Active load prediction method and device for low-voltage distribution transformer area

Abstract

The invention provides an active load prediction method and device for a low-voltage distribution transformer area, relates to the field of transformer area load prediction, and solves the technical problem that a prediction error can be sharply increased in a non-linear response facing an air temperature abrupt change influence. The method comprises the following steps: collecting and cleaning multi-source data of a transformer area to obtain electrical measurement data; obtaining weather forecast data of a transformer area location, and performing tensor conversion to obtain a three-dimensional tensor; and obtaining an electric regulation and control policy, performing time sequence alignment on the effective time of the electric regulation and control policy and the three-dimensional tensor to form a bimodal input matrix, inputting the bimodal input matrix and the electric measurement data into the multi-factor coupling probability prediction model, and outputting a load prediction value. The method is used in the active load prediction process of the low-voltage distribution transformer area. According to the method, meteorological parameters, policy texts and other unstructured data are converted into quantifiable features, and high-precision load prediction within 48 hours, 72 hours and other time periods is realized through a multi-factor coupling prediction model.

Description

Technical Field

[0001] This application relates to the field of transformer area load forecasting, and more particularly to a method and apparatus for forecasting active power load in a low-voltage transformer area. Background Technology

[0002] With the accelerated construction of new power systems, low-voltage distribution substations are facing unprecedented operational pressure. Against the backdrop of rapid urbanization and electrification, the proportion of impact loads such as central air conditioning and energy storage devices in commercial complexes, data centers, and smart building clusters continues to rise. In some coastal cities, the peak-to-valley difference rate of daily active power load in low-voltage substations can reach 3-5 times the rated capacity. Especially during peak summer electricity consumption periods, regional voltage dips occur frequently, directly causing significant economic losses such as the downtime of precision equipment and production line shutdowns. Even more serious is the large-scale integration of new loads such as distributed photovoltaic power and electric vehicle charging piles, which has transformed the traditional distribution network structure, primarily based on unidirectional power supply, into a multi-source, bidirectional interactive mode. The three-phase imbalance of active power exhibits a more complex spatiotemporal distribution trend, with the negative sequence voltage imbalance in typical substations reaching over 4.2%. Under continuous 20% overload conditions, this leads to an additional 12%-25% increase in losses for distribution transformers and a reduction in equipment lifespan by an average of 3-5 years.

[0003] Mainstream load forecasting methods, such as ARIMA and support vector regression, perform well under stable load scenarios, but their prediction errors increase sharply to over 12% when faced with the nonlinear response of air conditioning loads affected by sudden temperature changes. At the state assessment level, static power flow analysis methods cannot capture the impact of millisecond-level load fluctuations. Summary of the Invention

[0004] This application provides a method and apparatus for predicting the active load of a low-voltage distribution substation, which solves the technical problem that the prediction error will increase sharply when the nonlinear response to the influence of sudden temperature changes is faced with the existing technology.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for predicting active power load in low-voltage distribution substation areas is provided, including: Multi-source data from the transformer substation area is collected and cleaned to obtain electrical measurement data; Meteorological forecast data for the location of the weather station area is obtained and tensor transformation is performed to obtain a three-dimensional tensor; the meteorological forecast data includes temperature, humidity and wind speed. The electricity regulation policy is obtained, and semantic vectors and the effective time of the electricity regulation policy are extracted through the BERT model. The effective time of the electricity regulation policy is temporally aligned with the three-dimensional tensor to form a dual-modal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probability prediction model, and the output includes the load prediction value and the confidence interval boundary; wherein, the multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

[0006] Based on the above technical solution, in the active load prediction method for a low-voltage distribution area provided in this application, unstructured data such as meteorological parameters and policy texts are transformed into quantifiable features, and then a high-precision load prediction within a future time n is achieved through a multi-factor coupled prediction framework.

[0007] In conjunction with the first aspect above, in one possible implementation, the method for acquiring the electrical measurement data includes: Multi-source heterogeneous data is acquired by integrating smart meters; wherein, the multi-source heterogeneous data includes voltage, current and active power; The adaptive Kalman filter algorithm is used to align and calibrate the spatiotemporal dimensions of multi-source heterogeneous data to obtain calibration data; The calibration data is processed through a hybrid data cleaning pipeline to obtain electrical measurement data; wherein the hybrid data cleaning pipeline includes the Isolation Forest algorithm, K-Nearest Neighbors, wavelet transform, hash time-locked contract, and knowledge graph.

[0008] It should be noted that the electrical measurement data is a dynamic device correlation map integrating data, topology, and reliability, providing a high-quality input basis for subsequent predictions.

