CN104361159A - Time-space parallel simulation method for transient stability of large-scale power system - Google Patents

Abstract

The invention provides a time-space parallel simulation method for the transient stability of a large-scale power system. The method comprises the following steps: establishing a model for an alternating current-direct current system; performing space decoupling on the alternating current-direct current system, and performing time decoupling on a direct current system; performing parallel simulation calculation on the alternating current-direct current system. The method disclosed by the invention uses a CPU/GPU (central processing unit/graphic processing unit) heterogeneous platform to realize the transient simulation of a large-scale alternating current-direct current interconnected power system, according to the independence of the alternating-current system and the direct-current system on the space and the characteristic that the alternating-current system and the direct-current system are easy to decouple in an electrical respect, a calculation task corresponding to the alternating-current system is assigned to a CPU, a calculation task corresponding to the direct-current system is assigned to a GPU, and parallelism on the space is realized; a traveling time window technique is used, the waste of GPU calculation resources is avoided, and the utilization efficiency of the GPU is improved.

Description

A kind of large-scale electrical power system transient stability time and space parallel simulation method

Technical field

The present invention relates to a kind of emulation mode, be specifically related to a kind of large-scale electrical power system transient stability time and space parallel simulation method.

Background technology

The task of multilayer output feedback network judges that electric system suffers comparatively after large disturbances by numerical computation method, and whether each generator can continue to keep synchronous operation.The electric system large disturbances that usual needs are considered comprises:

1, the unexpected change of load, the input of such as large-capacity user and cutting out;

2, to cut out or the main element of input coefficient, as generator, transformer or circuit;

3, be short-circuited fault.

Along with the development of power industry, interconnected more tight between electrical network, China's electrical network is forming extensive alternating current-direct current interconnecting electric power network.After the popularization of interconnected network, higher requirement be it is also proposed to the transient stability simulation velocity of ac and dc systems.Therefore the Study on Parallel Algorithm of multilayer output feedback network has practical significance.At present under parallel computation frame, three aspects are mainly paid close attention in simulation study:

1, straight-flow system Modeling Calculation method;

2, the interface between straight-flow system and AC system;

3, parallel computation hardware and software platform.

According to the level of detail to DC line and DC control simulation, modeling method can be divided into response model, detailed model and electrical-magnetic model.

The computing platform of transient stability has CPU, GPU, FPGA etc.Wherein, CPU is as traditional general-purpose computations processor, and monokaryon ability is strong, be applicable to various types of calculation task, but the core amounts comprised is usually few; GPU (GraphicProcessing Unit) is that a class is specifically designed to process graphical data, as a kind of more late device entering general-purpose computations field, has processor check figure numerous compared with CPU, and thread distributes and destroys feature rapidly.But its logic processing capability is relatively weak, be more applicable to Data-intensive computing.In recent years, GPU is also more and more general as a kind of way of the stream handle towards general-purpose computations.GPU may be used for multiple parallel computation task, such as Molecular Dynamics Calculation.They are very applicable to the very large calculating of data input and output amount.A large amount of data make GPU can utilize the vector calculation unit of GPU or the structure of single instruction multiple data fully.Calculating based on GPU has played increasing effect in large-scale calculating, in ten the strongest in the world supercomputers, has three advantages that all make use of GPU.The heterogeneous platform be made up of CPU and GPU, can merge the advantage of two kinds of processors, has higher computing power, but also requires that developer proposes the parallel algorithm of applicable this new platform.

In large-scale alternating current-direct current network system, the independence of straight-flow system itself is comparatively strong, easily divides out separately from whole system.Therefore in conjunction with ac and dc systems simulated properties, the AC portion in network system and direct current component can be divided and come, by its distribution of computation tasks on CPU and GPU of heterogeneous platform.Use CPU to calculate AC system part, use GPU to calculate straight-flow system part.Meanwhile, interacting between multiple straight-flow system is also relatively little, has higher natural concurrency.After considering the feature of direct current detailed model, by the time-domain-simulation task of straight-flow system according to time step divide, form the form of streamline, the powerful calculating ability of GPU can be made full use of, effectively improve the simulation calculation speed of straight-flow system.

Streamline refers to a kind of data processing method formed according to following rule, and the output of an element equals the input of another one.Article one, the element on streamline is often executed in parallel, according to the time mark off area dividing place.Many streamlines fully can expose the concurrency of tasks carrying, improve the utilization ratio of processor computational resource.

Transient stability problem refers to electric system and suffers comparatively after large disturbances, and whether each generator can continue the problem keeping synchronous operation.After multilayer output feedback network is and utilizes numerical computation method research to suffer disturbance, the change of system indices.

Parallel computation is a kind of calculating simultaneously can carrying out the multi-task.Its basic concept is that large problem usually can be divided into little problem, and these little problems can solve usually simultaneously.Parallel computation has various ways: bit-parallel, instruction level parallelism, data and tasks in parallel.Parallel computation has had applicating history for many years at high-performance computing sector.At civil area, due to the restriction of single core processor performance, the importance of parallel computation is also more and more paid attention to.

Summary of the invention

In order to overcome above-mentioned the deficiencies in the prior art, the invention provides a kind of large-scale electrical power system transient stability time and space parallel simulation method, independent based on large-scale electrical power system ac and dc systems space distribution, electrical relation is easy to the feature of decoupling zero, carried out on CPU by AC system partial simulation calculation task, it is parallel that straight-flow system partial simulation calculation task uses stream mode to realize on GPU; Utilize data transmission between heterogeneous platform CPU and GPU can the feature of asynchronous execution, open the calculating data transmission between multiple current control DC-AC system.

