JPWO2014199502A1 - Power system control apparatus, system and method - Google Patents

Abstract

It is an object of the present invention to reduce the load energy by giving a command that allows stochastic fluctuations as an output target to be supplied to the control device. Means for obtaining a value, means for setting an output target value of a control object linked to the power system based on the state value in a probability distribution, and means for obtaining a control parameter based on the set probability distribution Is provided.

Description

  The present invention relates to an apparatus, a system, and a method for stabilizing voltage or the like in a power system.

  Conventionally, due to fluctuations in the load connected to the power system, noise has been generated in the measurement signal of the sensor that acquires the system status, and the introduction of distributed power sources such as solar and wind power generation has progressed, and the fluctuation range has been increasing. It has become big. Therefore, a technique is known in which the fluctuation is grasped by estimating the state of the system, and the power system is optimally controlled based on the state estimation.

  For example, as a prior art, Patent Document 1 discloses a device configuration for controlling a voltage stabilizing device by estimating a unknown system impedance using a statistical method for the purpose of keeping the system voltage constant. ing. Patent Document 2 discloses a device configuration that estimates and predicts the state of photovoltaic power generation, which is a factor causing voltage fluctuation, using a statistical method.

Japanese Patent Laid-Open No. 4-229022 JP2011-087372

  However, in the above technique, the state of the target device is estimated in real time, and the control is performed by giving a definitive target value based on the state estimation. The load energy applied to the control equipment is large. When the load energy increases, problems such as early deterioration of the control device due to excessive number of times of control and control amount, and delay in calculation speed of the control amount may occur.

  Therefore, an object of the present invention is to reduce the load energy by giving a command that allows stochastic fluctuations as an output target given to the control device.

  In order to solve the above-described problem, in the power system control device according to the present invention, a means for obtaining a state value of the power system, and an output target value of a control object linked to the power system based on the state value Means with a probability distribution, and means for obtaining a control parameter based on the set probability distribution.

  According to the present invention, the target value is stochastically given as compared with the configuration in which the target value that is the characteristic of the prior art is definitely given, so that the necessary scale, operation amount, power amount, etc. of the control device can be obtained. This reduces the load energy required for control.

The figure which shows the apparatus structure of the whole this invention The figure which shows the relation of the probability density function of 1 input 1 output system and control constitution Diagram showing the combination of photovoltaic power generation and storage battery Diagram showing the combination of known and unknown quantities in the power system Diagram showing comparison between deterministic system and probabilistic system Diagram showing how to set stochastic characteristics (variance) Diagram showing how to set stochastic characteristics (variance) A diagram showing the passage of time to reflect the state of external equipment etc. in the variables of the power equation The figure which shows the configuration of voltage stabilization by the distributed control device

  Prior to a specific description of an embodiment of the present invention, a brief description will be given of a conventional technique for calculating a state value and an estimated value of a power system necessary for suitably implementing the embodiment.

  The power system is composed of a combination of L (inductance), C (capacitance), and R (resistance). The state values of the power system are V (voltage), I (current), P (active power), and Q (reactive power). ), Θ (phase), etc. These state values can be related by, for example, a power (tidal current) equation. Since the tidal current equation is a nonlinear simultaneous equation, it is difficult to solve it analytically, and numerical calculation is often used. One of them is the Newton-Raphson method, which obtains a solution by preparing a coefficient matrix called Jacobian and performing convergence calculation.

  The power flow equation is used not only to calculate the state value of the power system based on the measurement data, but also to calculate control parameters of the control device for stably operating the power system. As a control device for suppressing and stabilizing the fluctuation, for example, SVR (Step Voltage Regulator), SVC (Static Var Compensator), storage battery, and the like are used for voltage stabilization.

  When the state value of the target system cannot be acquired, a method of estimating an unknown state value using the state value that can be acquired is used.

  As such a state estimation technique, there is a method using a least square method. Further, a Kalman filter is known as a state estimation method that considers fluctuation factors stochastically. Here, the Kalman filter refers to a signal processing system that sequentially calculates an estimated value that minimizes the square error of the state vector of the system based on an observed value disturbed by noise (fluctuation component). The target system is represented by a combination of system equations and observation equations, and has a procedure for calculating both sequentially and repeatedly. When the system is linear and noise represented by a normal distribution is added additively, a stochastic characteristic is represented by adding a dispersion term to the equation. Since this linear system has a noise term expressed stochastically, it is also called a linear stochastic system. In addition, the present invention is also applied to a case where the target system has nonlinear characteristics, and when the system needs to be distinguished, it is called a nonlinear Kalman filter. Many variations of the Kalman filter have been proposed in derivation methods, calculation procedures, application fields, and the like. There is an interpretation based on Bayesian statistics. Nonlinear Kalman filters for nonlinear systems include an extended Kalman filter and an unscented Kalman filter. Further, a derivative method called a particle filter has been proposed that assumes that the statistical properties of noise are not known in advance. The present invention can be used as a Kalman filter including the derivation method as described above.

  Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The following examples show specific examples of the contents of the present invention, and the present invention is not limited to these examples, but by those skilled in the art within the scope of the technical idea disclosed in this specification. Various changes and modifications are possible.

  First, an example of the overall apparatus configuration of the present invention will be described with reference to FIG.

  The power system control apparatus 101 according to the present invention has means for acquiring or receiving system state values including voltage, current, and power value in the power system 102, and has means for storing the state value information. In addition, a part that calculates the state value information in a predetermined time zone as probability distribution information, the probability distribution information, and a target that is input from outside (ENERGY MANAGEMENT SYSTEM (hereinafter referred to as EMS), system operator, service provider, etc.) A value (including a threshold value and the like) is input to the control signal calculation unit, and the control signal calculation unit calculates a control parameter that allows stochastic fluctuation based on the information and issues the control parameter through the control signal issue unit. The control signal transmitted to the control device in the power system 102 is transmitted. Hereinafter, detailed portions including the control parameter calculation means will be described.

