CN105184027B - Power load modeling method based IMM Algorithm - Google Patents

Power load modeling method based IMM Algorithm Download PDF

Info

Publication number
CN105184027B
CN105184027B CN201510717484.6A CN201510717484A CN105184027B CN 105184027 B CN105184027 B CN 105184027B CN 201510717484 A CN201510717484 A CN 201510717484A CN 105184027 B CN105184027 B CN 105184027B
Authority
CN
China
Prior art keywords
model
set
state
power
time
Prior art date
Application number
CN201510717484.6A
Other languages
Chinese (zh)
Other versions
CN105184027A (en
Inventor
王振树
马阳阳
Original Assignee
山东大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东大学 filed Critical 山东大学
Priority to CN201510717484.6A priority Critical patent/CN105184027B/en
Publication of CN105184027A publication Critical patent/CN105184027A/en
Application granted granted Critical
Publication of CN105184027B publication Critical patent/CN105184027B/en

Links

Abstract

本发明公开了一种基于交互式多模型算法的电力负荷建模方法,包括将第一级模型集中每个模型在k‑1时刻的状态变量、状态协方差矩阵进行模型的状态变量与协方差矩阵交互,获得交互后的模型的状态变量以及模型的协方差矩阵;分别将各个模型对应的状态变量和协方差矩阵输入各自的扩展卡尔曼滤波器进行模型的状态估计,得到模型估计的k时刻的状态变量以及模型的协方差矩阵;利用模型状态估计过程的似然函数值及信息协方差矩阵计算模型的权重,对各个子模型的权重加权求和得到对应的模型的参数;模型的权重用于下一时刻的状态交互以及当前时刻的模型功率输出融合。 The present invention discloses a method of modeling power load based IMM algorithm comprises a first stage for each model in the model focus state variable time k-1, the state covariance matrix of the state variable model covariance matrix interactions, the model state variables obtained after the interaction and the covariance matrix of the model; state, respectively, corresponding to the respective model state variables and input covariance matrix for each of the extended Kalman filter model estimation, estimation of the model obtained at time k covariance matrix of state variables and model; like weight likelihood function value and the information covariance calculation model matrix reuse model state estimation process, on the right of each sub-model weighted summation parameter corresponding to the model; right model reusability model of the power output status information and the current time in the next moment of fusion. 负荷建模过程是递推实现的,协方差矩阵不需要存储,节省了存储空间。 Load recursive modeling process is implemented, the covariance matrix need not be stored, storage space is saved.

Description

一种基于交互式多模型算法的电力负荷建模方法 Power load modeling method based IMM Algorithm

技术领域 FIELD

[0001] 本发明涉及一种电力负荷建模方法,具体涉及一种基于交互式多模型算法的电力负荷建模方法。 [0001] The present invention relates to a power load modeling method, particularly relates to a power load modeling method based on interactive multiple model algorithm.

背景技术 Background technique

[0002] 电力负荷是电力系统的重要组成部分。 [0002] The power load is an important part of the power system. 在电力系统的设计、运行、分析中,建立准确的负荷模型是十分必要的,模型精度影响着电力系统数字仿真和分析的精度。 In power system design, operation, analysis, it is necessary to establish an accurate load model, model accuracy affects the accuracy of the system digital simulation and analysis of power. 电力系统主要由发电机、输电网络以及电力负荷三大部分构成。 Power system consists of a generator, a load and a power transmission network of three parts. 现阶段,发电机模型和电网络模型已相当成熟,然而负荷模型则相对简单。 At this stage, the electrical generator network model, and the model has been very mature, however, is relatively simple load model. 这已经严重制约了电力系统数字仿真结果精度和可信度的提高。 This has severely restricted the system digital simulation accuracy and reliability of electric power. 而且负荷模型对电力系统动态行为的定量计算,如潮流计算、短路计算、安全分析、电压稳定等也有一定影响。 Quantitative load model and the dynamic behavior of the power system is calculated, such as flow calculation, short circuit calculation, safety analysis, voltage stability, also have an impact. 在临界情况下,还有可能从根本上改变定性的结论。 In critical situations, it is also possible to change the qualitative conclusions fundamentally. 实际中的电力负荷具有复杂性、分散性、随机性、时变性等特点。 The electric load actual complexity, dispersion, random, time-varying characteristics. 因此,对实际负荷构建一个精确的固定的模型是很困难的,电力系统负荷建模是电力系统公认的难题之一。 Therefore, building an accurate model of the actual fixed load is very difficult, power system load modeling power system is recognized as one of the problems.

[0003] 电力负荷模型的本质是对实际电力系统中的成千上万的用电设备的总体特性的数学表述。 Nature of power load model [0003] is a mathematical description of the general characteristics of thousands of electrical equipment in the actual electric power system. 在电力负荷建模的历史中,许多杰出的学者构建了一系列的经典的负荷模型结构,如传统的机理动态负荷模型(Classic Load M〇dels,CLM),直接考虑配电网的综合负荷模型(Synthesis Load Models,SLM)以及考虑发电机模型的增广负荷模型(Generalized Load Models,GLM)等。 In the history of load modeling in power, many prominent scholars constructed a series of classic load model structure, such as the traditional mechanism of dynamic load model (Classic Load M〇dels, CLM), a comprehensive load model considering direct distribution network (Synthesis load models, SLM) and consider the augmented load model generator model (Generalized load models, GLM) and so on. 由于电力负荷本身的特点,电力负荷成分不是固定的,有可能随时发生变化,它的时变性和随机性是耦合在一起的。 Since the load characteristics of power itself, the component is not fixed power load, it is possible to change at any time, it is time-varying and randomness coupled together. 这也是负荷建模的困难之处。 This is also the difficulty of load modeling. 随着电力电子技术,计算机技术,现代控制理论等一系列新型的技术和理论在电力系统中的应用,一些风机,光伏电源等新型的分布式电源并入配电网,微电网技术得到大力发展,同时电气化铁路的换流器、变频器、可控整流装置等电力电子装置在电力系统负荷中大量应用,这些都对电力负荷的特性造成一定的影响,使得电力负荷的特性更为复杂。 With the application of a series of new technical and theoretical power electronics technology, computer technology, modern control theory in the power system, a number of fans, solar power and other new types of distributed power supply incorporated into the distribution network, micro-grid technology is to develop while the electric railway power electronics of the converter, the inverter, controlled rectifier devices widely used in power system load, these have a certain impact on the characteristics of the power load, such that the characteristics of the power load is more complex. 传统的负荷建模方法都是使用一个单一的模型来尽量拟合实际系统的动态过程。 Traditional load modeling methods use a single model to try to fit the actual system dynamics. 当实际的负荷成分发生变化之后, 传统的负荷模型势必会存在一定的误差。 When the actual load component change, the conventional load model is bound to certain errors.

发明内容 SUMMARY

[0004] 为解决现有技术存在的不足,本发明公开了一种电力负荷建模方法,具体为一种实现简单、计算量较小、精确度高的电力负荷建模方法。 [0004] To solve the disadvantages of the prior art, the present invention discloses a method of modeling power load, in particular to a simple, calculation amount is small, electric load modeling method of high accuracy. 本发明提出的方法构建的电力负荷模型克服了上述传统负荷模型在负荷成分发生变化存在误差的问题。 Power load model proposed method of the present invention overcomes the problems constructed in the above-described conventional load model change occurs in the component loading error exists. 本发明可以精确的建立反映实际电力负荷成分的时变性和随机性的负荷模型。 The present invention can accurately reflect the actual establishment of the time-varying power load component and a stochastic load models.

