A reactive voltage control limit adaptive setting method considering new energy fluctuation

Abstract

The application relates to the technical field of power system automation and smart grid control, and discloses a reactive power voltage control limit adaptive setting method considering new energy fluctuation. A multi-node voltage and controllable new energy reactive power injection sequence is first constructed, a joint state vector containing a voltage component and a reactive power component is formed, and historical voltage is cached; state transition is established according to nodes, voltage short-time prediction and reactive power short-time keeping are performed, and data-driven saturation suppression is introduced; an initial covariance based on a reference voltage deviation and a physical upper and lower bound is set; an observation model is constructed with voltage as an observation quantity; prior prediction and covariance propagation are completed, a gain is calculated according to a measurement error and a quantization step, and a posterior state is corrected; voltage and reactive power estimation is extracted, a sensitivity matrix is generated by combining a target voltage and a latest change index, and a reactive power adjustment vector is obtained; and boundary control is implemented to set the upper limit of each unit.

Description

Technical Field

[0001] This invention relates to the field of power system automation and smart grid control technology, specifically to an adaptive tuning method for reactive power voltage control limits that takes into account fluctuations in new energy sources. Background Technology

[0002] With the rapid increase in the proportion of new energy installed capacity, the reactive power and voltage control of the power system faces significant challenges. Taking the Jiangxi power grid as an example, the total installed capacity of new energy has exceeded 32 million kilowatts. The large-scale integration of distributed photovoltaic, hydropower, and wind power resources has greatly increased the operational complexity of the system. The output of new energy is characterized by significant volatility, randomness, and intermittency, especially during periods of abundant water resources, peak generation of distributed photovoltaic and hydropower, and periods of low load across the province. In some areas, such as the Wenfang area of ​​Yingtan, voltage spikes at 35kV and below are frequently observed. Due to insufficient reactor configuration in the 220kV and below power grids, traditional physical reactive power absorption methods are insufficient to alleviate such voltage problems in a timely and effective manner.

[0003] Currently, reactive power and voltage regulation in power systems mainly rely on the leading voltage capability of thermal power plants, reactive power compensation equipment at grid connection points (such as SVG), and transformer tap changer adjustments. Conventional Automatic Voltage Control (AVC) systems are often based on static power flow analysis of the power grid, setting setpoints in different zones and levels, and allocating reactive power resources through traditional dispatching rules or static optimal power flow models. While these methods are effective in power grids with relatively simple load and power supply structures and relatively stable operating conditions, they present the following prominent problems in scenarios with a high proportion of renewable energy: First, existing AVC and its parameter tuning methods are difficult to adapt to the real-time and random fluctuations in renewable energy output. Although the reactive power resource regulation capabilities of renewable energy plants have been greatly improved, such as photovoltaic inverters and wind turbines having good reactive power regulation ranges (power factor up to 0.95), these capabilities have not been fully coordinated and utilized. In actual operation, due to the lack of a unified, flexible, and dynamic limit setting mechanism, reactive power regulation resources are widely distributed and difficult to coordinate, resulting in voltages in some areas remaining above the upper limit for extended periods, while some reactive power regulation resources remain idle or inefficient. Secondly, existing technologies rely on steady-state models and static setpoints, making it difficult to reflect the dynamic process of AVC (Active Voltage Control). Conventional power system simulation software only supports static power flow and voltage calculations, failing to efficiently simulate the dynamic voltage evolution of power systems with a high proportion of renewable energy sources over short timescales. For complex scenarios such as drastic intraday load changes, significant fluctuations in photovoltaic output, and wind power peak shaving, there is a lack of fast and robust adaptive optimization methods for reactive power and voltage control setpoints, easily leading to overvoltage risks and reactive power waste during certain periods. Thirdly, existing technologies generally employ empirical methods or offline simulations and static sensitivity analysis for reactive power setpoint tuning, resulting in long adjustment cycles that cannot support high-frequency, real-time control requirements. When facing extreme conditions such as extreme weather, sudden load changes, or simultaneous large-scale output from multiple renewable energy plants, the voltage control strategy responds with lag, potentially leading to severe voltage exceedances at local nodes and even grid security and stability issues.

[0004] Therefore, this case aims to propose an adaptive setting method for reactive power voltage control limits that takes into account the fluctuations of new energy sources. It makes full use of the historical and current observation data of the voltage at multiple nodes of the power grid and the reactive power of controllable new energy sources. Through distributed state variable modeling, short-term voltage trend prediction, data-driven suppression rules, sensitivity adaptive correction and limit setting, it realizes intelligent and dynamic optimization of the upper limit of reactive power voltage regulation. Summary of the Invention

[0005] This invention provides an adaptive setting method for reactive power voltage control limits that takes into account fluctuations in renewable energy sources, thereby helping to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: an adaptive tuning method for reactive power voltage control limits considering renewable energy fluctuations, comprising:

[0007] Construct node voltage sequences and controllable renewable energy reactive power injection sequences, establish a joint state vector containing voltage components of each node and reactive power components of each controllable unit, and cache historical voltage data of each node.

[0008] Establish a state transition that is updated node by node, perform short-time voltage prediction and short-time reactive power retention, set data-driven saturation suppression rules, and form an overall state transition relationship;

[0009] Set the initial values ​​and covariance of the state estimation. The voltage part is based on the reference voltage and the initial deviation statistics, and the reactive power part is based on the physical upper and lower bound statistics to form the initial error covariance matrix.

[0010] An observation model is constructed, with voltage components selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector.

