Online gradient method has been widely used as a learning algorithm for training feedforward neural networks. Penalty is often introduced into the training procedure to improve the generalization performance and to de...Online gradient method has been widely used as a learning algorithm for training feedforward neural networks. Penalty is often introduced into the training procedure to improve the generalization performance and to decrease the magnitude of network weights. In this paper, some weight boundedness and deterministic con- vergence theorems are proved for the online gradient method with penalty for BP neural network with a hidden layer, assuming that the training samples are supplied with the network in a fixed order within each epoch. The monotonicity of the error function with penalty is also guaranteed in the training iteration. Simulation results for a 3-bits parity problem are presented to support our theoretical results.展开更多
Online gradient methods are widely used for training the weight of neural networks and for other engineering computations. In certain cases, the resulting weight may become very large, causing difficulties in the impl...Online gradient methods are widely used for training the weight of neural networks and for other engineering computations. In certain cases, the resulting weight may become very large, causing difficulties in the implementation of the network by electronic circuits. In this paper we introduce a punishing term into the error function of the training procedure to prevent this situation. The corresponding convergence of the iterative training procedure and the boundedness of the weight sequence are proved. A supporting numerical example is also provided.展开更多
This paper analyzes the advantages and disadvantages of three main online teaching methods, namely synchronous live broadcasting, asynchronous recording and broadcasting, and collaborative mixing of various means, as ...This paper analyzes the advantages and disadvantages of three main online teaching methods, namely synchronous live broadcasting, asynchronous recording and broadcasting, and collaborative mixing of various means, as well as the teaching effect questionnaire survey data, which provides support and reference for educators to explore online education rules and improve teaching quality.展开更多
Air environmental information plays an important role during plant growth and reproduction, prompt and accurate prediction of atmospheric environmental data is helpful for agricultural robots to make a timely decision...Air environmental information plays an important role during plant growth and reproduction, prompt and accurate prediction of atmospheric environmental data is helpful for agricultural robots to make a timely decision. For efficiency, an online learning method for predicting air environmental information was presented in this work. This method combines the advantages of convolutional neural network (CNN) and experience replay technique: CNN is used to extract features from raw data and predict atmospheric environmental information, experience replay technique can store environmental data over some time and update the hyperparameters of CNN. To validate the effects of this method, this online method was compared with three different predictive methods (including random forest, multi-layer perceptron, and support vector regression) using a public dataset (Jena). According to results, a suitable sample sequence size (e.g., 16) has a smaller number of training sessions and stable results, a larger replay memory size (e.g., 200) can provide enough samples to capture useful features, and 6 d of historical information is the best setting for training predictor. Compared with traditional methods, the method proposed in this study is the only method applied for various conditions.展开更多
This paper first analyzes the characteristics and current situation of the Advanced Mathematics course;secondly,it proposes a teaching model that integrates the outcome-based education(OBE)philosophy and blended teach...This paper first analyzes the characteristics and current situation of the Advanced Mathematics course;secondly,it proposes a teaching model that integrates the outcome-based education(OBE)philosophy and blended teaching method,reorganizing the teaching objectives,teaching content,and assessment evaluation process of the Advanced Mathematics course;lastly,through practice,it is proved that this approach can effectively improve students’mastery of course content,enhance students’ability to apply mathematical knowledge,and strengthen teaching effectiveness.展开更多
Precise states estimation for the lithium-ion battery is one of the fundamental tasks in the battery management system(BMS),where building an accurate battery model is the first step in model-based estimation algorith...Precise states estimation for the lithium-ion battery is one of the fundamental tasks in the battery management system(BMS),where building an accurate battery model is the first step in model-based estimation algorithms.To date,although the comparative studies on different battery models have been performed intensively,little attention is paid to the comparison among different online parameters identification methods regarding model accuracy,robustness ability,adaptability to the different battery operating conditions and computation cost.In this paper,based on the Thevenin model,the three most widely used online parameters identification methods,including extended Kalman filter(EKF),particle swarm optimization(PSO),and recursive least square(RLS),are evaluated comprehensively under static and dynamic tests.It is worth noting that,although the built model’s terminal voltage may well follow a measured curve,these identified model parameters may significantly out of reasonable range,which means that the error between measured and predicted terminal voltage cannot be seen as a gist to determine which model is the most accurate.To evaluate model accuracy more rigorously,battery state-of-charge(SOC)is further estimated based on identified model parameters under static and dynamic tests.The SOC prediction results show that EKF and RLS algorithms are more suitable to be used for online model parameters identification under static and dynamic tests,respectively.Moreover,the random offset is added into originally measured data to verify the robustness ability of different methods,whose results indicate EKF and RLS have more satisfactory ability against imprecisely sampled data under static and dynamic tests,respectively.Considering model accuracy,robustness ability,adaptability to the different battery operating conditions and computation cost simultaneously,EKF is recommended to be adopted to establish battery model in real application among these three most widely used methods.展开更多
