CN111313413A

CN111313413A - Power system state estimation method based on parallel acceleration of graphics processor

Info

Publication number: CN111313413A
Application number: CN202010201524.2A
Authority: CN
Inventors: 董树锋; 方睿; 唐坤杰; 毛航银
Original assignee: Zhejiang University ZJU; State Grid Zhejiang Electric Power Co Ltd
Current assignee: Zhejiang University ZJU; State Grid Zhejiang Electric Power Co Ltd
Priority date: 2020-03-20
Filing date: 2020-03-20
Publication date: 2020-06-19

Abstract

The invention discloses a power system state estimation method based on parallel acceleration of a graphic processor, which aims at the problem that the prior method needs to generate a Jacobian matrix for many times in an iterative process during power system state estimation, particularly when the system scale is large, the traditional serial generation method is difficult to meet the real-time requirement; when the system scale is large, the method can obviously improve the calculation efficiency and meet the real-time requirement of state estimation. Meanwhile, the method is high in operability and easy to implement.

Description

Power system state estimation method based on parallel acceleration of graphics processor

Technical Field

The invention relates to power system state estimation, in particular to a power system state estimation method based on parallel acceleration of a graphics processor.

Background

The state estimation of the power system is the basis of power grid dispatching operation, and along with the continuous expansion of the scale of the power grid, the demand of improving the calculation efficiency of the state estimation of the power system is improved so as to meet the calculation real-time performance. The method for solving the least square model by the Newton method which is widely applied at present is simple to operate and high in convergence speed, but a Jacobian matrix needs to be generated for many times in the iteration process, and when the system scale is large, the traditional serial generation method cannot meet the real-time requirement easily.

Disclosure of Invention

Aiming at the defects of the prior art, the technical problem to be solved by the invention is to provide a power system state estimation method based on the parallel acceleration of a graphics processor so as to improve the overall convergence performance and the calculation efficiency.

The invention adopts the following technical scheme:

a power system state estimation method based on graphics processor parallel acceleration comprises the following steps:

(1) obtaining state estimation information, comprising: measuring information, namely node voltage, node injection power, line power flow and transformer power flow, line parameter information and a measurement error variance matrix R, forming a measurement vector z and a node admittance matrix Y, and determining a variable x and a measurement function h (x) which need to be subjected to state estimation;

(2) setting initial values of state estimation: setting an initial value x of a variable to be state-estimated⁰The convergence precision epsilon of the residual vector target; the current iteration step number k is set equal to 0.

(3) And (3) calculating a Jacobian matrix: jacobian matrix H (x) for parallel computing current iteration step based on graphics accelerator^k)。

(4) Solving a correction equation set: the correction amount Deltax is obtained by solving a linear equation set expressed in the following formula^k。

Δx^k＝[H^T(x^k)R^-1H(x^k)]^-1H^T(x^k)R^-1[z-h(x^k)]

(5) And (3) state variable correction: calculated by the following formula

x^k+1＝Δx^k+x^k

(6) And (3) convergence judgment: if the following formula is satisfied, namely the infinite norm of the correction is smaller than the target convergence precision epsilon of the residual vector, the algorithm converges and ends; otherwise, adding 1 to the current iteration step number k, and then returning to the step (3).

|Δx^k|_∞≤ε

In the above technical solution, further, the jacobian matrix calculation in step (3) includes the following three substeps:

(301) and (3) parallel threads are started: the graphics processor starts a thread grid, wherein the number of thread blocks is set as the number of rows of the Jacobian matrix, and the number of threads in each thread block is set as the number of columns of the Jacobian matrix.

(302) And (3) calculating elements of a Jacobian matrix: for the jth thread in the ith thread block, the elements of the ith row and the jth column of the Jacobian matrix are calculated. The computational expressions for the elements of the jacobi matrix are shown in the appendix (for e.electric power system state estimation [ M ]. beijing: hydropower press, 1985.).

(303) Thread synchronization: waiting for all threads to complete the jacobian element calculation.

The invention has the beneficial effects that:

according to the technical scheme, the Jacobian matrix generation in the state estimation of the power system is accelerated in parallel by using the graphic accelerator, so that the calculation efficiency can be obviously improved when the system scale is large, and the real-time requirement of the state estimation is met. Meanwhile, the method is high in operability and easy to implement.