  A control parameter calculation method based on the probability distribution (probability density function) will be described below.

  When considering the introduction of natural energy such as sunlight and wind power, the occurrence of disasters and accidents, and fluctuations in measurement / communication means in an electric power system, it may be preferable to handle those behaviors stochastically and statistically. The present invention describes a power system as a stochastic system and realizes a means for performing numerical analysis incorporating stochastic and statistical fluctuation factors, thereby enabling methods and apparatus configurations such as model, simulation, state estimation, and control. Realize. The present invention is characterized in that, when fluctuations such as a target value and natural energy are given stochastically, a control parameter necessary for realizing the target value is calculated. Conventionally, a method for estimating a state value under a condition with a probabilistic variation has been proposed. An object of the present invention is not to estimate a state value but to perform control for realizing an output target value.

  The power system and the devices constituting the power system will be described as a specific control target of the present invention, but it can be diverted to various devices. In the following description, probability, statistics, fluctuation, noise, and the like may be used as terms meaning variation.

Here, the operation principle of the present invention will be described using a combination of probability density functions taking the one-input one-output system shown in the upper part of FIG. There are an input x1 and an output x3 on the outside, and an input x2 on the inside. Here, the probability density functions of the input x1 and the input x2 have normal distribution characteristics of y1 = N (μ1, σ1 ² ) and y2 = N (μ2, σ2 ² ), respectively. At this time, the probability density function y3 of the output x3 (= x1 + x2), which is the sum of the input x1 and the input x2, is given by Equation 1, and has a normal distribution characteristic of y3 = N (μ3, σ3 ² ). If the formula of the average value μ3 is modified, the formula 2 is obtained, and the first term is corrected by the difference between the average values μ1 and μ2 of the second term. Here, the coefficient applied to the second term is calculated from the variances σ1 ² and σ2 ² . Looking closely at this coefficient, it is a weighting coefficient ranging from 0 to 1 indicating which of μ1 and μ2 is important, which is determined by the ratio of the magnitudes of the variances σ1 ² and σ2 ² . By utilizing this property, when the signals x1 and x2 including noise are input, the output signal x3 can be obtained in consideration of the weight of each signal. Conventionally, a weighting factor is a widely known method, but a weighting factor with a basis can be determined by using variance. The weighting coefficient calculated on the basis of this variance is similar to a slider device that can arbitrarily adjust the position with the signals x1 and x2 at both ends.

The control configuration of the present invention is shown in the lower part of FIG. A target value of the output x3 is set stochastically, and a control parameter for realizing the target value is calculated. Here, if the input and output are deterministic, it is equivalent to the conventional feedback control. However, the present invention is characterized in that the output target value is given probabilistically. In the 1-input 1-output system, when there is an input x1 of y1 = N (μ1, σ1 ² ), a control signal (control parameter) is used to obtain an output target value x3 given by y3 = N (μ3, σ3 ² ). x2 is calculated. First, Equation 3 is obtained by modifying the above equation. Here, the average values μ1, μ2, and μ3 are values obtained by adding k signals x1 (t), x2 (t), and x3 (t) in a certain time zone and dividing by a constant k. Therefore, if the average values μ1, μ2, and μ3 are decomposed into time series signals and attention is paid to the time t, the relational expressions of the signals x1 (t), x2 (t), and x3 (t) are obtained as shown in Expression 4. Thus, the control signal x2 at time t can be calculated from x1 and x3. That is, in the configuration in which the input x1 is sequentially input, the stochastic characteristic N (μ3, σ3 ² ) of the output x3 is given in advance as the output target value, and the control signal x2 for realizing it can be obtained. In other words, when a constant value μ3 is given as the output target, the noise component remaining in the output can be managed as the variance σ3 ² .

For example, when the variance σ1 ² = 1 of the input signal x1 and the average μ3 of the output signal x3 is set to 0 and the variance σ3 ² = 0.5, the control signal x2 becomes x2 = −x1 and variance σ2 ² = 1. That is, the operation of canceling noise is performed by adding a signal in the reverse direction of the input signal. Here, the description has been made with the configuration of one input and one output, but it is easy to develop into a configuration of multiple inputs and multiple outputs.

FIG. 3 shows an example of a device configuration that compensates for fluctuations in photovoltaic power generation by charging and discharging a storage battery as a configuration example of the above-described one input and one output. The unit is power (watts), and the input is x1 for photovoltaic power generation, x2 is a storage battery, and x3 is a stabilized output. Although the amount of power generation of the photovoltaic power generation x1 varies depending on weather conditions, the purpose is to convert it into a stabilized output x3 by charging and discharging the storage battery x2. One method of stabilization is to control the charging / discharging of the storage battery so that x1 and x2 are reversed. However, as the storage battery is repeatedly charged and discharged, its capacity decreases and its life becomes shorter. Conventionally, the average charge / discharge capacity has been managed, but such fluctuation components including high frequency components have not been managed. In the long run, this will eventually work to increase the cost of control. Therefore, the present invention considers reducing the charge / discharge capacity of the storage battery by allowing stochastic fluctuations in the output x3. In general, fine fluctuations are high-frequency components, so they tend to be absorbed as a result of the impedance of the electric power system, or to have a smoothing effect and to be reduced as a whole. It is assumed that the stochastic fluctuation of solar power generation is represented by N (μ1, σ1 ² ). At this time, the stochastic variation allowed for the output is N (μ3, σ3 ² ). Under this condition, the charging / discharging control algorithm of the storage battery is expressed by Equation 3 or Equation 4 described above.

