Distribution network synchronous phasor measurement method and system based on situation awareness technology

Abstract

The invention discloses a distribution network synchronous phasor measurement method and system based on a situation awareness technology, and relates to the technical field of power system state estimation and uncertainty modeling, and the method comprises the steps: collecting and synchronously obtaining the multi-node phasor data of a power distribution network, and constructing a state input vector for situation prediction; a neural network model with uncertain output is constructed, and a predicted value and a confidence interval are obtained based on multiple times of sampling; an interlayer uncertainty regularization mechanism is introduced into model training, and the relative change range of standard deviation output by each layer is limited. According to the method, the synchronous phasor measurement data of the power system and the uncertainty perception modeling are deeply fused, so that the accurate prediction of the operation situation of the power distribution network can be realized, the confidence level information can be output, and the uncertainty propagation in the model can be dynamically controlled; the method greatly improves the stability, reliability and interpretability of a prediction result, and has good engineering application prospects and technical innovation values.

Description

Technical Field

[0001] This invention relates to the field of power system state estimation and uncertainty modeling technology, specifically to a distribution network synchronization phasor measurement method and system based on situational awareness technology. Background Technology

[0002] With the development of smart grids and the widespread integration of distributed power sources, the operating conditions of distribution networks are becoming increasingly complex, placing higher demands on voltage stability, frequency consistency, and fault detection capabilities. Synchronous phasor measurement units (PMUs), as a key means of achieving wide-area monitoring and situation estimation, have gradually expanded from transmission networks to distribution networks in recent years, with low-voltage side equipment such as μPMUs being deployed. Simultaneously, situation awareness theory, originating from advanced sensing and decision-making methods in fields such as aviation, has also been introduced into power system monitoring to enhance the system's ability to recognize, understand, and predict its operating conditions, providing a data and model foundation for adaptive distribution network control.

[0003] Although synchronous phasor measurement and situational awareness technologies have been applied to power system state estimation and prediction to some extent, current technical solutions still have several key shortcomings in practical engineering, making it difficult to support the highly dynamic and uncertain operating characteristics of distribution networks. First, current situational modeling methods are mostly based on deterministic neural network structures, lacking the ability to model fluctuations in the quality of input observation data or changes in data reliability under sparse PMU node layouts. This results in the model output being highly sensitive to input disturbances, making it difficult to quantify the confidence level of the prediction results. Second, while some studies have attempted to introduce uncertainty modeling methods, such as Bayesian networks or MC Dropout, these often only remain at the output layer level, neglecting the uncertainty propagation and inter-layer amplification effects within the neural network structure, leading to overall model instability and difficulty in practical application. Third, there is currently no regularization design mechanism for layer-by-layer control of neural network uncertainty in distribution network scenarios, making it impossible to constrain nonlinear distortions during model training. Furthermore, existing methods typically only provide a single prediction value, lacking confidence intervals or risk level output mechanisms that match the actual needs of power system operation, resulting in the inability of situational visualization platforms to perform confidence-level processing of prediction results. Therefore, existing technologies cannot achieve a distribution network situational awareness method with output reliability control capabilities, especially in terms of standard deviation control, confidence output, and structural stability, and it is difficult to achieve the technical effects achieved by this invention. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing distribution network synchronous phasor measurement and situation prediction methods lack the ability to model measurement input uncertainties, lack control mechanisms for inter-layer uncertainty propagation in the model structure, and cannot generate prediction results with confidence labels. The problem is how to realize an uncertain situation prediction model that can output confidence intervals and has stability control.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a distribution network synchronous phasor measurement method based on situational awareness technology, comprising collecting and synchronously acquiring phasor data of multiple nodes in the distribution network, constructing a state input vector for situational prediction; constructing a neural network model with uncertain output, obtaining predicted values ​​and confidence intervals based on multiple samplings; and introducing an inter-layer uncertainty regularization mechanism in model training to limit the relative variation range of the standard deviation of each layer's output.

