基于ANFIS网络水电机组控制系统建模

利用模糊神经网络ANFIS较强的非线性逼近能力建立辨识模型，对水电机组控制系统输入、输出特性进行了辨识。辨识采用离线训练ANFIS网络和在线辨识相结合的方法，模型能很好地辨识系统输入、输出特性，为研究智能水轮发电机组控制策略提供了有效的建模方法。     

基于神经网络预测控制的单元机组协调控制策略

利用BP神经网络的非线性映射能力对单元机组协调控制系统被控对象进行辨识,从而建立其动态模型;在这一模型的基础上对协调控制系统中的控制器参数优化进行研究,提出基于神经网络预测控制的协调控制策略.该方法很好地解决了协调控制系统中强耦合、非线性等问题.仿真实验表明该系统的跟踪速度加快、调节精度提高、并且具有较好的抗干扰性.图6参7     

热力系统局部非线性模型辨识

在多模型控制中,局部模型大多数是基于线性模型,其数量和精度影响多模型控制的效果。提出一种基于RBF神经网络的非线性模型辨识算法,采用G.B.Sentoni等人提出的非线性模型结构[2~3],利用径向基函数(Radial Basis Function,RBF)神经网络的逼近能力,实现热力系统非线性模型辨识。在RBF神经网络的学习过程中,根据性能函数调节学习率,可以加快学习的收敛过程。最后进行了仿真验证,基于2个局部非线性模型的多模型控制系统与基于5个局部线性模型的多模型控制系统相比,减少了切换时的震荡,控制精度有所提高。试验结果表明,该辨识算法能减少固定模型数量,从而减少模型搜索时间,并且能够提高模型预测精度。     

基于混沌RBF神经网络的汽油机瞬态工况油膜参数辨识研究

提出了基于混沌径向基(RBF)神经网络的汽油机瞬态工况油膜参数辨识方法。利用混沌优化算法确定隐含层高斯函数径向基中心和输出层连接权值,使其达到全局最优,有效地提高RBF神经网络的收敛速度;同时,利用混沌算法训练RBF神经网络,使目标函数取全局最小值或逼近全局最小值,有效地提高辨识模型的辨识精度,并与BP神经网络模型及最小二乘法辨识进行了分析和比较。仿真结果表明:混沌RBF神经网络模型收敛速度快,具有更强的非线性辨识能力,能够有效地提高油膜动态参数的辨识精度,进而得出不同工况下的油膜参数动态特征。     

基于果蝇优化算法的GRNN水电机组状态趋势预测

针对水电机组振动的非线性、非平稳特性,提出了一种基于果蝇优化算法（FOA）的广义回归神经网络（GRNN）模型（FOAGRNN ）,实现了GRNN分布参数的优化选择,并对四川省新政航电工程3台机组5个不同部位的振动序列峰峰值进行了预测,与BP神经网络预测结果的均方误差（MSE）对比结果表明,FOAGRNN预测精度较高。     

基于混合算法下RBF神经网络的执行机构非线性特性在线辨识与补偿

针对控制系统中执行机构非线性特性在线辨识及补偿问题,研究了一种基于变步长核最小均方(SVSKLMS)和遗传算法结合的混合径向基(VHRBF)神经网络。利用径向基(RBF)神经网络不依赖于精确的数学模型即可得到被控对象信息的特点,建立了控制系统执行机构的非线性特性模型;为解决传统RBF神经网络辨识性能差的问题,使用遗传算法(GA)对神经网络的中心向量和方差进行优化,利用SVSKLMS算法对RBF神经网络模型中的权重进行优化,进而得到最佳的RBF神经网络。基于VHRBF神经网络及其逆模型补偿器对执行机构非线性特性进行在线辨识及补偿。仿真结果表明：与其他算法训练下的RBF神经网络相比,所提出的VHRBF神经网络能够精确辨识并补偿执行机构的非线性特性,并且具有更快的收敛速度、更优的收敛性能。     

基于贝叶斯优化的LSTM火电机组一次调频能力辨识方法

提升火电机组的一次调频能力辨识有助于辅助电网的调度,保证电网的安全稳定运行。提出一种基于贝叶斯优化算法(Bayesian optimization, BO)的长短期记忆网络(long short term memory, LSTM)一次调频能力辨识方法,实现火电机组的一次调频能力精确建模。首先对机组机理及参数之间的相关性进行分析,确立模型的输入特征变量,再利用贝叶斯算法对LSTM网络结构进行优化,得到一次调频能力辨识模型。以某600 MW燃煤火电机组为研究对象,将该模型与传统BP神经网络模型、未优化LSTM网络模型进行对比。结果表明：所提出的网络模型均方根误差分别降低了66.51%和34.83%,具有更高的模型精度。     