[0011] Perform state prediction and covariance propagation, calculate voltage prior estimates by node, calculate reactive power prior estimates by unit, and generate prior estimates and corresponding covariances of joint state vectors;

[0012] Calculate the gain and perform state correction, give the measurement error covariance matrix and quantization step size, and then update the a priori state and a priori covariance.

[0013] Calculate the voltage sensitivity to reactive power and solve for the reactive power adjustment amount, extract the current voltage and reactive power estimate, obtain the target voltage of each node, obtain the recent change index of each unit, generate the sensitivity matrix and form the suggested reactive power adjustment vector;

[0014] The reactive power upper limit is set and boundary control is implemented to obtain the final upper limit setting value and setting vector of each unit.

[0015] Optionally, the construction of the node voltage sequence and the controllable new energy reactive power injection sequence, establishing a joint state vector containing the voltage components of each node and the reactive power components of each controllable unit, and caching the historical voltage data of each node, specifically includes:

[0016] Select and number multiple voltage monitoring nodes, install a standard voltage measuring device at each node, and set a fixed sampling period of seconds; acquire the voltage measurement value of each node sequentially according to the sampling time.

[0017] Multiple new energy power plants with reactive power regulation capabilities were selected and numbered, and the reactive power injection value of the corresponding unit was obtained at each sampling time.

[0018] The joint state vector is composed of the node voltage components and the unit reactive components, with the first part being the node voltage components and the second part being the unit reactive components.

[0019] A historical voltage cache is established for each node. The initial value is written at the first sampling time, and then the most recent voltage measurement value is updated according to time.

[0020] Optionally, the establishment of node-by-node state transitions, the execution of short-time voltage prediction and short-time reactive power retention, and the setting of data-driven saturation suppression rules to form an overall state transition relationship specifically include:

[0021] For each node, a short-time voltage prediction is performed, and the predicted value is obtained by the ratio of the current voltage to the recent voltage change of this node.

[0022] When calculating the local change ratio of this node, the change ratio is generated according to the incremental relationship of the three most recent voltage measurements. When the measurement is insufficient or the denominator is zero, it is set to zero.

[0023] Perform data-driven saturation suppression, giving an upper bound on the amplitude based on historical incremental amplitude and quantization resolution, maintaining direction by sign and limiting amplitude by minimum value;

[0024] Keep the reactive power injection of each power station constant within a short time scale, and use it as the reactive power prediction value for the next moment.

[0025] The updates of each node and each unit are merged to form an overall state transition relationship. The input is the state of the previous time step, and the output is the state of the next time step.

[0026] Optionally, the initial values ​​and covariance of the state estimation are set, with the voltage part based on the reference voltage and initial deviation statistics, and the reactive power part based on the physical upper and lower bound statistics, to form an initial error covariance matrix, specifically including:

[0027] When the system starts up, the initial estimates of each component of the joint state are set to the first acquisition values;

[0028] Construct an initial error covariance matrix in the form of diagonal blocks. The variance of the voltage part is given based on the statistical analysis of the reference voltage and initial deviation of each node, and the variance of the reactive power part is given based on the statistical analysis of the half-amplitude of the physical upper and lower bounds of each unit.

[0029] Optionally, the construction of the observation model, in which voltage components are selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector, specifically includes:

[0030] At each sampling time, an observation vector is constructed, and the observation components are the voltage measurement values ​​of each node;

[0031] The observation matrix is ​​set up to directly map the voltage components in the joint state to the observation vector, while non-voltage components do not participate in the mapping.

[0032] Optionally, the execution state prediction and covariance propagation, which calculates voltage prior estimates by node and reactive power prior estimates by unit, and generates prior estimates and corresponding covariances for the joint state vector, specifically includes:

[0033] Prior estimates of voltage are calculated at nodes, prior estimates of reactive power are calculated at units, and these are combined to form a joint state vector prior estimate.

[0034] Construct the Jacobian structure of the state function with respect to the state: voltage corresponds to one plus the local change ratio for the diagonal terms, reactive power corresponds to one for the diagonal terms, and the remaining terms are zero;

[0035] The prior covariance is calculated using the Jacobian and the posterior covariance of the previous time step, and the process noise covariance is set as a zero matrix.

[0036] Optionally, the calculation of gain and state correction, providing the measurement error covariance matrix and quantization step size, followed by post-abstract state and posterior covariance update, specifically includes:

[0037] Set the measurement error covariance matrix. The single-channel measurement variance is constructed according to the voltage quantization step size, which is zero kilovolts.

[0038] Calculate the gain matrix based on the prior covariance and the observation matrix;

[0039] The prior state is corrected using the gain matrix to obtain the posterior state, and the covariance is corrected according to the standard format to obtain the posterior covariance.

[0040] Optionally, the calculation of voltage sensitivity to reactive power and the solution of reactive power adjustment amount, extraction of current voltage and reactive power estimate, acquisition of target voltage for each node, acquisition of recent change index for each unit, generation of sensitivity matrix and formation of suggested reactive power adjustment vector specifically includes:

[0041] The voltage estimates of each node and the reactive power estimates of each unit are obtained from the posterior state, and the deviation sequence between the target voltage and the estimated voltage of each node is calculated.

[0042] For each reactive power unit, find the discrete index of the most recent reactive power change; if it does not exist, mark it as empty.

[0043] The sensitivity matrix elements are set according to the current and historical voltage and reactive power changes, and are set to zero when the denominator is zero or the index does not exist.

[0044] Construct a voltage deviation column vector and a sensitivity matrix. When the matrix is ​​invertible, obtain the suggested reactive power adjustment vector. When the sensitivity matrix is ​​not invertible, output the zero vector.

