A transformer area fusion terminal state estimation method, system, device and medium

Abstract

This invention belongs to the field of transformer substation fusion terminal state estimation technology, and discloses a method, system, device, and medium for transformer substation fusion terminal state estimation to solve the problems of existing technologies in simultaneously addressing the challenges of sample screening and retention, directional fusion of multi-source measurements, and rapid recovery from sudden changes in operating conditions. The method includes: acquiring measurement data for the current sampling period; using a default value as the initial state estimate for the first period; otherwise, calculating the surprise factor and determining the update step size based on the measurement data and the state estimation result of the previous period; when the surprise factor exceeds a preset threshold, generating an initial state estimate based on the measurement-state mapping relationship and historical snapshots; otherwise, using the result from the previous period; performing weighted least squares calculations starting from the initial state estimate to obtain a high-precision state estimate result; when low-precision measurement channel data arrives, generating a correction coefficient based on its deviation from the high-precision state estimate result and correcting along the deviation direction to obtain the final state estimate result.

Description

Technical Field

[0001] This invention belongs to the field of integrated terminal status estimation technology, specifically relating to a method, system, device and medium for integrated terminal status estimation. Background Technology

[0002] The low-voltage distribution network substation fusion terminal is the core carrier of edge sensing in the distribution network. It is responsible for aggregating multi-source measurement data from transformer outlet current transformers, voltage transformers, branch monitoring units, smart meters, etc., and for real-time sensing of the substation's operating status. With the large-scale integration of distributed photovoltaic power and frequent topology reconfiguration, the terminal needs to achieve high-precision, low-latency, and adaptive state estimation on resource-constrained embedded platforms. However, the power consumption, heat dissipation, and real-time constraints of embedded platforms mean that their computing power and storage bandwidth are far lower than those of cloud servers, making it difficult to directly deploy complex algorithms that rely on high computing power.

[0003] Currently, the weighted least squares state estimation method is mainly used for distribution network state estimation. This method solves independently for each section, does not remember historical state distributions, cannot track temporal evolution trends, lacks a memory filtering mechanism, and cannot distinguish between valuable abnormal operating conditions and random measurement noise. Furthermore, it requires three to five nonlinear iterations, with a single section delay of five to ten milliseconds, making it difficult to meet the second-level real-time requirements. Arithmetic weighted fusion is prone to losing the directional information of multi-source measurement covariance. In addition, the measurement channel weights are fixed, and abnormal channels cannot be automatically weighted down. There is a lack of fast and usable cold start recovery strategies when operating conditions change abruptly.

[0004] To address the problem of bad data in measurements, existing technologies have further developed bad data detection and timed reestimation methods based on statistical tests. However, these methods rely on the assumption of residual normality, resulting in a high rate of missed detection for non-Gaussian anomalies. They lack memory filtering logic, making it easy to store random noise in historical samples. They can only make binary judgments and cannot reduce weights channel by channel, causing the directional information of high-precision measurement covariance to be diluted by average. Furthermore, the timed reestimation cycle is fixed, resulting in wasted computing power in steady state, delayed response during periods of rapid change, and the invalidation of original time-series samples after topology changes, lacking a reliable cold start recovery mechanism.

[0005] Meanwhile, the machine learning field has proposed test-time adaptive methods based on online distribution estimation and attempted to introduce these ideas into state estimation. However, this method assumes that the feature dimensions are independent, ignoring the physical coupling between distribution network measurements; it does not set up memory partitions, so it cannot distinguish between special operating conditions and daily noise; the pseudo-label weights depend only on the classifier confidence, and abnormal measurements are prone to contaminating the distribution estimation; it uses simple weighted average fusion, which loses the covariance direction features; and it does not involve the power system measurement function, accuracy matrix, and topology matrix, so it cannot be transferred to the distribution terminal, nor does it have a cold start recovery architecture adapted to the distribution area scenario.

[0006] In summary, the above methods share three shortcomings: the memory dimension lacks a sample screening and management mechanism, making it impossible to distinguish between valuable abnormal operating conditions that should be retained and random daily noise that should be forgotten; the fusion dimension uses arithmetic average weighted fusion, which causes the directional information of high-precision measurement covariance to be blurred and lost; and the cold start dimension lacks a rapid recovery mechanism when the original historical samples become invalid after topology reconstruction, extreme weather, or large-scale switching of distributed photovoltaics. Summary of the Invention

[0007] Based on the aforementioned shortcomings and deficiencies in the prior art, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the prior art. In other words, one of the objectives of this invention is to provide a method, system, device, and medium for estimating the state of a converged terminal in a distribution area that meets one or more of the aforementioned requirements, so as to achieve low latency, low storage consumption, and adaptive state estimation on a resource-constrained embedded platform.