[0009] In conjunction with the first aspect above, in one possible implementation, the processing of calibration data through a hybrid data cleaning pipeline includes: The isolated forest algorithm is used to detect and remove outliers from the calibration data. The K-nearest neighbor algorithm is used to fill in the missing points in the calibration data after removing outliers. Finally, wavelet transform is used to filter the filled calibration data to obtain the corrected data. The corrected data is processed by a hash time-locked contract to obtain data blocks with trust labels; a device association model is constructed based on a knowledge graph, and the transformer area topology relationship is analyzed by a graph attention network to obtain the transformer area physical topology; the data blocks with trust labels are bound to the transformer area physical topology to obtain electrical measurement data.

[0010] In conjunction with the first aspect above, in one possible implementation, the method for obtaining the three-dimensional tensor includes: The Z-score standardization method was used to eliminate the dimensional differences in meteorological forecast data, and the meteorological forecast data were sorted according to the collection time to obtain the forecast sequence; pass The predicted sequence is processed to obtain a three-dimensional tensor; where, , , They represent the first Standardized temperature, humidity, and wind speed for hours, where n is the future time of the weather forecast data, t∈n.

[0011] In conjunction with the first aspect above, in one possible implementation, the construction method of the multi-factor coupled probability prediction model includes: A spatiotemporal feature extractor is constructed based on the Transformer-XL architecture and by introducing relative position encoding and gated attention mechanisms. The thermodynamic equations are transformed into differentiable constraint terms using a physical information neural network to construct a residual loss function. A load change inflection point layer is constructed based on the LSTM-CRF model and by introducing the historical feature difference of the lag time window. A joint probability distribution layer of meteorological, social, and load is constructed based on the Gaussian Copula function; The computational output layer is constructed based on Monte Carlo sampling.

[0012] It should be noted that long-term temporal dependencies can be captured by introducing a positional encoding formula.

[0013] The gating attention mechanism enables the model to adaptively adjust feature importance according to the characteristics of the prediction period; the temporal correlation features captured by the LSTM-CRF model are used as "additional constraints" and incorporated into subsequent multi-factor coupling modeling to ensure the temporal logic consistency of the coupled prediction.

[0014] In conjunction with the first aspect above, in one possible implementation, the relative position is encoded as follows: , where A i,j q represents the attention score from position i to j; i and j are the indices of the i-th and j-th positions in the sequence; i Let k be the query vector at the i-th position; j Let R be the key vector at the j-th position; i-j This is a relative position encoding matrix, where w represents the relative distance between encoding positions i and j. R It is a learnable relative position weight vector.

[0015] In conjunction with the first aspect above, in one possible implementation, the gate-controlled attention mechanism is as follows: ; The weight matrix is ​​a learnable matrix. This is a vector concatenated from meteorological and policy characteristics. It is the sigmoid activation function.

[0016] It should be noted that, This is a hidden state related to meteorological characteristics. This refers to a hidden state related to policy characteristics.

[0017] In conjunction with the first aspect above, in one possible implementation, the residual loss function is: , Let be the partial derivative of the load with respect to temperature, and k be the thermal conductivity coefficient. For the outside temperature, To set the temperature, This is the regularization coefficient, which is the weight that balances prediction error and physical constraints.

[0018] In conjunction with the first aspect mentioned above, one possible implementation also includes: obtaining a load change risk warning signal based on the load forecast value.

[0019] In conjunction with the first aspect above, in one possible implementation, obtaining the load change risk warning signal based on the load forecast value includes: The load current signal is decomposed into three levels using wavelet packet transform to obtain time-frequency energy sequences of several sub-bands. Constructing a multi-scale detection index based on different frequency bands—energy gradient change rate. The normalized rate of change of energy gradient is calculated by weighted summation. Among them, the rate of change of energy gradient Through calculation formula Calculated; For time intervals; For the k-th sub-band at t- Energy at time t; Let Ek,t be the variance of the energy sequence of the kth sub-band over the entire time range [1,T]. Let k be the rate of change of the energy gradient of the k-th sub-band. The energy gradient change rate index for node i; Construct a two-input fuzzy inference: using the gradient mean with a window length of x seconds. and variance As input, a dynamic threshold fuzzy rule base is defined, and the fuzzy output is a dynamically updated threshold that is updated every minute, defuzzified using the centroid method. Define a fuzzy rule base for risk levels, based on dynamic thresholds and energy gradient change rates. Matching risk level L; By establishing a risk propagation equation, the spatiotemporal diffusion pattern of mutation signals in the transformer area topology is quantified to obtain the risk source node; The risk level L, the propagation speed v, and the impact range S should each be assigned a color tone. saturation and transparency The dynamic color code for early warning is obtained; wherein, the propagation speed v is calculated using the formula... The following values ​​are obtained: dij is the electrical distance between nodes i and j; Δtij is the time difference between the first time the rate of change of the energy gradient between nodes i and j exceeds a set threshold; N is the number of neighboring nodes of node j; and the influence range S is the number of nodes whose risk influence score is greater than a set threshold.

