Electric power system bad data identification method and system adaptive to multiple types of errors

Abstract

The invention relates to the technical field of electric power system state estimation, in particular to an electric power system bad data identification method and system suitable for multiple types of errors, and the method comprises the steps: carrying out the preliminary screening of measurement and topology bad data based on a Lagrange multiplier method, screening out suspicious measurement and topology bad data, and carrying out the calculation of the measurement and topology bad data. According to the method, the problem of low solving efficiency caused by excessive integer variables of the model is solved, the identification of measurement and topology errors is realized based on the MINLP optimization model, the identification and correction of measurement mapping data are realized, the limitation of a traditional method on robust performance is solved, and technical support is provided for realizing multi-source data checking and repairing of a power system. According to the bad data identification method for the electric power system, measurement and topology errors can be processed at the same time, the problems of missed judgment and misjudgment possibly caused by a traditional method are effectively avoided, bad data of measurement mapping are corrected and reserved, loss of system observability is avoided, and the safety and stability of operation of the electric power system are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system state estimation, in particular to a power system bad data identification method and system suitable for multiple types of errors. BACKGROUND

[0002] Due to the insufficient accuracy of measurement equipment meters, communication error codes and other reasons, the telemetry and remote signaling data of the power system may have bad data with large errors, which has a great impact on the accuracy of state estimation, and in serious cases, it may cause the state estimation to fail to converge. The traditional residual identification method identifies bad data based on statistical rules, and is invalid for lever measurement bad data. The optimization identification method has high identification accuracy, but due to the complex nonlinear characteristics of the model, the solving efficiency is a great challenge. Robust state estimation models such as SHGM estimation model and M estimation model can suppress the influence of lever measurement bad data to a certain extent, but cannot eliminate it, and do not have the ability to identify and correct measurement and reference data. For power systems with insufficient observability, the adaptability is insufficient.

[0003] In terms of MINLP problem solving algorithms, for a large number of repeated structures in constraints and objective functions, the traditional automatic differentiation processing is completed by traversing each node in the graph, and the derivative is propagated step by step along the edge. This method is low in computational efficiency. For the processing of integer variables, traditional MINLP solving methods usually rely on the relaxation of integer variables, but in many practical problems, functions are not defined or discontinuous at non-integer points, which makes these methods unable to be applied or difficult to converge.

[0004] Therefore, an effective power system bad data identification method suitable for multiple types of errors is needed, which simultaneously considers automatic differentiation processing to achieve fast solving, has data identification and correction capability for measurement and reference scenarios, and can improve the observability of the power system compared to the traditional "one-size-fits-all" identification and rejection. SUMMARY

[0005] The purpose of the present application is to provide a power system bad data identification method and system suitable for multiple types of errors, which solves the problem that telemetry and remote signaling data are stored in multiple types of business systems during the construction and development of the power system, which may lead to multiple types of data errors and is difficult to identify and correct.

[0006] The present application is achieved by the following technical solutions:

[0007] In a first aspect, the present application provides a power system bad data identification method suitable for multiple types of errors, comprising:

[0008] Obtaining measurement data, preprocessing all the measurement data based on a Lagrange multiplier to obtain a plurality of Lagrange multipliers corresponding to the measurement data, setting a judgment threshold, marking the measurement data with the Lagrange multiplier greater than the judgment threshold and storing in a target set;

[0009] Establishing an identification model of discrete multi-type measurement bad data, calculating the identification model for the target set based on an optimization solver to obtain a bad data identification result;

[0010] Based on the bad data identification result, establishing a power system data correction model of measurement and inverse, performing secondary identification on the bad data in the bad data identification result, and correcting the measurement and inverse of the power system data, and outputting a bad data list.

[0011] Preferably, preprocessing all the measurement data based on a Lagrange multiplier comprises:

[0012] Setting an initial objective function and a constraint condition, the constraint condition comprising a measurement constraint, a structure constraint and an operation constraint;

[0013] Classifying the Lagrange multiplier, the measurement value, the measurement equation, the Jacobian matrix and the covariance matrix based on the constraint condition, and constructing a first Lagrange function based on the constraint condition and the initial objective function.