[0008] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for estimating the state of a converged terminal in a distribution area, comprising the following steps: Obtain measurement data for the current sampling period; If the current sampling period is the first period, then the default value is used as the initial state estimate; otherwise, The surprise factor is calculated based on the measurement data and the state estimation results of the previous period. The surprise factor is used to characterize the degree to which the current measurement data deviates from the historical state distribution. The update step size for the state estimation is determined based on the surprise level; the higher the surprise level, the larger the update step size. Determine whether the surprise level exceeds a preset threshold. If it does, start the cold start mode and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and historical snapshots in the long-term memory. If it does not exceed the threshold, use the state estimate result of the previous cycle as the initial state estimate. Starting from the initial state estimate, weighted least squares calculation is performed using real-time data from the high-precision measurement channel to obtain the high-precision state estimate result; When the data from the low-precision measurement channel arrives, a low-precision state estimation result is calculated based on the data from the low-precision measurement channel. The deviation between the high-precision state estimation result and the low-precision state estimation result is calculated. A correction coefficient is generated based on the deviation. The high-precision state estimation result is corrected along the deviation direction using the correction coefficient to obtain the final state estimation result. If the data from the low-precision measurement channel does not arrive, the high-precision state estimation result is used as the final state estimation result.

[0009] In a second aspect, the present invention provides a system for estimating the state of converged terminals in a distribution area, for implementing the method described in the first aspect, comprising: The data acquisition module is used to acquire measurement data for the current sampling period; The initialization module is used to use the default value as the initial state estimate when the current sampling period is the first period. The surprise factor calculation and dual-memory management module is used to calculate the surprise factor based on the measurement data and the state estimation result of the previous period when the current sampling period is not the first period, and to determine the update step size of the state estimation based on the surprise factor. The surprise factor is used to characterize the degree to which the current measurement data deviates from the historical state distribution. The higher the surprise factor, the larger the update step size. The dual-memory management includes a short-term memory bank and a long-term memory bank. The short-term memory bank is used to temporarily store recent samples, and the long-term memory bank is used to store historical snapshots. The long-term memory bank is managed for elimination based on the surprise factor, the duration of storage, and external scene tags. The cold start module is used to activate the cold start mode when the surprise level exceeds a preset threshold, and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and the historical snapshots in the long-term memory. The high-precision estimation module is used to perform weighted least squares calculations using real-time data from the high-precision measurement channel, starting from the initial state estimate, to obtain the high-precision state estimation result. The differential fusion module is used to calculate the low-precision state estimation result when the data from the low-precision measurement channel arrives, and generate a correction coefficient based on the deviation between the low-precision state estimation result and the high-precision state estimation result, correct the high-precision state estimation result along the deviation direction, and obtain the final state estimation result; and when the data from the low-precision measurement channel has not arrived, the high-precision state estimation result is used as the final state estimation result.

[0010] Thirdly, the present invention provides an electronic device, the electronic device including a memory, a processor and a computer program, wherein the computer program, when executed by the processor, implements the method described in the first aspect.

[0011] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention uses surprise as a unified driving index, while simultaneously controlling the update step size, cold start judgment, and dual-memory management of state estimation. Since surprise characterizes the degree to which current measurement data deviates from the historical state distribution, surprise is low under steady-state conditions, automatically reducing the update step size to resist random measurement noise; under high-surprise conditions such as photovoltaic sudden changes and switch position changes, the update step size is automatically increased to quickly track the new state. As a result, steady-state invalid calculations are reduced by approximately 95%, and latency after drastic changes in operating conditions is shortened from minutes to seconds. Compared to existing technologies with fixed measurement channel weights and the inability to automatically reduce the weight of abnormal channels, this invention achieves a scene-adaptive update rate.

[0013] 2. This invention, upon receiving low-precision measurement channel data, does not employ an arithmetic average to fuse high-precision and low-precision estimation results. Instead, it calculates the deviation between the two, generates a correction coefficient based on the deviation, and performs a unidirectional correction on the high-precision state estimation result along the deviation direction. Since the correction direction is always based on the high-precision measurement, the directional information carried by the high-precision measurement covariance matrix is ​​fully preserved, avoiding the problem of directional information being diluted by arithmetic averaging. Therefore, the RMSE of the end-point sparse measurement state estimation is reduced by 18-25% compared to the weighted average.

