阅读说明：本技术 基于隐马尔科夫模型与粒子群优化的木材含水率控制方法 (Wood moisture content control method based on hidden Markov model and particle swarm optimization ) 是由 张冬妍 蒋大鹏 李丹丹 李鸿博 曹军 于 2020-04-26 设计创作，主要内容包括：基于隐马尔科夫模型与粒子群优化的木材含水率控制方法,它属于木材干燥过程控制技术领域。本发明解决了采用现有方法在减速干燥阶段对木材含水率的控制精度低的问题。本发明首先测量出当前时刻木材的含水率、温度和湿度数据,再将测量的当前时刻数据输入隐马尔科夫模型,获得模型输出的下一时刻木材的含水率、温度和湿度数据。将当前时刻的木材温度、湿度数据、模型输出的下一时刻木材含水率、温度和湿度数据以及含水率基准数据代入适应度函数,对适应度函数进行优化得到当前时刻干燥系统需要设定的最佳温度和湿度,实现对干燥过程中木材含水率的控制,且通过本发明方法可以显著提高减速干燥阶段中对木材含水率的控制精度。本发明可以应用于木材干燥过程含水率控制。(A method for controlling the moisture content of wood based on hidden Markov model and particle swarm optimization belongs to the technical field of wood drying process control. The invention solves the problem that the control precision of the water content of the wood is low in the deceleration drying stage by adopting the existing method. The method comprises the steps of firstly measuring the moisture content, temperature and humidity data of the wood at the current moment, inputting the measured current moment data into a hidden Markov model, and obtaining the moisture content, temperature and humidity data of the wood at the next moment output by the model. The wood temperature and humidity data at the current moment, the wood moisture content, the temperature and humidity data at the next moment output by the model and the moisture content reference data are substituted into the fitness function, the fitness function is optimized to obtain the optimal temperature and humidity required to be set by the drying system at the current moment, the control of the wood moisture content in the drying process is realized, and the control precision of the wood moisture content in the deceleration drying stage can be obviously improved by the method. The invention can be applied to the control of the water content in the wood drying process.)

1. The method for controlling the water content of the wood based on the hidden Markov model and particle swarm optimization is characterized by comprising the following steps of:

step one, measuring the moisture content M of the wood at the time t by using a measuring instrument_tWood temperature T at time T_tWood moisture H at time t_tAnd the moisture content M of the wood at the time of t +1_t+1Wood temperature T at time T +1_t+1Wood moisture H at time t +1_t+1；

Wood moisture content M at time t_tWood temperature T at time T_tAnd wood moisture H at time t_tAs independent variable, the water content M of the wood at the t +1 moment_t+1Wood temperature T at time T +1_t+1And wood moisture H at time t +1_t+1Establishing a sample set consisting of independent variables and dependent variables for the dependent variables;

training an HMM prediction model by using the established sample set, and stopping training until the difference value of parameters of the HMM prediction model obtained by two continuous iterations is smaller than a threshold value Q to obtain a trained HMM prediction model;

step two, establishing a fitness function F;

step three, after measuring the moisture content, the temperature and the humidity of the wood at the current moment, inputting the moisture content, the temperature and the humidity data of the wood at the current moment into the HMM prediction model trained in the step one, and obtaining the moisture content, the temperature and the humidity data of the wood at the next moment output by the HMM prediction model;

substituting the measured wood temperature and humidity data at the current moment, the wood moisture content, temperature and humidity data at the next moment output by the model and the wood moisture content reference value at the next moment into a fitness function F, and calculating the optimal temperature and humidity required to be set by the wood drying system at the current moment according to the fitness function F;

step four, after the temperature and the humidity of the wood drying system at the current moment are respectively set to be the optimal temperature and the optimal humidity obtained in the step three, the moisture content, the temperature and the humidity data of the wood at the next moment are measured after the wood is dried;

adding the moisture content, temperature and humidity data of the wood at the current moment in the third step and the moisture content, temperature and humidity data of the wood at the next moment in the fourth step into the sample set, and continuously updating the sample set;

step six, if the difference value between the next moment wood water content data measured in the step four and the next moment wood water content data output by the HMM prediction model in the step three is larger than or equal to a threshold value, taking the parameters of the HMM prediction model trained in the step one as initialization input parameters, retraining the HMM prediction model by using the updated sample set, and executing the step seven by using the retrained HMM prediction model;

otherwise, continuing to use the HMM prediction model trained in the first step to execute a seventh step;

and step seven, repeatedly executing the process from the step three to the step six, and realizing the control of the water content in the wood drying process.

2. The method for controlling moisture content of wood based on hidden markov models and particle swarm optimization according to claim 1, wherein in the first step, the training algorithm adopted by the HMM prediction model is an EM algorithm.