[0014] Preferably, the initial objective function comprises:

[0015]

[0016] In the formula, minJ(x) is an initial objective function output value, r m is a measurement residual, R m and P are covariance matrices of measurement values and prior states, and are state variable estimated values and prior values, respectively;

[0017] The measurement constraint, the structure constraint and the operation constraint comprise:

[0018]

[0019] In the formula, z m is a measurement value, is a measurement equation, represents a zero injection power constraint, represents information provided by a switch state.

[0020] Preferably, the construction of the first Lagrange function based on the constraint condition and the initial objective function comprises:

[0021]

[0022] wherein, is a Lagrange function, λ m , λ s , λ o are Lagrange multipliers corresponding to measurement constraint, structure constraint and operation constraint respectively, λ is a set of Lagrange multipliers of all constraints, and T is a transpose.

[0023] Preferably, the establishing the identification model of discrete multi-type measurement bad data comprises:

[0024]

[0025] wherein, m1 is a number of suspicious measurements, m2 is a number of non-suspicious measurements, d i is a 0 / 1 integer variable, M cons is a constant, m spc_i is a residual term of suspicious bad data, m good_j is a residual term of non-suspicious bad data, is a set of switch branches, i is a node number i, j is a node number j, s ij is a binary variable, is a switch topology telecontrol state.

[0026] Preferably, the calculating the identification model for the target set based on the optimization solver comprises:

[0027] defining a second Lagrange function and a penalty function;

[0028] defining an iteration point, and constructing a trust region quadratic subproblem;

[0029] estimating a prediction improvement ratio, and updating a trust region radius according to the prediction improvement ratio;

[0030] updating the iteration point and a Hessian matrix, constructing a convergence condition, and judging whether the convergence condition is met;

[0031] if the convergence condition is met, terminating the iteration, and if the convergence condition is not met, redefining the iteration point, constructing the trust region quadratic subproblem, and continuing the iteration until the convergence condition is met.

[0032] Preferably, the estimating the prediction improvement ratio comprises:

[0033]

[0034] wherein, r k is the prediction improvement ratio, x k , y k are values of the iteration point, P σ (x k , y k) represents the penalty function at point (x k ,y k ), P σ (x k +d k ,y k +e k ) represents the penalty function at point (x k +d k ,y k +e k ), where d k and e k represent the update steps of the continuous variable x and the integer variable y, respectively, φ k (0) represents the actual improvement of the objective function value at the current solution (x k ,y k ), ignoring the integer variable y, φ k (d k ,e k ) represents the predicted improvement of the objective function after considering the update steps, as follows:

[0035] φ k (0) = σ k ||g(x k )|| ∞

[0036]

[0037] where g(x k ) = g(x k , 0), is the first order derivative of the objective function at (x k ,y k ), B k is the Hessian matrix approximation of the current solution, used to represent the quadratic approximation of the objective function, σ k is a parameter to control the strength of the penalty, and is the constraint that the first order derivative of g(x, y) at (x k ,y k ) is zero.

[0038] Preferably, the constructing the convergence condition comprises:

[0039] ‖g(x k ,y k ) - ‖ ∞ <ε

[0040] φ k (0) - φ k (d k ,e k ) < ε

[0041] where ε is a convergence threshold, g(x k ,y k ) - is a negative part of constraint violation, defined as follows:

[0042]

[0043] where g j (x k ,y k ) represents the jth constraint, m e is the number of equality constraints.

[0044] Preferably, the power system data correction model for measurement inversion is established by comprising:

[0045]

[0046] where b i is a 0 / 1 integer variable representing whether the measurement is inverted, m bad_i is the ith suspicious bad measurement.

[0047] In a second aspect, the application further provides a power system bad data identification system suitable for multiple types of errors, characterized in that the power system bad data identification system suitable for multiple types of errors is used to execute the power system bad data identification method suitable for multiple types of errors, and further comprises:

[0048] The bad data preliminary screening module is configured to acquire measurement data, pre-process all the measurement data based on Lagrange multipliers, obtain a plurality of Lagrange multipliers corresponding to the measurement data, and set a judgment threshold; measurement data with a Lagrange multiplier greater than the judgment threshold is marked and stored in a target set.

