A method and system for distributed photovoltaic voltage management based on advanced control

Abstract

The application relates to the technical field of distributed photovoltaic power distribution network voltage treatment, and discloses a distributed photovoltaic voltage treatment method and system based on advanced control, wherein the method comprises the following steps: acquiring real-time active power output time sequences of nodes and detecting mutation events; superimposing index decay Laplace noise associated with round times at mutation moments and performing distributed communication; performing variance decay processing on a disturbance time sequence of multiple cloud shadow events by using a weighted moving average filter; fitting a cloud shadow propagation speed vector by using a robust weighted least square method; generating a feedforward disturbance prediction vector of each downstream node; calculating a feedforward reactive power reserve; fusing feedforward and feedback to generate comprehensive reactive power control instructions; and adaptively adjusting a noise decay rate parameter.

Description

Technical Field

[0001] This invention relates to the field of voltage management technology for distributed photovoltaic power distribution networks, and more specifically, to a distributed photovoltaic voltage management method and system based on advanced control. Background Technology

[0002] In low-voltage distribution networks with high-penetration distributed photovoltaic (PV) systems, as clouds move along the feeder direction, each PV grid-connected node experiences a sudden change in active power output, forming a spatial propagation chain of voltage disturbances. To suppress voltage overshooting at downstream nodes, it is necessary to collaboratively estimate the direction and velocity of cloud shadow propagation, thereby providing feedforward early warnings for downstream nodes and enabling the early deployment of reactive power reserves.

[0003] Existing technologies typically employ a method of directly sharing active power output change data from each node to collaboratively estimate cloud shadow propagation parameters. However, in mixed-use commercial and residential areas, distributed photovoltaic systems belong to different owners or operators. The timing and magnitude of active power output changes directly reflect commercially sensitive information such as installed capacity, shading ratio, and power generation efficiency, and each property owner is unwilling to expose their actual power generation data. If privacy noise with a fixed variance is directly added to the output change data, the time difference information from this noise pollution will cause significant deviations in the estimation of cloud shadow propagation speed and direction. This leads to misalignment of the disturbance arrival time predicted by the downstream node's feedforward, resulting in reactive power pre-regulation arriving earlier or later than the actual disturbance arrival time, thus exacerbating the risk of voltage exceeding limits. Therefore, existing technologies struggle to maintain the accuracy of cloud shadow propagation parameter estimation while protecting the commercially sensitive information of multiple property owners, compromising the effectiveness of feedforward voltage management. Summary of the Invention

[0004] This invention provides a distributed photovoltaic voltage management method and system based on advanced control, which solves the technical problems in related technologies such as the difficulty in predicting voltage fluctuations of distributed photovoltaic grid-connected nodes under cloud shadow disturbances and the inability to balance feedforward compensation and privacy protection.

[0005] This invention discloses a distributed photovoltaic voltage mitigation method based on advanced control, comprising the following steps: The system acquires real-time active power output time series and geographic coordinate information of each distributed photovoltaic grid-connected node in the distribution network. Each node independently detects local active power output abrupt events and records the abrupt event time. With the magnitude of mutation ; Each node at the mutation time The time value of the disturbance generated by superimposed Laplace noise The scale parameter of the Laplace noise As the rounds of collaborative estimation are set to decay exponentially, each node sends the disturbance time value to its neighboring nodes through distributed communication. Each node collects the perturbation time sequence sent by multiple upstream neighbors in a series of cloud shadow events, uses a weighted moving average filter to perform variance attenuation processing, and outputs the filtered time difference and the residual variance of each node. Based on the filtered time difference and the geographical distance vector between nodes, the robust weighted least squares method is used to fit the estimated value of the cloud shadow propagation speed vector and its confidence interval. Based on the confidence interval of the cloud shadow propagation velocity vector estimate and the geographical location of downstream nodes, the disturbance arrival time interval of each downstream node is generated. The predicted output change amplitude of each downstream node is calculated by combining the upstream mutation amplitude and the propagation attenuation coefficient, and a feedforward disturbance prediction vector is formed. Based on the feedforward disturbance prediction vector, the feedforward reactive power reserve increment is linearly allocated within the disturbance arrival time interval and fused with the feedback reactive power compensation amount of the local real-time voltage deviation to generate a comprehensive reactive power control command which is sent to the inverter for execution.

[0006] Furthermore, the scale parameter The calculation method is as follows: ,in These are the initial scale parameters. For decay rate parameters, This is the current round number for collaborative estimation. It is a natural constant.

[0007] Furthermore, the weighted moving average filter applies to neighboring nodes. The estimated time of the disturbance The calculation method is as follows: ,in The number of historical cloud shadow events that participated in the filtering. For the first The filter weights corresponding to this event For the first Neighboring nodes in the secondary cloud shadow event The transmitted disturbance time value; the filter weights The result is obtained by normalizing the inverse of the noise variance for the corresponding round, i.e. ,in For the first The collaborative estimation round number corresponding to the next cloud shadow event.

