Due to the diversity of network data and the interference of illegal network intrusion, the response time of the current designed network stability controller is often too long. A design method of network stability controller based on machine learning is proposed. The controller is mainly composed of development board, control circuit and machine learning module. Machine learning module monitors the network controlled objects by specific learning methods, and the monitoring results will be transmitted to the control circuit for virtual control of various learning behaviors. The development board receives the virtual control results, and screens out the optimal control strategy for the network controlled objects. After evaluating the optimal control strategy, the machine learning module implements stable control to the network controlled object. The experimental results show that the designed controller can obtain excellent response time and strong response ability while maintaining effective control of the network, and can better achieve the design objectives. Machine learning is an artificial intelligence project interwoven with statistical probability, concave-convex test and approximation theory. It is widely used in analysis and control, and is the core of artificial intelligence. Network is the most important medium of information transmission in today’s society, and it is also a tool to assist all walks of life in design, analysis, identification and other work [1?3]. The stability of the network is the basic guarantee for the above work. Machine learning is applied to the design of network stability controller, which can reduce the response time of the controller while maintaining the effective control of the network through practical learning. This is also a new scientific research project [4?6], which is currently being studied by scientific research organizations. The network stability controller based on machine learning is mainly composed of development board, control circuit and machine learning module. It also includes various functional modules, such as sensor module, thermostatic element computer control module, storage module and transmission module. Among them, the development board and control circuit are the hardware end of the controller, and the machine learning module is the realization end of the controller. The whole controller takes the realization end as the core, and integrates the hardware end for auxiliary control, so as to achieve the design goal of reducing the response time of the controller. Fig. 1 is the overall design of the controller.
From Figure 1, we can see that the machine learning module directly contacts with the network controlled object, and supervises the network controlled object by using specific learning methods. The function of the sensor module is to induce the parameters of the network controlled object, and convert the useful parameters into analog/digital ones for the use of the machine learning module. The control circuit of the network stability controller based on machine learning is an adaptive linear circuit, which can effectively avoid the oscillation effect of the controller at low frequency, and has a transitional effect on the target of reducing the response time of the controller. The circuit design is shown in Figure 2. From Figure 2, it can be seen that the function of converter is to convert DC to AC voltage of the monitoring result of machine learning module. T1 and T2 represent the external transformer of converter.
There is a certain linear relationship between the output value of control circuit and the voltage parameters of the controlled object of induction network. Computer control module, development board and machine learning module are directly connected with the control circuit. This means that the control circuit of network stability controller based on machine learning is actually a serial communication circuit endowed with control ability. The development board of network stability controller selection based on machine learning is Mini2440 development board with four layers structure. It has the advantages of low energy consumption, strong practicability and low price. It is a development board with the highest cost performance in China. The design structure of Mini2440 development board adopts high-performance Harvard structure and built-in Samsung S3C2440 microprocessor.
Its wiring method adopts equal length wiring, which effectively guarantees the signal receiving ability of the controller and can reduce the response time of the controller to a certain extent. The structure diagram of Mini2440 development board is shown in Figure 3. From Figure 3, we can see that the Mini2440 development board is mainly composed of LCD screen, network port, JTAG interface and bus expander. Its working voltage is divided into two kinds, 1.8 V and 3.3 V, respectively. The level type of microprocessor is TTL. Considering the possibility of RS 232 level, the level support chip should be connected from the serial port when needed. In the designed network stability controller based on machine learning, the key design contents are: through the effective analysis of network data, the deep mining of data characteristics is realized, so that the controller has the ability of autonomous learning, which can be used to balance network stability and obtain excellent response time of the controller. In the machine learning module, there are three ways of machine learning, namely, supervisory, non-supervisory and semi-supervisory. The learning process is shown in figs. 4-6. From figs.
4 to 6, it can be seen that the supervisory mode of machine learning is the process of training and reconstructing the data aggregation model for the characteristic data of the network controlled objects. Reconfiguration refers to the supervision of the network controlled objects, whose supervision is realized by the reconfigurer. The reconstructed result is the model fine-tuning data obtained by comparing the standard and actual characteristics of the network controlled object. Usually, the more times the data set model is trained, the more accurate the reconstructed results are. Evaluation learning is a very important computational process in machine learning. It pays more attention to the influence of machine learning environment on reconstructed results and optimizes the results of various machine learning methods. Evaluation learning can adjust the reconstructed results to the optimal state by specific means of action, reduce the control error of network stability controller based on machine learning, and then shorten its response time. In the process of evaluation and learning, because the historical characteristic parameters of the network controlled object are not considered, only the future state and behavior of the network controlled object are estimated. It is assumed that the state characteristics of network controlled object data are behavioral characteristics, single-phase control error evaluation function and state characteristics adjustment function. In the formula, it is the learning factor under the network fluctuation state, and its value is always greater than 1. The response time of the network stability controller based on machine learning is analyzed through experiments. There are three objects selected to compare with the controller in this paper. They are network stability controller based on neuron, network stability controller based on fuzzy control and network stability controller based on single chip computer. Firstly, the control level of the four controllers is adjusted to the same level. That is to say, four controllers are given the same control time and control results. Three groups of experiments are carried out under the condition of changing the transmission power of the controllers, and the response time of the four controllers is counted.
Table 1 shows the experimental limits of the power transmitted by the controller. Figures 7-9 show the response time comparison curves of four controllers in three experiments. From figs. 7 to 9, it can be seen that the lower the absolute value of the transmission power of the controller, the smaller the fluctuation of the transmission power, the faster the response time curve is stable, and the response time of the controller will be reduced accordingly. For the network stability controller based on fuzzy control, the response time curve of the controller is the slowest among the four controllers, which proves that the response ability of the controller to the network is weak and the network stability can not be maintained. For the network stability controller based on neuron, its response time is lower than that based on fuzzy control, but its response time curve fluctuates most obviously. It proves that its response ability is far lower than that of the network stability controller based on single chip computer and the controller in this paper, and it can not obtain excellent response time. The response time of the network stability controller based on single chip computer is short and the response time curve fluctuates smoothly. However, under the high power of 500 kW in experiment 3, the response ability of the controller is much lower than that under the low power, which proves that the response time of the controller is not stable. The response time curve of the controller is very stable under three groups of experiments, and the response time curve is the fastest. It proves that the controller can obtain excellent response time while maintaining effective control of the network, and has strong response ability. This paper presents a design method of network stability controller based on machine learning. The controller is mainly composed of development board, control circuit and machine learning module. Machine learning module monitors the network controlled objects by specific learning methods, and the monitoring results will be transmitted to the control circuit for virtual control of various learning behaviors. The development board receives the virtual control results, and screens out the optimal control strategy for the network controlled objects. After evaluating the optimal control strategy, the machine learning module implements stable control to the network controlled object. The system gives the workflow of the machine learning module, and the evaluation of the optimal control strategy is realized by the evaluation learning objective function based on machine learning. The experimental results show that the designed controller achieves excellent response time and strong response ability while maintaining effective control of the network. It can achieve the design objectives well.