[0007] As a preferred embodiment of the distribution network synchronous phasor measurement method based on situational awareness technology described in this invention, the method of collecting and synchronously acquiring multi-node phasor data of the distribution network includes: collecting voltage, current, frequency, and phase angle information using synchronous phasor measurement devices configured at substations, feeder branches, or load ends; uniformly time-marking the data using GPS or IEEE 1588; aligning all sampled data to the same time axis; generating an original state matrix by sorting the data in the system according to node number and sampling time; and converting the original state matrix into a state input vector using a fixed sliding window strategy after outlier removal, linear interpolation completion, and normalization, with a window length of [missing information]. The step size is 1.

[0008] As a preferred embodiment of the distribution network synchronization phasor measurement method based on situational awareness technology described in this invention, the construction of the state input vector for situational prediction includes extracting feature vectors for each sampling time from the original state matrix. These features include voltage amplitude, voltage phase angle, current amplitude, current phase angle, frequency, and inter-node phase angle difference, and are organized into a continuous time series. Frame input, input format is: in, It is a tensor form of the state input vector, representing the sample data input to the model. For feature dimension, Let represent the set of real numbers. The model adopts a two-branch neural network structure. One branch outputs the predicted situation value at time H in the future, and the other branch uses Monte Carlo Dropout to propagate forward multiple times and output the distribution of the prediction results. The mean is used for situation estimation, and the standard deviation is used for uncertainty calculation. The number of samplings is no less than 20, and the Dropout rate is fixed at 0.2.

[0009] As a preferred embodiment of the distribution network synchronization phasor measurement method based on situational awareness technology described in this invention, the constructed neural network model with uncertain output includes an input layer, two GRU hidden layers, and a dual-branch output layer. The activation function in the GRU layer is Tanh, and Dropout is enabled in both the training and inference phases. The Dropout position is located inside the GRU unit and at the output connection. In the inference phase, 20 forward propagations are performed, and the predicted output is recorded each time. The mean of the final output is used as the predicted value, and the standard deviation is used as the confidence interval boundary. The training loss function includes both the prediction mean square error term and the negative log-likelihood term.

[0010] As a preferred embodiment of the distribution network synchronization phasor measurement method based on situational awareness technology described in this invention, the introduction of an inter-layer uncertainty regularization mechanism includes: calculating the standard deviation of the activation value distribution after each hidden layer output, and comparing it proportionally with the standard deviation of the previous layer; when the standard deviation of the current layer is greater than a set proportional coefficient... When multiplying by the standard deviation of the previous layer, a penalty term is added to the total training loss function. Represented as: Total loss function Represented as: in, For the prediction error term, For the first The stratified standard deviation ranges from [1.1, 1.5]. This is the regularization factor.

[0011] As a preferred embodiment of the distribution network synchronization phasor measurement method based on situational awareness technology described in this invention, the introduction of the inter-layer uncertainty regularization mechanism further includes: when the output standard deviation of a certain hidden layer exceeds the upper limit of the threshold in K consecutive training iterations, the system automatically triggers a structure compression mechanism, including reducing the number of neurons in the current layer by 20%-50% or replacing the activation function with linear activation, and only taking effect under the condition that the validation set loss does not decrease significantly within a specified number of iterations; the standard deviation changes during the compression operation are recorded in real time to the log for analysis of the uncertainty change trend during training; a complete validation is performed after each structure compression of the model.