[0005] 为实现上述目的,本发明的具体方案如下: [0005] To achieve the above object, according to the present invention, the following specific embodiments:

[0006] —种基于交互式多模型算法的电力负荷建模方法,包括以下步骤: [0006] - Power Load Modeling Based species interacting multiple model algorithm, comprising the steps of:

[0007] 步骤一:确定电力负荷模型集中的模型的基本结构为SLM模型; [0007] Step a: determining the power load model set as a basic structure model SLM model;

[0008] 步骤二:设定两级模型集,第一级模型集的模型所包含的模型数比较少用来表征负荷成分,每个模型分配一个权重,第一级模型集的每个模型对应于第二级模型集中的一个子模型集,第二级的模型集用来调整第一级的模型,设定一个阈值,如果第一级模型的误差小于阈值,则保持模型不变,如果模型的误差在一定时间内大于阈值,则利用第二级模型集中的子模型集对对应的第一级模型进行调整; [0008] Step 2: Setting two model set, the model number of the first model level model set contains relatively few components used to characterize the load, each model is assigned a weight, each model corresponding to a first set level model in a second stage set of sub-models model set, the model set of the second stage is used to adjust the first stage model, set a threshold, if the error is less than a first threshold level model, the model remains unchanged, if the model the error is greater than a threshold value within a certain time, corresponding to the first stage of the model is adjusted by a second stage set of sub-models model set;

[0009] 步骤三:将第一级模型集中各个模型在k-1时刻的状态变量、状态协方差矩阵进行交互运算,获得交互后的每个模型的状态变量以及模型的协方差矩阵; [0009] Step 3: first-stage concentrate of each model in the model interact state variable calculation time k-1, the state covariance matrix, the covariance model for each state variable and the model obtained after interaction matrix;

[0010] 步骤四:分别将第一级模型集中的各个模型对应的状态变量和协方差矩阵输入各自的扩展卡尔曼滤波器进行模型的状态估计,得到模型估计的k时刻的状态变量以及模型的协方差矩阵; [0010] Step Four: the first stage, respectively corresponding to each of the models in the collection of state variables and a respective input covariance matrix for the extended Kalman filter state estimation model, and a model to obtain a model state variables estimated time k covariance matrix;

[0011] 步骤五:利用步骤四中模型状态估计过程的似然函数值及信息协方差矩阵计算第一级模型集中的各个模型的权重; [0011] Step Five: using a state estimation model of the Step 4 likelihood function value and weight information covariance matrix calculated for each model of the first stage of the process model set weight;

[0012] 步骤六:用模型估计的状态变量计算模型的功率,并通过各个模型的权重值计算各个模型的功率加权的和作为模型集的综合输出,模型的权重用于下一时刻的状态交互以及当前时刻的模型功率输出融合。 [0012] Step Six: the state with the model estimated state variable calculation power model and power weighting each model by the weighting value of each model is calculated and a set of models of the total output, weight Model reused for the next time interaction and a model of the power output of the current time of integration.

[0013] 进一步的,根据模型状态估计过程的似然函数值判断第一级模型是否需要调整, 如果模型需要调整则利用对应的第二级模型集中的子模型集进行调整,调整后的模型参数用于下一时刻模型的计算;如果模型不需要调整,则保持不变。 [0013] Further, the need to adjust the model parameters if the model needs to be adjusted using a model corresponding to the sub-set of the second set level model is adjusted, the adjusted state of the model of the estimation process likelihood function value of the first level model is determined model for calculating the next time; if the model does not require adjustment, it remains unchanged.

[0014] 进一步的,利用子模型集对相应的第一级模型进行调整时,设定一个阈值,当第一级模型的上一时刻的似然函数值小于阈值,则认为模型合适,不需要调整。 [0014] Further, when the respective first level model is adjusted by the sub-set of models, a set threshold value, when the likelihood function value is less than a threshold value on the time of the first level model is considered a suitable model, no Adjustment. 如果在一定时间内大于阈值,则需要用对应的第二级的子模型集调整对应的第一级模型,在k时刻进行模型的遴选,设rgl 〇〇是第j个模型在k时刻的新息向量,则此模型的似然函数值为: If it exceeds the threshold within a certain time, the need to use a first set of sub-model adjustment level model corresponding to a second stage corresponding to, for the selection of the model at time k, is a new set rgl thousand and the j-th time k Model innovation vector, the likelihood function of this model is:

[0015] [0015]

[0016] 式中, [0016] In the formula,

Figure CN105184027BD00061

[0017] Sgj (k) =rgj (k) · rgj (k) τ〇 [0017] Sgj (k) = rgj (k) · rgj (k) τ〇

[0018] 进一步的,设子模型集中的模型概率转移矩阵为Π,各个模型权重的计算: [0018] Further, the sub-set model set to [pi model probability transition matrix, weight each model calculated weight:

Figure CN105184027BD00062

[0022] 其中,Ug1是第i个模型的权重值,Cgj是第j个模型的归一化常数,Agj是第j个模型的似然函数值。 [0022] wherein, Ug1 is a weight value of the i th model, Cgj is a normalization constant j-th model, Agj likelihood function is the j-th value model.

[0023] 进一步的,对各个子模型的权重加权求和得到对应的模型的参数: [0023] Further, on the right of each sub-model parameters corresponding to the weight weighted sum obtained model:

[0024] [0024]

Figure CN105184027BD00063

[0025] 其中,thetai (k)为k时刻子模型集中的第i个模型的参数,Theta⑹为k时刻对应上一级模型的参数。 [0025] wherein, thetai (k) is the i-th model parameter k set time submodel, Theta⑹ parameters corresponding to a model for time k.

[0026] 进一步的,模型交互时,具体为: [0026] Further, when the model of the interaction, in particular:

Figure CN105184027BD00071

[0027] 设第一级模型集中模型的马尔科夫状态转移矩阵为Mark,计算模型间的混合概率: [0027] Markov model state concentration provided the first stage of the transfer matrix model Mark, mixing between the probability model calculation:

[0028] [0028]

[0029] 式中,1,]_ = 1,2,3此^^』为模型1到模型」的转移概率,归一化常数(^可由下式求得: [0029] In the formula, 1,] _ ^^ = 1,2,3 this "as the transition probability model to the model 1", the normalization constant (^ by the following formula:

[0030] [0030]

Figure CN105184027BD00072

[0031] 计算输入混合状态估计值: [0031] The calculation of the input estimate mixed state:

Figure CN105184027BD00073

[0034] 其中,Xi(k-1)为模型i在k_l时刻的状态变量; [0034] where, Xi (k-1) i is the model state k_l time variable;

Figure CN105184027BD00074

为混合后模型j的状态变量;Pi (k-Ι)为模型i在k-Ι时刻的状态协方差矩阵: After mixing model is a state variable j; Pi (k-Ι) in the state i of the model covariance matrix k-Ι time:

Figure CN105184027BD00075

为混合后模型j的状态协方差矩阵。 After mixing matrix is ​​the covariance model state j.