[0020] In conjunction with the first aspect above, in one possible implementation, the risk source node is the node corresponding to the maximum risk impact score; wherein, the risk impact score is calculated using the formula... The calculation yields Rj(t), where Rj(t) is the risk influence score of node j at time t, and N(j) is the set of neighboring nodes of node j. Let be the electrical coupling weight from node i to j, tonset be the mutation start time, and α be the time decay factor. It should be noted that by overlaying the SCADA system onto the geographic wiring diagram, second-level visualization is achieved, providing a solution from micro-signal perception to macro-risk early warning for high-density commercial load scenarios.

[0021] In conjunction with the first aspect above, in one possible implementation, the risk source node is the node corresponding to the maximum risk impact score; wherein, the risk impact score is calculated using the formula... The calculation yields Rj(t), where Rj(t) is the risk influence score of node j at time t, and N(j) is the set of neighboring nodes of node j. Let be the electrical coupling weight from node i to j, tonset be the mutation start time, and α be the time decay factor.

[0022] Secondly, an electronic device is provided, comprising: a communication unit and a processing unit; the communication unit is used to collect multi-source data of a transformer substation area; acquire meteorological forecast data of the substation area location; and acquire power control policies; The processing unit is used to clean the multi-source data of the transformer area to obtain electrical measurement data; and to perform tensor transformation on the meteorological forecast data to obtain a three-dimensional tensor. The semantic vectors and effective time of the electricity regulation policy are extracted by the BERT model. The effective time of the electricity regulation policy is then temporally aligned with the three-dimensional tensor to form a bimodal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probability prediction model, and the output includes the load prediction value and the confidence interval boundary; wherein, the multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

[0023] Thirdly, this application provides a processing apparatus, comprising: a processor and a storage medium; the storage medium includes instructions, and the processor is configured to execute the instructions to implement the methods described in the first aspect and any possible implementation thereof. The processing apparatus may be an electronic device or a chip within an electronic device.

[0024] Fourthly, this application provides an active load forecasting system for a low-voltage distribution transformer area, comprising: an information acquisition module, a data processing module, and a forecasting module; wherein, the information acquisition module is used to acquire multi-source data of the transformer area.

[0025] Fifthly, this application provides a computer-readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0026] Sixthly, this application provides a computer program product containing instructions that, when run on an electronic device, cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0027] This application provides a method and device for active load forecasting in low-voltage distribution transformer areas. It breaks down traditional data barriers through a multi-source heterogeneous data fusion sensing system, deploying multi-source terminals such as smart meters and environmental sensors at edge computing nodes. An adaptive Kalman filter algorithm is used to achieve spatiotemporal data alignment, and a blockchain hash time-locked contract ensures data credibility. Graph Attention Network (GAT) is used to analyze the transformer area topology, constructing a dynamic device association map with credibility labels, providing a high-precision data foundation for upper-layer models. Based on this, a meteorological-social multi-dimensional load forecasting model innovatively integrates the Transformer-XL architecture and Physical Information Neural Network (PINN), transforming unstructured data such as meteorological parameters and policy texts into quantifiable features. An LSTM-CRF model captures load lag effects, constructing a multi-factor coupled forecasting model based on Copula theory, achieving high-precision load forecasting within time periods such as 48 and 72 hours. Its forecasting dimension expands from single electricity consumption to comprehensive environmental and social impact analysis.

[0028] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0029] Figure 1 A system architecture diagram of an active load prediction system for a low-voltage distribution transformer area provided in this application embodiment; Figure 2 A flowchart illustrating a method for predicting the active load of a low-voltage distribution substation, provided in an embodiment of this application; Figure 3 This is a schematic diagram of the process for acquiring electrical measurement data provided in an embodiment of this application; Figure 4 This is a schematic diagram comparing the load probability prediction effects provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the hardware structure of a processing device provided in an embodiment of this application. Detailed Implementation

[0030] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0031] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0032] The active load prediction method for low-voltage distribution substations provided in this application can be applied to, for example... Figure 1 In the active load prediction system 100 of the low-voltage distribution substation shown, such as Figure 1 As shown, the communication system includes: an integrated smart meter 101, an environmental monitoring device 102, and a cloud computer 103.