[0012] As a preferred embodiment of the distribution network synchronization phasor measurement method based on situational awareness technology described in this invention, the limitation of the relative variation range of the standard deviation of each layer's output includes setting a set of risk level thresholds. When the standard deviation is less than the first threshold A time marker is considered a reliable state when the standard deviation is within the first threshold. With the second threshold The time interval is marked as a medium confidence state when the standard deviation is greater than the second threshold. The time stamp is used to indicate a high uncertainty situation. The status stamp is pushed to the distribution network situation awareness visualization platform via a unified data interface in the form of a structured message. The message structure includes the measurement point number, confidence level, prediction standard deviation, and timestamp field. Based on the received confidence level stamp information, the visualization platform renders the corresponding measurement point on the situation layer using different color codes: green for a reliable situation, orange for a medium-reliability situation, and red for a high uncertainty situation. Layer rendering uses an SVG graphics interface to achieve color overlay. The platform will detect situations where the confidence level exceeds the warning level threshold. When the status is reached, the status is automatically written to the risk event log. The log records include fields such as measurement point ID, predicted value, standard deviation, level label, occurrence time, and response status. The system provides filtering by confidence level, filtering by time interval, and measurement point tracking functions.

[0013] Another objective of this invention is to provide a distribution network synchronization phasor measurement system based on situational awareness technology.

[0014] As a preferred embodiment of the distribution network synchronous phasor measurement system based on situational awareness technology described in this invention, it includes: a vector construction module, a sampling processing module, and a distribution network risk analysis module; the vector construction module is used to collect and synchronously acquire phasor data of multiple nodes in the distribution network to construct a state input vector for situational prediction; the sampling processing module is used to construct a neural network model with uncertain output, and obtain predicted values ​​and confidence intervals based on multiple samplings; the distribution network risk analysis module is used to introduce an inter-layer uncertainty regularization mechanism in model training to limit the relative variation range of the standard deviation of the output of each layer.

[0015] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a distribution network synchronization phasor measurement method based on situational awareness technology.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a distribution network synchronization phasor measurement method based on situational awareness technology.

[0017] The beneficial effects of this invention are as follows: The distribution network synchronization phasor measurement method based on situational awareness technology provided by this invention deeply integrates the power system synchronization phasor measurement data with uncertainty perception modeling. It can not only achieve accurate prediction of the distribution network operation status, but also output confidence level information and dynamically control the propagation of uncertainty within the model. This greatly improves the stability, reliability and interpretability of the prediction results, and has good engineering application prospects and technological innovation value. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 The above is an overall flowchart of a distribution network synchronization phasor measurement method based on situational awareness technology provided in Embodiment 1 of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Example 1, referring to Figure 1 As an embodiment of the present invention, a distribution network synchronization phasor measurement method based on situational awareness technology is provided, comprising: S1: Collect and synchronously acquire phasor data of multiple nodes in the distribution network to construct a state input vector for situation prediction.

[0022] Furthermore, the acquisition and synchronous acquisition of multi-node phasor data in the distribution network includes using synchronous phasor measurement devices configured at substations, feeder branches, or load ends to collect voltage, current, frequency, and phase angle information, and performing unified time stamping via GPS or IEEE 1588. All sampled data are uniformly aligned to the same time axis, and after being sorted by node number and sampling time in the system, an original state matrix is ​​generated. After outlier removal, linear interpolation completion, and normalization, the original state matrix is ​​converted into a state input vector using a fixed sliding window strategy, with a window length of [missing information]. The step size is 1.

[0023] It should be noted that constructing the state input vector for situation prediction involves extracting the feature vector for each sampling moment from the original state matrix. The features include voltage magnitude, voltage phase angle, current magnitude, current phase angle, frequency, and inter-node phase angle difference, and are organized into a continuous time series. Frame input, input format is: in, It is a tensor form of the state input vector, representing the sample data input to the model. For feature dimension, Let represent the set of real numbers. The model adopts a two-branch neural network structure. One branch outputs the predicted situation value at time H in the future, and the other branch uses Monte Carlo Dropout to propagate forward multiple times and output the distribution of the prediction results. The mean is used for situation estimation, and the standard deviation is used for uncertainty calculation. The number of samplings is no less than 20, and the Dropout rate is fixed at 0.2.