[0035] 进一步的,采用扩展卡尔曼滤波器对电机的状态变量进行估计,为了简化扩展卡尔曼滤波器的结构,同时电机的定子电阻对系统的动态特性影响较小,将电机的定子电阻Rs确定为0。 [0035] Further, using the extended Kalman filter for state variable estimation of the motor, in order to simplify the structure of the extended Kalman filter, while the motor stator resistance has little effect on the dynamics of the system, the motor stator resistance Rs determined to be 0.

[0036] 进一步的,电机的状态预测方程为: [0036] Further, the state machine prediction equation is:

Figure CN105184027BD00076

[0040] 式中,Ed是电机的次暂态电势在d轴的分量,Eq是电机的次暂态电势在q轴的分量,s 是电机的转差率;TS为采样周期,WB为同步角速度;TdO为暂态开路时间常数,X为转子开路电抗,X '为转子不动时短路电抗;K1为电机的负荷率,a为电机恒定转矩系数,MP与转速有关的转矩的方次;Uid、Uu分别为母线电压在d、q轴的分量。 [0040] In the formula, Ed is a component of sub-transient potential of the motor in the d-axis, Eq views transient potential is the q-axis component of the motor, s is the slip of the motor; the TS is the sampling period, WB is synchronized angular velocity; TDO is a transient open time constant, X is the open rotor reactance, X 'is a short-circuit does not move the rotor reactance; Kl rate of the motor as a load, a is a constant motor torque coefficient, MP related to the rotational speed side torque views; Uid, Uu respectively bus voltage in the d, q axis components.

[0041] 进一步的,状态协方差预测方程: [0041] Further, the predicted state covariance equations:

[0042] P (k I k-1) =F (k-1) P (k-1) F (k-1) T+Q (k-1) [0042] P (k I k-1) = F (k-1) P (k-1) F (k-1) T + Q (k-1)

[0043] 其中,Q为过程噪声协方差矩阵, [0043] wherein, Q is the process noise covariance matrix,

Figure CN105184027BD00081

[0045] 进一步的,量测预测方程为: [0045] Further, the measurement prediction equation is:

Figure CN105184027BD00082

[0048] 进一步的,新息协方差方程为: [0048] Further, the innovation covariance equation:

[0049] S ⑹=H ⑹ P (k I k-1) *H ⑹ T+R ⑹ [0049] S ⑹ = H ⑹ P (k I k-1) * H ⑹ T + R ⑹

[0050] 式中,R为量测噪声协方差矩阵, [0050] In the formula, R is the measurement noise covariance matrix,

Figure CN105184027BD00083

[0052] 进一步的,滤波增益矩阵方程: [0052] Further, the filter gain matrix equation:

[0053] W (k) =P (k I k-1) H (k) tS (k) —1 [0053] W (k) = P (k I k-1) H (k) tS (k) -1

[0054] 进一步的,状态更新方程与状态协方差更新方程 [0054] Further, a state update equation of state covariance update equation

[0055] [0055]

Figure CN105184027BD00084

[0056] 进一步的,新息方程为: [0056] Further, the new rate equation is:

Figure CN105184027BD00085

[0058] Pm ⑹=Pl ⑹-Pd ⑹-Ps ⑹ [0058] Pm ⑹ = Pl ⑹-Pd ⑹-Ps ⑹

[0059] Qm ⑹=Ql ⑹-Qd ⑹-Qs ⑹ [0059] Qm ⑹ = Ql ⑹-Qd ⑹-Qs ⑹

[0060] 式中,PM、QM分别是通过对量测数据计算获得的电机的有功无功数值分别是量测的总体的负荷有功无功数值;Pd、Qd分别是配电网的有功无功数值;Ps、Qs分别是静负荷的有功无功数值;ST为电机容量基准值与系统容量基准值的比值,Ipm为电机的比例,Plo负荷初始功率,Uuj是初始电压。 [0060] In the formula, PM, QM, respectively, by reactive active numerical calculation of the motor obtained measured data are measured overall values ​​of active and reactive load; Pd, Qd are the active and reactive power distribution network value; Ps, Qs are the static load values ​​of active and reactive; ST motor capacity reference value ratio of the reference value of the system capacity, Ipm a ratio motor, Plo initial power load, Uuj initial voltage.

Figure CN105184027BD00086

[0061] 进一步的,状态协方差更新方程 [0061] Further, the state covariance update equation

[0062] P (k) =P (k I k-1) -ff (k) S (k) ff (k) τ [0062] P (k) = P (k I k-1) -ff (k) S (k) ff (k) τ

[0063] 进一步的,模型权重计算公式如下: [0063] Further, the model weights calculated as follows:

[0064] 似然函数计算: [0064] The likelihood function is calculated:

Figure CN105184027BD00091

[0067]其中, [0067] wherein,

Figure CN105184027BD00092

[0069] γι是第i个模型的信息误差,S1是第i个模型的新息协方差矩阵,A1是第i个模型的似然函数值,Uj是第j个模型的权重。 [0069] γι is the i-th error information model, S1 is the covariance matrix of the innovation i-th model, A1 is the likelihood function value of the i th model, Uj is the j-th right model weight.

[0070] 进一步的,模型综合输出公式如下: [0070] Further, the total output model equation as follows:

[0071] 通过模型的状态变量计算各个模型的功率吸收的公式如下: [0071] The power absorption of each model calculated by the state variable model of the following formula:

Figure CN105184027BD00093

[0074] P ⑹=Pm (k I k) +Pd ⑹ +Ps ⑹ [0074] P ⑹ = Pm (k I k) + Pd ⑹ + Ps ⑹

[0075] Q ⑹=Qm (k I k) +Qd ⑹ +Qs ⑹ [0075] Q ⑹ = Qm (k I k) + Qd ⑹ + Qs ⑹

[0076] 其中,PM、QM分别为计算出的电机的有功无功;Pd、Qd分别为计算出的配电网的有功无功;Ps、Qs分别为计算出的静负荷的有功无功;P、Q分别为计算出的整个模型的有功无功。 [0076] wherein, PM, QM respectively calculated motor of active and reactive; Pd, Qd are calculated distribution network active and reactive; Ps, Qs are calculated static active and reactive load; P, Q are calculated for the entire model of active reactive.

[0077] 进一步的,通过对各个模型的功率的加权求和获得模型集的功率。 [0077] Further, by weighting each model of the power obtained by summing the power of the model set.

[0078] [0078]

Figure CN105184027BD00094

[0079] 式中, [0079] In the formula,

[0080] [0080]

Figure CN105184027BD00095

Pi、Qi分别为第i个模型的有功无功;PZ、QZ为整个模型集的功率。 Pi, Qi are active and reactive power of the i-th model; PZ, QZ model for the power of the entire set.

[0081] 本发明的有益效果: [0081] Advantageous effects of the invention:

[0082] 1.负荷建模过程是递推实现的,协方差矩阵不需要存储,节省了存储空间。 [0082] 1. Load recursive modeling process is implemented, the covariance matrix need not be stored, storage space is saved.

[0083] 2.状态估计算法采用扩展卡尔曼滤波器,负荷模型的状态变量的估计值可以十分精确,模型的误差很小。 [0083] 2. The state estimation algorithm estimates the value of the extended Kalman filter, state variables of the load model may be very precise, very small error model.