[0049] The bad data identification module is configured to establish an identification model for discrete multiple types of bad measurement data, calculate the identification model for the target set based on an optimization solver, and obtain a bad data identification result.

[0050] The bad data correction module is configured to establish a power system data correction model for measurement inversion based on the bad data identification result, perform secondary identification on the power system bad data in the bad data identification result, correct the power system data for measurement inversion, and output a bad data list.

[0051] The technical scheme of the application has at least the following advantages and beneficial effects:

[0052] 1. The method provided by the application comprises the following steps: screening suspicious measurement and topology bad data based on the Lagrange multiplier method, to avoid low solving efficiency caused by too many integer variables in the model; then, identifying measurement and topology errors based on the MINLP optimization model; finally, identifying and correcting the measurement and topology errors based on the above steps, to solve the limitations of traditional methods in robustness, and to provide technical support for multi-source data checking and repairing of the power system. The bad data identification method of the power system provided by the method can handle measurement and topology errors, effectively avoid the missed judgment and misjudgment problems caused by traditional methods, correct and retain the bad data of the measurement and topology, avoid the loss of system observability, and ensure the safety and stability of the power system operation.

[0053] 2. The method can handle multiple types of errors such as measurement and topology errors, has unified identification and processing capability, can greatly improve the quality of the state estimation calculation result, and can greatly improve the solving efficiency of the algorithm. The structure-aware automatic differentiation and trust region SQP method can construct and solve the MIQP sub-problem at the integer point, can handle non-relaxable MINLP problems, and is more robust and has fewer function calls than traditional methods, and is suitable for engineering optimization problems containing integer variables. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation to the scope. Other related drawings can also be obtained by those skilled in the art without creative labor.

[0055] Figure 1 The flowchart of the bad data identification of the application;

[0056] Figure 2 The flowchart of the structure-aware automatic differentiation of the application;

[0057] Figure 3 The flowchart of the solving process of the MINLP problem of the application. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solutions and advantages of the embodiments of the application more clear, the following will combine the drawings in the embodiments of the application to clearly and completely describe the technical solutions in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, not all the embodiments. The components of the embodiments of the application described and shown in the drawings can be arranged and designed in various different configurations.

[0059] The independent described module or sub-module can be physically separated or not physically separated, can be software implemented or hardware implemented, and part of the module or sub-module can be implemented by software, and the function of the part of the module or sub-module is called by a processor, and the other part of the module or sub-module is implemented by hardware, for example, by hardware circuit. In addition, part or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the present application.

[0060] Please refer to Figure 1 A power system bad data identification method suitable for multiple types of errors, comprising:

[0061] S101: Obtain measurement data, preprocess all the measurement data based on the Lagrange multiplier, obtain a plurality of Lagrange multipliers corresponding to the measurement data, and set a judgment threshold, mark the measurement data with the Lagrange multiplier greater than the judgment threshold and store it in a target set;

[0062] Among them, the measurement data includes telemetering data and telecontrol data, the measurement and topology bad data are preliminarily screened, all telemetering and telecontrol data sets are preprocessed based on the Lagrange multiplier method, and the telemetering and telecontrol data with the multiplier index exceeding the threshold are marked and stored in the target suspicious set.

[0063] Specifically, since the number of integer variables will affect the solving efficiency of the MINLP model, when the system scale increases, the number of telemetering and telecontrol will also increase, and the solving time of the model is positively correlated with the number of 0-1 integer variables. Therefore, in step S101, the measurement and topology bad data are screened, wherein the topology bad data can be regarded as the measurement data with the measurement value of 0, and the measurement here is the set of measurement values and topology telecontrol data. The Lagrange multipliers of all measurements and topologies are calculated, and the measurements exceeding the threshold are put into the suspicious measurement and topology set. Subsequently, only the 0 / 1 integer variables of the measurements in the suspicious set are set, thereby reducing the number of 0 / 1 variables and improving the calculation efficiency.