[0024] It should also be noted that by deploying synchronous phasor measurement devices at key nodes of the distribution network (such as substations, feeder branches, and load ends), real-time acquisition of electrical physical quantities such as voltage, current, frequency, and phase angle is achieved. The data is then time-stamped using a high-precision time synchronization protocol (such as GPS or IEEE 1588), ensuring temporal consistency and spatial matching of the measurement data. Subsequently, the measurement data is organized into a state matrix, and after outlier removal, interpolation completion, and normalization, it is converted into a state input vector using a sliding time window. This ensures that the input data has good structural integrity, continuity, and usability, and effectively reflects the dynamic correlation between the temporal and spatial relationships of each measurement point. S2: Construct a neural network model with uncertain output, and obtain the predicted value and confidence interval based on multiple samplings.

[0025] Furthermore, the neural network model with uncertain output is constructed by including an input layer, two GRU hidden layers, and a dual-branch output layer. The activation function in the GRU layer is Tanh, and Dropout is enabled in both the training and inference phases. The Dropout position is located inside the GRU unit and at the output connection. In the inference phase, 20 forward propagations are performed, and the predicted output is recorded each time. The mean of the final output is used as the predicted value, and the standard deviation is used as the confidence interval boundary. The training loss function includes both the prediction mean square error term and the negative log-likelihood term.

[0026] It should be noted that by constructing a dual-branch neural network structure, one branch outputs the situation prediction value at future time points, while the other branch introduces Monte Carlo Dropout technology during the training and inference phases. By retaining Dropout and performing multiple forward propagations, distributed prediction results are obtained at the output end. The mean of these results is used for situation estimation, and the standard deviation is used for uncertainty calculation. At least 20 sampling operations are performed during inference to ensure statistical accuracy. This establishes a coupled expression mechanism between predicted values ​​and uncertainties in distribution network situation prediction, enabling the quantification of the model's prediction reliability under current data input conditions. S3: Introduce an inter-layer uncertainty regularization mechanism during model training to limit the relative variation range of the standard deviation of each layer's output.

[0027] Furthermore, an inter-layer uncertainty regularization mechanism is introduced, which includes calculating the standard deviation of the activation value distribution after each hidden layer output and comparing it proportionally with the standard deviation of the previous layer. If the standard deviation of the current layer is greater than a set proportional coefficient, the mechanism will be applied. When multiplying by the standard deviation of the previous layer, a penalty term is added to the total training loss function. Represented as: Total loss function Represented as: in, For the prediction error term, For the first The stratified standard deviation ranges from [1.1, 1.5]. This is the regularization factor.

[0028] It should be noted that the introduction of the inter-layer uncertainty regularization mechanism also includes the following: when the output standard deviation of a certain hidden layer exceeds the upper limit of the threshold in K consecutive training iterations, the system automatically triggers the structure compression mechanism, including reducing the number of neurons in the current layer by 20%-50% or replacing the activation function with linear activation, and this only takes effect when the validation set loss does not decrease significantly within a specified number of iterations; the standard deviation changes during the compression operation are recorded in real time to the log for analysis of the uncertainty change trend during training; and a complete validation is performed after each structure compression of the model.

[0029] It should also be noted that limiting the relative variation range of the standard deviation of each layer's output includes setting a set of risk level thresholds. When the standard deviation is less than the first threshold A time marker is considered a reliable state when the standard deviation is within the first threshold. With the second threshold The time interval is marked as a medium confidence state when the standard deviation is greater than the second threshold. The time stamp is used to indicate a high uncertainty situation. The status stamp is pushed to the distribution network situation awareness visualization platform via a unified data interface in the form of a structured message. The message structure includes the measurement point number, confidence level, prediction standard deviation, and timestamp field. Based on the received confidence level stamp information, the visualization platform renders the corresponding measurement point on the situation layer using different color codes: green for a reliable situation, orange for a medium-reliability situation, and red for a high uncertainty situation. Layer rendering uses an SVG graphics interface to achieve color overlay. The platform will detect situations where the confidence level exceeds the warning level threshold. When the status is reached, the status is automatically written to the risk event log. The log records include fields such as measurement point ID, predicted value, standard deviation, level label, occurrence time, and response status. The system provides filtering by confidence level, filtering by time interval, and measurement point tracking functions.