[0045] Optionally, the process of setting the reactive power upper limit and implementing boundary control to obtain the final upper limit setting value and setting vector for each unit specifically includes:

[0046] For each reactive power unit, generate an untrunculated upper limit setting value based on the current reactive power estimate and the suggested reactive power adjustment.

[0047] Perform boundary truncation: when the uncrunted value is below the physical lower bound, take the lower bound; when it is above the physical upper bound, take the upper bound; when it is within the range, retain the uncrunted value.

[0048] Summarize the final upper limit tuning values ​​of each unit and output the tuning vector.

[0049] The present invention has the following beneficial effects:

[0050] 1. This method incorporates multi-node voltage and reactive power injection from multiple renewable energy sources into a unified joint state vector, and establishes a corresponding historical voltage data cache. This multi-dimensional joint state modeling approach overcomes the isolation observation mode of traditional single-node or single-device models, and can reflect the interactive influence between nodes and units in state estimation and subsequent analysis. The data caching mechanism ensures the full utilization of historical information, enabling the timely extraction of short-term dynamic characteristics, providing basic data support for subsequent state prediction, sensitivity analysis, and other stages, effectively improving the depth and timeliness of data-driven analysis.

[0051] 2. This scheme employs node-by-node differential state transition representation and innovatively introduces a data-driven saturation suppression rule to dynamically limit the local voltage change rate at nodes, preventing prediction distortion caused by noise interference or extreme abrupt changes. Compared to traditional static prediction or empirical limit methods, this adaptive suppression mechanism can automatically converge the prediction amplitude based on real-time historical increments and quantization resolution, balancing the sensitivity and safety margin of the voltage response, and effectively reducing the risk of misjudgment and over-adjustment.

[0052] 3. This scheme no longer uses a single empirical initial value. Instead, it constructs a diagonal block-type error covariance matrix based on the reference voltage and historical deviation statistics of each node, as well as the physical upper and lower bound statistics of each unit. This differentiated and block-based setting of initial values ​​and covariance fully considers the measurement uncertainties and adjustment capability boundaries of each state component of the system, improves the reliability and physical fit of the state estimation, and avoids the problems of state estimation deviation or slow convergence caused by traditional simple assumptions, laying the foundation for high-quality estimation.

[0053] 4. This scheme designs a direct observation mapping relationship for voltage components. The observation matrix precisely controls the coupling relationship between the state vector and the observation data, which helps maintain the simplicity and interpretability of the observation model. Compared with common integrated observation functions or indirect measurement methods, this design reduces the uncertainty of the observation mapping, makes the data processing more efficient, reduces the risk of error propagation caused by inconsistencies between measurement and model, and is suitable for large-scale multi-node observation scenarios.

[0054] 5. In the state prediction and covariance propagation stages, the scheme employs step-by-step prediction of voltage and reactive power components, combined with adjustments to the Jacobian matrix structure based on historical local variation ratios. The entire propagation process strictly follows the prior-posterior sequence of actual power grid operation, achieving dynamic tracking of short-term trends and changes. Compared to state propagation methods with fixed structures or those that do not distinguish between components, this method more realistically reflects the local variation characteristics of the power grid, improves prediction accuracy and dynamic response capabilities, and facilitates early detection and emergency response to abnormal fluctuations.

[0055] 6. By closely integrating the measurement error covariance with the actual voltage measurement quantization step size, and constructing the optimal fused Kalman gain through the least mean square criterion, optimal a posteriori correction of the joint state is achieved. Unlike conventional static or empirical weighting correction methods, this scheme can dynamically adapt to the measurement accuracy and actual fluctuations of the power grid, improving the reliability of estimation and its applicability in practical engineering, effectively eliminating the influence of measurement noise, and improving the accuracy of the final limit calculation.

[0056] 7. In multi-node, multi-reactive-unit scenarios, by tracing the most recent change index of each unit and current and historical data, a voltage-to-reactive-force sensitivity matrix is ​​constructed in real time, and the least squares principle is used to adaptively generate suggested reactive-force adjustment amounts. This method overcomes the limitations of existing methods that rely on fixed sensitivity coefficients or offline simulations, and can quickly reflect the latest characteristics of the system's operating state, improving the scientific nature and real-time performance of reactive-force distribution and voltage regulation. It is suitable for dynamic optimization under the conditions of large-scale and rapid fluctuations in renewable energy.

[0057] 8. This scheme dynamically generates the untruncation upper limit for each unit based on posterior estimation and sensitivity adjustment results, and strictly truncates the limit through physical boundary criteria, ultimately outputting the reactive power upper limit setting matrix. Unlike traditional methods that rely on experience or offline uniform setting of limits, this method can tailor the optimal limit for each unit according to the real-time status at each moment, maximizing the release of available regulation capacity, preventing protection failure and overload risks, achieving safe, flexible and efficient coordinated control of the power grid, and improving the absorption of new energy and voltage stability. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example, refer to Figure 1 An adaptive tuning method for reactive power voltage control limits that takes into account fluctuations in renewable energy sources, comprising:

[0061] Construct node voltage sequences and controllable renewable energy reactive power injection sequences, establish a joint state vector containing voltage components of each node and reactive power components of each controllable unit, and cache historical voltage data of each node.

[0062] Establish a state transition that is updated node by node, perform short-time voltage prediction and short-time reactive power retention, set data-driven saturation suppression rules, and form an overall state transition relationship;

[0063] Set the initial values ​​and covariance of the state estimation. The voltage part is based on the reference voltage and the initial deviation statistics, and the reactive power part is based on the physical upper and lower bound statistics to form the initial error covariance matrix.