[0064] S102: Establish an identification model of discrete multiple types of measurement bad data, calculate the identification model for the target set based on an optimization solver, and obtain a bad data identification result;

[0065] S103: Based on the bad data identification result, establish a power system data correction model of measurement and inversion, perform secondary identification on the power system bad data in the bad data identification result, correct the power system data of measurement and inversion, and output a bad data list.

[0066] The method provided by the application comprises the following steps: performing preliminary screening on the measurement and topology bad data based on the Lagrange multiplier method, screening out suspicious measurement and topology bad data, avoiding low solving efficiency caused by too many integer variables of the model; then, identifying the measurement and topology errors based on the MINLP optimization model; finally, on the basis of the above steps, identifying and correcting the measurement and topology error data, solving the limitations of the traditional method in the robustness performance, and providing technical support for realizing the power system multi-source data checking and repairing. The bad data identification method of the power system provided by the method can handle the measurement and topology errors, effectively avoids the missed judgment and misjudgment problems caused by the traditional method, corrects and retains the bad data of the measurement and topology, avoids the loss of system observability, and ensures the safety and stability of the power system operation.

[0067] In an example embodiment of the application, the pre-processing of all the measurement data based on the Lagrange multiplier comprises the following steps:

[0068] S201: setting an initial objective function and constraint conditions, wherein the constraint conditions comprise measurement constraints, structure constraints and operation constraints;

[0069] S202: classifying the Lagrange multiplier, measurement value, measurement equation, Jacobian matrix and covariance matrix based on the constraint conditions, and constructing a first Lagrange function based on the constraint conditions and the initial objective function.

[0070] Specifically, the initial objective function comprises:

[0071]

[0072] In the formula, minJ(x) is the output value of the initial objective function, r m is the measurement residual, R m and P are the covariance matrices of the measurement value and the prior state, and are the state variable estimated value and the prior value, respectively;

[0073] The measurement constraints, structure constraints and operation constraints comprise:

[0074]

[0075] In the formula, z m is the measurement value, is the measurement equation, represents the zero injection power constraint, represents the information provided by the switch state.

[0076] Secondly, according to the above three types of constraints, the Lagrange multiplier, measurement value, measurement equation, Jacobian matrix and covariance matrix are classified, and there are:

[0077]

[0078] z = [z m 0 0] T

[0079]

[0080] In the formula, λ is the set of all constrained Lagrange multipliers, λ m , λ s , λ o Let H be the Lagrange multipliers corresponding to the measurement constraints, structural constraints, and operational constraints, respectively; H be the Jacobian matrix; and R be the set of covariance matrices for all constraints. m R s R0 and R0 are the covariance matrices corresponding to the measurement constraints, structural constraints, and operational constraints, respectively. Specifically, constructing the first Lagrangian function based on the constraints and the initial objective function includes:

[0081]

[0082] In the formula, Let λ be the Lagrange function, λ be the Lagrange multiplier, and T be the transpose.

[0083] The KKT conditions for the objective function can be expressed as:

[0084]

[0085] In the formula, This represents the gradient of the Lagrange function with respect to the variable x, i.e. The first derivative of the variable x, P -1 It is the inverse of the prior state covariance matrix.

[0086] The above expression can be equivalent to:

[0087]

[0088] Substituting the above equation, we get:

[0089]

[0090] In the formula, Represents the Lagrange function with respect to the variable λ m The gradient, i.e. For variable λ m The first derivative.

[0091] Therefore, the equality constraint can be rewritten as:

[0092]

[0093] That is,

[0094]

[0095] Therefore, the first-order necessary condition of the objective function can be expressed as:

[0096]

[0097] Further, it can be expressed in matrix form as:

[0098]

[0099] In the formula, z is the measured value, and the vector λ is the Lagrange multiplier containing the measurement, operation and structural constraints, which can be used to detect and identify the inconsistency of the mathematical model under the state estimation constraints. Since the method has different sensitivities to each parameter, the Lagrange multiplier needs to be normalized, so the normalized Lagrange multiplier is further defined as:

[0100]

[0101] In the formula, v ii is the i-th row diagonal element of the Lagrange covariance matrix V, is the normalized Lagrange multiplier of the i-th constraint, and λ i is the Lagrange multiplier of the i-th constraint, which can be solved by the following formula:

[0102]

[0103] In the formula, the dimensions of the matrices Σ, C and V correspond one by one to P, H and R.