[0008] Furthermore, the method of fitting the cloud shadow propagation speed vector estimate using robust weighted least squares includes: adjusting the geographic distance vector between nodes. After performing decimal scaling and normalization, the following was obtained: ,in As a calibration factor; an auxiliary slowness vector is introduced. Establish observation equations ,in The time difference after filtering. For residual terms; the residual variance of each node. The reciprocal of the weight is used as the weight. Solving the weighted least squares problem yields... The cloud shadow propagation velocity vector estimate was recovered from the auxiliary slowness vector. ,in The square of the modulus of the auxiliary slowness vector estimate.

[0009] Furthermore, the robust weighted least squares method further includes: during the iteration process, resetting the weights of node pairs whose absolute residual value exceeds three times the median of the residuals and resolving the problem until no new abnormal node pairs are removed.

[0010] Furthermore, the calculation method for the disturbance arrival time interval is as follows: ,in For downstream nodes Relative to the normalized distance vector of the upstream reference node, and These are the auxiliary slowness vectors corresponding to the boundaries of the confidence intervals.

[0011] Furthermore, the calculation method for the predicted output change amplitude is as follows: ,in The observed abrupt change magnitude of the upstream reference node. The propagation attenuation coefficient, For downstream nodes The normalized distance along the propagation direction between the reference node and the reference node; the propagation attenuation coefficient. The results were obtained by fitting the mutation magnitude and normalized distance of each node in multiple historical cloud shadow events to a log-linear equation system using the least squares method.

[0012] Furthermore, the calculation method for the linear allocation feedforward reactive power reserve increment is as follows: ,in For the current moment, The target reactive power compensation amount, For downstream nodes The local voltage sensitivity coefficient at that location.

[0013] Furthermore, it also includes an adaptive adjustment step: after the cloud shadow event ends, each node calculates the actual arrival time of the disturbance. Prediction bias relative to the prediction interval When the prediction deviation Exceeding the preset deviation threshold At that time, the decay rate parameter is adjusted to ,in To slow down the factor; when the cumulative privacy budget is depleted When the preset budget limit is exceeded, the decay rate parameter will be adjusted to... .

[0014] This invention discloses a distributed photovoltaic voltage mitigation system based on advanced control, used to execute the aforementioned distributed photovoltaic voltage mitigation method based on advanced control, comprising: The mutation detection module is used to acquire the real-time active power output time series and geographic coordinate information of each distributed photovoltaic grid-connected node, independently detect local active power output mutation events and record the mutation time and mutation magnitude. The privacy noise injection and communication module is used to generate perturbation time values ​​by superimposing exponentially decaying Laplace noise associated with rounds at abrupt change times, and then sending them to neighboring nodes through distributed communication. The filtering module is used to collect the disturbance time sequence sent by multiple upstream neighbors in a series of cloud shadow events, perform variance attenuation processing using a weighted moving average filter, and output the filtered time difference and the residual variance of each node pair. The propagation parameter estimation module is used to fit the estimated value of the cloud shadow propagation speed vector and its confidence interval based on the filtered time difference and the geographical distance vector between nodes using the robust weighted least squares method. The feedforward prediction module is used to generate the disturbance arrival time interval based on the confidence interval of the cloud shadow propagation speed vector estimate and the geographical location of the downstream nodes. It combines the upstream mutation amplitude and the propagation attenuation coefficient to calculate the predicted output change amplitude of each downstream node and outputs the feedforward disturbance prediction vector. The feedforward reactive power calculation module is used to linearly allocate the feedforward reactive power reserve increment within the disturbance arrival time interval based on the feedforward disturbance prediction vector. The control command fusion module is used to fuse the feedforward reactive power reserve with the feedback reactive power compensation of the local real-time voltage deviation, generate a comprehensive reactive power control command, and send it to the inverter for execution. The adaptive adjustment module is used to compare the actual arrival time of the disturbance with the prediction interval after the cloud shadow event ends, calculate the prediction deviation statistics, and adaptively adjust the decay rate parameter of the Laplace noise scale parameter in the next round.

[0015] The present invention has the following beneficial effects: Applying round-correlated exponentially decaying Laplace noise to the moment of abrupt change allows property owners to participate in distributed communication without exposing real active power output changes, thus protecting commercially sensitive information such as installed capacity and power generation efficiency.

[0016] Since the Laplace noise scale parameter decays exponentially with each round, the noise intensity of later communications gradually decreases. The weighted moving average filter uses the temporal redundancy information of multiple cloud shadow events to perform variance attenuation processing on the time difference data of noise pollution, thereby restoring a reliable time difference estimate between nodes and ensuring the convergence accuracy of propagation parameter estimation.

[0017] Robust weighted least squares fitting suppresses the interference of node pairs with large residual noise on the estimation of cloud shadow propagation velocity vector by inverse weighting of residual variance and removing abnormal node pairs, thus making the estimation results of propagation direction and velocity robust.

[0018] The intervalized disturbance arrival time enables the feedforward reactive power pre-regulation window to cover the actual disturbance arrival time range. Even if there is a deviation in the single propagation parameter estimation, the feedforward regulation window can still cover the actual disturbance arrival time, avoiding the exacerbation of voltage over-limit caused by the misalignment of reactive power pre-regulation timing.