[0064] An observation model is constructed, with voltage components selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector.

[0065] Perform state prediction and covariance propagation, calculate voltage prior estimates by node, calculate reactive power prior estimates by unit, and generate prior estimates and corresponding covariances of joint state vectors;

[0066] Calculate the gain and perform state correction, give the measurement error covariance matrix and quantization step size, and then update the a priori state and a priori covariance.

[0067] Calculate the voltage sensitivity to reactive power and solve for the reactive power adjustment amount, extract the current voltage and reactive power estimate, obtain the target voltage of each node, obtain the recent change index of each unit, generate the sensitivity matrix and form the suggested reactive power adjustment vector;

[0068] The reactive power upper limit is set and boundary control is implemented to obtain the final upper limit setting value and setting vector of each unit.

[0069] By constructing a joint state vector of multi-node voltage components and controllable renewable energy reactive power injection components, and performing a series of operations on this vector, including element-by-element update, state prediction, observation mapping, gain correction, sensitivity analysis, and limit setting, this method solves the problems of isolated data for each node and unit and statically fixed limits in traditional reactive power and voltage control. In existing technologies, systems often only adjust a single node or a single renewable energy device, lacking coordinated optimization of multiple nodes and reactive power units across the entire network, leading to local overvoltage or excessive reactive power dispatch. This scheme achieves dynamic modeling of the grid's operating state by caching historical voltage data and real-time reactive power injection data; avoids estimation divergence caused by extreme fluctuations through data-driven saturation suppression; improves state estimation accuracy by fusing prediction and observation data through Kalman gain correction; quantifies the voltage impact of each unit through sensitivity matrix calculation, providing a precise basis for limit adjustment; and finally, implements boundary control to ensure that the limits neither exceed physical upper and lower bounds while fully releasing available regulation capacity. This method integrates prediction, estimation, and optimization of limits, overcoming the problems of lack of linkage, over-protection, or over-adjustment in traditional methods. It achieves voltage and reactive power coordinated adaptive setting, improving the grid's response speed, adjustment accuracy, and safety margin to new energy fluctuations.

[0070] The construction of the node voltage sequence and the controllable new energy reactive power injection sequence establishes a joint state vector containing the voltage components of each node and the reactive power components of each controllable unit, and caches the historical voltage data of each node, specifically including:

[0071] Select and number multiple voltage monitoring nodes, install a standard voltage measuring device at each node, and set a fixed sampling period of seconds; acquire the voltage measurement value of each node sequentially according to the sampling time.

[0072] Multiple new energy power plants with reactive power regulation capabilities were selected and numbered, and the reactive power injection value of the corresponding unit was obtained at each sampling time.

[0073] The joint state vector is composed of the node voltage components and the unit reactive components, with the first part being the node voltage components and the second part being the unit reactive components.

[0074] A historical voltage cache is established for each node. The initial value is written at the first sampling time, and then the most recent voltage measurement value is updated according to time.

[0075] Further specific implementation steps include:

[0076] Selecting in the power grid There are 1 voltage monitoring node, numbered as follows: One standard voltage measuring device is installed at each node, and the sampling period is set. (Unit: seconds), each record is as follows: , ;in, The total number of voltage-affected nodes included in the estimation and control; Index for the controlled node; A fixed time interval between two adjacent sampling times; For time indexing; This refers to the system startup time. For the first The time of sampling; For the first Node at time The voltage measurement value;

[0077] At the same time, select A new energy power station with reactive power regulation capability, numbered as follows: Let it be at time 1 The reactive power injection is: ;in, The total number of controllable new energy units with reactive power regulation capability; Index for controllable new energy units; For the first Unit at time The reactive power injection value;

[0078] Construct the joint state vector: ;in, For at any time The system status, before The first component is voltage, then... Each component is reactive, and the dimension is [missing information]. ; For dimension The real vector space;

[0079] Construct the first The node's historical voltage cache is .

[0080] This method addresses the problems of inconsistent voltage and reactive power data sources, inconsistent data timing, and missing historical information across nodes in existing technologies. By selecting multiple voltage monitoring nodes in the power grid and numbering them sequentially, and installing standardized voltage measurement equipment at each node with a fixed second-level sampling period, this method obtains the reactive power injection values ​​from several renewable energy power plants with reactive power regulation capabilities. Finally, a joint state vector is formed by concatenating the node voltage components and unit reactive power components, and a historical voltage cache is established for each node. In traditional systems, voltage data is often collected separately through different devices and protocols, and historical data retention is limited, failing to provide effective support for short-term dynamic prediction. This method, through synchronous sampling and historical cache design, enables subsequent state transition and filtering algorithms to perform collaborative analysis of voltage and reactive power at the same time point, ensuring data timing consistency and reliability.

[0081] The process of establishing a node-by-node state transition, performing short-time voltage prediction and short-time reactive power retention, setting data-driven saturation suppression rules, and forming an overall state transition relationship specifically includes:

[0082] For each node, a short-time voltage prediction is performed, and the predicted value is obtained by the ratio of the current voltage to the recent voltage change of this node.

[0083] When calculating the local change ratio of this node, the change ratio is generated according to the incremental relationship of the three most recent voltage measurements. When the measurement is insufficient or the denominator is zero, it is set to zero.

[0084] Perform data-driven saturation suppression, giving an upper bound on the amplitude based on historical incremental amplitude and quantization resolution, maintaining direction by sign and limiting amplitude by minimum value;

[0085] Keep the reactive power injection of each power station constant within a short time scale, and use it as the reactive power prediction value for the next moment.