Further, consider a case where the average value μ3 = 0 of the output x3 is set. This means that solar power generation is used only for charging the storage battery, but a stochastic fluctuation N (μ3 = 0, σ3 ² ) is allowed in the output x3. In other words, fine fluctuations are input / output from / to the outside, and fine fluctuations are removed from charge / discharge of the storage battery. Since this reduces the charge / discharge capacity and the number of times of charge / discharge, the life of the storage battery can be improved. In addition, a charging / discharging circuit corresponding to fine fluctuations, that is, high-frequency components is not necessary. Thus, there is an effect of reducing the cost for the control. For example, when the variance σ1 ² = 1 of the input signal x1 and the average μ3 of the output signal x3 is set to 0 and the variance σ3 ² = 0.25, the control signal x2 is x2 = −0.333 · x1, and the variance σ2 ² = 0.333. As a result, the charge / discharge capacity can be reduced to one-third compared to the case where all input signals are used for charge / discharge. In addition, there is a feature that the stochastic characteristic of fluctuation remaining in the output can be set in advance.

  Moreover, the technology of the switching converter developed in the field of power electronics can be utilized for the supply to the system side of the electric power generated by solar power, the control of the charge / discharge capacity of the storage battery, and the like. The stabilization of the power system can be realized by controlling active power and reactive power.

  In general, the purpose of control is to reduce an error between a target value given to a target system and an actual output signal. For this reason, for example, if the noise is canceled by adding a waveform signal opposite to the noise, the output can be kept constant. However, if noise cannot be measured from the outside, or if noise enters during measurement, it cannot be canceled. Further, in order to create a waveform signal for canceling noise when a wide frequency component is included in the noise, high-speed signal processing is required. Although an error (noise) may remain in the canceled result, the noise characteristic is not a control target.

  In contrast, the present invention creates a stochastic system that allows the output signal to contain stochastic variations. It is assumed that there is a stochastic fluctuation (noise) in the system. Then, an output target including fluctuation is set stochastically. As a result of the control, a preset stochastic variation is allowed to remain in the output signal. This characteristic behavior can be confirmed by comparing the input and output signal waveforms (ie, mean and variance) of the stochastic system.

  The use of the Kalman filter will be described. As shown in Equation 5, if the voltage is V, the current is I, and the admittance matrix is Y, I = YV from Ohm's law. If both sides are multiplied by V to replace the relational expression of power, VI = VYV, which is called a power equation (power flow equation). Since power depends on the phase relationship between voltage and current, it is convenient to express it as a complex number. For example, as shown in FIG. 4, there are active power (P), reactive power (Q), voltage (V), and phase (δ) as known amounts and unknown amounts set in the power equation. Based on the known amount of the node (node), it is called PQ designation, PV designation, etc., and the known amount is set. For example, an unknown quantity can be calculated by specifying a PQ or PV value measured using a sensor as a node and solving a power equation.

  Since the power equation is non-linear, it is common to use a numerical analysis method such as Newton-Raphson method. In the Newton-Raphson method, focusing on only low-order components when a nonlinear equation is Taylor-expanded, it is replaced with a linear equation to perform a convergence operation. In order to apply the Newton-Raphson method to the power equation, as shown in Equation 6, a linear simultaneous equation using a Jacobian matrix J obtained by partial differentiation of the admittance matrix Y as a coefficient is replaced. In this example, P2, Q2, P3, and Q3 of PQ designation nodes are set as known quantities on the left side, and P4 and V4 of PV designation nodes are set until the unknown voltage (e, f) on the right side is stabilized. Perform convergence calculation. Obtaining a stable state as a result of the convergence calculation (the state change is small even if the calculation is repeated further) is to solve the equation, and the state variable at this time is used as a solution. It is convenient to express the power system in a complex number format, where e is the real part of the complex voltage V and f is the imaginary part.

An example of solving the power equation using the Newton-Raphson method is shown. As an initial setting, a power equation and an admittance matrix Y are prepared. Also, a node specified value (PQ specified value or the like) and an initial voltage value (e, f) are set. In the convergence calculation, a PQ estimation value is obtained from the current voltage estimation value and the admittance matrix Y, and a difference ΔPQ between the PQ designated value and the PQ estimation value is obtained. Also, Jacobian J is calculated. Then, as shown in Expression 7, Δ (e, f) = J ⁻¹ · ΔPQ is solved, and the estimated voltage is updated using the obtained Δ (e, f).

  The convergence is determined by these calculations. Here, if Δ (e, f) is sufficiently small, the process is terminated as converged. If not, the process returns to the convergence calculation and the calculation is repeated.

By the way, the above system is a deterministic system in which the input / output is related by the admittance matrix Y or the Jacobian J. On the other hand, the above-described Kalman filter is known as a method for estimating the state of a system including stochastic fluctuations. Here, when the noise characteristic is given by the normal distribution N (μ, σ ² ) (where μ is an average and σ ² is a variance), the Kalman filter sequentially performs an operation so as to follow the characteristics of the input signal. A sequential operation of the Kalman filter is characterized in that it has a procedure for calculating an update value using a Kalman gain. Here, it is noted that the format for calculating the updated value is similar in both the Newton-Raphson method and the Kalman filter. Furthermore, as a derivation method of the Kalman filter, there is an extended Kalman filter for nonlinear systems, which incorporates a mechanism for performing linear approximation using a differential coefficient.