[0104] For the measurement and topology data, when bad data occurs, the NLM values related to the error information will be significantly higher than the corresponding values of normal measurements. By setting thresholds, the measurements and topology values greater than these thresholds are stored in the suspicious set. For the measurement and topology data whose NLM does not exceed the threshold, it is considered as "good data" or correct topology information, and the 0-1 integer variable is not set, thereby reducing the number of 0-1 variables in the whole, and finally improving the calculation efficiency.

[0105] In an example embodiment of the present application, the optimal state matching under the minimum cost of measurement and topology is considered based on the Kirchhoff's current law (KCL), and a bad data identification model based on MINLP is constructed as follows:

[0106]

[0107] In the formula, m1 is the number of measurements, d i is a 0 / 1 integer variable, M cons is a constant, and mspc_i is a residual term of suspicious bad data, m good_j is a residual term of non-suspicious bad data, is a set of switch branches, i, j are node numbers, s ij is a binary variable, is a switch topology telecommunication state.

[0108] c(x) = 0

[0109] d i = |b i |

[0110]

[0111] z i ∈ M spc

[0112] z j ∈ M good1

[0113] In the formula, c(x) is a set of equality constraints, b i is an integer variable representing the quality of measurement data, and the value range is [-1, 0, 1], b i = 1 represents bad data, b i = 0 represents good data, b i = -1 represents bad data, z i is a measurement value in a target set, h i (x) is an i-th measurement equation expression, R ii is a square of an i-th measurement standard deviation, R jj is a square of a j-th measurement standard deviation, z j is a measurement value in a normal measurement set, h j (x) is a j-th measurement equation expression, I i is a current flowing into a node i, I ij is a current amplitude variable of a branch i j, is a phase current of the branch i j, is a phase voltage of the node i, is a phase voltage of the node j, y ij is an admittance of the branch i j, M spc is a measurable set, M good1 is a normal measurement set.

[0114] An example embodiment of the present application, as shown in Figure 2 the identification model for the target set calculated based on the optimization solver includes:

[0115] The polynomial duplicate processing method based on structure-aware automatic differentiation includes the most critical operation for nonlinear problems, which is to calculate the Jacobian matrix of the constraint and the Hessian matrix of the Lagrange function. When solving the problem, if the differential calculation is performed on the structure by traversal, a large amount of computing power and time will be wasted.

[0116] To solve such problems, the constraints and objective functions can be grouped by structure before solving the problem. The hash value calculated by the hash algorithm is used to determine whether it is the same structure. If it is a hash value that has not been inserted before, it means a new structure, which is used as a representative structure of a group. The automatic differentiation calculation function is generated thereafter, avoiding repeated operations.

[0117] The traditional automatic differentiation processing is completed by traversing each node in the graph, and the derivative is propagated step by step along the edge. Structure-aware automatic differentiation tracks the execution process of reverse automatic differentiation in advance, and generates compact binary code to directly calculate the derivative. This is much more efficient than directly traversing the calculation graph.

[0118] When the structure grouping is completed, the representative of each group is generated by generating a CppAD automatic differentiation function structure, and then the binary code of the automatic differentiation function and its pointer are generated by the just-in-time compilation technology, which are passed to the underlying solver for use. When processing the integer part, it is processed by relaxation and used as a continuous variable to perform automatic differentiation using CppAD, and then the branch and bound method is used to process and approximate the integer solution. Thus, the structure-aware automatic differentiation of the mixed integer nonlinear programming problem is completed.

[0119] The mixed-integer trust region SQP method.