[0086] The updates of each node and each unit are merged to form an overall state transition relationship. The input is the state of the previous time step, and the output is the state of the next time step.

[0087] Further specific implementation steps include:

[0088] A voltage prediction expression is established for each node, specifically as follows:

[0089] ;in, For the predicted sequence of the first Voltage at the node at the next moment; For the first Node at time The local change ratio coefficient is as follows:

[0090] and to Perform data-driven saturation suppression:

[0091] ,

[0092] ;in, For the first The upper bound of the node's amplitude suppression; For voltage measurement resolution; For a sign function, when Return 1 when Returns 0 when return ; Formal parameter independent variable; This serves as a time cursor index for traversal;

[0093] For each power station The reactive power injection value is set to remain stable for a short period of time. ;

[0094] The state transition function is constructed by merging all state updates, specifically as follows:

[0095] ;in, This is the state transition function, with the input being... The output is .

[0096] By establishing short-time voltage prediction models for each monitoring node and employing a local change ratio coefficient adaptively calculated from historical voltage increments to implement data-driven saturation suppression of the prediction results, and setting the reactive power injection of each renewable energy power station as a constant within a short period, this method solves the problem of existing technologies that rely on a single formula for predicting voltage change trends and do not consider extreme fluctuation suppression. Traditional short-term voltage prediction typically uses fixed autoregressive models or manual empirical coefficients, which are prone to divergence under abnormal fluctuations or measurement noise. This method addresses these issues by: 1) calculating the change ratio using the voltage difference between adjacent nodes, which is based on historical data and closely approximates the expected fluctuations; 2) introducing a saturation threshold determined by both measurement resolution and historical amplitude to limit the upper bound of the prediction amplitude; and 3) maintaining stable reactive power injection to avoid errors caused by the superposition of reactive power fluctuations and voltage fluctuations in multivariate prediction. This step-by-step prediction scheme combined with suppression rules can capture short-time voltage change trends and effectively suppress abnormal jumps and measurement noise, providing more reliable input data for the dynamic operation of the power grid.

[0097] The initial values ​​and covariance of the state estimation are set up. The voltage part is based on the reference voltage and the initial deviation statistics, and the reactive power part is based on the physical upper and lower bound statistics, forming the initial error covariance matrix, which specifically includes:

[0098] When the system starts up, the initial estimates of each component of the joint state are set to the first acquisition values;

[0099] Construct an initial error covariance matrix in the form of diagonal blocks. The variance of the voltage part is given based on the statistical analysis of the reference voltage and initial deviation of each node, and the variance of the reactive power part is given based on the statistical analysis of the half-amplitude of the physical upper and lower bounds of each unit.

[0100] Further specific implementation steps include:

[0101] The initial state estimate is set according to the first data acquisition value, specifically as follows:

[0102] , ;in, For the first Initial estimates of each state component; For component index;

[0103] The initial error covariance matrix is ​​constructed as follows: ;in, This is the initial estimate of the error covariance matrix; Construct operators for diagonal block matrices; , They are respectively Rank, An identity matrix of order; The initial mean square of the voltage deviation; For the first Node reference voltage; The mean square of half the range of the nonfunctional interval; , The first The lower and upper limits of the unit's reactive physical parameters are constrained by the equipment nameplate.

[0104] This method addresses the problems of arbitrary initial value setting and uniform error covariance assumptions in traditional state estimation by setting the initial estimates of each component of the joint state vector to the first acquired values ​​during system startup, and constructing a diagonal block-form initial estimation error covariance matrix based on the historical deviation statistics of the reference voltage of each node and the half-amplitude statistics of the physical upper and lower bounds of each unit. In existing technologies, filters are often initialized using experience or default small variances, which cannot accurately reflect the measurement accuracy and physical variation range of each component. This method proposes: first, the initial variance of the voltage component is generated by statistical analysis of the deviation between the actual reference voltage and the observation, which better reflects the accuracy of on-site measurements; second, the variance of the reactive power component is calculated from the upper and lower bounds provided by the equipment nameplate, reflecting the uncertainty of the regulation capability boundary; and third, a block diagonal form is adopted to manage the covariance of the voltage and reactive power components separately, avoiding disordered propagation of error covariance between different components. In this way, the filter can be weighted based on a more realistic error distribution during the first iteration, improving convergence speed and estimation accuracy.

[0105] The construction of the observation model, in which voltage components are selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector, specifically includes:

[0106] At each sampling time, an observation vector is constructed, and the observation components are the voltage measurement values ​​of each node;

[0107] The observation matrix is ​​set up to directly map the voltage components in the joint state to the observation vector, while non-voltage components do not participate in the mapping.

[0108] Further specific implementation steps include:

[0109] At each sampling time The observable variable is the voltage component. Let the observation vector be the first Each component , And the observations are grouped into vectors: ;in, For at any time The observation vector;

[0110] Construct the observation function as an identity mapping and then construct the observation matrix. ;in, for Zero-dimensional matrix; observation matrix State The former Each voltage component is directly mapped to the observation.

[0111] By directly using voltage measurements as components of the observation vector at each sampling time and constructing an observation matrix with an identity mapping, the first few dimensions of the joint state vector are directly mapped to the observation data. This method solves the problems of mapping errors and model uncertainties caused by indirect measurement or transformation in traditional observation models. In existing technologies, complex observation functions or mapping matrices containing redundant components are often used to adapt to different measurement devices or achieve multi-parameter observation, resulting in high complexity for model calibration and online computation. This method simplifies the observation model by including only the truly measurable voltage components in the observation, reducing computational load and eliminating errors that may be introduced by virtual mapping components. The observation matrix uses a zero-block and unit-block concatenation method to ensure that the filter only corrects the voltage components in the observation update step, while the reactive power components remain in the predicted state, thereby improving online update efficiency and estimation quality. It is particularly suitable for new energy power grids with large grid-connected capacity and a large number of nodes.