In order to realize foregoing invention object, the present invention takes following technical scheme:

The invention provides a kind of large-scale electrical power system transient stability time and space parallel simulation method, said method comprising the steps of:

Step 1: set up ac and dc systems model;

Step 2: carry out spatially decoupled to ac and dc systems, and Time Decoupling is carried out to straight-flow system;

Step 3: ac and dc systems parallel artificial calculates.

In described step 1, ac and dc systems model comprises AC system dynamic model, straight-flow system model and alterating and direct current pessimistic concurrency control;

Described AC system dynamic model comprises generator model, excitation system model and primemover system model; Described straight-flow system model comprises direct current cables model, DC control system model and transverter model.

Described direct current cables model adopts T-shaped model, and the dynamic equation that direct current cables model is corresponding is as follows:

${\stackrel{\·}{I}}_{\mathrm{dcr}}=(V_{\mathrm{dcr}}-V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dcr}})/(L_{\mathrm{sr}}+L_{\mathrm{line}}/2)$

${\stackrel{\·}{I}}_{\mathrm{dcr}}=({-V}_{\mathrm{dci}}+V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dci}})/(L_{\mathrm{si}}+L_{\mathrm{line}}/2)$

${\stackrel{\·}{V}}_c=I_c/C_{\mathrm{cap}}=(I_{\mathrm{dcr}}-I_{\mathrm{dci}})/C_{\mathrm{cap}}$

Wherein, V _dcrfor the DC voltage of rectification side, V _dcifor the DC voltage of inverter side; I _dcrfor rectification side obtains DC current, I _dcifor the DC current of inverter side; V _cand I _cfor capacitance voltage and electric current; R _dc, L _lineand C _capbe respectively direct current cables resistance, inductance and electric capacity; L _srfor rectification side smoothing reactor inductance, L _sifor inverter side smoothing reactor inductance.

Described DC control system model is divided into main control level and pole controlled stage from top to bottom;

Described main control level comprises urgent power controller, changes DC current or value and power reference when the connected AC system voltage of straight-flow system or frequency generation larger fluctuation;

Described pole controlled stage is divided into following three links:

1) current limiting low-voltage controller: it is input as the current reference value of main control level setting, when DC voltage is too low, limits the current reference value of this setting, and current reference value is inputed to current gain controller;

2) current gain controller: the current reference value that current limiting low-voltage controller inputs is converted to Trigger Angle reference value by it, exports to gamma kick device;

3) gamma kick device: Trigger Angle reference value is converted to actual trigger pulse.

Described transverter model is transverter quasi steady state model, is expressed as:

$V_{\mathrm{dcr}}=N[\frac{3\sqrt{2}{T}_r}{\mathrm{\π}{T}_{\mathrm{tap}}}{V}_{\mathrm{adcr}}\cos \mathrm{\α}-\frac{3X_c{I}_{\mathrm{dcr}}}{\mathrm{\π}}-2R_c{I}_{\mathrm{dcr}}]$

$\cos (\mathrm{\δ}+\mathrm{\α})=\cos \mathrm{\α}-\frac{\sqrt{2}{T}_{\mathrm{tap}}{I}_{\mathrm{dcr}}{X}_c}{T_r{V}_{\mathrm{adcr}}}$

$P_{\mathrm{acr}}=N\frac{\sqrt{18}{T}_r}{2\mathrm{\π}{T}_{\mathrm{tap}}}{I}_{\mathrm{dcr}}{V}_{\mathrm{adcr}}[\cos \mathrm{\α}+\cos (\mathrm{\α}+\mathrm{\δ})]$

$I_{\mathrm{adcr}}=\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}+Q_{\mathrm{dcr}}{V}_{\mathrm{adcry}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}+j\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcry}}-Q_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}$

Wherein, V _dcrand I _dcrbe respectively DC voltage and the DC current of rectification side, N is series rectifier bridge number, T _tapfor rectification side load tap changer position, I _adcrfor rectification side injects AC system electric current, V _adcrfor rectification side transformer alternating side voltage, V _adcrxand V _adcrybe respectively rectification side transformer alternating side voltage real part and imaginary part, X _cand R _cbe respectively rectification side transformer reactance and resistance, α is rectification side Trigger Angle, and δ is commutation overlap angle, T _rfor the basic no-load voltage ratio of rectification side transformer, P _acrand Q _acrbe respectively active power and the reactive power of rectification side absorption, for power-factor angle.

In described step 2, spatially decoupled detailed process carried out to the ac and dc systems set up as follows:

AC system dynamic model expression is:

${\stackrel{\·}{x}}_{\mathrm{ac}}=f(x_{\mathrm{ac}},V_{\mathrm{ac}})$

Wherein, x _acand V _acbe respectively AC system state variable and exchange node voltage vector with pure;

Straight-flow system model representation is:

${\stackrel{\·}{x}}_{\mathrm{dc}}=H(x_{\mathrm{dc}},V_{\mathrm{adc}})$

Wherein, x _dcand V _adcbe respectively the node voltage vector at straight-flow system state variable and straight-flow system incoming transport system place;

Alterating and direct current pessimistic concurrency control is expressed as:

0＝I-Y _NV

Wherein, I is ac and dc systems node Injection Current, and it comprises V is ac and dc systems node voltage, Y _nfor network admittance matrix;

Due to AC system state variable and straight-flow system state variable separate, and be associated by network equation, decoupling zero is carried out according to AC system and straight-flow system position relationship spatially, straight-flow system adopts Norton equivalent access ac and dc systems electricity grid network, dispose AC system part at CPU and calculate data, i.e. variable initial value and parameter value in AC system dynamic equation and network equation, GPU disposes straight-flow system part and calculates data, i.e. variable initial value and parameter value in straight-flow system dynamic equation.

