基于神经网络的传感器故障在线诊断系统

介绍一种在线性或非线性情况下应用神经网络对大型机械设备的传感器故障进行检测、分离和恢复的系统。整个系统由Ｎ个神经网络组成，Ｎ是安装在大型机械设备上不重复检测同一参数的传感器个数，对机械系统动态情况作了特殊处理，使其也可应用于机械系统的动态情况。数字仿真结果表明，该系统是十分有效的。     

PEMFC的Elman神经网络建模与模糊神经网络控制

从质子交换膜燃料电池（PEMFC）实际应用的角度出发，采用Elman动态神经网络对PEMFC系统进行建模，以实验中采样到的PEMFC系统的工作温度输入输出数据训练网络，并采用动态反向传播学习算法根据误差不断调整网络参数直至达到要求精度。设计了一种适应模糊神经网络控制器，根据经验确定了初始隶属度函数和模糊规则，并采用自适应学习算法不断调整隶属度函数与模糊规则参数，使控制系统获得理想的输出。仿真实验以Elman神经网络模型为参考模型，使用自适应神经网络控制算法取得了较好的控制效果。总之，所设计的控制系统适合于控制PEMFC这样一类复杂非线性系统。     

基于神经网络的改进扰动观察法MPPT控制

为进一步提高光伏发电系统的输出效率和动态调节性能,设计基于神经网络的改进扰动观察法最大功率点跟踪（MPPT）控制。该方法在扰动观察法基础上引入神经网络算法,使得在环境剧烈变化时,系统动态调节速度加快。通过系统仿真和硬件实测,得出该改进扰动观察法相比于传统扰动观察法和神经网络法,在一定程度上能加快动态调节速度和减小稳态误差的结论。     

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。     

锅炉过热器系统的动态仿真模型

鉴于机理模型很难再现锅炉过热器系统的复杂动态特性,本文利用机理模型为主导,动态神经网络在线修正,从而显著地提高了仿真模型的精度。仿真结果表明,此建模方法为复杂大系统的动态仿真提供了一个理想的建模途径。     

基于动态神经网络的温度调节系统控制方法研究

利用动态模糊神经网络(DFNN)对室内温度系统进行辨识与控制。首先,研究现有控制方式中存在的问题,接着,根据这些问题提出了一种动态模糊神经网络的结构和算法。最终,通过仿真结果与实测温度进行比较,证明所提出方法的有效性。     

基于深度神经网络的燃煤电站SCR脱硝系统时序建模研究

选择性催化还原(SCR)脱硝系统可以通过控制氨的注入，有效减少氮氧化物的排放，但能源结构、负荷波动、反应器动态特性和系统延迟等因素均会影响氨注入量的精确控制。为了实现高精度的氮氧化物排放预测，提出基于时间序列特征的深度神经网络建模方法，用于预测反应器动态特性和系统延迟。以1台660 MW燃煤锅炉为例，利用连续3 d的50 000多个采样数据建立深度神经网络。结果显示：模型实现了对t+1时刻SCR出口NO_x的精确估算，测试集上的最大绝对误差仅为1.6 mg/m³。     

基于动态RBF神经网络的锅炉短期负荷预测研究

由于锅炉的热惯性，造成整个热力系统调节相对滞后，影响系统调峰和优化运行。因此，根据系统参数对锅炉短期负荷进行预测变得尤为重要。在基本RBF神经网络的基础上，提出了一种动态RBF神经网络，并定义了样本差异和样本局部差异两个相关参数，对新样本进行有效性判别，同时，给出了输入层灵敏度系数，以实现大差异样本的负荷预测。为了验证所建模型的正确性，以某电厂实际运行数据为基础验证了这一模型，使用动态RBF神经网络对未来锅炉负荷加以预测，同时对预测结果与实验结果进行了比较。结果表明，这种网络具有很强的适应性，能够对锅炉进行准确的负荷预测，具有很好的应用前景。     

用多变量模糊神经网络实现双抽汽轮机热电负荷的解耦控制

由于经典控制理论和现代控制理论分别在双抽汽轮机电、热负荷协调控制应用中的不足，提出了用模糊神经网络控制来实现解耦控制。以15MW双抽汽轮机为例，建立系统动态数学模型，并进行了模糊神经网络解耦控制的数字仿真。     

基于对角递归神经网络整定的PID解耦单元机组负荷控制系统

针对火电厂单元机组具有多变量强耦合、非线性及参数时变的受控对象，提出了基于对角递归神经网络整定的PID解耦控制方法，其主要特点是能够提供一个对角递归神经网络来辩识系统模型，进而对PID控制器参数进行整定，实现多变量解耦控制。通过对火电机组负荷控制系统的设计和仿真研究，表明系统达到了动态近似解耦、静态完全解耦和无静差跟踪，并具有响应速度快，鲁棒性好等特点。图5参6     

自适应记忆递归神经网络预测控制器及其应用

提出一种自适应学习率记忆递归神经网络预测控制器及自适应学习率方法，它由用于预测和控制的子神经网络组成，预测子网络向控制子网络提供控制灵敏度信号；并证明了记忆递归神经网络学习的收敛性和稳定性条件．仿真结果表明控制器在线实时控制具有非线性、时变、多变量特性的水轮发电机组，对各种工况具有良好的性能．     

An approach to sensorless operation of the permanent-magnet synchronous motor using diagonally recurrent neural networks