[0112] The execution state prediction and covariance propagation process calculates voltage prior estimates by node and reactive power prior estimates by unit, generating prior estimates and corresponding covariances for the joint state vector. Specifically, this includes:

[0113] Prior estimates of voltage are calculated at nodes, prior estimates of reactive power are calculated at units, and these are combined to form a joint state vector prior estimate.

[0114] Construct the Jacobian structure of the state function with respect to the state: voltage corresponds to one plus the local change ratio for the diagonal terms, reactive power corresponds to one for the diagonal terms, and the remaining terms are zero;

[0115] The prior covariance is calculated using the Jacobian and the posterior covariance of the previous time step, and the process noise covariance is set as a zero matrix.

[0116] Further specific implementation steps include:

[0117] Execution state prediction, specifically:

[0118] S501, , ;in, For the first Each node at time... The prior estimate of the voltage; For the first Each node at time... The posterior estimate of the voltage;

[0119] S502, , ;in, For the first Each controllable new energy unit at any time The prior estimate of reactive power; For the first Each controllable new energy unit at any time The reactive power posterior estimate;

[0120] S503, ;in, Let be the prior vector of the joint state;

[0121] Constructing the Jacobian matrix of the state function Each non-zero term is:

[0122] S504, Regarding voltage components: ;in, Let be the Jacobian matrix of the state transition function with respect to the state, with dimension . ;

[0123] S504. Reactive power injection: ;

[0124] All other partial derivatives are 0;

[0125] Constructing covariance predictions, specifically:

[0126] , ;in, The prior error covariance; The zero-triggered matrix is ​​used to represent the process noise covariance. for Zero-dimensional matrix.

[0127] This method addresses the problems of inaccurate covariance propagation in existing technologies by performing prior voltage estimation at each node and prior reactive power estimation at each unit, then concatenating these estimates to form a joint state vector. The covariance is then propagated using the Jacobian matrix of the actual state transition function. Traditional Kalman filtering often treats the entire state vector as a single, unified prediction, neglecting the different dynamic and physical constraints of voltage and reactive power, leading to deviations in prior covariance estimation. This method explicitly defines the partial derivative structure of the state functions for each component: the diagonal Jacobian term for the voltage component is one plus the short-term change ratio; the diagonal term for the reactive power component is fixed at one, with the remaining terms being zero. Furthermore, the process noise covariance is explicitly set as a zero matrix, aligning with the design principle of not actively introducing artificial noise into reactive power prediction. This block-based, component-based covariance propagation not only conforms to the physical characteristics of the power grid but also accurately quantifies prediction uncertainty, effectively integrating observational information in subsequent corrections and improving filter performance.

[0128] The calculation of gain and state correction, providing the measurement error covariance matrix and quantization step size, followed by post-abstract state and posterior covariance updates, specifically includes:

[0129] Set the measurement error covariance matrix. The single-channel measurement variance is constructed according to the voltage quantization step size, which is zero kilovolts.

[0130] Calculate the gain matrix based on the prior covariance and the observation matrix;

[0131] The prior state is corrected using the gain matrix to obtain the posterior state, and the covariance is corrected according to the standard format to obtain the posterior covariance.

[0132] Further specific implementation steps include:

[0133] Construct the measurement error covariance matrix: , , ;in, The measurement error covariance matrix; For single-channel measurement variance; Quantization step size for voltage measurement;

[0134] The Kalman gain matrix is ​​constructed and calculated as follows: ; Optimal fusion of priors and observations in the least mean square sense;

[0135] Updated state estimate:

[0136] ;

[0137] Correcting for covariance: ;in, for An identity matrix of order; Let be the posterior error covariance.

[0138] By constructing a measurement error covariance matrix and combining the single-channel measurement variance with the voltage quantization step size, the Kalman gain matrix is ​​dynamically calculated. This gain matrix is ​​then used to correct for prior states and covariance. This method addresses the problem of fixed filter gain or reliance on empirical parameters in existing technologies. In traditional practices, it is often assumed that measurement noise is fixed, leading to either over-reliance on predictions or over-reliance on observations during the fusion process, affecting estimation accuracy. This method proposes using the actual quantization step size, i.e., the minimum resolution of the measurement equipment, as the basis for constructing the error covariance, enabling the observation update weights to automatically adjust with changes in measurement accuracy. Simultaneously, the optimal gain is calculated using the prior covariance and the observation matrix to achieve minimum mean square error fusion. In this way, the filter can adaptively adjust its trust in the observation data under different operating conditions, effectively suppressing the interference of measurement noise or prediction bias on the estimation results, thereby improving the accuracy and stability of the overall state estimation.

[0139] The calculation of voltage sensitivity to reactive power and the solution of reactive power adjustment amount, extraction of current voltage and reactive power estimate, acquisition of target voltage for each node, acquisition of recent change index for each unit, generation of sensitivity matrix and formation of suggested reactive power adjustment vector, specifically includes:

[0140] The voltage estimate of each node and the reactive power estimate of each unit are obtained from the posterior state, and the deviation sequence between the target voltage and the estimated voltage of each node is calculated.

[0141] For each reactive power unit, find the discrete index of the most recent reactive power change; if it does not exist, mark it as empty.

