CN108711866B - A control system for reactive voltage of new energy power station - Google Patents

Abstract

The invention discloses a control system for reactive voltage of a new energy power station, which is characterized in that: the control system comprises a prediction model, a typical scene set model and an MPC optimization control model; and meanwhile, predicting a bus voltage value and a transformer substation load by using an autoregressive moving average method, generating a discretized scene for the predicted value through Latin hypercube sampling, acquiring a typical scene set by using a K-means clustering algorithm, establishing an optimal control model with an objective function being the minimum value of the internal network loss and the voltage offset in a future time window, and solving by using a second-order cone programming method. By adopting the method, the reactive power output can be reasonably arranged, the voltage stability is maintained, and the regional network loss is reduced.

Description

A control system for reactive voltage of new energy power station

technical field

The invention relates to reactive power optimization in a grid area, in particular to a control system for reactive power voltage of a new energy power station.

Background technique

With the large-scale connection of wind farms and photovoltaic power stations, new problems will be brought to the safe and reliable operation of the power grid. Due to the intermittent and random characteristics of new energy power stations, which are greatly affected by uncertain environmental factors such as wind speed and light, the problem of how to effectively deal with the voltage fluctuation of the power station grid-connected point has become increasingly prominent. Secondly, the goal of reactive power optimization in traditional power grids is local balance. Reactive power transmission is not allowed on the transmission line between the power plant and the grid, and at the same time, the reactive power flow should be reduced as much as possible. Therefore, how to coordinate reactive power compensation between the two is becoming increasingly important. .

Compared with traditional voltage regulation measures such as adjusting main transformer taps and switching capacitors, new energy power stations have deployed static var compensators or static var generators to deal with voltage fluctuations at grid-connected points. These two devices can continuously Smooth output reactive power. In addition, the generator sets of the new energy power station also have certain reactive power adjustment measures. At present, domestic wind turbines mostly use doubly-fed induction wind turbines, which can emit or absorb reactive power within a certain range. The smart inverter in the photovoltaic power station can still be connected to the grid at a power factor of 0.9 under the condition of rated active power output.

Therefore, how to coordinate and control the reactive power compensation measures in the power grid and the power station, ensure the voltage stability of the grid-connected point, and reduce the transmission loss at the same time is the key to the voltage and reactive power control of the new energy power station.

Contents of the invention

The technical problem to be solved by the present invention is to provide a control system for reactive power voltage of a new energy power station, which can ensure the voltage stability of the grid-connected point and reduce transmission loss at the same time.

In order to achieve the above purpose, the technical solution of the present invention is a control system for reactive power and voltage of new energy power plants, characterized in that: the control system includes three parts: prediction model, typical scene set model and MPC optimization control model composition; at the same time, the autoregressive moving average method is used to predict the bus voltage value and substation load, and then the predicted value is generated through Latin hypercube sampling to generate discretized scenes, and the K-means clustering algorithm is used to obtain typical scene sets, so as to establish the objective function It is an optimal control model with the minimum network loss value and voltage offset value in the future time window, and is solved by the second-order cone programming method.

The prediction model of the control system includes two parts: wind power/photovoltaic power prediction and load/bus voltage prediction. Wind power/photovoltaic power prediction error is fitted by beta distribution, and load/bus voltage is predicted by autoregressive moving average method.

In the typical scene set model of the control system, first, the predicted values of new energy output, bus voltage, and substation load in the electric island are generated into discretized scenes through Latin hypercube sampling, and then K-means clustering algorithm is used to obtain typical scenes set.

In the MPC optimal control model of the control system, the optimal control model is established with the objective function being the minimum network loss value and voltage offset value in the future time window, and is solved by a second-order cone programming method.

In the wind power/photovoltaic power prediction, when the wind power prediction value is P _t ^pred , the probability density function of the unit output active value x is:

In the above formula, B(α, β) is the beta function, and α and β are the morphological parameters of the beta distribution. Their values are related to the expected μ and variance σ ² of the beta distribution. The calculation formula is:

Calculate the beta distribution shape parameters α and β, and then obtain the probability density function of the wind power forecast error distribution;

In the optical power prediction, the output power of the photovoltaic array also obeys the beta distribution, and its probability density formula is as follows:

In the formula, Γ is the Gamma function, α and β are the morphological parameters of the beta distribution, and p _max is the maximum output power of the photovoltaic array; the wind power prediction distribution and the optical power prediction distribution are equally divided from the beta distribution, and the wind power prediction model and the optical power Predictive models are unified.

The generation of the typical scene set described above requires a large number of random variables according to a certain probability distribution to be sampled to obtain a scene data set reflecting the characteristics of the random variable, that is, the continuous probability model is discretized, and then the scenes are streamlined and clustered to obtain a typical scene Set, to achieve the characteristics of the original scene with fewer scenes.