[0030] It should also be noted that by setting a trigger condition that the standard deviation of the output of a hidden layer exceeds a set upper limit during K consecutive training iterations, network compression operations are automatically performed, including reducing the number of neurons in that layer or replacing it with a linear activation function. This operation only takes effect when the loss on the validation set shows no decreasing trend, thus realizing a mechanism that dynamically links model structure adjustment with uncertain behavior. Simultaneously, standard deviation change logs are recorded in real time for use in training process analysis. This establishes a structure-adaptive model evolution strategy with uncertainty awareness capabilities, ultimately improving the model's generalization ability, adaptability to complex data distributions, and training efficiency and stability.

[0031] Example 2, an embodiment of the present invention, provides a distribution network synchronization phasor measurement system based on situational awareness technology, including a vector construction module, a sampling processing module, and a distribution network risk analysis module.

[0032] The vector construction module is used to collect and synchronously acquire phasor data of multiple nodes in the distribution network to construct a state input vector for situation prediction; the sampling processing module is used to construct a neural network model with uncertain output and obtain the predicted value and confidence interval based on multiple samplings; the distribution network risk analysis module is used to introduce an inter-layer uncertainty regularization mechanism in the model training to limit the relative variation range of the standard deviation of each layer's output.

Claims

1. A method for distribution network synchronized phasor measurement based on situation awareness technology, characterized in that, The application relates to a power distribution network risk prediction method based on a neural network model with uncertainty output. The application comprises the following steps: Collecting and synchronously acquiring phasor data of multiple nodes of a power distribution network to construct a state input vector for situation prediction; Constructing a neural network model with uncertainty output, and obtaining a prediction value and a confidence interval based on multiple samplings; 2. The method of claim 1, wherein the method is based on a situational awareness technology-based synchronization of distribution network phasor measurement. Introducing an inter-layer uncertainty regularization mechanism in model training to limit the relative change range of the standard deviation of each layer output. The collecting and synchronously acquiring phasor data of multiple nodes of a power distribution network comprises the following steps: The original state matrix is processed by outlier rejection, linear interpolation completion and normalization, and then converted into state input vector according to fixed sliding window strategy, with window length of and step length of 1.

3. The method of claim 2, wherein the method further comprises: Synchronous phasor measurement devices configured at a transformer substation, a feeder branch or a load end are used to collect voltage, current, frequency and phase angle information, and unified time marking is performed through GPS or IEEE 1588; all sampling data are uniformly aligned to the same time axis; and an original state matrix is generated after the sampling data are sorted according to node numbers and sampling times in the system; The feature vectors of each sampling time are extracted from the original state matrix, including voltage amplitude, voltage phase angle, current amplitude, current phase angle, frequency and phase angle difference between nodes, and organized into continuous frame input, input format: wherein, is a tensor form of state input vector, representing sample data input to the model, is a feature dimension, represents a set of real numbers, the model adopts a double-branch neural network structure, one branch outputs the predicted state value at future time H, and the other branch adopts Monte Carlo Dropout mode for multiple forward propagation, outputs the predicted result distribution, the mean is used for state estimation, the standard deviation is used for uncertainty calculation, the sampling number is not less than 20 times, and the Dropout rate is fixed at 0.

2.