[0014] 3. This invention generates an initial state estimate in cold-start mode through incremental associative memory. Specifically, a first estimate is obtained by performing matrix multiplication on the current measurement data through the measurement-state mapping relationship, and this estimate is combined with historical snapshots in the long-term memory that meet the cosine similarity standard to generate an initial state estimate. Since this initial state estimate is close to the true state, the number of iterations for weighted least squares calculation is reduced from 3-5 times to 1-2 times, and the cold-start latency is reduced from 25ms to 8ms. Compared with existing technologies where time-series samples become invalid after topology changes and a reliable cold-start recovery mechanism is lacking, this invention eliminates the dependence on historical time-series change patterns.

[0015] 4. The long-term memory eviction management system of this invention uses a weighted average of three factors—surprise level, duration, and external scenario tags—to calculate a comprehensive score, with the lowest-scoring item being deleted first. Since rare voltage boundaries, fault precursors, and other operating conditions have high surprise levels and scenario importance tags, these snapshots can be retained in the long-term memory for an extended period. Meanwhile, routine, repetitive steady-state snapshots are automatically cleaned up due to their low surprise level and low scenario tag weight. Therefore, storage space is controllable, and the problem of forgetting critical operating condition modes is avoided.

[0016] 5. The entire algorithm of this invention only includes deterministic operations such as matrix factorization, matrix multiplication, and rank-1 update. It does not require backpropagation of neural networks, gradient calculations, or a large number of labeled samples. Therefore, the single-cycle time is 2ms-3ms, which can run in parallel with multiple tasks such as load forecasting and fault alarm, and is suitable for resource-constrained embedded terminals.

[0017] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, 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 This is a flowchart illustrating the method for estimating the state of converged terminals in the distribution area as described in Embodiment 1 of the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of the electronic device described in Embodiment 3 of the present invention.

[0021] Icon labels: 200. Electronic devices; 201. Processor; 202. Communication bus; 203. User interface; 204. Network interface; 205. Memory. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0023] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0024] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0025] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0026] The transformer substation fusion terminal state estimation method described in this specification is applied to transformer substation fusion terminals in low-voltage distribution networks. The transformer substation fusion terminal is the core carrier of edge sensing in the distribution network, responsible for aggregating multi-source measurement data from transformer outlet current transformers, voltage transformers, branch monitoring units, smart meters, etc., and for real-time sensing of the substation's operating status. With the large-scale integration of distributed photovoltaic systems and frequent topology reconfiguration, transformer substation fusion terminals need to achieve high-precision, low-latency, and adaptive state estimation on resource-constrained embedded platforms. In these scenarios, the application of the transformer substation fusion terminal state estimation method aims to drive the state update step size and dual-memory management through surprise factor unification, retain high-precision measurement covariance directionality information by combining unidirectional difference correction, and achieve rapid recovery after sudden changes in operating conditions using incremental associative memory.

[0027] The following is a brief explanation of the transformer fusion terminal, sampling period, measurement data, initial state estimate, default value, surprise factor, update step size, measurement-state mapping relationship, long-term memory, short-term memory, high-precision measurement channel, weighted least squares calculation, high-precision state estimation result, low-precision measurement channel, bias, and cosine similarity involved in the various embodiments of this specification: The distribution area fusion terminal is the core carrier of edge sensing in the distribution network. It is responsible for aggregating multi-source measurement data from transformer outlet current transformers, voltage transformers, branch monitoring units, smart meters, etc., and for real-time sensing of the operating status of the distribution area. It runs on a resource-constrained embedded platform.

[0028] The sampling period refers to the time interval between sampling and state estimation of multi-source measurement data by the integrated terminal of the distribution area. Each sampling period executes a complete state estimation process.

[0029] Measurement data refers to the multi-source measurement data collected by the transformer substation fusion terminal during the current sampling period, including three-phase electrical quantities at the transformer outlet, branch monitoring unit data, and smart meter measurements.

[0030] The initial state estimate serves as the starting point for the weighted least squares calculation iteration. Its sources include default values, the state estimate results from the previous cycle, or the estimate generated under cold start mode.