  From the above viewpoint, consider solving the power equation in the form of a Kalman filter. A coefficient matrix (Jacvian) used when solving the power equation by the Newton-Raphson method is used to calculate the Kalman gain of the Kalman filter. The power equation is a deterministic system that does not take into account variations, whereas the Kalman filter is a stochastic system that takes into account variations. The difference between the two is due to the dispersion function representing the fluctuation characteristics.

  Formula 8 shows the Kalman filter processing procedure. There are various Kalman filter notation methods, and this is not the only one. In addition, detailed subscripts not used in the explanation are simplified. First, a nonlinear equation of state is shown in the initial setting procedure (0) (hereinafter, procedure (0)). The upper part is the system equation, and the lower part is the observation equation. System noise v (k) and observation noise w (k) are noises represented by a normal distribution. Next, the convergence calculation procedure (1) (hereinafter, procedure (1)) is a time update step, and the convergence determination procedure (2) (hereinafter, procedure (2)) is an observation update step. Steps (1) and (2) are repeated to use the state variable asymptotically asymptotic to a stable state variable.

  One method applied to the power equation is shown. First, since the system equation of the procedure (0) does not consider dynamics here, the function f becomes a unit constant (if a transient characteristic or the like is incorporated, a function representing dynamics may be prepared in the system equation). And let an observation equation be an above-mentioned electric power equation. That is, y is a node designation value and a target value, x is a voltage, and h is an admittance matrix Y. Each variance is set as a target value. In the observation matrix H of the procedure (2) (a), the admittance matrix Y is a nonlinear observation matrix h, and the Jacobian J obtained by partial differentiation of Y is the observation matrix H. Then, the Kalman gain K is calculated in steps (2) and (b), and the signal incorporating the stochastic characteristics is updated in steps (2) and (c). By repeating the update, it gradually approaches a stable state, and the state variable at that time is used as a solution. Thus, the power equation can be solved in the form of a Kalman filter (extended Kalman filter).

  Note that as a characteristic characteristic of the power equation, a Kalman filter is sequentially calculated by preparing an arithmetic means in a complex number format. In addition, since a stable solution may not be reached, a procedure for determining the signal range is prepared in the repeated calculation, and if it is determined to be an abnormal value, it is escaped and some abnormality countermeasures are taken.

  If a known amount is set in advance and the power equation is solved in the form of a Kalman filter, an unknown state variable can be estimated. Therefore, the present invention can be used as a so-called state estimation solution considering stochastic fluctuations. Furthermore, the present invention sets the stochastic target value as the node designation value and solves the power equation in the form of the Kalman filter so that these target values are stably established, so that it is necessary to realize the set target value. The state value that becomes is calculated. For example, active power P and reactive power Q are set as target values for active power P and reactive power Q as PQ designated nodes, or active power P and voltage V as PV designated nodes. At the same time, each variance is set as a target value. The state in which the sequential calculation is performed and stabilized in this manner is a state for realizing the above target value, and a necessary control parameter can be extracted from this state.

  The variance set as the stochastic target value is calculated from the measurement result in terms of the definition of statistics, but the present invention provides means for setting this variance by the operator or some program regardless of the measurement result. Prepare. Furthermore, it is characterized in that switching between a deterministic system and a stochastic system is performed seamlessly by setting of dispersion. Specific setting methods and means will be described later.

  From the above results, the relationship of P, Q, V, etc. of each node for realizing the target value set is clarified. The set target value is realized by operating control devices such as SVC, PCS, SVR, and LRT so as to establish these relationships. Although the operating principle of these control devices is not mentioned, the switching converter technology in the power electronics field or the transformer technology in the power field is helpful.

  Solar power generation, wind power generation, and the like can operate as a distributed power supply that supplies power to the power system. This supply amount varies depending on conditions such as weather conditions. Along with this, the voltage distribution of the system also fluctuates. Conventionally, voltage control devices have been used for the purpose of stabilizing the voltage distribution of the system. For example, there are voltage control devices such as SVR, SVC, and storage battery. Furthermore, a method for adjusting the load by controlling the power consumption of the equipment installed in the consumer is being studied. As shown in the upper part of FIG. 5, these control methods generally set a deterministic target value of the system state and feed back a difference (error) between the target value and the actual output value as a control signal on the input side. Feedback control is performed. And it is supposed that the controllability is better as the difference from the target value is smaller, that is, as the error is smaller and the settling time is shorter. On the other hand, as shown in the lower part of the figure, the present invention gives a stochastic target value for the voltage control device as described above, and allows an average value while allowing a variation managed by the magnitude of dispersion. Perform asymptotic control. For example, control methods such as voltage stabilization of the power system by injecting reactive power and phase stabilization by injecting active power are known. The present invention does not limit these methods.

  Thus, the present invention can calculate the control parameter that asymptotically approaches the output target as time progresses, and can control the target device. By the way, the Newton-Raphson method and the Kalman filter have the following differences in the iterative calculation method. The Newton-Raphson method performs a convergence operation of a linearly approximated equation at a certain time t, while the Kalman filter performs a sequential operation that proceeds from a certain time t to the next time t + 1.

  In the present invention, when the power equation is solved in the form of a Kalman filter, it is possible to manage the progress of the time t by appropriately combining the above two repeated operations. Increasing the number of iterations of the Newton-Raphson method at each time can improve the accuracy of the computation results. Then, the time may be advanced using a Kalman filter by appropriate time management such as synchronization with the sampling interval of the sensor. For example, if you know that the sensor measurement is at 30-minute intervals, there is no update of the observation signal during that period, so iterative calculations related to the convergence calculation of the Newton-Raphson method approximate equation may be completed within that period. It ’s fine. This iterative calculation may be performed by sequential calculation of the Kalman filter. Then, if it is determined that it has sufficiently converged, the iterative calculation may be interrupted and a standby state may be entered.