Drawings

FIG. 1 is a schematic flow diagram of the process of the present invention.

FIG. 2 is a schematic diagram of the mode of operation of the graphics accelerator in the method of the present invention.

FIG. 3 is a diagram illustrating the correspondence of a Jacobian matrix to a graphics accelerator thread block in the method of the present invention.

Detailed Description

The technical scheme of the invention is further explained in detail by combining the drawings in the specification.

As shown in FIG. 1, the method mainly comprises six steps of state estimation information import, state estimation initial value setting, Jacobian matrix calculation, correction equation set solution, state variable correction and convergence judgment.

Step S1: importing state estimation information: and importing measurement information, line parameter information and a weight matrix R, and determining variables needing state estimation.

Step 2: setting initial values of state estimation: setting an initial value x of a variable to be state-estimated⁰And residual vector target convergence accuracy epsilon. And setting the current iteration step number k of the algorithm to be equal to 0.

And step 3: and (3) calculating a Jacobian matrix: jacobian matrix H (x) for parallel computing current iteration step based on graphics accelerator_k)。

And 4, step 4: solving a correction equation set: the correction amount Deltax is obtained by solving a linear equation set expressed in the following formula^k。

Δx^k＝[H^T(x^k)R^-1H(x^k)]^-1H^T(x^k)R^-1[z-h(x^k)]

And 5: and (3) state variable correction: calculated by the following formula

x_k+1＝Δx+x_k

Step S6: and (3) convergence judgment: if the following formula is satisfied (namely the infinite norm of the correction is smaller than the target convergence precision epsilon of the residual vector), the algorithm converges, and the operation is finished; otherwise, the current iteration step number k is added by 1, and then the process returns to step S3.

|Δx^k|_∞≤ε

As shown in fig. 1, step S3 includes three substeps:

step 301: and (3) parallel threads are started: as shown in FIG. 2, the graphics processor starts a thread grid, where the number of thread blocks is set as the number of rows of the Jacobian matrix and the number of threads in each thread block is set as the number of columns of the Jacobian matrix.

Step 302: and (3) calculating elements of a Jacobian matrix: for the jth thread in the ith thread block, the elements of the ith row and the jth column of the Jacobian matrix are calculated. The computational expressions for the elements of the jacobian matrix are shown in the appendix of the reference (for er. electric power system state estimation [ M ]. beijing: hydropower press, 1985.).

Step 303: thread synchronization: waiting for all threads to complete the jacobian element calculation.

Claims

1. A power system state estimation method based on graphics processor parallel acceleration is characterized by comprising the following steps:

1) obtaining state estimation information, comprising: measuring information, namely node voltage, node injection power, line power flow and transformer power flow, line parameter information and a measurement error variance matrix R, forming a measurement vector z and a node admittance matrix Y, and determining a variable x and a measurement function h (x) which need to be subjected to state estimation;

2) setting initial values of state estimation: setting an initial value x of a variable to be state-estimated⁰The convergence precision epsilon of the residual vector target; setting the current iteration step number k to be equal to 0;

3) and (3) calculating a Jacobian matrix: jacobian matrix H (x) for parallel computing current iteration step based on graphics accelerator^k) (ii) a The method comprises the following steps:

① starting parallel threads, namely starting a thread grid by the graphics processor, wherein the number of thread blocks is set as the row number of a Jacobian matrix, and the number of threads in each thread block is set as the column number of the Jacobian matrix;

② calculation of Jacobian matrix elements, i, j, column elements of Jacobian matrix are calculated for j, th thread in i, th thread block;

③ thread synchronization, waiting for all threads to complete Jacobian matrix element calculation;

4) solving a correction equation set: the correction amount Deltax is obtained by solving a linear equation set expressed in the following formula^k；

Δx^k＝[H^T(x^k)R^-1H(x^k)]^-1H^T(x^k)R^-1[z-h(x^k)]

5) And (3) state variable correction: calculated by the following formula

x^k+1＝Δx+x^k

And (3) convergence judgment: if the following formula is satisfied, namely the infinite norm of the correction is smaller than the target convergence precision epsilon of the residual vector, the algorithm converges and ends; otherwise, the current iteration step number k is added with 1, and then the step (3) is returned to

|Δx^k|_∞≤ε。