[0033] Among them, the integrated smart meter 101 is used to collect multi-source data of the transformer area and send it to the cloud computer 103; Environmental monitoring equipment 102 is used to acquire meteorological forecast data of the location of the station area and send it to cloud computer 103; The cloud computer 103 is used to clean multi-source data from the transformer substation to obtain electrical measurement data; and to perform tensor transformation on meteorological forecast data to obtain three-dimensional tensors. The semantic vectors and effective time of the electricity regulation policy are extracted by the BERT model. The effective time of the electricity regulation policy is then temporally aligned with the three-dimensional tensor to form a bimodal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probability prediction model, and the output includes the load prediction value and the confidence interval boundary; wherein, the multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

[0034] To address the technical problem that prediction errors in existing technologies increase dramatically when faced with nonlinear responses to sudden temperature changes, this application provides a method for predicting the active load of a low-voltage distribution transformer area. The method includes: collecting and cleaning multi-source data from the transformer area to obtain electrical measurement data. Meteorological forecast data for the location of the weather station area is obtained and tensor transformation is performed to obtain a three-dimensional tensor; the meteorological forecast data includes temperature, humidity and wind speed. The electricity regulation policy is obtained, and semantic vectors and the effective time of the electricity regulation policy are extracted through the BERT model. The effective time of the electricity regulation policy is temporally aligned with the three-dimensional tensor to form a dual-modal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probabilistic prediction model, which outputs load forecasts and confidence interval boundaries. Based on this, high-precision load prediction within a future time period n is achieved.

[0035] like Figure 2 As shown in the embodiment of this application, an active power load prediction method for a low-voltage distribution transformer area is provided, including: S201. Collect and clean multi-source data from the transformer substation to obtain electrical measurement data.

[0036] Among them, multi-source heterogeneous data includes voltage, current and active power.

[0037] In some implementations, multi-source heterogeneous data is acquired by integrating smart meters; wherein, the multi-source heterogeneous data includes voltage, current and active power; The adaptive Kalman filter algorithm is used to align and calibrate the spatiotemporal dimensions of multi-source heterogeneous data to obtain calibration data; The calibration data is processed through a hybrid data cleaning pipeline to obtain electrical measurement data; wherein the hybrid data cleaning pipeline includes the Isolation Forest algorithm, K-Nearest Neighbors, wavelet transform, hash time-locked contract, and knowledge graph.

[0038] Furthermore, the processing of calibration data through a hybrid data cleaning pipeline includes: The isolated forest algorithm is used to detect and remove outliers from the calibration data. The K-nearest neighbor algorithm is used to fill in the missing points in the calibration data after removing outliers. Finally, wavelet transform is used to filter the filled calibration data to obtain the corrected data. The corrected data is processed by a hash time-locked contract to obtain data blocks with trust labels; a device association model is constructed based on a knowledge graph, and the transformer area topology relationship is analyzed by a graph attention network to obtain the transformer area physical topology; the data blocks with trust labels are bound to the transformer area physical topology to obtain electrical measurement data.

[0039] It should be noted that electrical measurement data is a dynamic device correlation map integrating data, topology, and reliability, providing a high-quality input basis for subsequent predictions.

[0040] S202. Obtain meteorological forecast data for the location of the substation area and perform tensor transformation to obtain a three-dimensional tensor.

[0041] The meteorological forecast data includes temperature, humidity, and wind speed.

[0042] In some implementations, the Z-score normalization method is used to eliminate the dimensional differences in meteorological forecast data, and the meteorological forecast data is sorted according to the collection time to obtain the forecast sequence; pass The predicted sequence is processed to obtain a three-dimensional tensor; where, , , They represent the first Standardized temperature, humidity, and wind speed for hours, where n is the future time of the weather forecast data, t∈n.

[0043] S203. Obtain the electricity regulation policy, and extract the semantic vector and the effective time of the electricity regulation policy through the BERT model. Align the effective time of the electricity regulation policy with the three-dimensional tensor in time to form a dual-modal input matrix.

[0044] It should be noted that, based on the policy effective time window, the semantic vector is time-series aligned with the standardized meteorological data three-dimensional tensor to ensure that policy features and meteorological features match in the time dimension, forming a "meteorological-social" dual-modal input matrix.

[0045] S204. Input the dual-modal input matrix and electrical measurement data into the multi-factor coupled probability prediction model, and output the load prediction value and confidence interval boundary.