[0084] 3.构建子模型集来调整相应的模型,使得在负荷成分改变的情况下,模型的参数可以自适应调整,因此,模型可以很好的反映负荷的时变性与随机性。 [0084] 3. The set of adjusted model constructs corresponding model, such that in a case where the load component change, the model parameters can be adaptively adjusted so that the model can well reflect the randomness of the load and degeneration.

[0085] 4.算法是递推实现的且计算量比较小。 [0085] 4. The recursive algorithm is implemented and a relatively small amount of calculation. 因此,如果计算机的运算速度足够快的话, 负荷模型可以实时建立。 Thus, if the computer's operating speed is fast enough, the load model may be established in real time.

[0086] 5.本发明之基于交互式多模型算法的电力负荷建模方法适用于针对变电站负荷时变性和随机性的负荷建模,也适用于考虑地域分散性的不同变电站负荷的建模,能够解决负荷时变性造成建模困难的问题,本方法极具工程实用价值。 The [0086] 5. The present invention is based on a power load modeling method of interacting multiple model algorithm for the load applied to the substation and the load modeling random variability, consideration also applies to the geographical dispersion of modeling different substations load, difficulties caused by degeneration of the modeling problem can be solved load, this method of great practical value.

附图说明 BRIEF DESCRIPTION

[0087] 图1本发明之基于交互式多模型算法的电力负荷建模方法结构图; [0087] The method of the invention, power load modeling based on the structure of FIG IMM Algorithm FIG 1;

[0088] 图2通过子模型对相应的模型进行参数调整的算法的结构图; [0088] FIG 2 is a configuration diagram of a parameter adjustment algorithm for the corresponding sub-model by model;

[0089] 图3模型集中负荷模型的结构的示意图。 [0089] The schematic structure of a concentrated load model 3 model of FIG.

具体实施方式: Detailed ways:

[0090] 下面结合附图对本发明进行详细说明: DRAWINGS The invention is described in detail [0090] below with:

[0091] —种基于交互式多模型算法的电力负荷建模方法通过递推运算的形式进行建模, 包括以下步骤: [0091] - Power Load Modeling species interacting multiple model algorithm based on recursive computation modeled by the form, comprising the steps of:

[0092] 步骤一:确定电力负荷模型集中的模型的结构为SLM; [0092] Step a: determining the structure of the power load model centralized model for the SLM;

[0093] 步骤二:设定两级模型集,第一级的模型所包含的模型数比较少用来表征负荷成分,每个模型分配一个权重。 [0093] Step 2: Setting two model set, the model number of the first stage model contains relatively few components used to characterize the load, a weight assigned to each model. 第一级的每个模型对应于第二级模型集中的一个子模型集,第二级的模型集用来调整第一级的模型;设定一个阈值,如果第一级模型的误差小于阈值,则保持模型不变,如果模型的误差在一定时间内大于阈值,则利用第二级模型集中的子模型集对对应的第一级模型进行调整; Each model corresponds to the second stage of the first stage set of a model set of sub-model, the second model set the stage for adjusting the first stage model; set a threshold, if the error is less than a first threshold level model, model remains unchanged, if the model error is greater than a threshold within a certain time, corresponding to the first stage of the model is adjusted by a second stage set of sub-models model set;

[0094] 步骤三:通过马尔科夫状态转移矩阵以及模型的权重,对第一级的各个模型的状态变量和协方差矩阵进行阵进行交互运算; [0094] Step three: the Markov state transition matrix by weight and the weight of the model, each state variable model of the first stage matrix and a covariance matrix calculation interact;

[0095] 步骤四:用扩展卡尔曼滤波器对各个模型的状态变量进行状态估计; [0095] Step IV: Extended Kalman Filter state variables of each model state estimation;

[0096] 步骤五:通过每个模型的状态估计误差计算各个模型的权重值。 [0096] Step Five: an error calculating the weight value of each model is estimated via a state of each model.

[0097] 步骤六:用模型估计的状态变量计算模型的功率,并通过各个模型的权重值计算各个模型的功率加权的和作为模型集的综合输出。 [0097] Step Six: a state variable model estimation model calculation power, and weight values ​​by each individual power weighting calculation model and a model of the total output of the model set.

[0098] 模型集中是设定为两级的,第一级模型直接表示负荷的特性,第一级中含有较少的模型。 [0098] concentration is set to be two models, the first model level directly a characteristic of the load, the first stage contains less model. 第二级的模型集用于调整第一级的模型,因此,每个第一级的模型都对应于一个第二级模型集中的一个子模型集。 Model set for adjusting the second stage the first stage model, therefore, each model of the first stage of the second stage corresponds to a model of a sub-set of models concentrated. SLM是模型的具体结构,第一级和第二级的模型集中的每个模型都是这种结构的。 SLM is a detailed model of the first stage and second stage of each model is a model of such a centralized structure.

[0099] 所述步骤一中,第一级模型集中的模型记为:1〇如1_1^、1〇如1_(:、1〇如1_1?。每个模型对应一个确定的第二级模型集中的子模型集,模型集的区别主要在于电机比例的差别, 其中,Mode 1_L对应于小电机比例的负荷模型集;Model_C对应于中电机比例的负荷模型集; Model_R对应于大电机比例的负荷模型集。每个第二级的子模型集在建立的时候根据参数的变化范围进行离散化,然后再根据参数的具体情况去掉不合理的模型,选取合理的模型, 减少模型的个数,同时模型集中还包含一些极端模型。通过具体的分析,确定各个子模型集。 [0099] a step of the first stage referred to as a model set of models: 1〇 such 1_1 ^, _ 1 as 1〇 (eg :, 1〇 1_1 ?. second stage corresponding to a model of each model to determine the concentration of a sub-set of models, the difference between the model set mainly differences motors ratio, wherein, Mode 1_L corresponding to the small motor proportional load model set; Model_C corresponding to the motor proportional to the load model set; Model_R corresponding to the large motor proportional load models set. each of the second stage sub-sets in the creation of the model according to the change ranges of the parameters discretization, then remove unreasonable model parameters depending on the circumstances, selecting appropriate model, reduce the number of models, and models It also includes a number of terminal models. by detailed analysis, to determine the respective sub-set of models.

[0100] 所述步骤二中,设定一个阈值,当第一级的模型的上一时刻的似然函数值小于阈值,则认为模型合适,不需要调整。 [0100] In step two, a threshold value is set, when the likelihood function on a time model of the first stage is smaller than the threshold value, is considered a suitable model, no adjustment. 如果似然函数值在一定时间内大于阈值,需要用第二级的子模型集调整对应的上一级模型。 If the likelihood function value is greater than the threshold value within a certain time, the need to use a second stage sub-model on a set of adjustments corresponding to the model. 子模型集包含7个合适的模型。 Set contains seven sub-models suitable models. 在k时刻进行模型的遴选,设rgj⑹是第j个模型在k时刻的新息向量,则此模型的似然函数值为: Model selection performed at time k, j-th set rgj⑹ new information model in the vector at time k, the model likelihood function is:

[0101] [0101]

Figure CN105184027BD00111

[0102] 式中, [0102] In the formula,

[0103] Sgj (k) = rgj (k) · rgj (k) τ [0103] Sgj (k) = rgj (k) · rgj (k) τ

[0104] 根据模型集上一次调整时的子模型的权重情况分配各个模型的位置,权重值越大的位置,对应的模型似然函数值越大。 [0104] model according to the weights assigned respective sub-model when the first model set weight adjustment the position, the greater the weight value of the position, the greater the likelihood function corresponding to the model values.