In described step 2, carry out Time Decoupling to straight-flow system, detailed process is as follows:

Subscript is omitted to straight-flow system dynamic equation, the direct current differential equation can be obtained, have:

$\stackrel{\·}{x}=H(x,V)$

At interchange integration step h _acin, the method that when the direct current differential equation adopts single, step order solves, calculates h _ac/ h _dcstep, wherein h _dcfor d.c. integration step-length; Use hiding-trapezium integral method, if subscript n represents t, n+1 represents t+h _acin the moment, the differencing equation of the direct current differential equation is:

$x_n=x_{n-1}+\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)],n=1,...,h_{\mathrm{ac}}/h_{\mathrm{dc}}$

Wherein, x _nfor the state variable of straight-flow system t, x _n-1for straight-flow system t-h _acthe state variable in moment, V _nfor the node voltage vector at straight-flow system t incoming transport system place, V _n-1for straight-flow system t-h _acthe node voltage vector at incoming transport system place;

Definition intermediate vector R _n, it is expressed as:

$R_n=x_n-x_{n-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)]$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

$R_n^{k-1}=-\mathrm{J\Δ}{x}_n^k$

Wherein, k is iterations, be straight-flow system state variable difference in twice iteration, J is the Jacobian matrix of kth time iteration in Newton-Raphson approach, be expressed as with J:

$\mathrm{\Δ}{x}_n^k=x_n^k-x_n^{k-1}$

$J=\frac{\∂R_n}{\∂x_n}{|}_{x_n=x_n^{k-1}}$

Wherein, for the state variable of straight-flow system t in kth time iteration, for the state variable of straight-flow system t in kth-1 iteration;

So obtain wherein V ₀for step gained magnitude of voltage during a upper interchange.

Described step 3 specifically comprises the following steps:

Step 3-1: ac and dc systems carries out initialization, adopt OpenMP multithreading on CPU, open two threads, be respectively thread A and thread B, thread A completes the relevant calculating of AC system on CPU, thread B for control GPU, to GPU allocating task, sending controling instruction; Then ac and dc systems enters time-domain-simulation process, according to the propelling of time, calculates ac and dc systems state variable corresponding to each moment, arranges emulation moment t=0, if advance step-length to be a h of step when exchanging _ac;

Step 3-2: exchange frontier point voltage and current value between AC system and straight-flow system;

Step 3-3: the implicit trapezoidal rule method adopting alternating iteration, the CPU in heterogeneous platform calculates AC system state variable;

Step 3-4: the calculating completing multiple straight-flow system state variable on the GPU of heterogeneous platform, specifically comprises the following steps:

Step 3-4-1: judged whether many DC line, if so, then opens multiple GPU and flows, each stream calculation DC line; If not, then a GPU stream is only used;

Step 3-4-2: each DC line uses the time pipeline parallel method method in conjunction with Moving Window to calculate, and in conjunction with magnitude of voltage, judges the number of streamline;

Step 3-4-3: after calculating convergence, calculates the electric current that straight-flow system injects AC system;

Step 3-5: after the whole DC line Injection Current of acquisition, arranges emulation moment t=t+h _ac, enter subsequent time;

Step 3-6: repeat step 3-1 to step 3-4, until emulation terminates.

In described step 3-4-2, be multiple time window by this emulation Time segments division, and each time window comprise n _wwalk time individual, multiple time window calculates in order, but in each time window, the calculation task walked is assembled into the form of streamline, realizes parallel computation time multiple;

For time window 1, n _wthe difference equation that time individual, step is corresponding is:

$\left[\begin{array}{c}{R}_1\\ R_2\\ \·\\ \·\\ \·\\ R_{n_w}\end{array}\right]=\left[\begin{array}{c}{x}_1-x_0-\frac{h_{\mathrm{dc}}}{2}[H(x_1,V_1)+H(x_0,V_0)]\\ x_2-x_1-\frac{h_{\mathrm{dc}}}{2}[H(x_2,V_2)+H(x_1,V_1)]\\ \·\\ \·\\ \·\\ x_{n_w}-x_{n_w-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n_w},V_{n_w})+H(x_{n_w-1},V_{n_w-1})]\end{array}\right]=\left[\begin{array}{c}0\\ 0\\ \·\\ \·\\ \·\\ 0\end{array}\right]$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

Wherein: $J_{n,n-1}^{k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},J_{n,n}^{k-1}=\frac{\∂R_n}{\∂x_n}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},n=1,...,n_w;$

For the n in time window _wwalk time individual, in the first iteration, when streamline P1 carries out, walk the first time iterative of 1, during acquisition, walk the state variable of 1 in second time iteration, when streamline P2 carries out walk 2 first time iterative with the second time iterative walking 1 in time; The like, realize the Parallel implementation to step time whole in actual time window; Solution formula is:

$\mathrm{\Δ}{x}_n^k=\left\{\begin{array}{cc}-{\left(J_{n,n}^{*k-1}\right)}^{-1}{R}_n^{k-1}& ,n=1\\ -{\left(J_{n,n}^{*k-1}\right)}^{-1}[R_n^{k-1}+J_{n,n-1}^{*k}\mathrm{\Δ}{x}_{n-1}^k]& ,n=2,...,n_w\end{array}\right.$

Wherein, $J_{n,n-1}^{*k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}},J_{n,n}^{*k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}}.$

Compared with prior art, beneficial effect of the present invention is:

1, CPU/GPU heterogeneous platform is adopted to achieve the transient emulation of extensive alternating current-direct current interconnected electric power system.