[0120] Consider the general optimization problem of minimizing the objective function f under nonlinear equality and inequality constraints:

[0121] min f(x,y)

[0122] g j (x,y)=0,j=1,…,m e

[0123]

[0124] x l ≤x≤x u

[0125] y l ≤y≤y u

[0126] Where x and y are vectors representing continuous and integer variables, respectively. Assume the problem function is f(x,y) and g... j (x,y),j=1,…,m, in all Under the condition that it is continuously differentiable, where x l Let x be the lower bound of variable x. u Let x be the upper limit of variable y l Let y be the lower bound of the variable y. u m is the upper limit of the variable y. e The number of equality constraints. For n c A real space of dimension n c Let x be the dimension of the continuous variable. For n i Integer space of dimension n i The dimension of the integer variable y is represented by m, and the total number of model constraints is m.

[0127] like Figure 3 As shown, the solution process for this MINLP problem is as follows:

[0128] Define the second Lagrange function and the penalty function.

[0129] Define the Lagrange function as:

[0130]

[0131] The penalty function is:

[0132] P σ (x,y)=f(x,y)+σ·‖g(x,y) - || ∞

[0133] Where, if j≤m e Then g j (x,y) - =g j (x,y), otherwise g j (x,y) - =min(0,g j (x,y)); σ>0 is the penalty factor, u j To constrain g j The Lagrange multipliers corresponding to (x,y), L(x,y,u) are Lagrange functions, and P σ (x,y) is the penalty function.

[0134] Construct a trust region quadratic subproblem, defining the current point as (x k ,y k The problem can be transformed into the following mixed-integer quadratic programming problem:

[0135]

[0136] where d is the continuous variable direction increment, e is the integer variable increment, B k is the approximate Hessian matrix of the current Lagrangian, and are the radii of the trust region for the continuous and integer variables respectively, δ represents the variable perturbation, d y represents the change of the integer variable y.

[0137] Estimate the predicted improvement ratio:

[0138]

[0139] where r k is the predicted improvement ratio, x k , y k are the values of the iteration point, P σ (x k , y k ) represents the penalty function at the point (x k , y k ), P σ (x k + d k , y k + e k ) represents the penalty function at the point (x k + d k , y k + e k ), where d k and e k represent the update steps of the continuous variable x and the integer variable y respectively, φ k (0) represents the actual improvement of the objective function value at the current solution (x k , y k ) after ignoring the integer variable y, φ k (d k , e k ) represents the predicted improvement of the objective function after considering the update steps, as follows:

[0140] φ k (0) = σ k || g(x k ) || ∞

[0141]

[0142] where g(x k ) = g(x k , 0), is the objective function value at the point (xk ,y k ) lower bound of the trust region radius k is the Hessian matrix approximation of the current solution, which is used to represent the quadratic approximation of the objective function, σ k is a parameter to control the penalty strength, is the lower bound of the trust region radius k ,y k ) lower bound of the trust region radius

[0143] Trust region adjustment, update the trust region radius according to the value of r k

[0144]

[0145] is the trust radius, and the integer radius Set the lower bound to prevent the step size from becoming too small and freezing the integer variable.

[0146] Update the iteration point and the Hessian matrix:

[0147] If r k is good enough, accept:

[0148] x k+1 = x k + d k ,y k+1 = y k + e k

[0149] Update B k with BFGS for the next iteration:

[0150]

[0151] where s k is the update step size, is the first derivative of the updated Lagrangian function, is the first derivative of the updated Lagrangian function.

[0152] Construct a convergence condition to determine whether to converge, if the convergence condition is met, terminate the iteration, if the convergence condition is not met, redefine the iteration point, construct a trust region quadratic subproblem, continue iteration until the convergence condition is met.

[0153] The construction of the convergence condition includes:

[0154] ‖g(x k ,y k ) - ‖∞<ε

[0155] φ​k (0)-φ k (d k ,e k )<ε

[0156] In the formula, ε is a convergence threshold, g(x k ,y k ) - is a negative part of constraint violation, defined as follows:

[0157]

[0158] In the formula, g j (x k ,y k ) represents the jth constraint, m e is the number of equality constraints.

[0159] In an embodiment, on the basis of the measurement and topological identification result, the measurement bad data is identified again in step S102, the bad data of the measurement tag is screened, and correction is performed.