[0105] 设子模型集模型之间的概率转移矩阵为Π。 Probability between [0105] provided submodel set model transition matrix Π. 各个模型权重的计算: Each model to calculate weights:

Figure CN105184027BD00112

[0109] 其中,Ug1是第i个模型的权重值,Cgj是第j个模型的归一化常数,Agj是第j个模型的似然函数值。 [0109] wherein, Ug1 is a weight value of the i th model, Cgj is a normalization constant j-th model, Agj likelihood function is the j-th value model.

[0110] 对各个子模型的参数加权求和得到对应的模型的参数: [0110] The weighting parameters of the respective sub-model summed parametric model corresponding to:

[0111] [0111]

Figure CN105184027BD00113

[0112] 其中,thetai(k)为k时刻子模型集中的第i个模型的参数,Theta⑹为k时刻对应上一级模型的参数。 [0112] wherein, thetai (k) is the i-th model parameter k set time submodel, Theta⑹ parameters corresponding to a model for time k.

[0113] 所述步骤三中模型交互的公式为: [0113] Step III the Formula interaction model are:

[0114] 设第一级模型的马尔科夫状态转移矩阵为Mark,计算模型间的混合概率: [0114] provided a first stage the Markov state transition matrix models Mark, mixing between the probability model calculation:

[0115] [0115]

Figure CN105184027BD00114

[0116] 式中,1,」_ = 1,2,3此^^』为模型1到模型」的转移概率。 [0116] In the formula, 1, "this ^^ _ = 1,2,3" as the transition probability model 1 to model "of. 归一化常数(^可由下式求得: A normalization constant (^ by the following formula:

[0117] [0117]

Figure CN105184027BD00115

[0118] 计算输入混合状态估计值: [0118] mixing the input state estimate is calculated:

Figure CN105184027BD00116

[0121] 其中,Xi(k-1)为模型i在k-1时刻的状态变量; [0121] where, Xi (k-1) i is the model state variable time k-1;

Figure CN105184027BD00117

为混合后模型j的状态变量;Pi (k-Ι)为模型i在k-Ι时刻的状态协方差矩阵 After mixing model is a state variable j; Pi (k-Ι) as the model covariance matrix of state i k-Ι time

Figure CN105184027BD00118

为混合后模型j的状态协方差矩阵。 After mixing matrix is ​​the covariance model state j.

[0122] 所述步骤四中采用扩展卡尔曼滤波器对电机的状态变量进行估计。 [0122] in the Step 4 using the extended Kalman filter for estimating the state variables of the motor. 为了简化扩展卡尔曼滤波器的结构,同时电机的定子电阻对系统的动态特性影响较小,将电机的定子电阻Rs确定为0。 In order to simplify the structure of the extended Kalman filter, while the motor stator resistance has little effect on the dynamics of the system, the motor stator resistance Rs is determined to be 0.

[0123] 电机的状态预测方程为: Status [0123] Motor prediction equation is:

Figure CN105184027BD00121

[0127] 式中,Ed是电机的次暂态电势在d轴的分量,Eq是电机的次暂态电势在q轴的分量,s 是电机的转差率;TS为采样周期,WB为同步角速度;TdO为暂态开路时间常数,X为转子开路电抗,X '为转子不动时短路电抗;K1为电机的负荷率,a为电机恒定转矩系数,MP与转速有关的转矩的方次;Uid、Uu分别为母线电压在d、q轴的分量。 [0127] In the formula, Ed is a component of sub-transient potential of the motor in the d-axis, Eq views transient potential is the q-axis component of the motor, s is the slip of the motor; the TS is the sampling period, WB is synchronized angular velocity; TDO is a transient open time constant, X is the open rotor reactance, X 'is a short-circuit does not move the rotor reactance; Kl rate of the motor as a load, a is a constant motor torque coefficient, MP related to the rotational speed side torque views; Uid, Uu respectively bus voltage in the d, q axis components.

[0128] 状态协方差预测方程: [0128] Covariance prediction equation of state:

[0129] P (k I k-1) =F (k-1) P (k-1) F (k-1) T+Q (k-1) [0129] P (k I k-1) = F (k-1) P (k-1) F (k-1) T + Q (k-1)

[0130] 其中,Q为过程噪声协方差矩阵, [0130] wherein, Q is the process noise covariance matrix,

Figure CN105184027BD00122

[0132] 量测预测方程为: [0132] Measurement prediction equation is:

Figure CN105184027BD00123

[0135] 新息协方差方程为: [0135] innovation covariance equation:

[0136] S ⑹=H ⑹ P (k I k-1) *H ⑹ T+R ⑹ [0136] S ⑹ = H ⑹ P (k I k-1) * H ⑹ T + R ⑹

[0137] 式中,R为量测噪声协方差矩阵, [0137] In the formula, R is the measurement noise covariance matrix,

Figure CN105184027BD00124

[0139] 滤波增益矩阵方程 [0139] filter gain matrix equation

[0140] W ⑹=P (k I k-1) H ⑹ tS ⑹―1 [0140] W ⑹ = P (k I k-1) H ⑹ tS ⑹-1

[0141] 状态更新方程与状态协方差更新方程 [0141] update equation of state covariance update equation

[0142] [0142]

Figure CN105184027BD00131

[0143] 新息方程为: [0143] New information equation:

[0144] [0144]

Figure CN105184027BD00132

[0145] Pm (k) = Pl (k) -Pd (k) -Ps (k) [0145] Pm (k) = Pl (k) -Pd (k) -Ps (k)

[0146] Qm ⑹=Ql ⑹-Qd ⑹-Qs ⑹ [0146] Qm ⑹ = Ql ⑹-Qd ⑹-Qs ⑹

[0147] 式中,PM、QM分别是通过对量测数据计算获得的电机的有功无功数值分别是量测的总体的负荷有功无功数值;Pd、Qd分别是配电网的有功无功数值;Ps、Qs分别是静负荷的有功无功数值;ST为电机容量基准值与系统容量基准值的比值 [0147] In the formula, PM, QM, respectively, by reactive active numerical calculation of the motor obtained measured data are measured overall values ​​of active and reactive load; Pd, Qd are the active and reactive power distribution network value; Ps, Qs are the static load values ​​of active and reactive; ST motor capacity reference value ratio of the reference value of the system capacity

Figure CN105184027BD00133

· Kpm为电机的比例,Plo负荷初始功率,Uuj是初始电压。 · Kpm a ratio motor, Plo initial power load, Uuj initial voltage.