[0031] The default value is the initial state estimate used when the current sampling period is the first period. In this embodiment, the per-unit voltage value of each node is 1 and the phase angle is 0.

[0032] Surprise is used to characterize the degree to which the current measurement data deviates from the historical distribution. In this embodiment, the squared Mahalanobis distance is used as the metric. The higher the surprise, the more abnormal the current operating condition.

[0033] Update step size: The degree of confidence of the state estimate in the new measurement is determined by the surprise factor. The higher the surprise factor, the larger the update step size. Small step size is more resistant to noise in steady-state conditions, while large step size is more responsive to rapid changes in conditions.

[0034] Measurement-state mapping relationship: The measurement-state mapping matrix trained online is used to quickly generate initial values ​​for iteration during the cold start phase through matrix multiplication, and incremental updates are performed using the Delta rule.

[0035] The long-term memory is used to store historical snapshots. Each historical snapshot contains the state mean, covariance matrix factors, and scene importance labels. The maximum storage capacity is 500 sets. When storage overflows, the snapshots are deleted in descending order of importance.

[0036] A first-in-first-out queue with a short-term memory capacity of 60 is used to temporarily store recent samples. When the queue is full, the oldest sample is popped out.

[0037] The high-precision measurement channel includes a 0.2-class current transformer / voltage transformer, a branch box monitoring unit, and high-frequency sampling.

[0038] Weighted least squares calculation uses the power flow equation as the measurement function and solves the state estimation method of node voltage magnitude and phase angle by iteratively minimizing the weighted residual sum of squares.

[0039] The high-precision state estimation result is obtained by weighted least squares calculation using real-time data from the high-precision measurement channel, and includes the state mean and covariance matrix.

[0040] The low-precision measurement channel includes a Class 1 household smart meter, which uploads data every 15 minutes.

[0041] To measure the difference between the high-precision and low-precision state estimation results, this embodiment uses Mahalanobis distance as the metric.

[0042] Cosine similarity is a metric used to measure the degree of similarity between historical snapshots and current measurement data.

[0043] Example 1: like Figure 1 As shown, this embodiment provides a method for estimating the state of a converged terminal in a distribution area, including the following steps: Step S1: Obtain measurement data for the current sampling period: Set up a converged terminal in the transformer area N Key monitoring nodes include transformer outlets, branch boxes, and end-user locations. NThe standard value is 15 to 20. In this embodiment, the substation fusion terminal acquires measurement data for the current sampling period through high-precision and low-precision measurement channels. The high-precision measurement channel includes a 0.2-level current transformer / voltage transformer and a branch box monitoring unit, using high-frequency sampling; the low-precision measurement channel includes a 1-level household smart meter, uploading data every 15 minutes. Measurement Vector This includes three-phase electrical quantities at the transformer outlet, data from branch monitoring units, and measurements from smart meters. The value ranges from 20 to 40. In this embodiment, the effective measurement is 30, and the high-precision channel is 14.

[0044] Specifically, in this embodiment, the preferred state dimension is 36, and the expression for the state vector is: (1), In formula (1): N The total number of key monitoring nodes is selected as the optimal number. N =18; V i Represents a node i The per-unit voltage value; Represents a node i The voltage phase angle, and its unit is rad.

[0045] Specifically, the state vector The quantity to be solved, the measurement vector The two quantities are known and are related through a nonlinear measurement function. The purpose of this embodiment is to find the optimal estimate of the state vector based on the measurement vector of the current sampling period.

[0046] Step S2: If the current sampling period is the first period, then the default value is used as the initial state estimate: When the current sampling period is the first period, there is no state estimation result from the previous period, the long-term memory is empty, and the measurement-state mapping relationship has not been initialized. At this time, the default value is used as the initial state estimate, that is, the per-unit voltage value of each node is 1 and the phase angle is 0. The first period skips the surprise calculation, step size update, and cold start judgment, and directly uses the default value as the initial state estimate to enter the subsequent weighted least squares calculation.

[0047] Step S3: When the current sampling period is not the first period, calculate the surprise factor based on the measurement data and the state estimation result of the previous period. The surprise factor is used to characterize the degree to which the current measurement data deviates from the historical state distribution. Specifically, compared with the traditional fixed forgetting and timed refresh strategies, this solution uses the squared Mahalanobis distance as the surprise factor to adaptively manage short-term and long-term memory, dynamically adjust the update step size, preserve rare abnormal operating conditions, and filter normal noise.