[0046] The multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

[0047] In some implementations, the construction methods of multi-factor coupled probability prediction models include: A spatiotemporal feature extractor is constructed based on the Transformer-XL architecture and by introducing relative position encoding and gated attention mechanisms. The thermodynamic equations are transformed into differentiable constraint terms using a physical information neural network to construct a residual loss function. A load change inflection point layer is constructed based on the LSTM-CRF model and by introducing the historical feature difference of the lag time window. A joint probability distribution layer of meteorological, social, and load is constructed based on the Gaussian Copula function; The computational output layer is constructed based on Monte Carlo sampling.

[0048] It should be noted that a relative position encoding formula is introduced to address the problem that traditional Transformers are insufficient in capturing long-term load dependencies; where is a learnable relative position matrix and is a learnable relative position weight vector. The gating attention mechanism enables the model to adaptively adjust feature importance based on the characteristics of the prediction period. To address the lag effect of load changes (e.g., air conditioning requires 1-2 hours of spatiotemporal calibration to reach a stable load), a Long Short-Term Memory (LSTM)-Spatial-Dimensional Calibration Conditional Random Field (LSTM-CRF) model is employed. This model incorporates the difference in historical load features from 6 hours of spatiotemporal calibration, and uses sequence probability calculations in the spatiotemporal calibration layer of the CRF to accurately pinpoint the inflection point of load changes, thus solving the problem of traditional models neglecting load response delays. The temporal correlation features captured by the LSTM-CRF model are used as "additional constraints" and integrated into subsequent multi-factor coupling modeling to ensure the temporal logical consistency of the coupled predictions.

[0049] The relative position encoding is as follows: , where A i,j q represents the attention score from position i to j; i and j are the indices of the i-th and j-th positions in the sequence; i Let k be the query vector at the i-th position; j Let R be the key vector at the j-th position; i-j This is a relative position encoding matrix, where w represents the relative distance between encoding positions i and j. R It is a learnable relative position weight vector.

[0050] Furthermore, the gate-controlled attention mechanism is as follows: ; The weight matrix is ​​a learnable matrix. This is a vector concatenated from meteorological and policy characteristics. It is the sigmoid activation function.

[0051] Furthermore, the residual loss function is , Let be the partial derivative of the load with respect to temperature, and k be the thermal conductivity coefficient. For the outside temperature, To set the temperature, This is a regularization coefficient. This content forces the model to "learn physical logic" during training—avoiding the output of purely data-driven models that violate physical common sense due to historical data noise (such as abnormal meter readings or short-term load fluctuations). For example, it can lead to contradictory predictions such as "temperature rises sharply but air conditioning load decreases." This ensures that the prediction results are consistent with the physical logic of actual electricity consumption scenarios.

[0052] Based on the above technical solution, the active load prediction method for a low-voltage distribution substation provided in this application further includes: obtaining a load change risk warning signal based on the load prediction value.

[0053] In some implementations, obtaining the load change risk warning signal based on the load forecast value includes: The load current signal is decomposed into three levels using wavelet packet transform to obtain time-frequency energy sequences of several sub-bands. Constructing a multi-scale detection index based on different frequency bands—energy gradient change rate. The normalized rate of change of energy gradient is calculated by weighted summation. Among them, the rate of change of energy gradient Through calculation formula Calculated; For time intervals; For the k-th sub-band at t- Energy at time t; Let Ek,t be the variance of the energy sequence of the kth sub-band over the entire time range [1,T]. Let k be the rate of change of the energy gradient of the k-th sub-band. The energy gradient change rate index for node i; Construct a two-input fuzzy inference: using the gradient mean with a window length of x seconds. and variance For input, define a dynamic threshold fuzzy rule base (e.g., when the gradient mean...). and variance When all values ​​are greater than 3, the dynamic alarm threshold is 9. The dynamic threshold is updated minute-by-minute by defuzzifying the output using the centroid method. Define a fuzzy rule base for risk levels (e.g.: >8 and When the value is greater than 4, the risk level is 8, based on the dynamic threshold and the rate of change of the energy gradient. Matching risk level L; By establishing a risk propagation equation, the spatiotemporal diffusion pattern of mutation signals in the transformer area topology is quantified to obtain the risk source node; The risk level L, the propagation speed v, and the impact range S should each be assigned a color tone. saturation and transparency The dynamic color code for early warning is obtained; wherein, the propagation speed v is calculated using the formula... Get, d ij Let Δt be the electrical distance between nodes i and j. ij The time difference between the first time the rate of change of the energy gradient of nodes i and j exceeds a set threshold is defined as N, where N is the number of neighboring nodes of node j; and the influence range S is the number of nodes whose risk influence score is greater than the set threshold.