  Many Kalman filters have been proposed, such as Extended Kalman Filter, Iterated Extended Kalman Filter, Unscented Kalman Filter, and Particle Filter. Yes. These methods are devised for application when the variation factor is not a normal distribution, when the target system is nonlinear, or when a model of the target system cannot be obtained in advance. The present invention is not limited as long as it follows the principle of the Kalman filter.

  The stabilization control using the voltage (current) equation will be described. The power system uses a power equation to relate state quantities to analyze, estimate, and evaluate characteristics. On the other hand, electronic circuits are often solved by establishing equations with voltage and current as state quantities based on Ohm's law and Kirchhoff's law. The difference between the two concepts is that the electric system pays attention to signal transmission while the electric power system uses energy transmission.

  Since the power equation becomes nonlinear with respect to voltage, current, etc., the solution itself becomes a problem. For example, a solution using the Newton-Raphson method is known. However, since it is an approximate solution that converges to a solution by repeating local linear approximation, it does not always converge to a true solution.

  In general, stabilization control is required to shift to a stable state when the state changes or is in some unstable state. Here, if the control signal is calculated based on the approximate solution of the power equation described above, the transition to the stable state cannot be made if the validity of the control signal based on the solution is not certain.

  When focusing on stabilization of voltage fluctuations in an electric power system, instead of using an electric power equation based on an energy viewpoint, consider using an equation relating to voltage (or current) based on Ohm's law and Kirchhoff's law instead. The voltage (or current) equation relates the voltage and current of the power system. Here, the admittance (impedance) matrix that relates the voltage and the current may be the same as the matrix used in the above-described power equation. The voltage (or current) equation can be expressed in a complex form with the phase of voltage and current as variables.

  By the way, some devices linked to the power system operate based on electric power. In the first place, the reason why the power equation has been widely used in the past is to explicitly express the power input and output. For example, there are a power source that inputs and outputs power, a device that performs power control by an inverter, and the like. These control such that the power converges to a target power depending on the voltage, current, and phase of the location linked to the power system. In order to include such a device in the state equation, the system equation describes an operation for performing power control based on voltage, current, and phase. Then, as described above, the voltage and current are related to the observation equation by an admittance (impedance) matrix. Both are repeated sequentially. In this way, the state quantity of the power system interconnecting the power devices can be calculated. The system equation and the observation equation are respectively

  The merit of using the voltage (or current) equation is that the power system can be handled by a linear equation. A solution can be calculated by using a general simultaneous equation solution method without using an approximate solution method. Since the solution method is simplified and there is no possibility of falling into a local solution, a control signal based on the obtained exact solution can be used reliably.

  It can be expected that the calculation results based on the conventional power equation and the proposed voltage (current) equation are almost the same in a steady state. On the other hand, in an unstable state where the voltage fluctuation is large, the proposed method gives a stable result for the reason described above. This property can be used to distinguish the conventional method and the proposed method from the outside, and can be distinguished from each other by observing the output signal when a test signal is input.

  In the voltage (current) equation, a stochastic system equation and an observation equation can be described by adding noise (variation component) in a distributed manner. This stochastic equation can be solved using a Kalman filter. The present invention sets voltages, currents, and phases that allow stochastic fluctuations as target values, and calculates voltage, current, and phase control signals for realizing these target values.

  In order to calculate this control signal, first, a voltage (current) equation is described in the Kalman filter format, a target value is set as a variable, and a variable to be controlled is solved. Since the Kalman filter is a sequential calculation procedure with time, the control signal can be calculated while sequentially changing the target value and the current state value with time.

  In order to solve the power equation with the Kalman filter, since the power equation is a nonlinear equation, a nonlinear Kalman filter is used. As the nonlinear Kalman filter, as described above, there is an extended Kalman filter. However, since this method has a linear approximation procedure in the solution, it eventually becomes an approximate solution.

  On the other hand, according to the present invention, by using a linear voltage (current) equation, an exact solution can be obtained using a linear Kalman filter, and a control signal based on this solution can be calculated. Thus, there is an advantage that stable control can be realized.

The present invention calculates a control signal by regarding a power system as a stochastic system. At this time, it is characterized in that a stochastic state variable setting method specific to the power system is performed. The stochastic state variables of the Kalman filter include A. F. System fluctuation characteristics N (μ1, σ1 ² ) (hereinafter referred to as A) represented by a normal distribution; Fluctuation characteristics N (μ2, σ2 ² ) (hereinafter referred to as B) of the observation signal represented by a normal distribution, C.I. There is a variance-covariance matrix (hereinafter referred to as C) representing the relationship between state variables. The variance of C is a value that is updated by sequential calculation. Therefore, A and B are set in advance.

  The role of dispersion is shown by focusing on the Kalman gain of the Kalman filter. If the variance covariance is a unit diagonal matrix for simplicity, the Jacobian derived from the power equation and the variance remain in the denominator. Here, when the variance is 0, only the Jacobian remains in the Kalman gain, which is the same as the convergence calculation of the Newton-Raphson method for solving the original power equation. The variance of 0 means that the observation signal does not contain an error, and the simulation of the conventional power system is calculated under this condition. On the contrary, if the variance is sufficiently large, the Kalman gain approaches 0 and the observation signal is not used and the estimated value is not updated. A large variance corresponds to low reliability of the observation signal or low accuracy. If the variance is appropriately set for these, the estimated value is sequentially updated based on the Kalman gain.