The objective function of the control system for the optimal control of the voltage of the new energy power station is:

In the formula, ρ _s is the probability corresponding to the scene w _s ; P _t (w _s ) is the active power loss of the electrical island at time t in the prediction period t _pre ; V _t (w _s ) is the bus voltage t in the prediction period t _pre Deviation value at time; α, β are weight coefficients. P _it (w _s ), Q _it (w _s ), V _it (w _s ) are the active power, reactive power and voltage value of node i at time t in the prediction period t _pre respectively; V _is is the voltage setting value of node i .

The control system incorporates the reactive power that DFIG can output into optimal control, wherein the range of reactive power is:

In the formula, P _s is the active value generated by the stator; U _s is the stator side terminal voltage; X _s is the stator side leakage reactance; X _m is the excitation reactance; I _Imax is the maximum current value of the rotor converter.

The MPC optimization control model includes discrete and continuous variables such as capacitor switching, gear adjustment, new energy output, SVC reactive power compensation, etc., and the second-order cone relaxation programming is used to relax the power flow equation convexly.

A control system for reactive power and voltage of a new energy power station. Due to the adoption of the above method, the present invention can reasonably arrange reactive power output, maintain voltage stability, and reduce regional network loss.

Description of drawings

Below in conjunction with accompanying drawing and specific embodiment the present invention is described in further detail;

FIG. 1 is a grid-connected structure diagram of a new energy power station in a control system for reactive voltage of a new energy power station according to the present invention;

Fig. 2 is a generation diagram of a typical scene set in a control system for reactive power voltage of a new energy power station according to the present invention.

Detailed ways

The invention applies the model predictive control theory to the voltage and reactive power control of the new energy power station, utilizes the ultra-short-term power prediction value of the new energy power station, and simplifies the grid structure model to realize the optimal control of the voltage and reactive power. At present, my country's wind power stations are mainly connected to the grid through 110kV and above busbars. The structure is shown in Figure 1. The wind turbines are connected to the 35kV busbar in the station, and then connected to the grid through the booster station. Medium-sized and above photovoltaic power stations need to be connected to the power grid through 10kV and above busbars, and its structure is shown in Figure 1. It is worth noting that there may be no step-up station in the low-voltage photovoltaic power station, and the collection bus in the station is directly connected to the power grid. This structure can also be included in the research scope of this paper, but only some voltage regulation measures are missing.

In the present invention, the new energy power station and the substation on the power grid access side are taken as a whole area for optimal control. Due to the wide application of the AVC system in the power grid, the bus voltage fluctuation in the system is small, and the high-voltage bus of the grid-side substation can be regarded as an infinite voltage source, and the bus voltage value can be based on the historical value using the auto regression moving average method (auto regression moving average , ARMA) to predict. Do the same with the output load of this station. The predicted value of the new energy power station can be obtained through the power system in the station.

There is a certain deviation between the above predicted value and the actual value of the system. First, the uncertainty analysis must be carried out, and then combined with the simplified double-terminal power supply network model, the optimization with the minimum network loss as the objective function is carried out, and then the voltage and reactive power control is realized.

Wind power prediction error has certain kurtosis and skewness. If normal distribution is used to describe the error, the error will be larger. In this paper, beta distribution is used to fit wind power prediction error. When the predicted value of wind power is P _t ^pred , the probability density function of the unit output active value x is as follows:

In the above formula, B(α, β) is the beta function, and α and β are the morphological parameters of the beta distribution. Their values are related to the expected μ and variance σ ² of the beta distribution. The calculation formula is as follows:

Through the above two formulas, the beta distribution shape parameters α and β can be obtained, and then the probability density function of the wind power forecasting error distribution can be obtained.

The light intensity also approximately obeys the beta distribution in the hour-level time period. Considering that the output power of the photovoltaic array is linearly related to the area of the photovoltaic array and the photoelectric conversion efficiency, the output power of the photovoltaic array also obeys the beta distribution. The probability density formula is as follows:

In the formula, Γ is the Gamma function, α and β are the morphological parameters of the beta distribution, and p _max is the maximum output power of the photovoltaic array.

The wind power prediction distribution and the optical power prediction distribution are equally divided from the beta distribution, so that the wind power prediction model and the optical power prediction model can be unified.

In the present invention, the high-voltage side bus of the new energy power station connected to the substation is equivalent to an infinite power supply. Although the voltage value of the bus does not fluctuate much in a short period of time, considering the accuracy of optimal control, the autoregressive sliding average model is used to predict the voltage of the bus:

和θ是待求系数。In the formula, U(t _i ) is the predicted value at time t _i , U(t _ik ) is the historical measurement value at time t _ik , ε(t) is the forecast error at time t, p and q are the model autoregressive order number and moving average order,

and θ are coefficients to be sought.