[0054] It should be noted that by overlaying the SCADA system onto the geographic wiring diagram to achieve second-level visualization, a solution from micro-signal perception to macro-risk early warning is provided for high-density commercial load scenarios.

[0055] Furthermore, the risk source node is the node corresponding to the maximum risk impact score; wherein, the risk impact score is calculated using the formula... The calculation yields Rj(t), where Rj(t) is the risk influence score of node j at time t, and N(j) is the set of neighboring nodes of node j. Let be the electrical coupling weight from node i to j, tonset be the mutation start time, and α be the time decay factor.

[0056] To address power grid safety risks, the load mutation risk early warning system innovatively integrates wavelet packet transform and mutation detection theories, constructs multi-scale gradient monitoring indicators, combines a digital twin platform to perform risk propagation simulation, and develops an HSV color space early warning coding scheme to achieve a 10-second anomaly detection response.

[0057] For example, please refer to Figure 4 As shown in the figure, the actual load curve exhibited significant peak fluctuations from 18:00 on August 1st to 7:00 the following day (marked in gray) due to the typhoon's passage, with a maximum load value reaching 693MW, a surge of 38.6% compared to the baseline load. The Transformer-XL architecture's dynamic gating attention mechanism automatically increased the meteorological feature weight to 0.67 (policy feature weight 0.33) during the typhoon warning period. The predicted mean curve (red dashed line) accurately tracked the load change trend, showing an upward trend 5 hours before the abrupt change (13:00 on August 1st), with a peak prediction error of only 1.8%. Furthermore, it perfectly replicated the gradient decay characteristics caused by policy adjustments during the load decline phase. Notably, the 95% confidence interval (orange area) generated by the model through Copula probability prediction intelligently widened during the core typhoon period, reaching a maximum width of ±60MW. This accurately quantified the dual uncertainties brought about by sudden temperature changes and policy implementation lags, while traditional deterministic prediction methods had a fixed error band of ±20MW in this scenario, failing to reflect dynamic risk changes.

[0058] The foregoing primarily describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as an electronic device, includes at least one of the hardware structures and software modules corresponding to the execution of each function in order to achieve the aforementioned functions. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0059] This application embodiment can divide the electronic device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0060] When using integrated units, Figure 5 A possible structural schematic diagram of the electronic device (referred to as electronic device 50) involved in the above embodiments is shown. The electronic device 50 includes a processing unit 501 and a communication unit 502, and may also include a storage unit 503. Figure 5 The structural diagram shown can be used to illustrate the structure of the electronic device involved in the above embodiments.

[0061] when Figure 5 The schematic diagram shown is used to illustrate the structure of the electronic device involved in the above embodiments. The processing unit 501 is used to control and manage the operation of the electronic device, the communication unit 502 is used for the electronic device to communicate with other devices, and the storage unit 503 is used to store the program code and data of the electronic device.

[0062] For example, communication unit 502 is used to collect multi-source data of the transformer area; obtain meteorological forecast data of the transformer area; and obtain power control policies. The processing unit 501 is used to clean the multi-source data of the transformer area to obtain electrical measurement data; and to perform tensor transformation on the meteorological forecast data to obtain a three-dimensional tensor. The semantic vectors and effective time of the electricity regulation policy are extracted by the BERT model. The effective time of the electricity regulation policy is then temporally aligned with the three-dimensional tensor to form a bimodal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probabilistic prediction model, which outputs load prediction values ​​and confidence interval boundaries. The multi-factor coupled probabilistic prediction model is constructed based on a Physical Information Neural Network (PINN) and an LSTM-CRF model. The processing unit 501 can be a processor or a controller, and the communication unit 502 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 503 can be a memory. When the electronic device 50 is a chip, the processing unit 501 can be a processor or a controller, and the communication unit 502 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 503 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0063] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the electronic device 50 can be considered as the communication unit 502 of the electronic device 50, and the processor with processing functions can be considered as the processing unit 501 of the electronic device 50. Optionally, the device in the communication unit 502 that implements the receiving function can be considered as a communication unit. The communication unit is used to execute the receiving steps in the embodiments of this application, and the communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 502 that implements the transmitting function can be considered as a transmitting unit. The transmitting unit is used to execute the transmitting steps in the embodiments of this application, and the transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0064] Figure 5 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0065] Figure 5The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0066] This application also provides a hardware structure diagram of a processing device (referred to as processing device 60), see [link to diagram]. Figure 6 The processing device 60 includes a processor 601, and optionally, a memory 602 connected to the processor 601.