  FIG. 6 shows an apparatus configuration for setting the variance according to the purpose of use. First, the control signal calculation means 110 is a means for calculating the control signal of the present invention described above. For example, a means for executing a Kalman filter by software processing is implemented. The data collection means is means for inputting an external device or some situation as digital data using a sensor or the like. The average / variance calculation means calculates an average and variance used for performing stochastic statistical signal processing like a Kalman filter. The control signal issuing means is means for transmitting the calculated control signal to the control device, and is configured by appropriate means such as wired or wireless. The present invention has the following configuration as a dispersion setting method and means.

  Statistical calculation performed by the apparatus configuration 601 will be described.

  The average and variance of system noise A in the power system are due to fluctuations in the load connected to the system. The average and variance of B, which is an observation signal, are determined by factors such as variations in measured values by the sensor, reliability of the sensor itself, and noise of communication means that transmits the measurement signal. Therefore, the basis for setting the stochastic characteristic is to calculate the statistics at any time based on the measurement result of the observation signal or the external environment.

  Setting as a control target value performed from the apparatus configuration 602 will be described.

One of the setting methods of these stochastic characteristics is set as a control target value. For example, based on the voltage range of the low-voltage distribution system, the output target value is set so that the average μ is 101V and the variance σ ² is 3σ (99.7% range) is ± 6V. The average and variance of the voltage vary depending on the voltage class of the power system and also vary from country to country. For example, in the power equation described above, the average value and variance of V are set with the target node designated as PV.

  Switching between a deterministic system and a probability system performed by the apparatus configuration 701 in the upper part of FIG. 7 will be described.

  Another method for setting the stochastic characteristic is to switch the characteristics of both the deterministic system and the stochastic system by setting the statistical property (that is, variance) of the state variable of the target system. For example, the reliability of a signal acquired by a sensor is expressed by the magnitude of dispersion. For example, characteristics such as a sensor containing a lot of noise, a sensor with a rough quantization step, and a sensor with a long sampling interval are set by increasing the dispersion value. Conversely, reliable measurements reduce variance. Further, the reliability of the sensor signal communication means may be combined. Thus, while maintaining the power equation (that is, without changing the program), it is possible to adjust the rate of use of the observation signal by setting the dispersion.

  In the present invention, techniques related to sampling of measurement signals (for example, sensors, AD conversion, communication, filtering, interpolation, etc.) are appropriately used without limitation.

  In order to improve the stability, sensitivity, and resolution of state estimation values, control signals, etc. obtained in the Kalman filter format, suitable settings are made by repeatedly changing the type, number, quantization step, and time interval of observation signals as appropriate. The value can be determined. In order to evaluate in advance the effects of failures, accidents, malfunctions, etc. that occur in the power system or sensor data communication means, the range and magnitude of the effects can be determined by repeatedly calculating while changing the conditions as described above. Can be grasped in advance. If the variance value temporarily set and the resulting effect are stored as learning patterns at this time, the cause can be quickly found from the result when a similar failure, accident, malfunction, or the like occurs. In this way, countermeasures can be executed quickly.

  The use of the prediction signal performed by the apparatus configuration 702 in the lower part of FIG. 7 will be described.

  A case where a prediction signal can be obtained by some means will be described. For example, in solar power generation, when a change in solar radiation can be measured from a sensor placed in the vicinity, the solar power generation may be predicted by predicting the amount of solar radiation at the location where the solar power generation is installed using the measurement signal. The present invention does not limit the prediction method. However, in many cases, an error is included in the measurement signal, and an error is also included in the prediction signal obtained from the result. Therefore, in the present invention, the statistical property is used instead of the prediction signal itself. That is, the prediction signal is the average and variance of signals to be predicted. Thus, according to the present invention, the control system is calculated by describing the power system using a stochastic system and performing the stochastic setting of the system noise based on the predicted signal (signal mean and variance). In this way, the statistical properties related to the target system or its surroundings can be reflected without spending the time required for measurement, communication, and analysis.

  FIG. 8 shows the passage of time in which the state relating to the external device or the external environment is collected by the sensor and reflected in the variables of the power equation when sequentially calculating the power system of the present invention in the Kalman filter format. . The operation of sensor a and sensor b is shown in the upper part of the figure. Since the signal input from the sensor is affected by the response time of the sensor itself, the operation of the AD converter, the communication means, and the like, the period and the sampling time may not be specified. Even when a sampling time stamp function such as GPS is used, the delay time on the way cannot always be managed. Since the target system is dynamically changing, it may not be desirable to handle a sensor signal collected at a certain time with the same reliability regardless of the passage of time. Conventionally, a resampling (resampling) technique is known in which the sampling time is post-adjusted by interpolating the sampling data. However, the change in reliability due to the interpolation process is not taken into consideration.

  Therefore, the present invention treats the collected data of the sensor stochastically in combination with dispersion. Collected data shall be the most recently acquired data. Variance is the meaning of variance defined in statistics, and the smaller the value, the smaller the variance. If this variation is interpreted as reliability, the smaller the value, the higher the reliability. The distribution of the sensor a and the distribution of the sensor b in the figure shows a configuration example that increases based on the elapsed time since the collection data is updated. Furthermore, you may prepare the fixed time range used as a dead zone. Also, some trigger may be prepared so that the increase in variance starts when data that should be collected periodically is lost for some reason. In any case, a means for managing the magnitude of the dispersion is prepared so that the period not updated with new collection data increases as time elapses (it does not decrease even if there is a stop). The form of the function to be specifically used is not limited.

  In this way, the reliability of the sensor signal can be arbitrarily adjusted only by setting the dispersion without changing the configuration of the power equation, that is, the Kalman filter processing procedure. The variance is reflected in the Kalman gain, and an update process that places a weight on the corresponding sensor signal is performed as the variance is small, and an update process that does not depend on the sensor signal is performed as the variance is large.