[0019] The adaptive decay rate parameter adjustment dynamically balances the noise intensity based on the prediction deviation and the cumulative privacy budget consumption, so that the system can maintain the effectiveness of feedforward voltage management in long-term operation, while keeping the total privacy budget consumption within a preset upper limit. Attached Figure Description

[0020] Figure 1 This is a flowchart of a distributed photovoltaic voltage mitigation method based on advanced control provided in an embodiment of the present invention; Figure 2 This is a schematic diagram showing the distribution of mutation times and mutation magnitudes at each node provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the abrupt change times before and after privacy noise injection, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the residual variance and observation weights of each node pair after filtering, provided in an embodiment of the present invention. Figure 5 This is a schematic diagram of the confidence interval for estimating the cloud shadow propagation velocity vector provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the linear allocation curve of the feedforward reactive power reserve at node 5 provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the fusion of reactive power control commands at node 5 provided in an embodiment of the present invention; Figure 8 This is a schematic diagram comparing privacy budget consumption with the upper limit provided in an embodiment of the present invention; Figure 9 This is a schematic diagram comparing the installed capacity of each node with the predicted power output drop, provided in an embodiment of the present invention. Detailed Implementation

[0021] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0022] At least one embodiment of the present invention discloses a distributed photovoltaic voltage mitigation method based on advanced control, see [link to relevant documentation]. Figure 1 This includes the following steps: Step 1: Obtain real-time operating data of each node and detect sudden changes in active power output.

[0023] The system acquires real-time active power output time series, node voltage measurements, and geographic coordinates of each distributed photovoltaic grid-connected node in the distribution network. Each node independently performs abrupt change detection on its local active power output time series and records the abrupt change times. With the magnitude of mutation ,in For the first The timestamp of the step change in output detected by each node. Number the nodes. This represents the change in active power output at that moment.

[0024] It should be noted that the above-mentioned mutation detection is to perform sliding window differencing on the active power output time series. When the difference between the average output values ​​of adjacent windows exceeds the preset mutation threshold, the corresponding time is marked as the mutation time, and the difference between the average values ​​is used as the mutation amplitude.

[0025] Step 2: Apply round-related privacy noise to the mutation time and perform distributed communication.

[0026] Each node's local detection of the mutation time Adding Laplace noise generates privacy-preserving perturbation time values. The noise scaling parameter is associated with the current collaborative estimation round and is set according to an exponential decay law: ; in, For the first Laplace noise scaling parameter during round-trip collaborative estimation. These are the initial scale parameters. For decay rate parameters, This is the current round number for collaborative estimation. It is a natural constant. The time values ​​of the disturbances generated by each node are: ; in, Obey the scale parameter is The Laplace distribution. Each node will perturb the time value. Send to neighboring nodes through distributed communication.

[0027] In this embodiment of the application, in order to provide sufficient protection for commercially sensitive information in the early collaboration stage, the initial scale parameter... The value of makes the differential privacy budget for a single communication The preset minimum privacy protection level is met. As the number of rounds increases, the noise gradually decreases, ensuring the convergence accuracy of the later propagation parameter estimation.

[0028] Step 3: Filter the privacy protection disturbance moments of multiple cloud shadow events.

[0029] Each node collects privacy-preserving perturbation time sequences sent by multiple upstream neighbors during consecutive cloud shadow events. For the same neighbor node... In the The perturbation time value sent during the secondary cloud shadow event Variance attenuation is performed using a weighted moving average filter: ; in, Neighboring nodes after noise suppression The estimated time of the disturbance, The number of historical cloud shadow events that participated in the filtering. For the first The filter weights corresponding to this event This refers to the sequence number of the Cloud Shadow event. The disturbance time value for each event. Before participating in the weighted summation, the data must be aligned with the same reference time to ensure that the time values ​​of each event are under the same time reference before performing the weighted average calculation.

[0030] It should be noted that the filter weights The variance is obtained by normalizing the inverse of the noise variance for the corresponding round. The variance of the Laplace distribution is... Therefore, the first The corresponding round number for the Cloud Shadow event is: At that time, its unnormalized weight is ,in For the first The collaborative estimation round number corresponding to the cloud shadow event is assigned a larger weight to recent events with smaller noise variance and a smaller weight to earlier events with larger noise variance.

[0031] This step involves each neighbor node. By independently performing the above filtering, the estimated time of disturbance for each neighboring node is obtained. .node Estimated time of its own disturbance This data is obtained by processing multiple mutation times of the node itself using the same filtering process, and is compared with neighboring nodes. The processing method is the same. Node With nodes Their respective filtered residual variance and Calculate in the same way: ; in, For the first Unnormalized filter weights for the next event This represents the Laplace noise scaling parameter for the corresponding round of this event. The residual variance after filtering. The calculation method and the residual variance after filtering Similarly, in the above formula, each Replace with node The noise scale parameters corresponding to each event cycle can be obtained.