2, according to the feature of AC system and straight-flow system independence spatially and its easy decoupling zero in electric, by distribution of computation tasks corresponding for AC system on CPU, by distribution of computation tasks corresponding for straight-flow system on GPU, walking abreast on implementation space.

3, for the mutual independently feature of multiple straight-flow system, the calculation task of its correspondence is realized on GPU temporal parallel.

4, for the transient emulation of straight-flow system, its calculation task is divided into time window, is assembled into the form of many streamlines, utilize in multi-core parallel concurrent process time window of GPU each time step calculation task.

5, adopt the technology of moving time-window, avoid the waste of GPU computational resource, improve the utilization ratio of GPU.

Accompanying drawing explanation

Fig. 1 is large-scale electrical power system transient stability time and space parallel simulation method process flow diagram in the embodiment of the present invention;

Fig. 2 is direct current cables model schematic in the embodiment of the present invention;

Fig. 3 is DC control system model schematic in the embodiment of the present invention;

Fig. 4 is transverter model schematic in the embodiment of the present invention;

Fig. 5 is time window sequence diagram in the embodiment of the present invention;

Fig. 6 is streamline and Moving Window schematic diagram in the embodiment of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described in further detail.

As Fig. 1, the invention provides a kind of large-scale electrical power system transient stability time and space parallel simulation method, said method comprising the steps of:

Step 1: set up ac and dc systems model;

Step 2: carry out spatially decoupled to ac and dc systems, and Time Decoupling is carried out to straight-flow system;

Step 3: ac and dc systems parallel artificial calculates.

In described step 1, ac and dc systems model comprises AC system dynamic model, straight-flow system model and alterating and direct current pessimistic concurrency control;

Described AC system dynamic model comprises generator model, excitation system model and primemover system model; Described straight-flow system model comprises direct current cables model, DC control system model and transverter model.

As Fig. 2, described direct current cables model adopts T-shaped model, and the dynamic equation that direct current cables model is corresponding is as follows:

${\stackrel{\·}{I}}_{\mathrm{dcr}}=(V_{\mathrm{dcr}}-V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dcr}})/(L_{\mathrm{sr}}+L_{\mathrm{line}}/2)---\left(1\right)$

${\stackrel{\·}{I}}_{\mathrm{dcr}}=({-V}_{\mathrm{dci}}+V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dci}})/(L_{\mathrm{si}}+L_{\mathrm{line}}/2)---\left(2\right)$

${\stackrel{\·}{V}}_c=I_c/C_{\mathrm{cap}}=(I_{\mathrm{dcr}}-I_{\mathrm{dci}})/C_{\mathrm{cap}}---\left(3\right)$

Wherein, V _dcrfor the DC voltage of rectification side, V _dcifor the DC voltage of inverter side; I _dcrfor rectification side obtains DC current, I _dcifor the DC current of inverter side; V _cand I _cfor capacitance voltage and electric current; R _dc, L _lineand C _capbe respectively direct current cables resistance, inductance and electric capacity; L _srfor rectification side smoothing reactor inductance, L _sifor inverter side smoothing reactor inductance.

As Fig. 3, described DC control system model is divided into main control level, pole controlled stage and valve controlled stage from top to bottom; In detailed model, transverter adopts quasi steady state model, therefore have ignored the effect of valve controlled stage;

Described main control level comprises urgent power controller, changes DC current or value and power reference when the connected AC system voltage of straight-flow system or frequency generation larger fluctuation;

Described pole controlled stage is divided into following three links:

1) current limiting low-voltage controller: it is input as the current reference value of main control level setting, when DC voltage is too low, limits the current reference value of this setting, and current reference value is inputed to current gain controller;

2) current gain controller: the current reference value that current limiting low-voltage controller inputs is converted to Trigger Angle reference value by it, exports to gamma kick device;

3) gamma kick device: Trigger Angle reference value is converted to actual trigger pulse.

As Fig. 4, described transverter model is transverter quasi steady state model, is expressed as:

$V_{\mathrm{dcr}}=N[\frac{3\sqrt{2}{T}_r}{\mathrm{\π}{T}_{\mathrm{tap}}}{V}_{\mathrm{adcr}}\cos \mathrm{\α}-\frac{3X_c{I}_{\mathrm{dcr}}}{\mathrm{\π}}-2R_c{I}_{\mathrm{dcr}}]---\left(4\right)$

$\cos (\mathrm{\δ}+\mathrm{\α})=\cos \mathrm{\α}-\frac{\sqrt{2}{T}_{\mathrm{tap}}{I}_{\mathrm{dcr}}{X}_c}{T_r{V}_{\mathrm{adcr}}}---\left(5\right)$

$P_{\mathrm{acr}}=N\frac{\sqrt{18}{T}_r}{2\mathrm{\π}{T}_{\mathrm{tap}}}{I}_{\mathrm{dcr}}{V}_{\mathrm{adcr}}[\cos \mathrm{\α}+\cos (\mathrm{\α}+\mathrm{\δ})]---\left(6\right)$

$I_{\mathrm{adcr}}=\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}+Q_{\mathrm{dcr}}{V}_{\mathrm{adcry}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}+j\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcry}}-Q_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}---\left(8\right)$

Wherein, V _dcrand I _dcrbe respectively DC voltage and the DC current of rectification side, N is series rectifier bridge number, T _tapfor rectification side load tap changer position, I _adcrfor rectification side injects AC system electric current, V _adcrfor rectification side transformer alternating side voltage, V _adcrxand V _adcrybe respectively rectification side transformer alternating side voltage real part and imaginary part, X _cand R _cbe respectively rectification side transformer reactance and resistance, α is rectification side Trigger Angle, and δ is commutation overlap angle, T _rfor the basic no-load voltage ratio of rectification side transformer, P _acrand Q _acrbe respectively active power and the reactive power of rectification side absorption, for power-factor angle.