  If the sequential calculation of the Kalman filter is set shorter than the sensor cycle and shorter than the output timing of the calculation result, the calculation using the latest state is always performed. The sequential calculation of the Kalman filter may be executed in combination with a convergence calculation that stabilizes the internal state.

  The sensor signal that is not updated for a long time has a sufficiently large variance, and the weight in the Kalman gain calculation is small. Regardless of whether the reason is a sensor failure, a communication path failure, or an accident, the feature is that it can be managed by a unified index called the size of dispersion. On the contrary, an arbitrary magnitude of dispersion may be set assuming these causes. A variance can be set by simulating some accident, and the state variable to be controlled at that time can be calculated by the Kalman filter. In addition, a method for dealing with the assumed cause can be examined in advance. Then, the causes and results, and the countermeasures can be summarized in a table. When an accident actually occurs, the cause, result, and coping method can be searched quickly by searching the above table.

  The control range / object will be described below.

  The present invention does not limit the target range to be controlled based on the stochastic target value, and does not limit the stabilizing device. Some control devices cannot set the control signal size, control time interval, etc. continuously. For example, SVR has a limited number of tap switching stages on the secondary side, and switching may take several to several tens of seconds. SVC can only control reactive power (Q), although its response speed is fast.

  The present invention does not limit the number of control devices that are driven with the calculated control parameters to one. Stabilization control may be performed by combining a plurality of control devices having different characteristics. For example, the overall control characteristics can be improved by combining the above-described SVR and SVC or other devices.

  There are cases where the operation timings of the devices or sensors constituting the power system differ from device to device. When the sampling time and cycle of the sensor are different for each device, the timing at which the collected data is reflected in the power equation is shifted. Conversely, the timing at which a voltage control device can be controlled may vary from device to device.

  As described above, by making the sequential calculation of the power equation in the form of the Kalman filter sufficiently earlier than the above period, the sensor signal is reflected in the system equation at the available timing, while the operation timing of the control device is By generating and outputting the control signal at, the latest state can be used in any case. This can be realized because the sequential calculation of the power equation in the Kalman filter format is discrete.

  A configuration of application to stabilization control of output of solar power generation and wind power generation that are distributed power sources using natural energy will be described. The output of these distributed power sources fluctuates depending on weather such as solar radiation and wind speed. Therefore, a stabilizing device using an inverter, a storage battery, or the like has been proposed.

  FIG. 9 shows a configuration example of a distributed power source, a voltage stabilizing device, and a control device arranged in a distributed manner.

  If the distributed power source is solar power generation, it generates power in proportion to the amount of solar radiation, so the power generation amount changes depending on the weather conditions. When surplus power is supplied to the grid, a voltage rise is likely to occur. However, this voltage fluctuation greatly depends on weather conditions. Therefore, for example, the distributed arrangement control device collects information such as the amount of photovoltaic power generation and the system voltage in the surrounding area. When clear weather continues, the amount of solar radiation stays high and the surplus power becomes larger. In the case of cloudy weather (rainy weather), the amount of solar radiation has stopped decreasing, and the surplus power is less. And the amount of solar radiation will fluctuate | varieously greatly in the conditions, such as sunny and cloudy or sunny and cloudy. The distributed arrangement control device calculates the probabilistic characteristics (average and variance) of the amount of solar radiation (or power generation amount or surplus power) based on such information. This calculation is performed at an appropriate cycle. The calculation result may be replaced with some learning rule or feature pattern and stored for later reuse.

  Then, when the surplus power supply tends to remain high due to continuous sunny weather, tap control is performed using a voltage control device such as SVR so as to prevent the system voltage from staying high. On the contrary, when surplus power supply tends to stop decreasing due to cloudy (rainy) continuation, tap control is performed using a voltage control device such as SVR to prevent the system voltage from stopping decreasing. Alternatively, when the surplus power supply tends to be cloudy when it is sunny or cloudy when it is sunny, tap control is performed so that the system voltage becomes an average intermediate range using a voltage control device such as SVR, or Although not shown, voltage stabilization is realized using a voltage control device such as SVC capable of high-speed voltage control.

  Although the above scenario has been described qualitatively, based on the power equation of the target power system, a signal measured by a sensor and a stabilization target value are prepared, and the control parameter calculation of the present invention described above is performed. A procedure can be used to generate a control signal.

  Naturally, the control devices arranged in a distributed manner may be provided with means capable of communicating with each other. Data on solar power generation in other areas can be collected and used to set stochastic characteristics.

  The Example linked with the power distribution system for every area is described. Here, although the size of the area is not limited, for example, it can be a low-voltage distribution system partitioned by a columnar transformer. Consumers in the region are equipped with devices that consume electricity and PV generators that generate electricity (power generation is considered a negative load) with an arbitrary number and capacity, and are operating with fluctuations. And Here, it is assumed that stabilization control is performed such that the load for each customer or region is set by a probability density function. The probability density function that is the load control result overlaps with the interconnected power systems and becomes the probability density function of the entire load of the power system. The method for controlling the load is not limited. For example, a DSM (Demand Side Management) technique for controlling equipment on the consumer side can be used.

Next, n consumers connected to the power system are considered. When the load of each customer overlaps in the power system and the fluctuation follows an independent normal distribution, if n is increased, the voltage concentrates on the average μ and the variance σ ² becomes 1 / n. . This property is known as the law of large numbers in the field of statistics. The same applies to the case of controlling the load in a region where a plurality of consumers are grouped. When the voltage stabilization control result for each region is represented by a normal distribution N (μ, σ ² ), the normal distribution N (μ, σ ² / n) is obtained in a voltage system connecting them. In general, the effect of 1 / n increases as the higher system is a combination of a plurality of systems.