[0032] Based on the filtered perturbation time estimates between each node pair, the node pair is calculated. Filtered time difference between: ; in, For nodes Relative to node The filtered disturbance arrival time difference, in seconds. Node pair The residual variance of the time difference after filtering is: ; Step 4: Estimate the cloud shadow propagation velocity vector using the robust weighted least squares method.

[0033] Based on the filtered time difference The cloud shadow propagation speed vector is fitted using a robust weighted least squares method with the geographic distance vector between nodes.

[0034] For node pairs Its geographic distance vector is ,in and They are nodes Relative to node The distance components in the east-west and north-south directions, in meters. This indicates the transpose. The cloud shadow propagation velocity vector to be estimated is... ,in and These are the components of the cloud shadow propagation speed vector in the east-west and north-south directions, respectively, in meters per second.

[0035] Before establishing the observation equation, the components of the geographic distance vector are normalized using decimal scaling to eliminate the influence of distance dimensions on the condition number of the equation's coefficient matrix. Let the scaling factor used for normalization be... (Unit: meters), the normalized distance vector is The corresponding auxiliary slowness vector is ,satisfy ,in Let the square of the magnitude of the cloud shadow propagation velocity vector be denoted by: Then the observation equation is: ; in, This is the residual term. Because... The unit is meters divided by (meters per second), i.e., seconds, and the time difference after filtering. The dimensions are consistent.

[0036] It should be noted that the above linearized observation equation can be rewritten in matrix form. Among them, the quantities to be determined The observation matrix is ​​a two-dimensional column vector. Each row corresponds to a normalized distance vector of a pair of nodes. Observation vector Each element corresponds to a filtered time difference. The residual variance of the time difference between each node pair after filtering, calculated in step 3, is then used. As the inverse of the observation weights, i.e., the weights This forms a diagonal weight matrix. The objective function of weighted least squares is about Minimize, whose analytical solution is: Therefore, the estimated value of the cloud shadow propagation velocity vector can be recovered: ; in, To provide the least squares estimate of the auxiliary slowness vector, The square of the modulus of the auxiliary slowness vector estimate.

[0037] It should be noted that the above equation is the inverse transformation from the auxiliary slow speed vector to the cloud shadow propagation speed vector. From It can be seen that, and It satisfies the vector reciprocal relation, that is ,therefore The dimension of this is meters multiplied by (seconds per meter squared) and then by meters, which simplifies to meters per second, consistent with the physical dimension of the cloud shadow propagation speed vector. The scaling factor introduced by decimal scaling normalization... It is completely eliminated in this inverse transformation, without the need for an auxiliary slowness vector estimate. Perform the inverse normalization transformation separately.

[0038] Covariance matrix based on weighted residuals Output cloud shadow propagation velocity vector estimate The confidence interval is used to reduce the fitting contribution of node pairs with large residual noise.

[0039] In this embodiment of the application, in order to further suppress the influence of individual abnormal node pairs, during the weighted least squares iteration process, for node pairs whose absolute residual value exceeds three times the median of the residual, their weights are reset to zero and the solution is recalculated until no new abnormal node pairs are removed.

[0040] Step 5: Generate the feedforward perturbation prediction vector for each downstream node.

[0041] Based on cloud propagation velocity vector estimation The confidence interval and the geographical locations of each downstream node are used to generate each downstream node. Estimation of the time interval of disturbance arrival ,in Number the downstream nodes. This is the lower bound of the expected arrival time interval of the disturbance. This is the upper bound of the expected arrival time interval of the disturbance. Downstream nodes The geographic distance vector relative to the upstream reference node from which the mutation has been observed is Before substituting the geographic distance vector into the calculation The same decimal scaling and normalization process as in step 4 is adopted, i.e. This yields the normalized distance vector. .

[0042] It should be noted that the confidence interval corresponds to the upper bound of the cloud shadow propagation velocity vector. and the lower world The covariance matrix output from step 4 Determine the boundary cloud shadow propagation velocity vectors that respectively minimize and maximize the estimated arrival time. Based on... Boundary cloud shadow propagation speed vector and The corresponding auxiliary slowness vectors are denoted as follows: and ,satisfy , The disturbance arrival time interval is calculated as follows: ; in, and These are the propagation velocity vectors of the boundary cloud shadows, respectively. and The corresponding auxiliary slowness vector, and This is the square of the magnitude of the corresponding auxiliary slowness vector. After normalization, The dimension is seconds, which is consistent with the dimension of the disturbance arrival time.

[0043] Based on the observed mutation magnitude and propagation attenuation characteristics of upstream nodes, the values ​​of each downstream node are calculated. Predicted output change range : ; in, The observed abrupt change magnitude of the upstream reference node. The propagation attenuation coefficient, For downstream nodes The normalized distance along the propagation direction between the reference node and the downstream node is calculated as follows: Estimated velocity vector of cloud shadow propagation between reference node and the reference node The projected distance in the direction is divided by a preset reference distance. Propagation attenuation coefficient. The measured mutation magnitudes of each node in multiple historical cloud shadow events were obtained through least-squares fitting: the measured mutation magnitudes of each node in the same cloud shadow event. and corresponding normalized distance according to ; Establish a system of linear equations, where Representing the natural logarithm operation, and solving it using the least squares method. The propagation attenuation coefficient is obtained by taking the exponent. The feedforward perturbation prediction vector for each downstream node is determined by the perturbation arrival time interval. and the predicted output change range composition.