This article presents a new two-axis solar tracker based on an online optimization algorithm so as to track the position of the sun without using its movement model.In this research,four well-known optimization algori...This article presents a new two-axis solar tracker based on an online optimization algorithm so as to track the position of the sun without using its movement model.In this research,four well-known optimization algorithms are employed to find the two unknown parameters named azimuth and zenith angles,which determine the position of the sun.The magnitude of the sunray is considered as the cost function of all algorithms.Then,several experiments are carried out to find the best optimization algorithm with optimal population size,number of iterations,and also the best initialization method.Uniform initialization leads to faster convergence compared to random initialization.The results clearly show that the particle swarm optimization algorithm with a population size of 15 and 7 iterations using uniform initialization method has better performance than the other algorithms,with a convergence time of less than 40 s.The average fitness value or voltage received by the tracker is 2.4 Volts in this method,which is higher than other methods.TLBO also performs well with a population size of 15 and 7 iterations.Afterward,the artificial neural network with one hidden layer and 20 neurons is employed to predict these two parameters in each day and moment in a year in Shiraz city according to the experimental data extracted from PSO.Number of the day from January and the time are inputs and zenith and azimuth angles are considered the output of neural network modeling.The performance of the proposed ANN model is evaluated using regression plots,demonstrating a strong correlation between predicted and target outputs.Finally,the outcomes reveal the feasibility of using online optimization algorithms and neural network modeling in an effort to bypass the complex mathematical model of mechatronic systems and predict the movement of the sun automatically.展开更多
Sufficient survey data are required to describe the stochastic behaviors of live loads.However,due to manual and on-site operation required by traditional survey methods,traditional surveys face challenges like occupa...Sufficient survey data are required to describe the stochastic behaviors of live loads.However,due to manual and on-site operation required by traditional survey methods,traditional surveys face challenges like occupant resistance,high costs,and long implementation periods.This study proposes a new survey method to access live load data online and automatically.Required samples are acquired from multi-source,open-access and dynamically updated data on the Internet.The change intervals,geometrical dimensions and object quantities are obtained from transaction information,building attributes and virtual reality models on real estate websites,respectively.The object weights are collected from commodity information on e-commerce websites.The integration of the aforementioned data allows for the extraction of necessary statistics to describe a live load process.The proposed method is applied to a live load survey in China,covering 20040 m^(2),with around 90000 samples acquired for object weights and load changes.The survey results reveal that about 70%−80%of the amplitude statistics are attributable to 1/6 of the total object types.展开更多
基金The NSF (10871220) of Chinathe Doctoral Foundation (Y080820) of China University of Petroleum
文摘Online gradient method has been widely used as a learning algorithm for training feedforward neural networks. Penalty is often introduced into the training procedure to improve the generalization performance and to decrease the magnitude of network weights. In this paper, some weight boundedness and deterministic con- vergence theorems are proved for the online gradient method with penalty for BP neural network with a hidden layer, assuming that the training samples are supplied with the network in a fixed order within each epoch. The monotonicity of the error function with penalty is also guaranteed in the training iteration. Simulation results for a 3-bits parity problem are presented to support our theoretical results.
文摘Online gradient methods are widely used for training the weight of neural networks and for other engineering computations. In certain cases, the resulting weight may become very large, causing difficulties in the implementation of the network by electronic circuits. In this paper we introduce a punishing term into the error function of the training procedure to prevent this situation. The corresponding convergence of the iterative training procedure and the boundedness of the weight sequence are proved. A supporting numerical example is also provided.
文摘This paper analyzes the advantages and disadvantages of three main online teaching methods, namely synchronous live broadcasting, asynchronous recording and broadcasting, and collaborative mixing of various means, as well as the teaching effect questionnaire survey data, which provides support and reference for educators to explore online education rules and improve teaching quality.
基金funded by the National Key Research and Development Program(Grant No.2022YFD2002202)Beijing Innovation Consortium of Agriculture Research System(BAIC08-2024-FQ04)+2 种基金Key Laboratory of Agricultural Sensors,Ministry of Agriculture and Rural Affairs(Grant No.PT2024-46)China Postdoctoral Science Foundation(Grant No.BX20230048)Postdoctoral fund of Beijing Academy of Agriculture and Forestry Sciences(Grant No.2023-ZZ-025).