[0148] 状态协方差更新方程 [0148] Covariance update equation of state

[0149] P ⑹=P (k I k-1) -W ⑹ S ⑹ W ⑹ τ [0149] P ⑹ = P (k I k-1) -W ⑹ S ⑹ W ⑹ τ

[0150] 所述步骤五中模型权重计算公式如下: [0150] The fifth step of weighting model is calculated as follows:

[0151] 似然函数计算: [0151] likelihood function calculated:

Figure CN105184027BD00134

[0154] 其中, [0154] wherein,

[0155] [0155]

Figure CN105184027BD00135

[0156] γι是第i个模型的信息误差,S1是第i个模型的新息协方差矩阵,A1是第i个模型的似然函数值,Uj是第j个模型的权重。 [0156] γι is the i-th error information model, S1 is the covariance matrix of the innovation i-th model, A1 is the likelihood function value of the i th model, Uj is the j-th right model weight.

[0157] 所述步骤六中模型综合输出公式如下: [0157] The total output step model equation six as follows:

[0158] 通过模型的状态变量计算各个模型的功率吸收的公式如下: [0158] the power absorption of each model calculated by the state variable model of the following formula:

Figure CN105184027BD00136

[0161] P ⑹=PM (k I k) +Pd ⑹ +Ps ⑹ [0161] P ⑹ = PM (k I k) + Pd ⑹ + Ps ⑹

[0162] Q ⑹=Qm (k I k) +Qd ⑹ +Qs ⑹ [0162] Q ⑹ = Qm (k I k) + Qd ⑹ + Qs ⑹

[0163] 其中,Pm、Qm分别为计算出的电机的有功无功;Pd、Qd分别为计算出的配电网的有功无功;Ps、Qs分别为计算出的静负荷的有功无功;P、Q分别为计算出的整个模型的有功无功。 [0163] where, Pm, Qm were calculated by the active and reactive electric machine; Pd, Qd are active distribution network calculated reactive power; Ps, Qs are calculated static active and reactive load; P, Q are calculated for the entire model of active reactive.

[0164] 通过对各个模型的功率的加权求和获得模型集的功率。 [0164] obtained by summing the power model set by weighting each model's power.

[0165] [0165]

Figure CN105184027BD00141

[0166] 式中, [0166] In the formula,

[0167] [0167]

Figure CN105184027BD00142

Pi、Qi分别为第i个模型的有功无功;PZ、QZ为整个模型集的功率。 Pi, Qi are active and reactive power of the i-th model; PZ, QZ model for the power of the entire set.

[0168] 所述步骤一中的模型结构选择SLM结构。 [0168] The model structure selected in step a SLM structure.

[0169] 所述步骤一中,模型集记为:[Model_L Model_C Model_R]。 [0169] In the step a, referred to as a model set: [Model_L Model_C Model_R].

[0170] 所述步骤二、三、四、五、六中,电力负荷的递推建模均在MATLAB仿真平台中完成。 [0170] The step two, three, four, five, six, the average power load recursive modeling in MATLAB simulation platform is completed.

[0171] 为了更好的说明本发明,更为详细的说明为: [0171] In order to better illustrate the present invention, as described in more detail:

[0172] 如图1所示,设在k时刻对电力负荷进行递推建模: [0172] As shown, provided in the time k to the power load for a recursive model:

[0173] XL (k-1)、XC (k-1)、XR (k-1)分别是第一级模型集中3个模型在k-1时刻的状态变量; Pl (k-1)、PC (k-1)、PR (k-Ι)分别是模型在k-Ι时刻的状态协方差矩阵;u (k-Ι)为模型在k-Ι时刻的权重。 [0173] XL (k-1), XC (k-1), XR (k-1) are first concentrated three level model variables model the state of the time k-1; Pl (k-1), PC (k-1), PR (k-Ι) are the model covariance matrix in the k-Ι state time; u (k-Ι) is the weight of the model in the time k-Ι weight. 通过步骤三中模型交互的公式进行模型的状态变量与协方差矩阵交互,获得交互后的模型的状态变量XLe(kl)、XCe(k_l)、XRe(k_l)以及模型的协方差矩阵PLe(k_l)、PCe (k-1)、PR(3 (k-1)。然后分别将各个模型对应的状态变量和协方差矩阵输入各自的扩展卡尔曼滤波器进行模型的状态估计,状态估计按步骤四中的公式进行。得到模型估计的k时刻的状态变量Xl⑹、Xc⑹、Xr⑹以及模型的协方差矩阵Pl⑹、Pc⑹、Pr⑹。协方差矩阵用于下一时刻的状态估计过程。模型状态估计过程的似然函数值是AUk)、Ac(k)、AR(k);信息协方差矩阵是SL(k)、SC⑹、SR(k)。 State variable covariance matrix interaction and coordination modeled by three mathematical expression model interaction step, a state variable XLe model the obtained interaction (kl), XCe (k_l), XRe (k_l) and a model covariance matrix PLe (k_l ), PCe (k-1), PR (3 (k-1). then each model state corresponds to the state variables of each model and the input covariance matrix for the respective extended Kalman filter estimated by the state estimation step four in formula. obtained state variable estimation model at time k Xl⑹, the covariance matrix Pl⑹ Xc⑹, Xr⑹ and model, Pc⑹, Pr⑹. covariance matrix for the next time state estimation process. like model state estimation process However, the function value is AUk), Ac (k), AR (k); the information is the covariance matrix SL (k), SC⑹, SR (k). 根据这些数据利用步骤五中的公式计算模型的权重。 Calculated according to the formula model using data weights step 5. 模型的权重用于下一时刻的状态交互以及当前时刻的模型功率输出融合。 Model of the power output of the model state right weight for next time interaction and integration of the current time. 根据各个模型状态估计过程的似然函数值AUk)、Ac(k)、Ar 00判断模型是否需要调整,如果模型需要调整则利用对应的第二级模型集中的子模型集通过步骤二的公式进行调整,调整后的模型参数用于下一时刻模型的计算;如果模型不需要调整,则保持不变。 The likelihood function value AUk like each model state estimation process), Ac (k), Ar 00 determines whether to adjust the model if the model needs to be adjusted using a model corresponding to a second set of sub-level model by the equation set in step two is performed adjusting the model parameters adjusted model is used to calculate the next time; if the model does not require adjustment, it remains unchanged. 根据步骤六中的公式利用各个模型的状态变量的估计值计算当前时刻的模型的功率。 Using the estimated value of each model according to the formula in step six state variables of the model to calculate the power of the current time. 根据各个模型的权重将各个模型的功率加权求和,所得结果作为模型集的整体功率吸收情况。 The power weighted sum weight of each model, the model obtained as a result of the overall power absorption set according to the weight of each model.

[0174] 图2所示为通过利用第二级中的子模型集对相应的第一级模型进行参数调整的算法的结构图。 [0174] FIG. 2 shows a configuration diagram for parameter adjustment algorithm corresponding to a first stage of the model by using the set of sub-models in the second stage. 如果模型误差在一段时间内大于阈值,则按图2所示对模型进行参数调整。 If the model error is greater than a threshold period of time, as shown in FIG press model parameter adjustment 2. X (k-Ι)是需要调整的模型在k-Ι时刻的状态变量,利用这一状态变量对子模型集中的每个模型求取功率并计算其相应的误差。 X (k-Ι) the model needs to be adjusted in the state variable k-Ι time, with each sub-model of the state variable model of centralized power and obtaining a corresponding calculated error. 根据模型集上一次调整时的子模型的权重情况分配各个模型的位置,权重值越大的位置,对应的模型似然函数值越大。 The weights assigned to each sub-model when the model of a weight adjustment model set the position, the greater the weight value of the position, the greater the likelihood function corresponding to the model values. 再根据步骤二中相应的公式计算各个模型的权重值,通过将各个子模型加权求和得到对应的上一级模型的参数。 Then calculating the weight value of each model according to the procedure corresponding formula II, by a weighted sum of each sub-model to obtain the model parameters corresponding to a.