In described step 2, spatially decoupled detailed process carried out to the ac and dc systems set up as follows:

Ac and dc systems dynamic model expression is:

${\stackrel{\·}{x}}_{\mathrm{ac}}=f(x_{\mathrm{ac}},V_{\mathrm{ac}})---\left(9\right)$

Wherein, x _acand V _acbe respectively AC system state variable and exchange node voltage vector with pure;

Straight-flow system model representation is:

${\stackrel{\·}{x}}_{\mathrm{dc}}=H(x_{\mathrm{dc}},V_{\mathrm{adc}})---\left(10\right)$

Wherein, x _dcand V _adcbe respectively the node voltage vector at straight-flow system state variable and straight-flow system incoming transport system place;

Alterating and direct current pessimistic concurrency control is expressed as:

0＝I-Y _NV (11)

Wherein, I is ac and dc systems node Injection Current, and it comprises V is ac and dc systems node voltage, Y _nfor network admittance matrix;

Due to AC system state variable and straight-flow system state variable separate, and be associated by network equation, decoupling zero is carried out according to AC system and straight-flow system position relationship spatially, straight-flow system adopts Norton equivalent access ac and dc systems electricity grid network, dispose AC system part at CPU and calculate data, i.e. variable initial value and parameter value in AC system dynamic equation and network equation, GPU disposes straight-flow system part and calculates data, i.e. variable initial value and parameter value in straight-flow system dynamic equation.

In described step 2, carry out Time Decoupling to straight-flow system, detailed process is as follows:

Subscript is omitted to straight-flow system dynamic equation, the direct current differential equation can be obtained, have:

$\stackrel{\·}{x}=H(x,V)---\left(12\right)$

At interchange integration step h _acin, the method that when the direct current differential equation adopts single, step order solves, calculates h _ac/ h _dcstep, wherein h _dcfor d.c. integration step-length; Use hiding-trapezium integral method, if subscript n represents t, n+1 represents t+h _acin the moment, the differencing equation of the direct current differential equation is:

$x_n=x_{n-1}+\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)],n=1,...,h_{\mathrm{ac}}/h_{\mathrm{dc}}---\left(13\right)$

Wherein, x _nfor the state variable of straight-flow system t, x _n-1for straight-flow system t-h _acthe state variable in moment, V _nfor the node voltage vector at straight-flow system t incoming transport system place, V _n-1for straight-flow system t-h _acthe node voltage vector at incoming transport system place;

Definition intermediate vector R _n, it is expressed as:

$R_n=x_n-x_{n-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)]---\left(14\right)$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

$R_n^{k-1}=-\mathrm{J\Δ}{x}_n^k---\left(15\right)$

Wherein, k is iterations, be straight-flow system state variable difference in twice iteration, J is the Jacobian matrix of kth time iteration in Newton-Raphson approach, be expressed as with J:

$\mathrm{\Δ}{x}_n^k=x_n^k-x_n^{k-1}---\left(16\right)$

$J=\frac{\∂R_n}{\∂x_n}{|}_{x_n=x_n^{k-1}}---\left(17\right)$

Wherein, for the state variable of straight-flow system t in kth time iteration, for the state variable of straight-flow system t in kth-1 iteration;

So obtain wherein V ₀for step gained magnitude of voltage during a upper interchange.

Described step 3 specifically comprises the following steps:

Step 3-1: ac and dc systems carries out initialization, adopt OpenMP multithreading on CPU, open two threads, be respectively thread A and thread B, thread A completes the relevant calculating of AC system on CPU, thread B for control GPU, to GPU allocating task, sending controling instruction; Then ac and dc systems enters time-domain-simulation process, according to the propelling of time, calculates ac and dc systems state variable corresponding to each moment, arranges emulation moment t=0, if advance step-length to be a h of step when exchanging _ac;

Step 3-2: exchange frontier point voltage and current value between AC system and straight-flow system;

Step 3-3: the implicit trapezoidal rule method adopting alternating iteration, the CPU in heterogeneous platform calculates AC system state variable;

Step 3-4: the calculating completing multiple straight-flow system state variable on the GPU of heterogeneous platform, specifically comprises the following steps:

Step 3-4-1: judged whether many DC line, if so, then opens multiple GPU and flows, each stream calculation DC line; If not, then a GPU stream is only used;

Step 3-4-2: each DC line uses the time pipeline parallel method method in conjunction with Moving Window to calculate, and in conjunction with magnitude of voltage, judges the number of streamline;

Step 3-4-3: after calculating convergence, calculates the electric current that straight-flow system injects AC system;

Step 3-5: after the whole DC line Injection Current of acquisition, arranges emulation moment t=t+h _ac, enter subsequent time;

Step 3-6: repeat step 3-1 to step 3-4, until emulation terminates.