  In the combination of solar power generation and storage battery described above, a method for setting a stochastic target value that allows variation as a stabilized output and performing storage battery control based on the setting is shown in such a power system. One reason is that an averaging effect can be expected.

  In the present invention, a probabilistic target value is set for each consumer, for each region, for each distribution feeder, or the like, using the above-mentioned large number law. For example, the procedure of sharing the target value required for the upper power system with the lower power system is performed hierarchically, and the control content of the lowest system is determined. At the lowest level, there can be devices in the customer's home, and high power quality can be ensured by setting the variance of the load target value small in the customer. At this time, a mechanism for balancing the control cost and power quality required to reduce the dispersion may be used in combination.

  In this way, performing control by setting a probabilistic target value for each region is advantageous in the construction of a control system. In the power equation described above, the device configuration is characterized by an admittance matrix, but the admittance matrix (or Jacobian) increases as the system increases. When solving the power equation, the Newton-Raphson method or the Kalman filter uses the inverse matrix of the admittance matrix (or Jacobian). However, in general, calculating an inverse matrix is problematic in terms of processing load and calculation accuracy, and this becomes more apparent as the matrix becomes larger. Therefore, it is effective to treat the system configuration hierarchically, to realize the stochastic power quality maintenance of the present invention for the power system in the lower region, and to use the above-mentioned 1 / n effect in the upper unit. is there.

  Although the above has described the stabilization of the load, it can be replaced with voltage, active power, current, phase, etc., by the method of specifying nodes in the power equation.

  As described above, the present invention calculates a control parameter by probabilistically giving a target value of a load or a voltage to a consumer, a region, or a distribution system that constitutes an electric power system, and controls each of them. It is characterized by combining individually set probabilistic target values based on the law of large numbers. Compared to the case where individual target values are set as deterministic fixed values, stochastic variations are allowed, and the effect of reducing the cost for control can be obtained.

  Explain the use of derivation methods in statistical methods.

  When it is difficult to describe a power system model with a power (power flow) equation, a derivative method such as an unscented Kalman filter or a particle filter can be used instead of the Kalman filter, the extended Kalman filter or the iterated Kalman filter. The unscented Kalman filter represents a target system using a feature amount called a sigma point. The particle filter is a technique that uses statistical sampling, and although there is an opinion that it is different from the Kalman filter in principle, it is treated as a derivation method here.

  The description so far has shown the configuration example in which the probability density function of the target voltage value is set in a normal distribution. However, the distribution is not limited to a normal distribution, and may be any asymmetric probability density function. An example of using an asymmetric probability density function is a system in which photovoltaic power generation is connected. Since solar power generation depends on the amount of solar radiation, the amount of power generation decreases when there is a cloud flow. When it is sunny and cloudy, the amount of solar radiation is often mixed with the decrease and recovery of the amount of solar radiation. As the amount of power generation, the power generation amount declines and recovers from time to time while the maximum power generation amount continues. Thus, the appearance of voltage may not be a symmetrical distribution.

  When any dead band is provided at either the input or output of the power system, the signal characteristics are non-differentiable nonlinear characteristics. A linear approximation method using a differential coefficient cannot be used.

  For example, an asymmetric probability density function may be created by combining a plurality of normal distributions. Further, as a method of not setting the probability density function in advance, the above particle filter or the like may be used. The present invention can calculate a control parameter for achieving a stochastic output target value by applying a Kalman filter derivation method capable of coping with these restrictions to these variations.

DESCRIPTION OF SYMBOLS 101 Power system control apparatus 102 Power system 110 Control signal calculation means 601, 602, 701, 702 Device configuration 901, 902, 903 Consumer

Claims (10)

In the power system control device that determines the control parameters of the power system,
Means for obtaining a state value of the power system;
Based on the state value, means for setting an output target value of a control target linked to the power system in a probability distribution;
Means for determining a control parameter based on the set probability distribution;
A power system control device comprising: In the electric power system control apparatus according to claim 1,
The power system control apparatus further comprising means for controlling the control object using the control parameter. In the electric power system control apparatus according to claim 1,
The power system control apparatus according to claim 1, wherein the probability distribution includes a normal distribution. In the electric power system control apparatus according to claim 1,
The state value includes a phase, voltage, current, or power. In the electric power system control apparatus according to claim 1,
The output target value is obtained using a signal processing method including a Kalman filter. In the electric power system control apparatus according to claim 1,
The output target value is obtained by changing a variance value based on reliability of the state value. In the electric power system control apparatus according to claim 1,
Means for setting the control target value as a final value;
The power system control device, wherein the output target value is selected and set from either a probability distribution or a definite value based on the reliability of the state value. In the electric power system control apparatus according to claim 1,
The control system includes a power source and a storage battery using natural energy including solar power generation or wind power generation. In the power system control system for controlling the power system,
Means for obtaining a state value of the power system;
Based on the state value, means for setting an output target value of a control target linked to the power system in a probability distribution;
Means for determining a control parameter based on the set probability distribution;
Means for controlling the control object using the control parameter;
A power system control system characterized in that an output target value having a small dispersion value is obtained by interconnecting output target values of a plurality of control targets in the power system, and each control target is controlled. In a power system control method for controlling a power system,
Obtaining a state value of the power system;
Based on the state value, setting a target output value to be controlled linked to the power system in a probability distribution;
Obtaining a control parameter based on the set probability distribution;
A power system control method comprising the step of controlling the control object using the control parameter.

Images