[0175] 图3所示为两层模型集中负荷模型的基本结构:SLM结构。 [0175] Figure 3 shows the basic structure of the model to two concentrated loads Model: SLM structure. 本模型结构计及了配电网的作用,包含电动机模型和静负荷模型。 This effect accounts for the model structure of the distribution network, comprising a motor model and static load model. 静模型采用幂函数模型,电机模型采用3阶动态模型。 Static power function model using the model, the motor model using the third-order dynamic model.

[0176] 上述虽然结合附图对本发明的具体实施方式进行了描述,但并非对本发明保护范围的限制,所属领域技术人员应该明白,在本发明的技术方案的基础上,本领域技术人员不需要付出创造性劳动即可做出的各种修改或变形仍在本发明的保护范围以内。 [0176] The combination with drawings of specific embodiments of the present invention have been described, but not limit the scope of the present invention, those skilled in the art should understand that, on the basis of the technical solution of the present invention, those skilled in the art without paying creative work to make various modifications or variations are still within the scope of the present invention.

Claims (9)

1. 一种基于交互式多模型算法的电力负荷建模方法,其特征是,包括以下步骤: 步骤一:确定电力负荷模型集中的模型的结构为SLM模型; 步骤二:设定两级模型集,第一级模型集的模型所包含的模型数比较少用来表征负荷成分,每个模型分配一个权重,第一级模型集的每个模型对应于第二级模型集中的一个子模型集,第二级的模型集用来调整第一级的模型:设定一个阈值,如果第一级模型集的模型的误差小于阈值,则保持模型不变,如果模型的误差在一定时间内大于阈值,则利用第二级模型集中的子模型集对对应的第一级模型进行调整; 步骤三:将第一级模型集中每个模型在k-1时刻的状态变量、状态协方差矩阵进行交互运算,获得交互后的各个模型的状态变量以及模型的协方差矩阵; 步骤四:分别将第一级模型集中的各个模型对应的状态变量和协方 1. A power load modeling method of interacting multiple model algorithm, characterized in that, comprising the following steps: Step one: determine the model of the electric load model focused SLM model structure; Step 2: Setting two sets Model , the number of first level model model model set contains relatively few components used to characterize the load, each model is assigned a weight, for each model of the first stage corresponding to a second model set level model a set of sub-model set, model set for the second stage of the first stage to adjust the model: setting a threshold value, if the error of the model of the first level model is less than the set threshold value, the model remains unchanged, if the model error is greater than a threshold within a certain time, a first stage corresponding to the model is adjusted by a second stage set of sub-models model set; step three: the first level model for each model in the concentrated state variable time k-1, the state covariance matrix interaction operation, covariance state of each model, and the model of interactive variables obtained matrix; step four: the first level model are set corresponding to the respective model state variables and covariance 矩阵输入各自的扩展卡尔曼滤波器进行模型的状态估计,得到模型估计的k时刻的状态变量以及模型的协方差矩阵; 步骤五:利用步骤四中模型状态估计过程的似然函数值及信息协方差矩阵计算第一级模型集中的各个模型的权重; 步骤六:用模型估计的状态变量计算模型的功率,并通过各个模型的权重值计算各个模型的功率的加权和,作为模型集的综合输出,模型的权重用于下一时刻的状态交互以及当前时刻的模型功率输出融合。 Covariance matrix of a respective input state matrix for the extended Kalman filter model estimation, estimation of the model obtained at time k and the state variable model; Step five: the step of utilizing four state model and likelihood function value information RA estimation process covariance matrix calculating a weight of each model collection to a first stage weight; step six: estimation model state variables calculated power model, and by the weight value of each model calculated weighted power of each model and the integrated output as model set the model output power state model of interaction right weight for the next time and the current time of integration.
2. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,根据模型状态估计过程的似然函数值判断模型是否需要调整,如果模型需要调整则利用相应的第二级模型集中的子模型集进行调整,调整后的模型参数用于下一时刻模型的计算;如果模型不需要调整,则保持不变。 2. one of the claims 1 to power load modeling method of interacting multiple model algorithm, characterized in that the value of the likelihood function determines whether to adjust the model based on estimated process model state, if the model needs to be adjusted using the set model respective submodel second concentration stage to adjust the model parameters for calculating the next time adjustment model; if the model does not require adjustment, remains unchanged.
3. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,对子模型集进行调整时,设定一个阈值,当模型的上一时刻的似然函数值小于阈值,则认为模型合适,不需要调整,如果似然函数值在一段时间内大于阈值,需要用第二级子模型集调整对应的第一级模型集的模型,在k时刻进行模型的遴选,设(k)是第j个模型在k时刻的新息向量,则此模型的似然函数值为: 3. one of the claims 1 to power load modeling method of interacting multiple model algorithm, characterized in that, when the sub-set of the model to be adjusted to set a threshold value, a time when the likelihood of the model function value is less than the threshold value, it is considered a suitable model, no adjustment, if the likelihood function value is greater than the threshold over time, the model needs to use a second set of a first-stage sub-model adjustment stage model set corresponding to the model at time k selection, set (k) is the j-th new information model in the vector at time k, the model likelihood function is:
Figure CN105184027BC00021
式中, In the formula,
Figure CN105184027BC00022
进一步的,设子模型集模型概率转移矩阵为Π,各个模型权重的计算: Further, the sub-set model set to [pi model probability transition matrix, heavy weights calculated for each model:
Figure CN105184027BC00023
其中,Ug1是第i个模型的权重值,Cgj是第j个模型的归一化常数,Agj是第j个模型的似然函数值。 Wherein, Ug1 is a weight value of the i th model, Cgj is a normalization constant j-th model, Agj likelihood function is the j-th value model.
4. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,对各个子模型的权重加权求和得到对应的模型的参数: 4. one of the claims 1 to power load modeling method of interacting multiple model algorithm, characterized in that, on the right of each sub-model parameters corresponding to the weight weighted sum obtained model:
Figure CN105184027BC00031
其中,thetai⑹为k时刻子模型集中的第i个模型的参数,Theta⑹为k时刻对应上一级模型的参数。 Wherein, thetai⑹ is the i-th model parameter k set time submodel, Theta⑹ parameters corresponding to a model for time k.
5. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,模型交互时,具体为: 设第一级模型集的模型的马尔科夫状态转移矩阵为Mark,计算模型间的混合概率: 5. The one of the power load of claim 1 interacting multiple model modeling algorithm, wherein, when the interactive model, specifically: the state of the Markov model is provided in a first stage of the transfer matrix model set as Mark, the probability of mixing between the model calculations:
Figure CN105184027BC00032
式中,i,j = 1,2,3 ;Mark (i,j)为模型i到模型j的转移概率,归一化常数Cj可由下式求得: Wherein, i, j = 1,2,3; Mark (i, j) is the transition probability model to the model i to j, Cj normalization constant determined by the following formula:
Figure CN105184027BC00033
计算输入混合状态估计值: Mixing the input state estimate is calculated:
Figure CN105184027BC00034
其中,Xi (k-1)为模型i在k-1时刻的状态变量; Where, Xi (k-1) i is the model state variable time k-1;
Figure CN105184027BC00035
为混合后模型j的状态变量; Pi (k-Ι)为模型i在k-Ι时刻的状态协方差矩阵; After mixing model is a state variable j; Pi (k-Ι) covariance matrix of the k-Ι state time for the model I;
Figure CN105184027BC00036
为混合后模型j的状态协方差矩阵。 After mixing matrix is ​​the covariance model state j.
6. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,采用扩展卡尔曼滤波器对电机的状态变量进行估计,为了简化扩展卡尔曼滤波器的结构,同时电机的定子电阻对系统的动态特性影响较小,将电机的定子电阻Rs确定为0。 6. one of the claims 1 to power load modeling method of interacting multiple model algorithm, characterized in that, using the extended Kalman filter for estimating the state variables of the motor, in order to simplify the extended Kalman filter structure, while the motor stator resistance has little effect on the dynamics of the system, the motor stator resistance Rs is determined to be 0.
7. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,模型权重计算公式如下: 似然函数计算: 7. The one of the claims 1 to power load modeling method Interacting Multiple Model algorithm, characterized in that the weight of the model is calculated as follows: calculated likelihood function:
Figure CN105184027BC00037
其中, among them,
Figure CN105184027BC00038
Y1是第i个模型的信息误差,S1是第i个模型的新息协方差矩阵,A1是第i个模型的似然函数值,叫是第j个模型的权重,归一化常数Cj。 Y1 is the information error of the i-th model, Sl is the innovation covariance matrix for the i-th model, A1 is the likelihood function value of the i th model, called is the weight of the j th model weight normalization constant Cj.
8. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,模型综合输出公式如下: 通过模型的状态变量计算各个模型的功率吸收的公式如下: 1 one of the electric load as claimed in claim interacting multiple model modeling algorithm, characterized in that the total output model equation is as follows: each model is calculated by the state variable model of the power absorption following formula:
Figure CN105184027BC00041
其中,Pm、Qm分别为计算出的电机的有功无功;Pd、Qd分别为计算出的配电网的有功无功; Ps、Qs分别为计算出的静负荷的有功无功;P、Q分别为计算出的整个模型的有功无功。 Wherein, Pm, Qm were calculated reactive power of active motor; Pd, Qd are active calculated reactive power distribution network; Ps, Qs are active static load calculated reactive power; P, Q They were calculated for the entire model of the active and reactive.
9. 如权利要求1所述的一种基于交互式多模型算法的电力负荷建模方法,其特征是,通过对各个模型的功率的加权求和获得模型集的功率; 1, one of the electric load as claimed in claim interacting multiple model modeling algorithm, characterized in that, by weighting each model of the power obtained by summing the power of the model set;
Figure CN105184027BC00042
式中, In the formula,
Figure CN105184027BC00043
,Pi、Qi分别为第i个模型的有功无功;PZ、QZ为整个模型集的功率,Ui⑹是第i个模型在k时刻的权重。 , Pi, Qi are active and reactive power of the i-th model; PZ, QZ model for the power of the entire set, Ui⑹ model is the weight of the i th weight at time k.
CN201510717484.6A 2015-10-29 2015-10-29 Power load modeling method based IMM Algorithm CN105184027B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510717484.6A CN105184027B (en) 2015-10-29 2015-10-29 Power load modeling method based IMM Algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510717484.6A CN105184027B (en) 2015-10-29 2015-10-29 Power load modeling method based IMM Algorithm