基于神经网络的热工过程动态建模

针对热交换器的滞后、时变、非线性特性,提出一种新的基于工作点的换热器神经网络动态建模方法,即利用静态网络加动态环节的方法建立换热器系统的神经网络动态模型,称为灰箱式动态模型。算法推导过程中采用矩阵形式,易于理解和编程,模型经过训练后,预测结果准确。     

水轮发电机组的神经网络可辨识性研究

在分析了水电机组作为准线性同构异参系统有关特性基础上，给出了神经网络可辨识性的定义，论证了ＢＩＢＳ稳定的水轮发电机的神经网络可辨识性，对仿真模型和辨识网络的计算结果的分析比较表明神经网络具有良好的映射性能及较强的容错性。     

基于神经网络模型的锅炉主蒸汽温度预测控制

为解决传统控制方法在火力发电机组蒸汽温度控制过程中存在的强非线性、大迟延的难题,提出了一种基于长短期记忆(LSTM)神经网络在线估计和粒子群(PSO)滚动优化的预测控制算法。该方法将常规串级控制系统的主回路控制器用预测控制器替代,采用LSTM神经网络建立主蒸汽温度控制系统的过程模型,通过多步预测实现了对复杂非线性系统模型的精确预测。利用PSO算法在线求解主蒸汽温度控制系统的最优预测控制律,避免了传统递推方法无法直接求解非线性优化问题。仿真结果表明：与传统主蒸汽温度串级控制策略相比,该控制算法明显改善了控制系统的快速性,抗扰能力较强,对主蒸汽温度这类具有非线性及模型不精确的被控对象有一定的参考价值。     

A neural net model-based multivariable long-range predictivecontrol strategy applied in thermal power plant control

A constrained multivariable control strategy along with its application in more efficient thermal power plant control is presented in this paper. A neural network model-based nonlinear long-range predictive control algorithm is derived, which provides offset-free closed-loop behavior with a proper and consistent treatment of modeling errors and other disturbances. A multivariable controller is designed and implemented using this algorithm. The system constraints are taken into account by including them in the control algorithm using real-time optimization. By running a simulation of a 200 MW oil-fired drum-boiler thermal power plant over a load-profile along with suitable PRBS signals superimposed on controls, the operating data is generated. Neural network (NN) modeling techniques have been used for identifying global dynamic models (NNARX models) of the plant variables off-line from the data. To demonstrate the superiority of the strategy in a MIMO case, the controller has been used in the simulation to control main steam pressure and temperature, and reheat steam temperature during load-cycling and other severe plant operating conditions     

Predictive control of SOFC based on a GA-RBF neural network model

Transients in a load have a significant impact on the performance and durability of a solid oxide fuel cell (SOFC) system. One of the main reasons is that the fuel utilization changes drastically due to the load change. Therefore, in order to guarantee the fuel utilization to operate within a safe range, a nonlinear model predictive control (MPC) method is proposed to control the stack terminal voltage as a proper constant in this paper. The nonlinear predictive controller is based on an improved radial basis function (RBF) neural network identification model. During the process of modeling, the genetic algorithm (GA) is used to optimize the parameters of RBF neural networks. And then a nonlinear predictive control algorithm is applied to track the voltage of the SOFC. Compared with the constant fuel utilization control method, the simulation results show that the nonlinear predictive control algorithm based on the GA-RBF model performs much better.     

Application of neural network for the modeling and control of evaporative condenser cooling load

Artificial neural network (ANN), in comparison with PID controllers which have broad applications in the highly complex HVAC systems, has recently received more attention. The present paper includes thermodynamic modeling of an evaporative condenser under steady state and transient state conditions for establishing control of thermal capacity, using Artificial neural network. To train the system under dynamic condition, predictive neural network, capable of understanding dynamic behavior and predicting the preset output is used. The principle operation of such neural networks is based on the reduction of gradients of errors existing between the predicted output and the actual output of the system. To control the system thermal capacity, neural controller based on training received from the reduction of gradients between the output controller and the ideal output, is used. Results obtained during present investigation indicate that artificial neural network controller is suitable substitute for PID controllers for thermal systems.     