[0048] The expression for surprise level is: (2), In formula (2): The measurement vector for the current sampling period; For nonlinear measurement functions consisting of tidal current and branch tidal currents, The measurement function predicts the value based on the state estimation results of the previous cycle; the new information covariance matrix. ,in For Jacobian matrices, The covariance matrix is ​​estimated for the state of the previous period. This is the diagonal matrix of measurement errors. Its diagonal elements are determined by the sensor's accuracy level; In engineering implementation, the Jacobian matrix adopts the CSR sparse storage format; the diagonal matrix R is directly added; and the diagonal values ​​are less than 1. e -12 This forces clamping to avoid numerical anomalies during matrix inversion; instead of directly inverting the matrix, it performs... Perform Cholesky decomposition, reducing complexity through previous and back-generation operations; if decomposition fails, switch to LDLT decomposition and save the alarm log.

[0049] Step S4: Determine the update step size for the state estimation based on the surprise level. The higher the surprise level, the larger the update step size. The update step size is determined by the degree of surprise. Obtained through sigmoid function mapping: (3), In formula (3): ; ; p =2; The chi-square threshold is set to 95% confidence for m degrees of freedom.

[0050] Surprise When the surprise factor is extremely high, the upper limit of the step size is set directly; when the surprise factor is extremely low, the minimum step size is locked to prevent floating-point overflow. Update step size The degree of confidence in new measurements for control state estimation; noise immunity with small steps in steady-state conditions; fast tracking with large steps in rapidly changing conditions.

[0051] Step S5: Determine whether the surprise level exceeds a preset threshold. If it does, start the cold start mode and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and historical snapshots in the long-term memory. If it does not exceed the threshold, use the state estimate result of the previous cycle as the initial state estimate.

[0052] When the surprise level exceeds the preset threshold ( If the cold start flag is set to True, it means that the current operating condition is out of the historical distribution, the time series prediction is invalid, and the associative memory cold start is enabled; otherwise, the cold start mode is disabled, and the state estimation result of the previous cycle is used as the initial state estimation value.

[0053] The generation of initial state estimates in cold start mode specifically includes: First, the first estimate is obtained by performing matrix multiplication on the current measurement data through the measurement-state mapping relationship: (4), In equation (4): Here is the measurement-state mapping matrix, where For the dimension of the state vector, N The total number of key monitoring nodes is selected as the optimal number. N =18, m The dimension of the measurement vector ranges from 20 to 40. This is the current measurement vector.

[0054] Secondly, all historical snapshots in the long-term memory are traversed, and the cosine similarity between each historical snapshot and the current measurement data is calculated. The historical snapshot with the highest cosine similarity is selected as the matching snapshot. After the measurement vector is normalized, the cosine similarity is calculated by traversing all long-term memory snapshots. The storage format of each snapshot in the long-term memory is as follows: ,in State mean (i.e., the state vector composed of the voltage amplitude and phase angle of all nodes in the distribution area under the operating condition corresponding to this historical snapshot). The Cholesky lower triangular factor is the covariance matrix. For the snapshot save time, This is a tag indicating the importance of the scene.

[0055] If the cosine similarity of the matched snapshot is greater than the first threshold, then the mean state value of the historical snapshot is directly used as the initial state estimate.

[0056] If the cosine similarity of the matched snapshots is between a first threshold and a second threshold, then the mean state value of the matched snapshots and the first estimated value are weighted and fused according to a preset fusion weight to generate an initial state estimate. In this embodiment, the preset fusion weight is 0.7 for the mean state value and 0.3 for the first estimated value.

[0057] If the cosine similarity of the matched snapshot is less than the second threshold, then the first estimated value is directly used as the initial state estimated value.

[0058] Preferably, the first threshold is 0.85 and the second threshold is 0.5.

[0059] When the long-term memory is empty and the measurement-state mapping relationship has not been initialized, the initial voltage value is set to 1 per unit by default, and the phase angle is set to zero.

[0060] Step S6: Starting from the initial state estimate, perform weighted least squares calculation using real-time data from the high-precision measurement channel to obtain the high-precision state estimate result.

[0061] Specifically, the initial value selection rule for weighted least squares iteration is as follows: when cold start is enabled, the initial value is taken from the associative memory output. Otherwise, inherit the results from the previous round of estimation. .