[0044] Step 6: Calculate the feedforward reactive power reserve for each downstream node.

[0045] Based on the feedforward disturbance prediction vector, the feedforward reactive power reserve of each downstream distributed photovoltaic inverter is calculated. The lower bound of the disturbance arrival time interval is used as the reference. Reactive power pre-adjustment is initiated at the earliest possible arrival time to disrupt the upper bound of the expected arrival time interval. The pre-adjustment window ends at the latest possible arrival time. Within the time interval... Within this range, the incremental feedforward reactive power reserve is linearly allocated according to the predicted change in power output: ; in, For downstream nodes At any moment The feedforward reactive power reserve, For the current moment, Based on the predicted output change amplitude The target reactive power compensation amount is calculated using the local voltage sensitivity coefficient. The calculation method is as follows: ; in, For downstream nodes The local voltage sensitivity coefficient at a given point represents the ratio of the voltage change at that node caused by a unit change in reactive power, and is obtained from the corresponding element of the distribution network power flow Jacobian matrix.

[0046] Step 7: Integrate feedforward and feedback to generate a comprehensive reactive power control command.

[0047] Feedforward reactive power reserve Real-time reactive power compensation calculated at each node based on local real-time voltage deviation The data is then integrated to generate a comprehensive reactive power control command. : ; in, This is the feedback reactive power compensation calculated using proportional-integral control based on the difference between the measured local node voltage and the rated voltage. (Integrated reactive power control command) The voltage regulation is then implemented by distributing the data to each distributed photovoltaic inverter.

[0048] Step 8: Adaptively adjust the noise attenuation rate parameter.

[0049] After the cloud shadow event ends, each node uses the locally observed arrival time of the disturbance. With the prediction interval Compare the results and calculate the prediction bias statistic: ; in, For downstream nodes The prediction bias This represents the actual observed arrival time of the disturbance.

[0050] Adaptive adjustment of the decay rate parameter of the next round of Laplace noise scale parameters When the prediction deviation Exceeding the set threshold At that time, the decay rate is slowed down to maintain the accuracy of the propagation parameter estimation, i.e. ,in As a slowing factor, This is a preset deviation threshold; when the cumulative privacy budget consumption rate exceeds the preset budget limit, the decay is accelerated to save privacy budget, i.e. .

[0051] In this embodiment of the application, in order to balance privacy budget consumption and estimation accuracy during long-term operation, the cumulative privacy budget is calculated as follows: ; in, For the front The cumulative privacy budget consumption of each round Estimate the total number of rounds for the collaboration that has been completed so far. For the first The single-round differential privacy budget, The preset total privacy budget cap is set when the accumulated privacy budget is consumed. Approaching the total privacy budget limit Accelerated decay is triggered at certain times.

[0052] On the other hand, the present invention also proposes a distributed photovoltaic voltage management system based on advanced control, comprising: The mutation detection module is used to acquire the real-time active power output time series, node voltage measurement value and geographical coordinate information of each distributed photovoltaic grid-connected node, independently detect local active power output mutation events and record the mutation time and mutation magnitude.

[0053] The privacy noise injection and communication module is used to generate perturbation time values ​​by superimposing exponentially decaying Laplace noise associated with rounds at abrupt change times, and then sending them to neighboring nodes through distributed communication.

[0054] The filtering module is used to collect the privacy-preserving perturbation time series sent by multiple upstream neighbors in a series of cloud shadow events, perform variance attenuation processing using a weighted moving average filter, and output the filtered time difference sequence and the residual variance of each node pair.

[0055] The propagation parameter estimation module is used to fit robust estimates and confidence intervals of the output cloud shadow propagation velocity vector based on the filtered time difference sequence and the geographical distance between nodes using the robust weighted least squares method.

[0056] The feedforward prediction module is used to generate a disturbance arrival time interval estimate based on the confidence interval of the cloud shadow propagation velocity vector and the geographical location of downstream nodes. It calculates the predicted output change amplitude of each downstream node based on the upstream mutation amplitude and propagation attenuation characteristics, and outputs the feedforward disturbance prediction vector.

[0057] The feedforward reactive power calculation module is used to calculate the feedforward reactive power reserve of each downstream inverter based on the feedforward disturbance prediction vector, and linearly allocate the feedforward reactive power reserve increment according to the predicted output change range within the disturbance arrival time interval.

[0058] The control command fusion module is used to merge the feedforward reactive power reserve with the real-time reactive power compensation of the local voltage deviation to generate a comprehensive reactive power control command and send it to the inverter for execution.

[0059] The adaptive adjustment module is used to compare the actual arrival time of the disturbance with the prediction interval after the cloud shadow event ends, calculate the prediction deviation statistics, and adaptively adjust the decay rate parameter of the Laplace noise scale parameter in the next round.

[0060] By way of example, the system proposed in this invention may also include various features and combinations thereof in the method embodiments, which will not be elaborated here.