[0142] The sensitivity matrix elements are set according to the current and historical voltage and reactive power changes, and are set to zero when the denominator is zero or the index does not exist.

[0143] Construct a voltage deviation column vector and a sensitivity matrix. When the matrix is ​​invertible, obtain the suggested reactive power adjustment vector. When the sensitivity matrix is ​​not invertible, output the zero vector.

[0144] Further specific implementation steps include:

[0145] Extracting the estimator: , ;in, For the state vector of the th Estimation of individual voltage components; For the state vector of the th An estimate of the reactive component;

[0146] Calculate the target voltage deviation ;

[0147] Construct the sensitivity matrix as follows:

[0148] S701, Obtain the first Discrete index of the most recent change in reactive power of the unit :

[0149] If the set is empty, then It does not exist;

[0150] S702. Set the sensitivity matrix elements, specifically as follows:

[0151] ;in, Let be the element of the sensitivity matrix, representing the th Unit reactive power change on the first The linear ratio of node voltage changes;

[0152] Calculate the reactive power adjustment:

[0153] S703, Order ,

[0154] ;in, This is a column vector of voltage deviations; The sensitivity matrix has dimensions [not specified]. ;

[0155] S704, Suggested Construction of Reactive Power Adjustment Vector Specifically:

[0156] ;

[0157] in, for The One portion, for Zero vector; This is the sensitivity matrix.

[0158] This method solves the problem of fixed sensitivity coefficients in existing technologies, which fail to reflect real-time operating conditions, by extracting voltage and reactive power estimates from the posterior state, obtaining the target voltage deviation of each node, retrieving the index of the most recent reactive power change for each renewable energy unit, setting sensitivity matrix elements based on current and historical data, and generating suggested reactive power adjustment vectors using matrix inversion or a backup zero vector strategy. Existing scheduling methods often rely on offline simulations or empirical coefficient tables, making it difficult to capture the impact of load changes and renewable energy output fluctuations in a timely manner. This method ensures that sensitivity calculations are based only on the latest valid change data through index retrieval, automatically reflects the real-time impact of each unit on voltage by calculating elements using the ratio of voltage deviation to historical reactive power changes, and outputs a zero vector in the case of matrix inversion to prevent adjustment anomalies caused by unstable calculations or insufficient data. This adaptive sensitivity analysis and adjustment strategy ensures both the timeliness and relevance of the adjustment scheme, avoids the adjustment risks caused by misuse of failure coefficients, and improves the scientific and real-time nature of reactive power allocation.

[0159] The process of setting the reactive power upper limit and implementing boundary control yields the final upper limit setting value and setting vector for each unit, specifically including:

[0160] For each reactive power unit, generate an untrunculated upper limit setting value based on the current reactive power estimate and the suggested reactive power adjustment.

[0161] Perform boundary truncation: when the uncrunted value is below the physical lower bound, take the lower bound; when it is above the physical upper bound, take the upper bound; when it is within the range, retain the uncrunted value.

[0162] Summarize the final upper limit tuning values ​​of each unit and output the tuning vector.

[0163] Further specific implementation steps include:

[0164] Set the first Untrunculated temporary upper limit setting value of the unit Specifically:

[0165] ;

[0166] Execute limit boundary truncation:

[0167] ;in, For the first The final upper limit setting value of the unit;

[0168] Final output tuning matrix: ;in, This is the limit tuning vector for all controllable units.

[0169] This method generates an untruncation upper limit setting value for each unit by adding the suggested reactive power adjustment to the current reactive power estimate. This value is then truncated according to physical upper and lower bounds. Finally, the values ​​are aggregated to form a network-wide reactive power limit setting vector. This method solves the problem that existing static limit settings cannot dynamically adapt to the grid's operating state. Traditionally, reactive power upper limits rely heavily on experience or fixed manufacturer recommendations, failing to meet real-time adjustment needs arising from renewable energy fluctuations and easily triggering protection actions or excessive power supply. This method precisely truncates the upper and lower bounds for each unit, ensuring the setting value is neither lower than the equipment's safety lower limit nor exceeds the maximum regulation capacity. Simultaneously, it outputs a setting vector based on the overall grid situation, enabling unified distribution and control across multiple units. This adaptive boundary control method not only improves the grid's response speed to reactive power demand but also maximizes the release of renewable energy regulation capacity while ensuring equipment safety, enhancing the grid's capacity to support high-proportion renewable energy integration and its operational stability.

[0170] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0171] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An adaptive tuning method for reactive power voltage control limits considering fluctuations in renewable energy sources, characterized in that, include: Construct node voltage sequences and controllable renewable energy reactive power injection sequences, establish a joint state vector containing voltage components of each node and reactive power components of each controllable unit, and cache historical voltage data of each node. Establish a state transition that is updated node by node, perform short-time voltage prediction and short-time reactive power retention, set data-driven saturation suppression rules, and form an overall state transition relationship; Set the initial values ​​and covariance of the state estimation. The voltage part is based on the reference voltage and the initial deviation statistics, and the reactive power part is based on the physical upper and lower bound statistics to form the initial error covariance matrix. An observation model is constructed, with voltage components selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector. Perform state prediction and covariance propagation, calculate voltage prior estimates by node, calculate reactive power prior estimates by unit, and generate prior estimates and corresponding covariances of joint state vectors; Calculate the gain and perform state correction, give the measurement error covariance matrix and quantization step size, and then update the a priori state and a priori covariance. Calculate the voltage sensitivity to reactive power and solve for the reactive power adjustment amount, extract the current voltage and reactive power estimate, obtain the target voltage of each node, obtain the recent change index of each unit, generate the sensitivity matrix and form the suggested reactive power adjustment vector; The reactive power upper limit is set and boundary control is implemented to obtain the final upper limit setting value and setting vector of each unit.

2. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 1, characterized in that, The construction of the node voltage sequence and the controllable new energy reactive power injection sequence establishes a joint state vector containing the voltage components of each node and the reactive power components of each controllable unit, and caches the historical voltage data of each node, specifically including: Select and number multiple voltage monitoring nodes, install a standard voltage measuring device at each node, and set a fixed sampling period of seconds; acquire the voltage measurement value of each node sequentially according to the sampling time. Multiple new energy power plants with reactive power regulation capabilities were selected and numbered, and the reactive power injection value of the corresponding unit was obtained at each sampling time. The joint state vector is composed of the node voltage components and the unit reactive components, with the first part being the node voltage components and the second part being the unit reactive components. A historical voltage cache is established for each node. The initial value is written at the first sampling time, and then the most recent voltage measurement value is updated according to time.

3. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 2, characterized in that, The process of establishing a node-by-node state transition, performing short-time voltage prediction and short-time reactive power retention, setting data-driven saturation suppression rules, and forming an overall state transition relationship specifically includes: For each node, a short-time voltage prediction is performed, and the predicted value is obtained by the ratio of the current voltage to the local change of the node. When calculating the local change ratio of this node, the change ratio is generated according to the incremental relationship of the three most recent voltage measurements. When the measurement is insufficient or the denominator is zero, it is set to zero. Perform data-driven saturation suppression, giving an upper bound on the amplitude based on historical incremental amplitude and quantization resolution, maintaining direction by sign and limiting amplitude by minimum value; Keep the reactive power injection of each power station constant within a short time scale, and use it as the reactive power prediction value for the next moment. The updates of each node and each unit are merged to form an overall state transition relationship. The input is the state at the previous time step, and the output is the state at the next time step.

4. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 3, characterized in that, The initial values ​​and covariance of the state estimation are set up. The voltage part is based on the reference voltage and the initial deviation statistics, and the reactive power part is based on the physical upper and lower bound statistics, forming the initial error covariance matrix, which specifically includes: When the system starts up, the initial estimates of each component of the joint state are set to the first acquisition values; Construct an initial error covariance matrix in the form of diagonal blocks. The variance of the voltage part is given based on the statistical analysis of the reference voltage and initial deviation of each node, and the variance of the reactive power part is given based on the statistical analysis of the half-amplitude of the physical upper and lower bounds of each unit.

5. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 4, characterized in that, The construction of the observation model, in which voltage components are selected as the observations, and the observation matrix directly maps the voltage components in the state to the observation vector, specifically includes: At each sampling time, an observation vector is constructed, and the observation components are the voltage measurement values ​​of each node; The observation matrix is ​​set to directly map the voltage components in the joint state to the observation vector, while non-voltage components do not participate in the mapping.

6. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 5, characterized in that, The execution state prediction and covariance propagation process calculates voltage prior estimates by node and reactive power prior estimates by unit, generating prior estimates and corresponding covariances for the joint state vector. Specifically, this includes: Prior estimates of voltage are calculated at nodes, prior estimates of reactive power are calculated at units, and these are combined to form a joint state vector prior estimate. Construct the Jacobian structure of the state function with respect to the state: voltage corresponds to one plus the local change ratio for the diagonal terms, reactive power corresponds to one for the diagonal terms, and the remaining terms are zero; The prior covariance is calculated using the Jacobian and the posterior covariance of the previous time step, and the process noise covariance is set as a zero matrix.

7. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 6, characterized in that, The calculation of gain and state correction, providing the measurement error covariance matrix and quantization step size, followed by post-abstract state and posterior covariance updates, specifically includes: Set the measurement error covariance matrix. The single-channel measurement variance is constructed according to the voltage quantization step size, which is zero kilovolts. Calculate the gain matrix based on the prior covariance and the observation matrix; The prior state is corrected using the gain matrix to obtain the posterior state, and the covariance is corrected according to the standard format to obtain the posterior covariance.

8. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 7, characterized in that, The calculation of voltage sensitivity to reactive power and the solution of reactive power adjustment amount, extraction of current voltage and reactive power estimate, acquisition of target voltage for each node, acquisition of recent change index for each unit, generation of sensitivity matrix and formation of suggested reactive power adjustment vector, specifically includes: The voltage estimates of each node and the reactive power estimates of each unit are obtained from the posterior state, and the deviation sequence between the target voltage and the estimated voltage of each node is calculated. For each reactive power unit, find the discrete index of the most recent reactive power change; if it does not exist, mark it as empty. The sensitivity matrix elements are set according to the current and historical voltage and reactive power changes, and are set to zero when the denominator is zero or the index does not exist. Construct a voltage deviation column vector and a sensitivity matrix. When the matrix is ​​invertible, obtain the suggested reactive power adjustment vector. When the sensitivity matrix is ​​not invertible, output the zero vector.

9. The adaptive tuning method for reactive power voltage control limits considering new energy fluctuations according to claim 8, characterized in that, The process of setting the reactive power upper limit and implementing boundary control yields the final upper limit setting value and setting vector for each unit, specifically including: For each reactive power unit, generate an untrunculated upper limit setting value based on the current reactive power estimate and the suggested reactive power adjustment. Perform boundary truncation: when the uncrunted value is below the physical lower bound, take the lower bound; when it is above the physical upper bound, take the upper bound; when it is within the range, retain the uncrunted value. Summarize the final upper limit tuning values ​​of each unit and output the tuning vector.

Images