In described step 3-4-2, be multiple time window by this emulation Time segments division, and each time window comprise n _wwalk time individual, multiple time window calculates in order, but in each time window, the calculation task walked is assembled into the form of streamline, realizes parallel computation time multiple;

As Fig. 5, for time window 1, n _wthe difference equation that time individual, step is corresponding is:

$\left[\begin{array}{c}{R}_1\\ R_2\\ \·\\ \·\\ \·\\ R_{n_w}\end{array}\right]=\left[\begin{array}{c}{x}_1-x_0-\frac{h_{\mathrm{dc}}}{2}[H(x_1,V_1)+H(x_0,V_0)]\\ x_2-x_1-\frac{h_{\mathrm{dc}}}{2}[H(x_2,V_2)+H(x_1,V_1)]\\ \·\\ \·\\ \·\\ x_{n_w}-x_{n_w-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n_w},V_{n_w})+H(x_{n_w-1},V_{n_w-1})]\end{array}\right]=\left[\begin{array}{c}0\\ 0\\ \·\\ \·\\ \·\\ 0\end{array}\right]---\left(18\right)$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

Wherein: $J_{n,n-1}^{k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},J_{n,n}^{k-1}=\frac{\∂R_n}{\∂x_n}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},n=1,...,n_w;$

Equation coefficient matrix in formula (19) is large-scale diagonal matrix, if again solution vector is substituted into next iteration after it being decomposed completely, performs iterative process successively, can not accelerate whole simulation process.In order to utilize the parallel ability of GPU, introduce pipelining below.

Because the calculating that successive ignition process is corresponding has front and back dependence, also there is dependence between step during each in same time window, namely calculated just can calculate afterwards but calculating can carry out, so such computation process also has larger parallel potentiality simultaneously.According to such thinking, can by whole process composition streamline form.

For the n in time window _wwalk time individual, in the first iteration, when streamline P1 carries out, walk the first time iterative of 1, during acquisition, walk the state variable of 1 in second time iteration, when streamline P2 carries out walk 2 first time iterative with the second time iterative walking 1 in time; The like, realize the Parallel implementation to step time whole in actual time window; Solution formula is:

$\mathrm{\Δ}{x}_n^k=\left\{\begin{array}{cc}-{\left(J_{n,n}^{*k-1}\right)}^{-1}{R}_n^{k-1}& ,n=1\\ -{\left(J_{n,n}^{*k-1}\right)}^{-1}[R_n^{k-1}+J_{n,n-1}^{*k}\mathrm{\Δ}{x}_{n-1}^k]& ,n=2,...,n_w\end{array}\right.---\left(20\right)$

Wherein, $J_{n,n-1}^{*k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}},J_{n,n}^{*k-1}=\frac{\∂R_n}{\∂x_{n-1}}{|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}}.$

Many streamline is realized by the block in CUDA, the corresponding block of each streamline; N-th time iterative process is as follows:

1) the straight-flow system state variable of (n-1)th iteration acquisition is got as the initial value of current iteration, wherein m=1,2 ..., n _w;

2) n is used _windividual block simulates n _wbar streamline, generates n simultaneously _wthe Jacobian matrix that time individual, step is corresponding and right-hand vector, complete system of linear equations to be solved;

3) system of linear equations formed in solution procedure 2, calculates n _wthe state variable of the straight-flow system of step when-1;

4) check whether convergence, if do not restrained, enter iteration next time.

In addition, in order to improve the efficiency of this method further, invention also uses the method for moving time-window to the state avoiding processor to be in wait.If time window is fixing, in first time window, comprise n _wwalk time individual, so n-th _wafter step calculating completes time individual, just can carry out the calculating of second time window.Such task assembly method also has the leeway optimized further undoubtedly, and the use of moving time-window is just in order to solve such problem.Moving time-window refer to once step 1 state variable iterative computation complete, reach convergence, immediately by belong to originally in future time window time step n _w+ 1 supplements into present streamline.

The increase walked due to time in the acute variation of system dynamic variable and a time window all can make the global convergence speed of time parallel algorithm decline.So whole simulation process, according to DC voltage value, is divided into failure phase and Restoration stage by this patent.If AC line voltage value is lower than 80% of steady-state value in transient state process, be defined as failure phase, algorithm opens n _warticle 6, streamline.If AC line voltage exceedes 80% of steady-state value, be defined as Restoration stage, open 12 streamlines.

If there are many DC line, every bar DC line uses a stream of GPU to realize, and Fig. 6 shows and is streamline and Moving Window schematic diagram.

Finally should be noted that: above embodiment is only in order to illustrate that technical scheme of the present invention is not intended to limit; those of ordinary skill in the field still can modify to the specific embodiment of the present invention with reference to above-described embodiment or equivalent replacement; these do not depart from any amendment of spirit and scope of the invention or equivalent replacement, are all applying within the claims of the present invention awaited the reply.

Claims (9)

1. a large-scale electrical power system transient stability time and space parallel simulation method, is characterized in that: said method comprising the steps of:

Step 1: set up ac and dc systems model;

Step 2: carry out spatially decoupled to ac and dc systems, and Time Decoupling is carried out to straight-flow system;

Step 3: ac and dc systems parallel artificial calculates.

2. large-scale electrical power system transient stability time and space parallel simulation method according to claim 1, it is characterized in that: in described step 1, ac and dc systems model comprises AC system dynamic model, straight-flow system model and alterating and direct current pessimistic concurrency control;

Described AC system dynamic model comprises generator model, excitation system model and primemover system model; Described straight-flow system model comprises direct current cables model, DC control system model and transverter model.