[0061] This invention applies exponentially decaying Laplace noise associated with each power output surge to the moment of power output change, enabling stakeholders to participate in distributed communication in the early coordination phase without exposing actual power generation data, thus protecting commercially sensitive information such as installed capacity and power generation efficiency. Because the Laplace noise scale parameter decays exponentially with each surge, the noise intensity in later communication gradually decreases. A weighted moving average filter utilizes the temporal redundancy information of multiple cloud shadow events to attenuate the variance of the noise-contaminated time difference data, thereby restoring reliable inter-node time difference estimates. Robust weighted least squares fitting, through inverse weighting of residual variance, suppresses interference from nodes with large residual noise on propagation parameter estimates, ensuring the accuracy of cloud shadow propagation velocity vector and propagation direction estimates. The intervalized disturbance arrival time ensures that the feedforward reactive power pre-regulation window covers the possible arrival time range of the actual disturbance. Even if there is a deviation in the estimation of a single propagation parameter, the feedforward regulation window can still cover the actual disturbance arrival time, avoiding the exacerbation of voltage over-limit caused by misaligned reactive power pre-regulation timing. The adaptive decay rate parameter adjustment dynamically balances noise intensity based on prediction bias and privacy budget consumption, ensuring the continued effectiveness of feedforward voltage management during long-term operation. Therefore, this embodiment maintains the estimation accuracy of cloud shadow propagation parameters while protecting the privacy of multiple stakeholders, supporting accurate feedforward voltage management.

[0062] See Figures 2-9 The following application examples are proposed: A low-voltage power distribution network in a mixed-use commercial and residential area contains multiple rooftop photovoltaic (PV) systems belonging to different owners. The network feeder runs east-west and has five distributed PV grid-connected nodes (nodes 1 to 5), spaced approximately 150 to 300 meters apart. Each node has an installed capacity between 30 and 80 kilowatts, belonging to different commercial and residential owners, none of whom wish to disclose their local power generation data to others. One summer afternoon, a large cloud moved from west to east, successively obscuring the PV arrays above each node, triggering a chain reaction of active power drops propagating along the feeder, causing continuous voltage drops at the nodes along the line. The power distribution network operator needs to deploy reactive power reserves in advance to prevent voltage exceedances before the cloud shadow reaches the downstream nodes.

[0063] Each node's edge controller collects local active power output time series in real time, with a sampling interval of 1 second. A sliding window differential processing is performed on the output time series, with a window length of 5 seconds and a sudden change threshold set to 15% of the installed capacity. When the cloud cluster passes sequentially through nodes 1 to 4, each node independently detects a sudden drop in output and records the moment of the change. and mutation magnitude Node 5 is located at the downstream end, and no mutation has been detected at this point, so it serves as the target node for feedforward prediction.

[0064] Table 1 Original records of mutation detection at each node

[0065] Node 1 detects the mutation first, and its mutation time is used as the reference time zero. The mutation times of nodes 2 to 4 are all recorded relative to this reference time.

[0066] This Cloud Shadow incident is the first Round-trip collaborative estimation. Initial scale parameters. Seconds, decay rate parameter Thus, the noise scale parameters for this round were obtained. Each node's local mutation time Superimposed noise that follows a Laplace distribution Generate disturbance time values And it is sent to neighboring nodes through distributed communication links. Mutation magnitude It does not participate in external communications, remains local at all times, and protects commercially sensitive information such as the installed capacity and power generation efficiency of each owner.

[0067] Table 2. Disturbance time values ​​before and after this round of privacy noise injection.

[0068] Each node will Send to neighboring nodes, real time Without leaving the local controller.

[0069] Each node collects three consecutive cloud shadow events (rounds 1 to 3, i.e.) The sequence of time-of-perturbation events sent by neighboring nodes in the node pair (node ​​1 and node 2). Let's take an example to illustrate the filtering process. After aligning the perturbation time values ​​of the three events with the first mutation time of node 1 as a unified reference time, a weighted moving average is performed on the three perturbation time values ​​of node 2: weights... More recent events with smaller noise variance (round 3) are given greater weight. After filtering, the estimated perturbation times for nodes 1 and 2 are obtained. and Then calculate the time difference after filtering. and residual variance The above filtering was repeated for all adjacent node pairs, and the results are summarized below.

[0070] Table 3. Filtered time difference and residual variance for each node pair

[0071] The reciprocal of the residual variance is the observation weight of each node pair. Nodes with more reliable time difference estimations contribute more to subsequent fitting.

[0072] Take the scaling factor The geographic distance vectors of each node pair are normalized using decimal scaling. For example, the eastward distance of node 2 relative to node 1 is 210 meters and the northward distance is 30 meters. After normalization... Construct an observation matrix for all 6 node pairs. Observation vector and diagonal weight matrix Solving the weighted least squares problem yields an auxiliary slowness vector estimate. Then through inverse transformation Recover the cloud shadow propagation velocity vector. During the iteration process, examine the residuals of each node pair. If the absolute value of the residual exceeds three times the median of the residual, reset its weights to zero and solve it again.