Publications (2)

Publication Number Publication Date
CN105184027A CN105184027A (en) 2015-12-23
CN105184027B true CN105184027B (en) 2018-04-06

Family

ID=54906106

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510717484.6A CN105184027B (en) 2015-10-29 2015-10-29 Power load modeling method based IMM Algorithm

Country Status (1)

Country Link
CN (1) CN105184027B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104502907A (en) * 2014-12-15 2015-04-08 西安电子工程研究所 Stable ground moving/static target tracking method for airborne radar
CN104573190A (en) * 2014-12-15 2015-04-29 东南大学 Target tracking method based on interactive multiple models
CN104898115A (en) * 2015-05-11 2015-09-09 电子科技大学 Multi-target tracking method after through-wall radar imaging

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7599861B2 (en) * 2006-03-02 2009-10-06 Convergys Customer Management Group, Inc. System and method for closed loop decisionmaking in an automated care system

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104502907A (en) * 2014-12-15 2015-04-08 西安电子工程研究所 Stable ground moving/static target tracking method for airborne radar
CN104573190A (en) * 2014-12-15 2015-04-29 东南大学 Target tracking method based on interactive multiple models
CN104898115A (en) * 2015-05-11 2015-09-09 电子科技大学 Multi-target tracking method after through-wall radar imaging

Also Published As

Publication number Publication date
CN105184027A (en) 2015-12-23

Similar Documents

Publication Publication Date Title
Chaturvedi et al. Load frequency control: a generalised neural network approach
Eidiani et al. FAD-ATC: a new method for computing dynamic ATC
US7979239B2 (en) System and method for monitoring and managing electrical power transmission and distribution networks
Chiang et al. A BCU method for direct analysis of power system transient stability
Glavic et al. Wide-area detection of voltage instability from synchronized phasor measurements. Part II: Simulation results
Kalcon et al. Small-signal stability analysis of multi-terminal VSC-based DC transmission systems
Zhang et al. Further results on delay-dependent stability of multi-area load frequency control
Huang et al. Quasi-Monte Carlo based probabilistic small signal stability analysis for power systems with plug-in electric vehicle and wind power integration
Pizano-Martinez et al. A new practical approach to transient stability-constrained optimal power flow
Venayagamoorthy Online design of an echo state network based wide area monitor for a multimachine power system
Maiti et al. An adaptive speed sensorless induction motor drive with artificial neural network for stability enhancement
Ray et al. Wide-area signal-based optimalneurocontroller for a UPFC
Li et al. State of charge estimation for LiMn2O4 power battery based on strong tracking sigma point Kalman filter
Jurado Modeling SOFC plants on the distribution system using identification algorithms
Gadoue et al. Stator current model reference adaptive systems speed estimator for regenerating-mode low-speed operation of sensorless induction motor drives
Kojabadi Active power and MRAS based rotor resistance identification of an IM drive
Biegel et al. Distributed low-complexity controller for wind power plant in derated operation
Nguyen et al. Sensitivity approaches for direct computation of critical parameters in a power system
Wang et al. Direct adaptive neural control for stabilization of nonlinear time-delay systems
CN102324743B (en) Screening method for expected fault in online transient safety and stability assessment of power system
CN101404412B (en) Method for static electric voltage stability analysis
Chowdhury et al. A novel aggregated DFIG wind farm model using mechanical torque compensating factor
CN100521443C (en) Excitation control method based on non linear robust power system stabilizer
CN101625389B (en) Power System Stability analysis method considering characteristics of the load change
CN100444494C (en) Coherency generating sets of electric power system and polymeric method of asynchronous motor group parameter

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
GR01