The autoregressive moving average model requires that the time series must be stationary, so the stationarity test of the historical values is required before forecasting, and the autoregressive order, moving average order, and coefficients to be found of the model are recalculated before each optimization process. The load value in the substation is also predicted according to Equation 6 autoregressive moving average model.

The power prediction model and bus load prediction model of the new energy power station are obtained above. The probability density function is continuous. If it is directly used for voltage and reactive power optimization control, the calculation time and complexity will be greatly increased, and it is not suitable for field applications. However, using a typical scene set that is representative and can reflect the probability distribution characteristics of the original data can effectively improve the calculation efficiency, reduce the calculation amount, and ensure the optimization accuracy.

The generation of a typical scene set first requires a large number of random variables according to a certain probability distribution to be sampled to obtain a scene data set that reflects the characteristics of the random variable, that is, the continuous probability model is discretized. The dots in Figure 2 are the original predicted values in Discretization at each time point, and each dot corresponds to the probability of possible occurrence. However, the number of scene sets obtained in this step is still huge, more than 10,000, which is not conducive to calculation. It is necessary to streamline and cluster the scenes to obtain typical scene sets, so that fewer scenes can be used to represent the characteristics of the original scene, and Figure 2 is obtained. The black heart dot in the .

In this paper, the Latin hypercube sampling (LHS) method is used to sample the new energy power prediction value and voltage load prediction value on each prediction interval t _pred N times, and obtain w _i (i=1,2,…, N) scenarios, and the probability corresponding to each scenario is ρ _i (i=1, 2, . . . , N). In this way, the original input distribution can be represented by a limited number of sampling times.

K-means clustering algorithm is a kind of unsupervised learning algorithm. It divides the original data set into K categories according to the characteristics of the data. Each category contains at least one data, and the data in the category have similar characteristics and attributes. Perform clustering analysis on the N scenes sampled above to obtain K classes, and K typical scene sets can be obtained. The calculation process is as follows:

1) Normalize the prediction value in the scene according to the following formula, and map its value to the interval [0,1];

2) Randomly select K initial cluster centers;

3) Calculate the Euclidean distance from each scene to the initial cluster center according to the following formula, and divide the scene into the nearest class;

4) Recalculate the arithmetic mean of each scene in the class in 2) as the new center of the class;

5) Repeat step 2) until the clustering result does not change.

The goal of the voltage optimization control of the new energy power station in the present invention is to ensure that the busbar voltage in the electric island composed of the new energy power station and the substation is within a reasonable range on the basis of the K typical scene sets obtained above, and at the same time reduce the grid voltage of the electric island. loss, the specific objective function is as follows:

In the formula, ρ _s is the probability corresponding to the scene w _s ; P _t (w _s ) is the active power loss of the electrical island at time t in the prediction period t _pre ; V _t (w _s ) is the bus voltage t in the prediction period t _pre Deviation value at time; α, β are weight coefficients. P _it (w _s ), Q _it (w _s ), V _it (w _s ) are the active power, reactive power and voltage value of node i at time t in the prediction period t _pre respectively; V _is is the voltage setting value of node i .

The scope of optimization of the present invention is a double-ended power supply network as shown in Figure 1, and the equality constraints and the inequality constraints are as follows:

P _it ＝P _i-1.t -P _Li.t

Q _it ＝Q _i-1.t -Q _Li.t

Q _BC.t ≥ 0

τ _it ≤ τ _imin

0≤ν _it ≤ν _imax

1≤σ _it ≤σ _imin

Q _svc.i.min ≤Q _svc.it ≤Q _svc.i.max

Equations 1-5 are the constraint conditions of the power flow equation of the network. In this paper, bus A is regarded as an infinite power source, that is, the balance node of the electrical island, bus D is a PV node, and other buses are PQ nodes. Equation 6 is the node voltage amplitude constraints. Equation 7 is the constraint condition of the maximum transmission capacity of the transmission line. Equation 8 is the constraint condition for reactive power transmission of transmission lines. Electric power companies generally require new energy power plants to achieve reactive power on-site balance, and the situation of reactive power transmission is not allowed. Equations 9-10 are the constraints on the allowed number of actions and action time intervals of the capacitor within the control period. Equations 11-12 are constraints on the number of actions allowed by the transformer gear within the control period and the action time interval. Equation 13 is the maximum and minimum output capacity constraints of the SVC in the new energy power station.