3. large-scale electrical power system transient stability time and space parallel simulation method according to claim 2, is characterized in that: described direct current cables model adopts T-shaped model, and the dynamic equation that direct current cables model is corresponding is as follows:

${\stackrel{.}{I}}_{\mathrm{dcr}}=(V_{\mathrm{dcr}}-V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dcr}})/(L_{\mathrm{sr}}+L_{\mathrm{line}}/2)$

${\stackrel{.}{I}}_{\mathrm{dcr}}=(-V_{\mathrm{dci}}+V_c-0.5R_{\mathrm{dc}}{I}_{\mathrm{dci}})/(L_{\mathrm{si}}+L_{\mathrm{line}}/2)$

${\stackrel{.}{V}}_c=I_c/C_{\mathrm{cap}}=(I_{\mathrm{dcr}}-I_{\mathrm{dci}})/C_{\mathrm{cap}}$

Wherein, V _dcrfor the DC voltage of rectification side, V _dcifor the DC voltage of inverter side; I _dcrfor rectification side obtains DC current, I _dcifor the DC current of inverter side; V _cand I _cfor capacitance voltage and electric current; R _dc, L _lineand C _capbe respectively direct current cables resistance, inductance and electric capacity; L _srfor rectification side smoothing reactor inductance, L _sifor inverter side smoothing reactor inductance.

4. large-scale electrical power system transient stability time and space parallel simulation method according to claim 2, is characterized in that: described DC control system model is divided into main control level and pole controlled stage from top to bottom;

Described main control level comprises urgent power controller, changes DC current or value and power reference when the connected AC system voltage of straight-flow system or frequency generation larger fluctuation;

Described pole controlled stage is divided into following three links:

1) current limiting low-voltage controller: it is input as the current reference value of main control level setting, when DC voltage is too low, limits the current reference value of this setting, and current reference value is inputed to current gain controller;

2) current gain controller: the current reference value that current limiting low-voltage controller inputs is converted to Trigger Angle reference value by it, exports to gamma kick device;

3) gamma kick device: Trigger Angle reference value is converted to actual trigger pulse.

5. large-scale electrical power system transient stability time and space parallel simulation method according to claim 2, is characterized in that: described transverter model is transverter quasi steady state model, is expressed as:

$V_{\mathrm{dcr}}=N[\frac{3\sqrt{2}{T}_r}{\mathrm{\π}{T}_{\mathrm{tap}}}{V}_{\mathrm{adcr}}\cos \mathrm{\α}-\frac{3X_c{I}_{\mathrm{dcr}}}{\mathrm{\π}}-2R_c{I}_{\mathrm{dcr}}]$

$\cos (\mathrm{\δ}+\mathrm{\α})=\cos \mathrm{\α}-\frac{\sqrt{2}{T}_{\mathrm{tap}}{I}_{\mathrm{dcr}}{X}_c}{T_r{V}_{\mathrm{adcr}}}$

$P_{\mathrm{acr}}=N\frac{\sqrt{18}{T}_r}{2\mathrm{\π}{T}_{\mathrm{tap}}}{I}_{\mathrm{dcr}}{V}_{\mathrm{adcr}}[\cos \mathrm{\α}+\cos (\mathrm{\α}+\mathrm{\δ})]$

$I_{\mathrm{adcr}}=\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}+Q_{\mathrm{dcr}}{V}_{\mathrm{adcry}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}+j\frac{P_{\mathrm{dcr}}{V}_{\mathrm{adcry}}-Q_{\mathrm{dcr}}{V}_{\mathrm{adcrx}}}{V_{\mathrm{adcrx}}^2+V_{\mathrm{adcry}}^2}$

Wherein, V _dcrand I _dcrbe respectively DC voltage and the DC current of rectification side, N is series rectifier bridge number, T _tapfor rectification side load tap changer position, I _adcrfor rectification side injects AC system electric current, V _adcrfor rectification side transformer alternating side voltage, V _adcrxand V _adcrybe respectively rectification side transformer alternating side voltage real part and imaginary part, X _cand R _cbe respectively rectification side transformer reactance and resistance, α is rectification side Trigger Angle, and δ is commutation overlap angle, T _rfor the basic no-load voltage ratio of rectification side transformer, P _acrand Q _acrbe respectively active power and the reactive power of rectification side absorption, for power-factor angle.

6. large-scale electrical power system transient stability time and space parallel simulation method according to claim 1, is characterized in that: in described step 2, carries out spatially decoupled detailed process as follows to the ac and dc systems set up:

AC system dynamic model expression is:

${\stackrel{.}{x}}_{\mathrm{ac}}=f(x_{\mathrm{ac}},V_{\mathrm{ac}})$

Wherein, x _acand V _acbe respectively AC system state variable and exchange node voltage vector with pure;

Straight-flow system model representation is:

${\stackrel{.}{x}}_{\mathrm{dc}}=H(x_{\mathrm{dc}},V_{\mathrm{adc}})$

Wherein, x _dcand V _adcbe respectively the node voltage vector at straight-flow system state variable and straight-flow system incoming transport system place;

Alterating and direct current pessimistic concurrency control is expressed as:

0＝I-Y _NV

Wherein, I is ac and dc systems node Injection Current, and it comprises V is ac and dc systems node voltage, Y _nfor network admittance matrix;

Due to AC system state variable and straight-flow system state variable separate, and be associated by network equation, decoupling zero is carried out according to AC system and straight-flow system position relationship spatially, straight-flow system adopts Norton equivalent access ac and dc systems electricity grid network, dispose AC system part at CPU and calculate data, i.e. variable initial value and parameter value in AC system dynamic equation and network equation, GPU disposes straight-flow system part and calculates data, i.e. variable initial value and parameter value in straight-flow system dynamic equation.