[0062] High-precision measurement channel G high This includes 0.2-class current transformers / voltage transformers, branch box monitoring units, and high-frequency sampling. The high-precision measurement incremental weighted least squares update process is as follows: Kalman gain: (5), State mean update: (6), Covariance update: (7), In equations (5) to (7): This is the Jacobian matrix corresponding to the high-precision channel; This is the diagonal matrix of measurement errors corresponding to the high-precision channel; Real-time data for high-precision measurement channels; For the measurement function corresponding to the high-precision channel; This is the updated state mean (high-precision state estimation result); This is the predicted state value, i.e., the initial state estimate of this round of weighted least squares. I It is an identity matrix.

[0063] When the matrix condition number is too high, the step size is halved and the solution is retried to obtain a high-precision state estimation result, including the state mean. With covariance matrix .

[0064] Step S7: When the data from the low-precision measurement channel arrives, calculate the low-precision state estimation result based on the data from the low-precision measurement channel, calculate the deviation between the high-precision state estimation result and the low-precision state estimation result, generate a correction coefficient based on the deviation, and use the correction coefficient to correct the high-precision state estimation result along the deviation direction to obtain the final state estimation result; if the data from the low-precision measurement channel does not arrive, then use the high-precision state estimation result as the final state estimation result.

[0065] Measurement channel grouping: The high-precision measurement channel includes 0.2-level current transformers / voltage transformers and branch box monitoring units, using high-frequency sampling; the low-precision measurement channel includes 1-level household smart meters, uploading data every 15 minutes. After the meters report valid data, high-precision estimates can be obtained respectively. Low-precision meter estimation .

[0066] Specifically, the Mahalanobis distance between the two sets of estimates is first calculated: (8), Then calculate the S-type correction factor. : (9), (10) In equations (8) to (10): For high-precision state estimation results The covariance matrix, Low-precision state estimation results The covariance matrix; The steepness parameter of the S-shaped function controls how quickly the correction coefficient changes with the Mahalanobis distance. This is a preset constant; The center threshold of the S-shaped correction function; The 0.9 quantile of a chi-square distribution with 2n degrees of freedom is a statistical critical value that can be obtained through a chi-square distribution table or numerical calculation.

[0067] The higher the surprise level, the higher the threshold. The smaller the value, the more reliable the deviation signal between the high and low estimates; the steady-state correction is conservative. The correction coefficient ranges from 0 to 1; the larger the deviation, the larger the correction coefficient. The closer it is to 1.

[0068] Then, one-way difference correction is performed: (11), (12), In equations (11) to (12): This is the corrected final state mean (final state estimation result). This is the final covariance matrix after correction.

[0069] Update the mean and covariance of the current state. When the two sets of estimates are nearly identical... If the result approaches 0, the result is fully accepted from the high-precision current transformer; if the deviation between the two is large, the high-precision estimate is corrected along the direction of the difference; unlike bidirectional averaging, it does not erase the information of the covariance direction.

[0070] If the meter measurement is missing, the differential fusion process is skipped, and the high-precision state estimation result is used as the final state estimation result.

[0071] Furthermore, this embodiment also updates the measurement-state mapping relationship based on the final state estimation result and the corresponding measurement data: Incremental update of the associative memory matrix using the Delta rule: (13) In equation (13): The square of the L2 norm of the current measurement vector. Used to prevent division by zero; Adaptive learning rate Controlled by surprise level: (14) In equation (14): lower limit of learning rate β min =0.01; Upper limit of learning rate β max =0.3; the higher the surprise factor, the faster the learning rate, and the faster the memorization of new working conditions; slow updates under steady state avoid noise overfitting.

[0072] The initialization steps for the measurement-state mapping relationship include: collecting only samples for the first 100 sampling cycles after the terminal is powered on; solving the initial matrix through ridge regression in the 100th cycle; and then switching to Delta incremental update mode. (15) In equation (15): This is a sample matrix representing the state estimation results for the first 100 sampling periods. This is a sample matrix of measurement data from the first 100 sampling periods; The regularization coefficient is . .

[0073] Furthermore, this embodiment also calculates a comprehensive score based on a weighted average of the surprise level, duration, and external scene tags, and the item with the lowest comprehensive score is deleted first: In terms of short-term memory management, the short-term memory is a 60-capacity first-in-first-out queue, storing... New samples are enqueued; if the queue is full, the oldest sample is popped. The most recent 30 samples are read. If the surprise index of all samples is below the chi-square threshold of 0.5, the region has entered a steady state. A steady-state snapshot with a surprise index close to the median is selected and stored in the long-term memory, while the short-term memory cache is cleared. If there are samples with high surprise perturbation within the interval, the short-term memory is cleared and no snapshot is saved. If there are fewer than 30 samples in the short-term memory, only enqueuing is performed.