The invention incorporates the outputtable reactive power of the DFIG into optimal control. Since the reactive power output and absorbed by the DFIG rotor is relatively small, it can be ignored, and the range of reactive power that can be output by the unit is mainly determined by the stator:

In the formula, P _s is the active value generated by the stator; U _s is the stator side terminal voltage; X _s is the stator side leakage reactance; X _m is the excitation reactance; I _Imax is the maximum current value of the rotor converter.

The inverter of photovoltaic power generation also has certain reactive power regulation capability. Considering that the inverter can operate at 1.1 times the rated power, its reactive power adjustment range is as follows:

In the formula, S _INV is the rated capacity of the inverter; P _PV is the power of photovoltaic power generation.

The optimization model includes discrete and continuous variables such as capacitor switching, gear adjustment, new energy output, and SVC reactive power compensation. It is a typical mixed integer non-convex nonlinear programming, and it is difficult to directly find the optimal solution. To solve this problem, second-order cone programming (SOCP) can be used to relax the convexity of the power flow equation, and the optimization problem can be transformed into a second-order cone programming problem with integer variables. At the same time, the optimization model in this paper is a two-terminal power supply network with fewer nodes, which can effectively improve the accuracy of optimization and the efficiency of calculation.

The present invention has been exemplarily described above in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above methods. As long as various improvements made by the technical solution of the present invention are adopted, or directly applied to other occasions without improvement, all Within the protection scope of the present invention.

Claims (4)

1. A control system for reactive voltage of a new energy power station is characterized in that: the new energy power station and the power grid access side transformer substation are used as an area to carry out optimization control integrally, and the control system comprises a prediction model, a typical scene set model and MPC optimization control; meanwhile, predicting a bus voltage value and a transformer substation load by using an autoregressive moving average method, generating a discretized scene for the predicted value through Latin hypercube sampling, acquiring a typical scene set by using a K-means clustering algorithm, establishing an optimal control model with an objective function being the minimum in a future time window internal network loss value and voltage offset value, and solving by using a second-order cone programming method;

the prediction model of the control system comprises a wind power/photovoltaic power prediction part and a load/bus voltage prediction part, wherein the wind power/photovoltaic power prediction error is fitted by adopting beta distribution, and the load/bus voltage is predicted by adopting an autoregressive moving average method;

the control system is characterized in that a typical scene set model of the control system firstly generates a discretized scene by pulling Ding Chao cubic samples of predicted values of new energy output, bus voltage and transformer substation load in an electric island, and then acquires a typical scene set by using a K-means clustering algorithm;

the MPC optimization control of the control system establishes an optimization control model with the minimum target function of the network loss value and the voltage offset value in a future time window, and solves the model by adopting a second order cone planning method; the MPC optimization control comprises capacitor switching constraint, transformer gear adjustment constraint and SVC reactive compensation constraint;

in the wind power/photovoltaic power prediction, when the wind power predicted value is P _t ^pred When the probability density function of the unit output active value x is:

where B (α, β) is the beta function, α and β are morphological parameters of the beta distribution, their values are the expected μ and variance σ of the beta distribution ² The calculation formula is as follows:

the beta distribution form parameters alpha and beta are obtained, and then a probability density function of wind power prediction error distribution can be obtained;

the output power of the photovoltaic array also obeys the beta distribution in the optical power prediction, and the probability density formula is as follows:

wherein Γ is Gamma function, alpha and beta are morphological parameters of beta distribution, p _max For maximum output power of photovoltaic array, p _solar The output power of the photovoltaic array; the wind power prediction distribution and the light power prediction distribution are equally divided from beta distribution, and the wind power prediction model and the light power prediction model are unified.

2. A control system for reactive voltage of a new energy power station according to claim 1, characterized in that: the generation of the typical scene set firstly carries out a large number of samples on random variables distributed according to a certain probability to obtain a scene data set reflecting the characteristics of the random variables, namely discretizing a continuous probability model, and then carrying out reduced clustering on the scenes to obtain the typical scene set so as to realize the purpose of representing the characteristics of the original scene by fewer scenes.

3. A control system for reactive voltage of a new energy power station according to claim 1, characterized in that: the control system aims at the objective function of the new energy power station voltage optimization control:

wherein ρ is _s Is scene w _s The corresponding probabilities; p (P) _t (w _s ) Is the electrical island at t _pre Predicting active loss at time t in the period; v (V) _t (w _s ) Is the bus voltage at t _pre A deviation value at time t in the prediction period; a. b is a weight coefficient; k is the number of typical scene sets。

4. A control system for reactive voltage of a new energy power station according to claim 1, characterized in that: the control system incorporates reactive power that can be output by the DFIG into an optimal control, wherein the reactive power ranges:

wherein P is _s An active value for the stator; u (U) _s Is the stator side terminal voltage; x is X _s Is stator side leakage reactance; x is X _m Is an excitation reactance; i _Imax For the maximum current value of the rotor converter.

Images