文摘Air environmental information plays an important role during plant growth and reproduction, prompt and accurate prediction of atmospheric environmental data is helpful for agricultural robots to make a timely decision. For efficiency, an online learning method for predicting air environmental information was presented in this work. This method combines the advantages of convolutional neural network (CNN) and experience replay technique: CNN is used to extract features from raw data and predict atmospheric environmental information, experience replay technique can store environmental data over some time and update the hyperparameters of CNN. To validate the effects of this method, this online method was compared with three different predictive methods (including random forest, multi-layer perceptron, and support vector regression) using a public dataset (Jena). According to results, a suitable sample sequence size (e.g., 16) has a smaller number of training sessions and stable results, a larger replay memory size (e.g., 200) can provide enough samples to capture useful features, and 6 d of historical information is the best setting for training predictor. Compared with traditional methods, the method proposed in this study is the only method applied for various conditions.
文摘This paper first analyzes the characteristics and current situation of the Advanced Mathematics course;secondly,it proposes a teaching model that integrates the outcome-based education(OBE)philosophy and blended teaching method,reorganizing the teaching objectives,teaching content,and assessment evaluation process of the Advanced Mathematics course;lastly,through practice,it is proved that this approach can effectively improve students’mastery of course content,enhance students’ability to apply mathematical knowledge,and strengthen teaching effectiveness.
基金supported by the State Grid Company Science and Technology Project(Grant No.5230HQ19000J).
文摘Precise states estimation for the lithium-ion battery is one of the fundamental tasks in the battery management system(BMS),where building an accurate battery model is the first step in model-based estimation algorithms.To date,although the comparative studies on different battery models have been performed intensively,little attention is paid to the comparison among different online parameters identification methods regarding model accuracy,robustness ability,adaptability to the different battery operating conditions and computation cost.In this paper,based on the Thevenin model,the three most widely used online parameters identification methods,including extended Kalman filter(EKF),particle swarm optimization(PSO),and recursive least square(RLS),are evaluated comprehensively under static and dynamic tests.It is worth noting that,although the built model’s terminal voltage may well follow a measured curve,these identified model parameters may significantly out of reasonable range,which means that the error between measured and predicted terminal voltage cannot be seen as a gist to determine which model is the most accurate.To evaluate model accuracy more rigorously,battery state-of-charge(SOC)is further estimated based on identified model parameters under static and dynamic tests.The SOC prediction results show that EKF and RLS algorithms are more suitable to be used for online model parameters identification under static and dynamic tests,respectively.Moreover,the random offset is added into originally measured data to verify the robustness ability of different methods,whose results indicate EKF and RLS have more satisfactory ability against imprecisely sampled data under static and dynamic tests,respectively.Considering model accuracy,robustness ability,adaptability to the different battery operating conditions and computation cost simultaneously,EKF is recommended to be adopted to establish battery model in real application among these three most widely used methods.
文摘This article presents a new two-axis solar tracker based on an online optimization algorithm so as to track the position of the sun without using its movement model.In this research,four well-known optimization algorithms are employed to find the two unknown parameters named azimuth and zenith angles,which determine the position of the sun.The magnitude of the sunray is considered as the cost function of all algorithms.Then,several experiments are carried out to find the best optimization algorithm with optimal population size,number of iterations,and also the best initialization method.Uniform initialization leads to faster convergence compared to random initialization.The results clearly show that the particle swarm optimization algorithm with a population size of 15 and 7 iterations using uniform initialization method has better performance than the other algorithms,with a convergence time of less than 40 s.The average fitness value or voltage received by the tracker is 2.4 Volts in this method,which is higher than other methods.TLBO also performs well with a population size of 15 and 7 iterations.Afterward,the artificial neural network with one hidden layer and 20 neurons is employed to predict these two parameters in each day and moment in a year in Shiraz city according to the experimental data extracted from PSO.Number of the day from January and the time are inputs and zenith and azimuth angles are considered the output of neural network modeling.The performance of the proposed ANN model is evaluated using regression plots,demonstrating a strong correlation between predicted and target outputs.Finally,the outcomes reveal the feasibility of using online optimization algorithms and neural network modeling in an effort to bypass the complex mathematical model of mechatronic systems and predict the movement of the sun automatically.
基金study by the Key Project of National Natural Science Foundation of China(Grant No.51538010)the National Natural Science Foundation of China(Grant No.52178151)is gratefully acknowledged.
文摘Sufficient survey data are required to describe the stochastic behaviors of live loads.However,due to manual and on-site operation required by traditional survey methods,traditional surveys face challenges like occupant resistance,high costs,and long implementation periods.This study proposes a new survey method to access live load data online and automatically.Required samples are acquired from multi-source,open-access and dynamically updated data on the Internet.The change intervals,geometrical dimensions and object quantities are obtained from transaction information,building attributes and virtual reality models on real estate websites,respectively.The object weights are collected from commodity information on e-commerce websites.The integration of the aforementioned data allows for the extraction of necessary statistics to describe a live load process.The proposed method is applied to a live load survey in China,covering 20040 m^(2),with around 90000 samples acquired for object weights and load changes.The survey results reveal that about 70%−80%of the amplitude statistics are attributable to 1/6 of the total object types.