[0074] Regarding long-term memory maintenance, the format of long-term memory array elements... ,in The mean of the state. The Cholesky lower triangular factor of the covariance matrix. For the snapshot save time, Use scene importance tags; the long-term memory can store a maximum of 500 snapshots.

[0075] When storage overflows, delete items in descending order of importance: (16) In equation (16): For the first j Importance score of each historical snapshot (retention weight); For the first j The surprise level of a snapshot is the surprise level value at the moment the snapshot is stored in long-term memory. The value at the time the snapshot is stored is recorded. As The higher the surprise factor, the rarer the snapshot and the higher its retention weight. =30; Represents the time interval between snapshots; high-priority conditions such as faults and voltage boundaries rely on scenario importance tags. Increase retention weight.

[0076] The inference controller receives topology changes and voltage limit exceedance alarms from the distribution automation master station and assigns scene weight labels to the long-term memory snapshots of operating conditions. Elimination strategies used for long-term memory.

[0077] In this embodiment, N=18, state dimension 36, effective measurement 30, and high-precision channel 14; each step of a cycle runs serially, with an overall average time of about 2.1ms; abnormal branches such as switching to LDLT when Cholesky decomposition fails, reducing the step size of the matrix due to ill-conditioning, and skipping fusion when the meter is missing are complete.

[0078] The entire algorithm only includes deterministic operations such as matrix decomposition, matrix multiplication, and rank-1 update, and does not require backpropagation of neural networks; the single cycle takes 2~3ms and can run in parallel with multiple tasks such as load forecasting and fault alarm, with strong hardware compatibility.

[0079] Example 2: This embodiment provides a converged terminal state estimation system for a distribution area, used to implement the method described in Embodiment 1, including: The data acquisition module is used to acquire measurement data for the current sampling period; The initialization module is used to use the default value as the initial state estimate when the current sampling period is the first period, and to trigger the skipping of surprise calculation and cold start judgment. The surprise factor calculation and dual-memory management module is used to calculate the surprise factor based on the measurement data and the state estimation result of the previous period when the current sampling period is not the first period, and to determine the update step size of the state estimation based on the surprise factor. The surprise factor is used to characterize the degree to which the current measurement data deviates from the historical state distribution. The higher the surprise factor, the larger the update step size. The dual-memory management includes a short-term memory bank and a long-term memory bank. The short-term memory bank is used to temporarily store recent samples, and the long-term memory bank is used to store historical snapshots. The long-term memory bank is managed for elimination based on the surprise factor, the duration of storage, and external scene tags. The cold start module is used to activate the cold start mode when the surprise level exceeds a preset threshold, and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and the historical snapshots in the long-term memory. The high-precision estimation module is used to perform weighted least squares calculations using real-time data from the high-precision measurement channel, starting from the initial state estimate, to obtain the high-precision state estimation result. The differential fusion module is used to calculate the low-precision state estimation result when the data from the low-precision measurement channel arrives, and generate a correction coefficient based on the deviation between the low-precision state estimation result and the high-precision state estimation result, correct the high-precision state estimation result along the deviation direction, and obtain the final state estimation result; and when the data from the low-precision measurement channel has not arrived, the high-precision state estimation result is used as the final state estimation result. An incremental associative memory module is used to update the measurement-state mapping relationship based on the final state estimation result and the corresponding measurement data.

[0080] Example 3: like Figure 2 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0081] The communication bus can be used to enable communication between the various components mentioned above.

[0082] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0083] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0084] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0085] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory may be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an estimation application program. The processor may be used to call the estimation application program stored in the memory and execute the steps of the substation convergence terminal state estimation method mentioned in the foregoing embodiments.