4. The method of claim 3, wherein the method further comprises: The constructing a state input vector for situation prediction comprises the following steps: The constructing a neural network model with uncertainty output comprises the following steps: An input layer, two GRU hidden layers and a double-branch output layer; the activation function in the GRU layer is Tanh; and Dropout is enabled in both the training and inference stages; the Dropout position is located inside the GRU unit and at the output connection; In the inference stage, 20 forward propagations are performed, the prediction output of each time is recorded, the mean value of the final output is taken as the prediction value, and the standard deviation is taken as the confidence interval boundary; 5. The distribution network synchronization phasor measurement method based on situational awareness technology as described in claim 4, characterized in that: The training loss function simultaneously contains a prediction mean square error term and a negative log-likelihood term. The standard deviation of the activation value distribution is calculated after the output of each hidden layer, and a proportional comparison is made with the standard deviation of the previous layer. When the standard deviation of the current layer is greater than a set proportional coefficient When the product of the standard deviation of the current layer and the standard deviation of the previous layer is greater than a set proportional coefficient, a penalty term is added to the total loss function of the training, and the penalty term is expressed as: Total loss function is represented as: wherein, is the prediction error term, is the first layer standard deviation, taking values in the range [1.1, 1.5], is the regularization factor.

6. The method of claim 5, wherein the method is based on a situational awareness technology-based network synchronization phasor measurement method. The introducing an inter-layer uncertainty regularization mechanism comprises the following steps: The introducing an inter-layer uncertainty regularization mechanism further comprises the following steps: When the output standard deviation of a certain hidden layer exceeds the upper limit of the threshold value for K consecutive times in the training, a structure compression mechanism is automatically triggered, including reducing the number of neurons in the current layer by 20%-50% or replacing the activation function with a linear activation function, and the mechanism only takes effect under the condition that the validation set loss does not obviously decrease within a specified number of iteration rounds; The standard deviation variation during the compression operation process is recorded in real time to a log, which is used for analyzing the uncertainty change trend in the training process; 7. The method of claim 6, wherein the method further comprises: determining a set of parameters for the set of sensors based on the set of parameters for the set of sensors and the set of parameters for the set of actuators. After the structure compression is performed once, a complete verification is performed once. Setting a set of risk level thresholds , when the standard deviation is less than a first threshold a credible posture, when the standard deviation is between the first threshold and a second threshold and a high uncertainty posture when the standard deviation is greater than the second threshold ; The limiting the relative change range of the standard deviation of each layer output comprises the following steps: The state marking is pushed to a power distribution network situation awareness visualization platform in a structured message form through a unified data interface; the message structure comprises a measuring point number, a confidence level, a prediction standard deviation value and a timestamp field; The platform automatically writes the state into a risk event log when it receives a state with a confidence level higher than a warning level threshold The log records include fields such as measurement point ID, predicted value, standard deviation, level mark, occurrence time, response state, etc. The system provides filtering by confidence level, time interval screening, and measurement point tracking functions.

8. A power distribution network synchronous phasor measurement system based on the situational awareness technology, adopting the power distribution network synchronous phasor measurement method based on the situational awareness technology according to any one of claims 1-7, characterized in that: The visualization platform renders the corresponding measuring point on a situation layer in different color coding modes according to the received confidence level marking information; green corresponds to a credible situation, orange corresponds to a medium credible situation, and red corresponds to a high uncertainty situation; and layer rendering is realized by using an SVG graphical interface to realize color overlay. The application comprises a vector construction module, a sampling processing module and a power distribution network risk analysis module. The vector construction module is used for collecting and synchronously acquiring phasor data of multiple nodes of a power distribution network to construct a state input vector for situation prediction. The sampling processing module is used for constructing a neural network model with uncertainty output, and obtaining a prediction value and a confidence interval based on multiple samplings. The power distribution network risk analysis module is used for introducing an inter-layer uncertainty regularization mechanism in model training to limit the relative change range of the standard deviation of each layer output. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The computer program is executed by the processor to implement the steps of the power distribution network synchronization phasor measurement method based on the situational awareness technology in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the power distribution network synchronization phasor measurement method based on the situational awareness technology in any one of claims 1 to 7.

Images