Due to the drawbacks associated with the use of rotor position sensors in permanent-magnet synchronous motor (PMSM) drives, there has been significant interest in the so-called rotor position sensorless drive. Rotor position sensorless control of the PMSM typically requires knowledge of the PMSM structure and parameters, which in some situations are not readily available or may be difficult to obtain. Due to this limitation, an alternative approach to rotor position sensorless control of the PMSM using a diagonally recurrent neural network (DRNN) is considered. The DRNN, which captures the dynamic behavior of a system, requires fewer neurons and converges quickly compared to feedforward and fully recurrent neural networks. This makes the DRNN an ideal choice for implementation in a real-time PMSM drive system. A DRNN-based neural observer, whose architecture is based on a successful model-based approach, is designed to perform the rotor position estimation on the PMSM. The advantages of this approach are discussed and experimental results of the proposed system are presented.     

An adaptive power system stabilizer based on recurrent neuralnetworks

Application of recurrent, neural networks in the design of an adaptive power system stabilizer (PSS) is presented in this paper. The architecture of the proposed adaptive PSS has two recurrent neural networks. One functions as a tracker to learn the dynamic characteristics of the power plant and the second one functions as a controller to damp the oscillations caused by the disturbances. In the proposed approach, the weights of the neural networks are updated on-line. Therefore, any new information available during actual control of the plant is considered. Simulation studies show that the artificial neural network (ANN) based PSS can provide very good damping over a wide range of operating conditions     

Design of an Adaptive PSS Based on Recurrent Adaptive Control Theory

Inspired by observing the similarity between adaptive control systems and recurrent neural networks (RNNs), a new control scheme, the recurrent adaptive control (RAC), is presented in this paper. Back propagation through time (BPTT), a learning algorithm for RNNs, can be exploited in RAC. Application of truncated BPTT to RAC is also discussed. Further, a new control algorithm for RAC, termed recursive gradient (RG), is developed to improve the performance of the original and truncated BPTT algorithms. Effectiveness of the RG control algorithm as a power system stabilizer is demonstrated.      

Artificial-Neural-Network-Based Sensorless Nonlinear Control of Induction Motors

In this paper, two architectures of artificial neural networks (ANNs) are developed and used to correct the performance of sensorless nonlinear control of induction motor systems. Feedforward multilayer perception, an Elman recurrent ANN, and a two-layer feedforward ANN is used in the control process. The method is based on the use of ANN to get an appropriate correction for improving the estimated speed. Simulation and experimental results were carried out for the proposed control system. An induction motor fed by voltage source inverter was used in the experimental system. A digital signal processor and field-programmable gate arrays were used to implement the control algorithm.     

Wind power forecasting using advanced neural networks models

In this paper, an advanced model, based on recurrent high order neural networks, is developed for the prediction of the power output profile of a wind park. This model outperforms simple methods like persistence, as well as classical methods in the literature. The architecture of a forecasting model is optimised automatically by a new algorithm, that substitutes the usually applied trial-and-error method. Finally, the online implementation of the developed model into an advanced control system for the optimal operation and management of a real autonomous wind-diesel power system, is presented     

Wheel-Individual Estimation of the Friction Potential for Split Friction and Changing Friction Conditions for the Application in an Automated Emergency Braking System

Including information of the current road surface conditions can significantly improve the effectiveness of an AEB （automated emergency braking） system to avoid accidents or reduce the injury severity in rear-end crashes. A method to estimate the friction potential based on on-board sensor information is shown in this work. This work expands the scope of existing investigations on whether the accuracy needed for the warning and intervention strategies of AEB can be reached with the proposed method. First, the bandwidth of surface conditions investigated is extended by including low friction surfaces comparable to ice. Second, situations of changing surface conditions and wheel-individual surface conditions were evaluated. Finally, estimation based on different sensor sets was conducted with regard to series application. The investigations are based on measurements performed on a proving ground. The main emphasis was placed on estimation during longitudinal driving conditions. The used sensors include advanced vehicle dynamics measurement equipment as well as standard on-board sensors of the vehicle.     

Data‐driven lithium‐ion battery states estimation using neural networks and particle filtering

The state of charge and state of health estimations are two of the most crucial functions of a battery management system, which are the quantified evaluation of driving mileage and remaining useful life of electric vehicles. This paper investigates a novel data‐driven–enabled battery states estimation method by combining recurrent neural network modeling and particle‐filtering–based errors redress. First, a recurrent neural network with long‐short time memory is employed to learn the long‐term nonlinear relation between batteries states and measurable signals of lithium‐ion batteries, such as current, voltage, and temperature. Second, to denoise the estimation errors of the neural network model, particle filtering is employed to smooth the state of charge estimation results. Third, the terminal voltage difference of battery is highly related to the internal resistance of the battery, which is thus taken as a new input to track the internal resistance of the battery. The performance of the proposed method is verified by multiple comparisons with conventional techniques under randomized loading profiles and different temperatures.     

基于关注指标和深度学习的台区配变重过载预警方法研究

针对现行配变重过载治理方法过于被动的问题,结合电力大数据和深度学习技术提出了一种适用于大规模配电网分析的台区配变重过载预测方法。首先基于配变负载特性建立重载关注指标并建立二级过滤系统筛选出重过载风险较高的台区配变及其日期作为预测对象。然后充分考虑内外影响因素,建立卷积神经网络-门限循环单元深度学习模型实现负载率预测并转化为预警等级。实例证明了所提方法的有效性。