[0160] The power system data correction model of the measurement tag is established, including:

[0161]

[0162] c(x)=0

[0163]

[0164] z i ∈M bad

[0165] z j ∈M good2

[0166] In the formula, m bad_i is a measurement residual term corresponding to the second stage suspicious data, M bad is the target set in step S101, M good2 is non-bad data measurement after identification in step S102, the bad data of the measurement tag can be corrected through the above model calculation, the power system robust state estimation calculation is realized, b i is a 0 / 1 integer variable representing whether the measurement tag is.

[0167] In summary, the main process of the application is as follows:

[0168] Based on the obtained original measurement information, including power, current, voltage measurement data telemetry information.

[0169] Based on the Lagrange multiplier method, the suspicious measurement and topological data are screened.

[0170] A measurement and topology bad data unified identification model is established, and structural perception automatic differentiation processing and mixed integer trust region SQP algorithm are used to realize efficient solving.

[0171] On the basis of the above process, a data correction model is established, and structural perception automatic differentiation processing and mixed integer trust region SQP algorithm are used to realize efficient solving.

[0172] A bad data list and global electrical quantity information are output.

[0173] The application further provides a power system bad data identification system suitable for multiple types of errors, which is used for executing the power system bad data identification method suitable for multiple types of errors and further comprises the following.

[0174] The bad data preliminary screening module is configured to acquire measurement data, pre-process all the measurement data based on Lagrange multipliers, obtain a plurality of Lagrange multipliers corresponding to the measurement data, set a judgment threshold, mark the measurement data with a Lagrange multiplier greater than the judgment threshold, and store the measurement data in a target set.

[0175] The bad data identification module is configured to establish an identification model of discrete multiple types of measurement bad data, calculate the identification model for the target set based on an optimization solver, and obtain a bad data identification result.

[0176] The bad data correction module is configured to establish a measurement and inverse power system data correction model based on the bad data identification result, perform secondary identification on the power system bad data in the bad data identification result, correct the measurement and inverse power system data, and output a bad data list.

[0177] In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0178] When the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. The computer software product stored in a storage medium includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0179] The preferred embodiments of the present application have been described above with the intent to enable those skilled in the art to make and use it. Various modifications to the embodiments described in this disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Accordingly, the true scope of the present application is not limited to the embodiments described herein but is only limited to the scope of the claims together with the full range of equivalents to which such claims are entitled.

Claims

1. A power system bad data identification method adaptive to multiple types of errors, characterized in that, include: Acquire measurement data, preprocess all the measurement data based on Lagrange multipliers to obtain several Lagrange multipliers corresponding to the measurement data, set a judgment threshold, mark the measurement data whose Lagrange multipliers are greater than the judgment threshold, and store them in the target set; Establish a discrete multi-type measurement bad data identification model, calculate the identification model for the target set based on the optimization solver, and obtain the bad data identification results; Based on the bad data identification results, a power system data correction model with reversed measurement standards is established. The bad data in the power system in the bad data identification results is re-identified, and the power system data with reversed measurement standards is corrected, and a bad data list is output.

2. The method for bad data detection of power system with multiple types of errors according to claim 1, characterized in that, The preprocessing of all the measurement data based on Lagrange multipliers includes: Set the initial objective function and constraints, including measurement constraints, structural constraints, and operational constraints; Based on the constraints, the Lagrange multipliers, measurements, measurement equations, Jacobian matrix, and covariance matrix are classified, and a first Lagrange function based on the constraints and the initial objective function is constructed.

3. The method for bad data detection of power system with multiple error types according to claim 2, characterized in that, The initial objective function includes: where minJ(x) is the initial objective function output value, r m is the measurement residual, R m and P are the covariance matrices of the measurement and the prior state, and are the state variable estimate and the prior value, respectively; Measurement constraints, structural constraints, and operational constraints include: where z m is a measured value, is a measurement equation, represents a zero-injection power constraint, represents information provided by the switch state.

