight is compared with the reference weight; if the difference is too large (e.g., actual temperature coefficient 0.9 vs. reference 0.7), the parameters of the differential equation module are adjusted, and the temperature response curve is remodeled after feedback, so that the final output weight accurately reflects the coal-to-electricity load mode. This includes obtaining the actual output weight of the gating fusion unit, extracting the reference weight corresponding to the power line properties from the weight parameter table, comparing the actual output weight with the reference weight through the degree of difference calculation function, and when the degree of difference exceeds the allowable threshold, adjusting the modeling parameters of the differential equation module and the delay integral module, and feeding the adjusted modeling parameters back to the differential equation module and the delay integral module.

[0092] Furthermore, in power load monitoring, the electricity price rule database stores provincial electricity price policies (such as peak-valley electricity price tables), and the current real-time electricity price rules are extracted by querying this database. Based on the rules, time-series adjustment parameters (such as time weight offsets) are generated to adjust the time-series dimension of the dynamic weight matrix. For example, during peak electricity price periods, the parameters are weighted with load to optimize the fusion output.

[0093] This application also provides a power load monitoring data fusion device, comprising:

[0094] The acquisition module acquires multi-source data from the power load monitoring system, including meteorological element data.

[0095] The module constructs a power physics constraint branch based on the meteorological element data;

[0096] The temperature module models the temperature response curve through the differential equation module in the electrical physics constraint branch, and the temperature response curve characterizes the temperature-power relationship.

[0097] The infiltration module simulates the precipitation infiltration process through the delay integration module in the electrical physics constraint branch, and the precipitation infiltration process characterizes the precipitation infiltration hysteresis effect.

[0098] The weighting module generates dynamic weights based on physical laws according to the temperature response curve and the precipitation infiltration process;

[0099] The signal module sets the dynamic weights of the physical laws as physical law-driven gating signals;

[0100] The output module controls the output weight of the gated fusion unit by driving the gated signal through the physical laws, thereby realizing the embedding of physical laws for load monitoring.

[0101] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0102] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the above-described method.

[0103] The above description of the embodiments is provided to enable those skilled in the art to understand and apply this application. Those skilled in the art will readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without inventive effort. Therefore, this application is not limited to the above embodiments, and any improvements and modifications made to this application based on the disclosure thereof should be within the scope of protection of this application.

Claims

1. A power load monitoring data fusion method, characterized by, The method comprises the following steps: acquiring multi-source data of a power load monitoring system, wherein the multi-source data comprises meteorological element data; constructing a power physical constraint branch according to the meteorological element data; modeling a temperature response curve through a differential equation module in the power physical constraint branch, wherein the temperature response curve represents a temperature-power relationship; simulating a precipitation infiltration process through a delay integral module in the power physical constraint branch, wherein the precipitation infiltration process represents a precipitation infiltration hysteresis effect; generating a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process; setting the physical law dynamic weight as a physical law driven gating signal; controlling an output weight of a gating fusion unit through the physical law driven gating signal to realize physical law embedding of load monitoring.

2. The power load monitoring data fusion method of claim 1, wherein, The modeling of the temperature response curve through the differential equation module in the power physical constraint branch comprises the following steps: converting the meteorological element data into a feature vector; solving a temperature response differential equation according to the feature vector to generate an output gradient parameter, wherein the gradient parameter is used to generate a physical law dynamic weight and represents a temperature-power relationship.

3. The power load monitoring data fusion method of claim 1, wherein, The simulation of the precipitation infiltration process through the delay integral module in the power physical constraint branch comprises the following steps: processing the meteorological element data to obtain a precipitation feature vector; performing time delay integral on the precipitation feature vector to obtain a time integral result; inputting the time integral result into a nonlinear function to generate a precipitation weight value, wherein the precipitation weight value is used to generate a physical law dynamic weight.

4. The power load monitoring data fusion method of claim 1, wherein, The generation of the physical law dynamic weight according to the temperature response curve and the precipitation infiltration process comprises the following steps: extracting a line load rate from grid operation data; superimposing the gradient parameter and the precipitation weight value to generate an initial dynamic weight; when the line load rate exceeds a preset threshold, constraining an upper limit of a precipitation weight component in the initial dynamic weight through a weight truncation function to output the constrained initial dynamic weight as a physical law dynamic weight.

5. The power load monitoring data fusion method of claim 1, wherein, The setting of the physical law dynamic weight as a physical law driven gating signal comprises the following steps: constructing a dynamic weight matrix according to the physical law dynamic weight, wherein the dynamic weight matrix maps an initial gating signal; extracting a current real-time electricity price rule from an electricity price rule library; generating a timing adjustment parameter based on the current real-time electricity price rule; integrating the timing adjustment parameter into the dynamic weight matrix through a gating remapping function to output a reconstructed dynamic weight matrix as a physical law driven gating signal.

6. The power load monitoring data fusion method of claim 1, wherein, The control of the output weight of the gating fusion unit through the physical law driven gating signal to realize physical law embedding of load monitoring comprises the following steps: acquiring an actual output weight of the gating fusion unit; extracting a reference weight corresponding to a power line property from a weight parameter table; comparing the actual output weight and the reference weight through a difference degree calculation function, and adjusting modeling parameters of the differential equation module and the delay integral module when the difference degree exceeds an allowable threshold; feeding back the adjusted modeling parameters to the differential equation module and the delay integral module.

7. The power load monitoring data fusion method of claim 1, wherein, The construction of the power physical constraint branch satisfies the following conditions: only when there is a coal-to-electricity load, the temperature response curve modeling is activated; The precipitation infiltration process simulation is activated only when there is an irrigation load.

8. An electric power load monitoring data fusion apparatus characterized by comprising: The method comprises the following steps: An acquisition module acquires multi-source data of a power load monitoring system, wherein the multi-source data comprises meteorological element data; A construction module constructs a power physical constraint branch according to the meteorological element data; A temperature module models a temperature response curve through a differential equation module in the power physical constraint branch, wherein the temperature response curve represents a temperature-power relationship; An infiltration module simulates a precipitation infiltration process through a delay integral module in the power physical constraint branch, wherein the precipitation infiltration process represents a precipitation infiltration hysteresis effect; A weight module generates a physical law dynamic weight according to the temperature response curve and the precipitation infiltration process; A signal module sets the physical law dynamic weight as a physical law driven gating signal; An output module controls an output weight of a gating fusion unit through the physical law driven gating signal, thereby realizing physical law embedding for load monitoring.

9. An electronic device, comprising: The method comprises the following steps:

10. A computer-readable storage medium, characterized in that, A computer program is stored in a memory, and a processor executes the computer program to realize the method according to any one of claims 1-7. A computer program is stored on a computer readable medium, and when the computer program is executed in a computer, the computer program causes the computer to execute the method according to any one of claims 1-7.

Images