[0067] In the first possible implementation, see Figure 6 The processing device 60 also includes a transceiver 603. The processor 601, memory 602, and transceiver 603 are connected via a bus. The transceiver 603 is used to communicate with other devices or communication networks. Optionally, the transceiver 603 may include a transmitter and a receiver. The device in the transceiver 603 that implements the receiving function can be considered as a receiver, which is used to perform the receiving steps in the embodiments of this application. The device in the transceiver 603 that implements the transmitting function can be considered as a transmitter, which is used to perform the transmitting steps in the embodiments of this application.

[0068] Based on the first possible implementation method Figure 6 The schematic diagram shown can be used to illustrate the structure of the processing device involved in the above embodiments.

[0069] in, Figure 6 Alternatively, the system chip in the processing device can be illustrated. In this case, the actions performed by the aforementioned processing device can be implemented by the system chip, and the specific actions performed can be found above, and will not be repeated here.

[0070] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

[0071] The processor in this application may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., which are various computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform calculations or processing. The processor may be a separate semiconductor chip or integrated with other circuits into a single semiconductor chip. For example, it may be integrated with other circuits (such as encoding / decoding circuits, hardware acceleration circuits, or various bus and interface circuits) to form a SoC (System-on-a-Chip), or it may be integrated as a built-in processor within an ASIC. The ASIC with the integrated processor may be packaged separately or together with other circuits. In addition to the cores for executing software instructions to perform calculations or processing, the processor may further include necessary hardware accelerators, such as field-programmable gate arrays (FPGAs), PLDs (programmable logic devices), or logic circuits that implement dedicated logic operations.

[0072] The memory in the embodiments of this application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable-only memory (EEPROM). In some scenarios, the memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto.

[0073] This application also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0074] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0075] This application also provides a chip including a processor and an interface circuit. The interface circuit is coupled to the processor. The processor is used to run computer programs or instructions to implement the above-described method. The interface circuit is used to communicate with other modules outside the chip.

[0076] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer 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, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0077] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0078] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A method for predicting active load in a low-voltage distribution transformer area, characterized in that, include: Multi-source data from the transformer substation area is collected and cleaned to obtain electrical measurement data; Obtain meteorological forecast data for the location of the substation area and perform tensor transformation to obtain a three-dimensional tensor; The electricity regulation policy is obtained, and semantic vectors and the effective time of the electricity regulation policy are extracted through the BERT model. The effective time of the electricity regulation policy is temporally aligned with the three-dimensional tensor to form a dual-modal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probability prediction model, and the output includes load prediction values; wherein, the multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

2. The active load prediction method for a low-voltage distribution transformer area according to claim 1, characterized in that, The methods for acquiring the electrical measurement data include: Acquire multi-source heterogeneous data by integrating smart meters; The spatiotemporal dimension calibration of multi-source heterogeneous data is performed using an adaptive Kalman filter algorithm to obtain calibration data; The calibration data is processed through a hybrid data cleaning pipeline to obtain electrical measurement data; wherein the hybrid data cleaning pipeline includes the Isolation Forest algorithm, K-Nearest Neighbors, wavelet transform, hash time-locked contract, and knowledge graph.

3. The active load prediction method for a low-voltage distribution transformer area according to claim 2, characterized in that, The process of processing calibration data through a hybrid data cleaning pipeline includes: The isolated forest algorithm is used to detect and remove outliers from the calibration data. The K-nearest neighbor algorithm is used to fill in the missing points in the calibration data after removing outliers. Finally, wavelet transform is used to filter the filled calibration data to obtain the corrected data. The corrected data is processed by a hash time-locked contract to obtain data blocks with trust labels; a device association model is constructed based on a knowledge graph, and the transformer area topology relationship is analyzed by a graph attention network to obtain the transformer area physical topology; the data blocks with trust labels are bound to the transformer area physical topology to obtain electrical measurement data.

4. The active load prediction method for a low-voltage distribution transformer area according to claim 1, characterized in that, The methods for obtaining the three-dimensional tensor include: The Z-score normalization method is used to eliminate the dimensional differences in meteorological forecast data. The meteorological forecast data is sorted according to the collection time to obtain a forecast sequence. The meteorological forecast data includes temperature, humidity and wind speed. pass The predicted sequence is processed to obtain a three-dimensional tensor; where, , , Let represent the standardized temperature, humidity, and wind speed at hour t, respectively, and n be the future time of the weather forecast data, where t∈n.