4. The method for bad data detection of power system with multiple error types according to claim 3, characterized in that, The construction of the first Lagrange function based on constraints and an initial objective function includes: wherein is a Lagrangian function, λ m , λ s , λ o are the Lagrange multipliers corresponding to the measurement constraints, the structure constraints and the operation constraints, respectively, λ is the set of Lagrange multipliers for all constraints, and T is the transpose.

5. The bad data identification method for power systems with multiple types of errors according to claim 1, wherein, The establishment of the identification model for discrete multi-type measurement bad data includes: where m1 is the number of suspicious measurements, m2 is the number of non-suspicious measurements, d i is a 0 / 1 integer variable representing whether a measurement is suspicious or not, M cons is a constant, m spc_i is the residual term of suspicious bad data, m good_j is the residual term of non-suspicious bad data, is a set of switch branches, i and j are both node numbers, s ij is a binary variable, is the status of switch topology telecommunication.

6. The method for bad data detection of power system with multiple types of errors according to claim 5, characterized in that, The identification model for the target set calculated based on the optimization solver includes: Define the second Lagrange function and the penalty function; Define the iteration point and construct the trust region quadratic subproblem; Estimate the predicted improvement ratio and update the trust region radius based on the predicted improvement ratio; Update the iteration points and the Hessian matrix, construct the convergence condition, and determine whether convergence has occurred; If the convergence condition is met, the iteration terminates. If the convergence condition is not met, the iteration point is redefined, a trust region quadratic subproblem is constructed, and the iteration continues until the convergence condition is met.

7. The method for bad data detection of power system with multiple types of errors according to claim 6, characterized in that, The estimated prediction improvement rate includes: In the formula, r k To predict the improvement rate, x k y k The values ​​of P are the iteration points, respectively. σ (x k ,y k ) indicates at point (x k ,y k The penalty function P under ) σ (x k +d k ,y k +e k ) indicates at point (x k +d k ,y k +e k The penalty function under ) where d k With e k φ represents the update step size for the continuous variable x and the integer variable y, respectively. k (0) indicates that in the current solution (x) k ,y k Under the condition that the integer variable y is ignored, the actual improvement in the objective function value is φ. k (d k ,e k The expression () indicates an improvement in the objective function's prediction after considering the updated step size, as follows: φ k (0) = σ k ||g(x k )|| ∞ where g(x k ) = g(x k , 0), is the first derivative of the objective function at (x k , y k ), B k is the Hessian matrix approximation of the current solution, which is used to represent the quadratic approximation of the objective function, σ k is a parameter that controls the strength of the penalty, is the first derivative of the constraint g(x, y) at (x k , y k ).

8. The method for bad data detection of power system with multiple types of errors according to claim 7, characterized in that, The convergence conditions include: ‖g(x k ,y k ) - ‖∞<ε φ k (0)-φ k (d k ,e k )<ε where ε is a convergence threshold, g(x k ,y k ) - is the negative part of the constraint violation, as follows: where g j (x k ,y k ) represents the jth constraint, m e is the number of equality constraints.

9. The bad data identification method for power systems with multiple types of errors according to claim 6, wherein, The power system data correction model for establishing the measurement standard includes: wherein b i is a 0 / 1 integer variable representing whether the measurement flag is bad or not, m bad_i is the ith suspicious measurement bad data.

10. A power system bad data identification system that accommodates multiple types of errors, characterized by, A method for identifying bad data in a power system adaptable to multiple types of errors, as described in any one of claims 1-9, includes: The bad data screening module is configured to acquire measurement data, preprocess all the measurement data based on Lagrange multipliers to obtain several Lagrange multipliers corresponding to the measurement data, set a judgment threshold, mark the measurement data whose Lagrange multipliers are greater than the judgment threshold, and store them in the target set. The bad data identification module is configured to establish an identification model for discrete multi-type measurement bad data, calculate the identification model for the target set based on the optimization solver, and obtain the bad data identification results. The bad data correction module is configured to establish a power system data correction model based on the bad data identification results, perform secondary identification of bad power system data in the bad data identification results, correct the power system data with reversed measurement standards, and output a bad data list.

Images