基于互信息和PID神经网络的SCR脱硝扰动补偿控制

针对火电厂选择性催化还原(Selective Catalytic Reduction,SCR)烟气脱硝系统机理复杂,工况变化时呈现的不确定性、强扰动等特点,提出了一种基于互信息和PID神经网络的SCR烟气脱硝扰动补偿控制方法。利用PID前向神经网络的学习性能逼近被控对象的逆构成扰动观测器对系统进行反馈补偿,以达到超前消除系统扰动的目的。选取观测扰动和系统扰动的互信息为目标函数,采用改进的帝国竞争算法实现PID神经网络权值的优化调整。设计鲁棒PID控制器来进一步克服被控对象存在的不确定性。仿真实验表明,该方法具有突出的抗干扰能力和较好的鲁棒性,控制品质优于常规的PID控制。     

A new RBFN based MPPT controller for grid-connected PEMFC system with high step-up three-phase IBC

Extraction of maximum power from a proton exchange membrane fuel cell (PEMFC) power source is necessary for its economical and optimal utilization. In this paper, a neural network based maximum power point tracking (MPPT) controller is proposed for the grid-connected PEMFC system. Radial basis function network (RBFN) algorithm is implemented in the neural network controller to extract the maximum power from PEMFC. A high step-up three-phase interleaved boost converter (IBC) is also designed in order to reduce the current ripples coming out from the PEMFC. Interleaving technique provides high power capability and reduces the voltage stress on the power semiconductor devices. The performance analysis of the proposed RBFN MPPT controller is analyzed in MATLAB/Simulink platform for both standalone as well as for the grid-connected PEMFC system.     

模糊自学习控制在水电机组控制中应用研究

采用具有自学习能力的自适应模糊控制器来控制水电机组运行。自适应模糊控制器将模糊控制和神经网络结合,根据运行情况在线调整模糊推理规则和隶属函数,使控制系统具有自适应学习的特性。学习中学习速率和平滑因子可根据误差情况在线修改,克服了网络学习速度慢和局部最优的缺点。仿真实验表明,设计的自适应模糊控制器具有良好的鲁棒性,可有效地改善水轮发电机组系统的动、静态性能。     

Laboratory implementation of a neural network trajectory controllerfor a DC motor

The laboratory implementation of a neural network controller for high performance DC drives is described. The objective is to control the rotor speed and/or position to follow an arbitrarily selected trajectory at all times. The control strategy is based on indirect model reference adaptive control (MRAC). The motor characteristics are explicitly identified through a multilayer perceptron type neural network. The output of the trained neural network is used to drive the motor in order to achieve a desired time trajectory of the controlled variable. The neural network controller is assembled in a commercially available PC-based real-time control system shell, using software subroutines. An H-bridge, DC/DC voltage converter is interfaced with the computer to generate the specified terminal voltage sequences for driving the motor. All software and hardware components are off the shelf. The versatility of the motor/controller arrangement is displayed through real-time plots of the controlled states     

基于改进的RAN网络的预测控制及其应用

提出一种基于改进的RAN网络的非线性预测控制方法。这种方法利用多输入多输出的、改进的RAN网络建立对象的多步预测模型,采用简化牛顿迭代法作为优化算法来求取预测控制量。以电厂过热汽温为对象进行了仿真研究。结果表明:采用该方法的神经网络预测模型精度高、泛化能力强,控制性能良好。图5参9     

Optimum Pareto design of non-linear predictive control with multi-design variables for PEM fuel cell

The control process of Proton Exchange Membrane Fuel Cells (PEMFCs) is a difficult task due to the non-linearities and uncertainties associated with the electrochemical processes governing it. Designing of a non-linear controller based on model predictive control for PEMFCs is presented in some previous works to regulate the cell voltage or power output based on just one of the input variables like hydrogen pressure or operating temperature respectively, but they use a constant signal for other important control variables due to computational limitations in on-line optimization of multivariable highly non-linear system. In this paper, by use of Approximate Predictive Control (APC) method based on neural network model, the whole control process is designed with three input variables. Operating temperature and hydrogen pressure are assumed as control design variables and current density as measured disturbance to manage the cell voltage. Moreover, multi-objective optimization based on multi-objective uniform-diversity genetic algorithm (MUGA) is used for optimal selection of the parameters of controller. The comparison of the obtained results with those in literature demonstrates the superiority of the results of this work.     

基于BP神经网络的温度控制系统

文中介绍了基于BP(Back Pmpagation)的神经网络气化炉温度控制系统。对BP神经网络控制算法作了详细的介绍，运用模糊逻辑控制概念赋予隐层含义，并决定其节点数，同时用高斯核函数作为节点激励函数，并做了仿真研究，叙述了系统的硬件与软件构成，试验表明所设计的系统操作方便、安全可靠，所选择的控制算法适应性强，控制效果良好。