5. The active load prediction method for a low-voltage distribution transformer area according to claim 1, characterized in that, The construction method of the multi-factor coupled probability prediction model includes: A spatiotemporal feature extractor is constructed based on the Transformer-XL architecture and by introducing relative position encoding and gated attention mechanisms. The thermodynamic equations are transformed into differentiable constraint terms using a physical information neural network to construct a residual loss function. A load change inflection point layer is constructed based on the LSTM-CRF model and by introducing the historical feature difference of the lag time window. A joint probability distribution layer of meteorological, social, and load is constructed based on the Gaussian Copula function; The computational output layer is constructed based on Monte Carlo sampling.

6. The active load prediction method for a low-voltage distribution transformer area according to claim 5, characterized in that, The relative position is encoded as Among them, A i,j q represents the attention score from position i to j; i and j are the indices of the i-th and j-th positions in the sequence; i Let k be the query vector at the i-th position; j Let R be the key vector at the j-th position; i-j This is a relative position encoding matrix, where w represents the relative distance between encoding positions i and j. R It is a learnable relative position weight vector.

7. The active load prediction method for a low-voltage distribution transformer area according to claim 5, characterized in that, The gate control attention mechanism is as follows: ; The weight matrix is ​​a learnable matrix. This is a vector concatenated from meteorological and policy characteristics. It is the sigmoid activation function.

8. The active load prediction method for a low-voltage distribution transformer area according to claim 5, characterized in that, The residual loss function is: , Let be the partial derivative of the load with respect to temperature, and k be the thermal conductivity coefficient. For the outside temperature, To set the temperature, is the regularization coefficient.

9. The active power load prediction method for a low-voltage distribution transformer area according to claim 1, characterized in that, Also includes: Early warning signals for load change risks are obtained based on load forecast values.

10. The active load prediction method for a low-voltage distribution transformer area according to claim 9, characterized in that, The step of obtaining a load change risk warning signal based on load forecast values ​​includes: The load current signal is decomposed into three levels using wavelet packet transform to obtain time-frequency energy sequences of several sub-bands. Constructing a multi-scale detection index based on different frequency bands—energy gradient change rate. The normalized rate of change of energy gradient is calculated by weighted summation. Among them, the rate of change of energy gradient Through calculation formula Calculated; For time intervals; For the k-th sub-band at t- Energy at time t; Let Ek,t be the variance of the energy sequence of the kth sub-band over the entire time range [1,T]. Let k be the rate of change of the energy gradient of the k-th sub-band. The energy gradient change rate index for node i; Construct a two-input fuzzy inference: using the gradient mean with a window length of x seconds. and variance As input, a dynamic threshold fuzzy rule base is defined, and the fuzzy output is a dynamically updated threshold that is updated every minute, defuzzified using the centroid method. Define a fuzzy rule base for risk levels, based on dynamic thresholds and energy gradient change rates. Matching risk level L; By establishing a risk propagation equation, the spatiotemporal diffusion pattern of mutation signals in the transformer area topology is quantified to obtain the risk source node; The risk level L, the propagation speed v, and the impact range S should each be assigned a color tone. saturation and transparency The dynamic color code for early warning is obtained; wherein, the propagation speed v is calculated using the formula... Get, d ij Let Δt be the electrical distance between nodes i and j. ij The time difference between the first time the rate of change of the energy gradient of nodes i and j exceeds a set threshold is defined as N, where N is the number of neighboring nodes of node j; and the influence range S is the number of nodes whose risk influence score is greater than the set threshold.

11. The active load prediction method for a low-voltage distribution transformer area according to claim 10, characterized in that, The risk source node is the node corresponding to the maximum risk impact score; whereby the risk impact score is calculated using the formula... The calculation yields R; where R is the largest known value. j (t) represents the risk impact score of node j at time t, and N(j) represents the set of neighboring nodes of node j. Let t be the electrical coupling weight from node i to j. onset α represents the mutation initiation time, and α is the time decay factor.

12. An electronic device, characterized in that, The device includes: a communication unit and a processing unit; The communication unit is used to collect multi-source data from the transformer substation area; obtain meteorological forecast data for the location of the transformer substation area; and obtain power control policies. The processing unit is used to clean the multi-source data of the transformer area to obtain electrical measurement data; and to perform tensor transformation on the meteorological forecast data to obtain a three-dimensional tensor. The semantic vectors and effective time of the electricity regulation policy are extracted by the BERT model. The effective time of the electricity regulation policy is then temporally aligned with the three-dimensional tensor to form a bimodal input matrix. The dual-modal input matrix and electrical measurement data are input into a multi-factor coupled probability prediction model, and the output includes the load prediction value and the confidence interval boundary; wherein, the multi-factor coupled probability prediction model is constructed based on the physical information neural network PINN and the LSTM-CRF model.

Images