[0073] Table 4 Estimation results of cloud shadow propagation velocity vector

[0074] The cloud shadow's propagation direction is approximately due east, about 7 degrees north of north, which coincides with the actual direction of cloud movement. The confidence interval is determined by the covariance matrix. This is determined and used for subsequent interval-based feedforward prediction.

[0075] Taking node 1 as the upstream reference node, its observed mutation magnitude kilowatts. Using the measured abrupt change amplitudes at various nodes of three historical cloud shadow events, through analysis of... Perform least squares fitting to obtain the propagation attenuation coefficient. The geographic distance vector of node 5 relative to node 1 is... Rice, after normalization The auxiliary slowness vector corresponding to the confidence interval boundary. and Substitute the values ​​into the disturbance arrival time interval formula to calculate the disturbance arrival time interval for node 5.

[0076] Table 5. Predicted feedforward perturbation vectors for node 5

[0077] The disturbance at node 5 is expected to arrive between 54.61 and 59.38 seconds after the reference time, with a predicted power drop of approximately 26.18 kilowatts.

[0078] Local voltage sensitivity coefficient of node 5 (kW / kvar), target reactive power compensation Thousands of kcal / kg. Feedforward reactive power reserve. In the pre-adjustment window The internal growth rate increases from zero to [a certain value] according to a linear law. This allows the inverter to complete reactive power reserve deployment before the actual arrival of the disturbance.

[0079] Table 6. Linear distribution of reactive power reserve at node 5 over time.

[0080] Reactive power pre-adjustment was initiated at 54.61 seconds, and the feedforward reactive power reserve reached the target value at 59.38 seconds, covering the entire time window during which the disturbance might actually occur.

[0081] The edge controller at node 5 continuously collects local voltage measurements at time [time missing]. At a certain time, the locally measured voltage was too low, and the reactive power compensation was fed back. Thousands of kcal / kg. At this time, the feedforward reactive power reserve is... Calculated using linear interpolation: Seconds in the window The progress within is ,correspond The two are combined to generate a comprehensive reactive power control command, which is then sent to the inverter.

[0082] Table 7. Fusion of Integrated Reactive Power Control Commands at Node 5

[0083] The integrated command combines feedforward prediction and feedback correction, enabling the inverter to be in a high reactive power output state before the disturbance arrives, thereby reducing the voltage drop.

[0084] After the cloud shadow event ends, node 5 records the actual arrival time of the disturbance. Seconds. The value falls within the prediction range. Within this range, therefore the prediction bias There is no need to slow down the decay rate. Simultaneously, verify the current cumulative privacy budget consumption. With the total privacy budget cap There is still a margin of safety, so there is no need to trigger accelerated decay. Therefore, the decay rate parameter for the next round... Keep the current value of 0.20 unchanged.

[0085] Table 8 Results of Adaptive Adjustment in Step 8

[0086] The logical chain of the entire data flow is as follows: Step 1: The original mutation time and geographical coordinates of each node are used as the initial state and enter Step 2. After adding privacy noise, the disturbance time value is generated and communicated externally. Step 3: Collect multiple rounds of disturbance time values, compress the noise variance to an acceptable level through weighted moving average, and output the filtered time difference and residual variance. Step 4: Perform robust weighted least squares fitting with the residual variance as weight to transform the time difference sequence into a cloud shadow propagation speed vector and its confidence interval. Step 5: Combine the geographical location of downstream nodes and the propagation attenuation coefficient to map the speed confidence interval to the disturbance arrival time interval and predicted output amplitude of node 5. Step 6: Transform the predicted vector into a feedforward reactive power reserve that grows linearly with time. Step 7: Combine with the real-time feedback reactive power to form a comprehensive control command sent to the inverter. Step 8: Compare the actual disturbance arrival time with the predicted interval, and call back the noise attenuation rate for the next round to form a closed-loop adaptive adjustment. The output of each step is strictly used as the input of the subsequent steps to ensure the continuity and physical consistency of the data flow.

[0087] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A distributed photovoltaic voltage mitigation method based on advanced control, characterized in that, Includes the following steps: The system acquires real-time active power output time series and geographic coordinate information of each distributed photovoltaic grid-connected node in the distribution network. Each node independently detects local active power output abrupt events and records the abrupt event time. With the magnitude of mutation ; Each node at the mutation time The time value of the disturbance generated by superimposed Laplace noise The scale parameter of the Laplace noise As the rounds of collaborative estimation are set to decay exponentially, each node sends the disturbance time value to its neighboring nodes through distributed communication. Each node collects the perturbation time sequence sent by multiple upstream neighbors in a series of cloud shadow events, uses a weighted moving average filter to perform variance attenuation processing, and outputs the filtered time difference and the residual variance of each node. Based on the filtered time difference and the geographical distance vector between nodes, the robust weighted least squares method is used to fit the estimated value of the cloud shadow propagation speed vector and its confidence interval. Based on the confidence interval of the cloud shadow propagation velocity vector estimate and the geographical location of downstream nodes, the disturbance arrival time interval of each downstream node is generated. The predicted output change amplitude of each downstream node is calculated by combining the upstream mutation amplitude and the propagation attenuation coefficient, and a feedforward disturbance prediction vector is formed. Based on the feedforward disturbance prediction vector, the feedforward reactive power reserve increment is linearly allocated within the disturbance arrival time interval and fused with the feedback reactive power compensation amount of the local real-time voltage deviation to generate a comprehensive reactive power control command which is sent to the inverter for execution.

2. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The scale parameter The calculation method is as follows: ,in These are the initial scale parameters. For decay rate parameters, This is the current round number for collaborative estimation. It is a natural constant.

3. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The weighted moving average filter for neighboring nodes The estimated time of the disturbance The calculation method is as follows: ,in The number of historical cloud shadow events that participated in the filtering. For the first The filter weights corresponding to this event For the first Neighboring nodes in the secondary cloud shadow event The transmitted disturbance time value; the filter weights The result is obtained by normalizing the inverse of the noise variance for the corresponding round, i.e. ,in For the first The collaborative estimation round number corresponding to the next cloud shadow event.

4. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The method of fitting the cloud shadow propagation velocity vector estimate using robust weighted least squares includes: geographic distance vector between nodes After performing decimal scaling and normalization, the following was obtained: ,in The scaling factor; Introducing an auxiliary slowness vector Establish observation equations ,in The time difference after filtering. For residual terms; residual variance of each node The reciprocal of the weight is used as the weight. Solving the weighted least squares problem yields... ; Recovering the cloud shadow propagation velocity vector estimate from the auxiliary slowness vector ,in The square of the modulus of the auxiliary slowness vector estimate.

5. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 4, characterized in that, The robust weighted least squares method further includes: during the iteration process, resetting the weights of node pairs whose absolute residual value exceeds three times the median of the residuals and solving the problem again until no new abnormal node pairs are removed.

6. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The method for calculating the expected arrival time interval of the disturbance is as follows: ,in For downstream nodes Relative to the normalized distance vector of the upstream reference node, and These are the auxiliary slowness vectors corresponding to the boundaries of the confidence intervals.

7. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The method for calculating the predicted output change amplitude is as follows: ,in The observed abrupt change magnitude of the upstream reference node. The propagation attenuation coefficient, For downstream nodes The normalized distance along the propagation direction between the reference node and the reference node; the propagation attenuation coefficient. The results were obtained by fitting the mutation magnitude and normalized distance of each node in multiple historical cloud shadow events to a log-linear equation system using the least squares method.

8. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 1, characterized in that, The calculation method for the linear allocation feedforward reactive power reserve increment is as follows: ,in For the current moment, The target reactive power compensation amount, For downstream nodes The local voltage sensitivity coefficient at that location.

9. The distributed photovoltaic voltage mitigation method based on advanced control according to claim 2, characterized in that, It also includes an adaptive adjustment step: after the cloud shadow event ends, each node calculates the actual arrival time of the disturbance. Prediction bias relative to the prediction interval When the prediction deviation Exceeding the preset deviation threshold At that time, the decay rate parameter is adjusted to ,in To slow down the factor; when the cumulative privacy budget is depleted When the preset budget limit is exceeded, the decay rate parameter will be adjusted to... .

10. A distributed photovoltaic voltage control system based on advanced control, used to execute the distributed photovoltaic voltage control method based on advanced control as described in any one of claims 1 to 9, characterized in that, include: The mutation detection module is used to acquire the real-time active power output time series and geographic coordinate information of each distributed photovoltaic grid-connected node, independently detect local active power output mutation events and record the mutation time and mutation magnitude. The privacy noise injection and communication module is used to generate perturbation time values ​​by superimposing exponentially decaying Laplace noise associated with rounds at abrupt change times, and then sending them to neighboring nodes through distributed communication. The filtering module is used to collect the disturbance time sequence sent by multiple upstream neighbors in a series of cloud shadow events, perform variance attenuation processing using a weighted moving average filter, and output the filtered time difference and the residual variance of each node pair. The propagation parameter estimation module is used to fit the estimated value of the cloud shadow propagation speed vector and its confidence interval based on the filtered time difference and the geographical distance vector between nodes using the robust weighted least squares method. The feedforward prediction module is used to generate the disturbance arrival time interval based on the confidence interval of the cloud shadow propagation speed vector estimate and the geographical location of the downstream nodes. It combines the upstream mutation amplitude and the propagation attenuation coefficient to calculate the predicted output change amplitude of each downstream node and outputs the feedforward disturbance prediction vector. The feedforward reactive power calculation module is used to linearly allocate the feedforward reactive power reserve increment within the disturbance arrival time interval based on the feedforward disturbance prediction vector. The control command fusion module is used to fuse the feedforward reactive power reserve with the feedback reactive power compensation of the local real-time voltage deviation, generate a comprehensive reactive power control command, and send it to the inverter for execution. The adaptive adjustment module is used to compare the actual arrival time of the disturbance with the prediction interval after the cloud shadow event ends, calculate the prediction deviation statistics, and adaptively adjust the decay rate parameter of the Laplace noise scale parameter in the next round.

Images