7. large-scale electrical power system transient stability time and space parallel simulation method according to claim 1, it is characterized in that: in described step 2, carry out Time Decoupling to straight-flow system, detailed process is as follows:

Subscript is omitted to straight-flow system dynamic equation, the direct current differential equation can be obtained, have:

$\stackrel{.}{x}=H(x,V)$

At interchange integration step h _acin, the method that when the direct current differential equation adopts single, step order solves, calculates h _ac/ h _dcstep, wherein h _dcfor d.c. integration step-length; Use hiding-trapezium integral method, if subscript n represents t, n+1 represents t+h _acin the moment, the differencing equation of the direct current differential equation is:

$x_n=x_{n-1}+\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)],n=1,...,h_{\mathrm{ac}}/h_{\mathrm{dc}}$

Wherein, x _nfor the state variable of straight-flow system t, x _n-1for straight-flow system t-h _acthe state variable in moment, V _nfor the node voltage vector at straight-flow system t incoming transport system place, V _n-1for straight-flow system t-h _acthe node voltage vector at incoming transport system place;

Definition intermediate vector R _n, it is expressed as:

$R_n=x_n-x_{n-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n-1},V_{n-1})+H(x_n,V_n)]$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

$R_n^{k-1}=-\mathrm{J\Δ}{x}_n^k$

Wherein, k is iterations, be straight-flow system state variable difference in twice iteration, J is the Jacobian matrix of kth time iteration in Newton-Raphson approach, be expressed as with J:

$\mathrm{\Δ}{x}_n^k=x_n^k-x_n^{k-1}$

$J={\frac{\∂R_n}{\∂x_n}|}_{x_n=x_n^{k-1}}$

Wherein, for the state variable of straight-flow system t in kth time iteration, for the state variable of straight-flow system t in kth-1 iteration;

So obtain wherein V ₀for step gained magnitude of voltage during a upper interchange.

8. large-scale electrical power system transient stability time and space parallel simulation method according to claim 1, is characterized in that: described step 3 specifically comprises the following steps:

Step 3-1: ac and dc systems carries out initialization, adopt OpenMP multithreading on CPU, open two threads, be respectively thread A and thread B, thread A completes the relevant calculating of AC system on CPU, thread B for control GPU, to GPU allocating task, sending controling instruction; Then ac and dc systems enters time-domain-simulation process, according to the propelling of time, calculates ac and dc systems state variable corresponding to each moment, arranges emulation moment t=0, if advance step-length to be a h of step when exchanging _ac;

Step 3-2: exchange frontier point voltage and current value between AC system and straight-flow system;

Step 3-3: the implicit trapezoidal rule method adopting alternating iteration, the CPU in heterogeneous platform calculates AC system state variable;

Step 3-4: the calculating completing multiple straight-flow system state variable on the GPU of heterogeneous platform, specifically comprises the following steps:

Step 3-4-1: judged whether many DC line, if so, then opens multiple GPU and flows, each stream calculation DC line; If not, then a GPU stream is only used;

Step 3-4-2: each DC line uses the time pipeline parallel method method in conjunction with Moving Window to calculate, and in conjunction with magnitude of voltage, judges the number of streamline;

Step 3-4-3: after calculating convergence, calculates the electric current that straight-flow system injects AC system;

Step 3-5: after the whole DC line Injection Current of acquisition, arranges emulation moment t=t+h _ac, enter subsequent time;

Step 3-6: repeat step 3-1 to step 3-4, until emulation terminates.

9. large-scale electrical power system transient stability time and space parallel simulation method according to claim 8, it is characterized in that: in described step 3-4-2, be multiple time window, and each time window comprises n by this emulation Time segments division _wwalk time individual, multiple time window calculates in order, but in each time window, the calculation task walked is assembled into the form of streamline, realizes parallel computation time multiple;

For time window 1, n _wthe difference equation that time individual, step is corresponding is:

$\left[\begin{array}{c}{R}_1\\ R_2\\ .\\ .\\ .\\ R_{n_w}\end{array}\right]=\left[\begin{array}{c}{x}_1-x_0-\frac{h_{\mathrm{dc}}}{2}[H(x_1,V_1)+H(x_0,V_0)]\\ x_2-x_1-\frac{h_{\mathrm{dc}}}{2}[H(x_2,V_2)+H(x_1,V_1)]\\ .\\ .\\ .\\ x_{n_w}-x_{n_w-1}-\frac{h_{\mathrm{dc}}}{2}[H(x_{n_w},V_{n_w})+H(x_{n_w-1},V_{n_w-1})]\end{array}\right]=\left[\begin{array}{c}0\\ 0\\ .\\ .\\ .\\ 0\end{array}\right]$

Adopt Newton-Raphson approach to above formula, kth time iterative formula is:

Wherein: $J_{n,n-1}^{k-1}={\frac{\∂R_n}{\∂x_{n-1}}|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},J_{n,n}^{k-1}={\frac{\∂R_n}{\∂x_n}|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^{k-1}}},n=1,...,n_w;$

For the n in time window _wwalk time individual, in the first iteration, when streamline P1 carries out, walk the first time iterative of 1, during acquisition, walk the state variable of 1 in second time iteration, when streamline P2 carries out walk 2 first time iterative with the second time iterative walking 1 in time; The like, realize the Parallel implementation to step time whole in actual time window; Solution formula is:

$\mathrm{\Δ}{x}_n^k=\left\{\begin{array}{c}-{\left(J_{n,n}^{*k-1}\right)}^{-1}{R}_n^{k-1},n=1\\ -{\left(J_{n,n}^{*k-1}\right)}^{-1}[R_n^{k-1}+J_{n,n-1}^{*k}\mathrm{\Δ}{x}_{n-1}^k],n=2,...,n_w\end{array}\right.$

Wherein, $J_{n,n-1}^{*k-1}={\frac{\∂R_n}{\∂x_{n-1}}|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}},J_{n,n}^{*k-1}={\frac{\∂R_n}{\∂x_{n-1}}|}_{\underset{x_n=x_n^{k-1}}{x_{n-1}=x_{n-1}^k}}.$