[0086] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figure 1 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0087] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The 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 specification 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 or transmitted through a computer-readable storage medium. The 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 such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0088] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0089] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0090] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0091] The foregoing description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include knowledge or means in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A method for estimating the state of a converged terminal in a distribution area, characterized in that, Including the following steps: Obtain measurement data for the current sampling period; If the current sampling period is the first period, then the default value is used as the initial state estimate; otherwise, The surprise factor is calculated based on the measurement data and the state estimation results of the previous period. The surprise factor is used to characterize the degree to which the current measurement data deviates from the historical state distribution. The update step size for the state estimation is determined based on the surprise level; the higher the surprise level, the larger the update step size. Determine whether the surprise level exceeds a preset threshold. If it does, start the cold start mode and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and historical snapshots in the long-term memory. If it does not exceed the threshold, use the state estimate result of the previous cycle as the initial state estimate. Starting from the initial state estimate, weighted least squares calculation is performed using real-time data from the high-precision measurement channel to obtain the high-precision state estimate result; When the data from the low-precision measurement channel arrives, a low-precision state estimation result is calculated based on the data from the low-precision measurement channel. The deviation between the high-precision state estimation result and the low-precision state estimation result is calculated. A correction coefficient is generated based on the deviation. The high-precision state estimation result is corrected along the deviation direction using the correction coefficient to obtain the final state estimation result. If the data from the low-precision measurement channel does not arrive, the high-precision state estimation result will be used as the final state estimation result.

2. The method according to claim 1, characterized in that, Generating initial state estimates in the cold start mode includes the following steps: The first estimated value is obtained by performing matrix multiplication on the current measurement data using the measurement-state mapping relationship; Traverse all historical snapshots in the long-term memory, calculate the cosine similarity between each historical snapshot and the current measurement data, and select the historical snapshot with the largest cosine similarity as the matching snapshot; If the cosine similarity of the matching snapshot is greater than the first threshold, then the state mean of the matching snapshot is used as the initial state estimate; the state mean is the state vector of all nodes in the transformer area under the corresponding working condition stored in a single historical snapshot, consisting of the voltage amplitude and phase angle of all nodes in the transformer area. If the cosine similarity of the matching snapshot is between the first threshold and the second threshold, then the state mean of the matching snapshot and the first estimated value are weighted and fused according to the preset fusion weight to generate an initial state estimated value. If the cosine similarity of the matched snapshot is less than the second threshold, then the first estimate is used as the initial state estimate.

3. The method according to claim 1, characterized in that: The measurement-state mapping relationship is incrementally updated using the Delta rule, and its learning rate is adaptively controlled by the surprise factor; the higher the surprise factor, the larger the learning rate.

4. The method according to claim 1, characterized in that, It also includes long-term memory eviction management steps: A comprehensive score is calculated based on the level of surprise, duration of existence, and external scene tags, with the item with the lowest comprehensive score being deleted first.

5. The method according to claim 1, characterized in that: The degree of surprise is measured using the squared Mahalanobis distance.

6. The method according to claim 1, characterized in that: The update step size is obtained by mapping the surprise level through a sigmoid function.

7. The method according to claim 1, characterized in that: The correction coefficient is calculated from the deviation using an S-shaped function, and its value ranges from 0 to 1. The larger the deviation, the closer the correction coefficient is to 1, and the higher the surprise factor, the smaller the correction threshold of the S-shaped function.

8. A system for estimating the state of a converged terminal in a distribution area, characterized in that, For implementing the method as described in any one of claims 1 to 7, comprising: The data acquisition module is used to acquire measurement data for the current sampling period; The initialization module is used to use the default value as the initial state estimate when the current sampling period is the first period. The surprise factor calculation and dual-memory management module is used to calculate the surprise factor and determine the update step size of the state estimation based on the measurement data and the state estimation result of the previous period when the current sampling period is not the first period. The dual-memory management includes a short-term memory bank and a long-term memory bank. The long-term memory bank is used to store historical snapshots. The long-term memory bank is managed for elimination based on the surprise factor, the duration of storage, and the external scene tag. The cold start module is used to activate the cold start mode when the surprise level exceeds a preset threshold, and generate an initial state estimate based on the pre-trained measurement-state mapping relationship and the historical snapshots in the long-term memory. The high-precision estimation module is used to perform weighted least squares calculations using real-time data from the high-precision measurement channel, starting from the initial state estimate, to obtain the high-precision state estimation result. The differential fusion module is used to calculate the low-precision state estimation result when the data from the low-precision measurement channel arrives, and generate a correction coefficient based on the deviation between the low-precision state estimation result and the high-precision state estimation result, correct the high-precision state estimation result along the deviation direction, and obtain the final state estimation result; and when the data from the low-precision measurement channel has not arrived, the high-precision state estimation result is used as the final state estimation result.

9. An electronic device, the electronic device comprising a memory, a processor, and a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

Images