3. The method for controlling the moisture content of the wood based on the hidden markov model and the particle swarm optimization according to claim 2, wherein in the second step, the fitness function F is specifically established as follows:

F＝(T′_t+1-T_t)²+(H′_t+1-H_t)²+(M_r(t+1)-M′_t+1)²

in the formula, T_tFor the measured wood temperature at time t, H_tFor the measured wood moisture at time t, M_r(t+1)Is a reference value of wood water content at time T +1, T'_t+1Timber temperature, H ', at time t +1 output by HMM predictive model'_t+1Wood moisture, M ', at time t +1 output by HMM predictive model'_t+1And (4) outputting the wood moisture content at the t +1 moment of the HMM prediction model.

4. The hidden markov model and particle swarm optimization-based wood water content control method according to claim 3, wherein in the third step, the optimal temperature and humidity to be set by the wood drying system at the current time are calculated according to the fitness function F, and the specific process is as follows:

and optimizing the fitness function F by using a PSO model, and calculating the corresponding drying system temperature T and the drying system humidity H when the fitness function F is minimum, wherein the T and the H are respectively the optimal temperature and humidity which are required to be set by the wood drying system at the current moment.

5. The hidden markov model and particle swarm optimization-based wood water content control method according to claim 1, wherein in the first step, the HMM prediction model parameters refer to an observed state matrix parameter, an initial hidden state matrix parameter, and a state transition matrix parameter.

6. The hidden Markov model and particle swarm optimization-based wood water content control method according to claim 5, wherein in the first step, the value of the threshold Q is 0.01.

Technical Field

The invention belongs to the technical field of wood drying process control, and particularly relates to a wood moisture content control method based on hidden Markov model and particle swarm optimization.

Background

The wood drying is an important technical measure for improving the physical and mechanical properties of the wood, reasonably using the wood, reducing the loss of the wood, and the like and improving the utilization rate of the wood, and is also one of the key technologies for ensuring the quality of the wood products. Because the demand of China on wood is continuously increased, the demand on wood drying equipment is also continuously increased, most of the existing drying equipment in China is old, simple and crude, operation personnel is required to control the drying equipment according to technological parameters by experience, and the loss of drying and the like is serious. Some manufacturers producing various drying devices can produce automatically controlled drying devices, and the product quality can basically meet the production requirements. Compared with the advanced foreign equipment, the method mainly has the problems of low detection and control precision, insufficient reliability, high energy consumption of wood drying equipment and the like. When designing automatic control equipment, not only an off-line simulation experiment needs an accurate model, but also the self-adaption, robustness and predictive control effects depend on the performance of the model, so that the problem of building a wood drying mathematical model is firstly considered in the application of solving the automatic control of wood drying. At present, the theory and the method for modeling the linear system are relatively mature, but the actual controlled object is mostly a time-varying and nonlinear dynamic system, the dynamic behavior of the system is very rich, and the traditional control method which needs to depend on an accurate mathematical model is difficult to obtain a good control effect. How to establish a model capable of accurately reflecting the input-output relationship of the system becomes a key for improving the control level of the nonlinear system.

The wood is a complex moisture-containing porous viscoelastic organism, the internal structure and characteristics of the wood can be greatly changed in the drying process, and the wood drying is complex and variable, namely the wood drying is a complex strong coupling nonlinear power system, and external interference and model uncertainty exist in the drying process, so that an ideal wood drying model is difficult to establish. Therefore, the model capable of accurately reflecting the wood drying process is established by selecting an effective modeling method, and the method is the first problem of researching the wood drying mechanism, realizing full-automatic control and comprehensively improving the performance of the wood drying equipment.

The wood drying belongs to an unsteady heat and mass transfer process with an ultra-fine structure, and the wood structure is complex and has diversity and variability, so that the established model is difficult to realize due to too complex structure or causes deviation due to too many simplified conditions, and is difficult to be applied to actual control. The model of the wood drying is modeled by using a machine learning method based on data, and a model capable of reflecting the macroscopic characteristics of the wood drying can be obtained only according to the input and output of the wood drying system which can be actually measured. The determination of the learning data is the first step of building a model of the wood drying system using machine learning theory. External factors affecting wood drying are many and mainly include medium temperature, medium humidity and wood surface air flow velocity. For a drying kiln with a full-speed fan, temperature and humidity data of the inside of the kiln can be used as input quantity during modeling, and the output quantity is the moisture content of wood. Then selecting a proper modeling method, using one kiln (random) in two or more kiln measured data as a learning sample training model, and verifying the accuracy of the model by other data.

Because the wood drying system is a dynamic system, the moisture content of the wood can be regarded as a non-stationary time sequence with nonlinear decrement in the drying process, and the moisture content of the wood at the current moment is not only related to the environmental intervention (mainly temperature, humidity and wood surface air flow rate) in the kiln, but also related to the moisture content at the previous moment or in a period of time.

When the moisture content of the wood is below the fiber saturation point (about 30%), the wood can generate dry shrinkage and wet expansion in the air, and the use of the wood is greatly influenced. In order to ensure the quality and prolong the service life of the wood and the wooden products, proper measures must be taken to reduce the moisture (water content) in the wood to a certain degree. To reduce the moisture content of the wood, the temperature of the wood is increased to evaporate and move the moisture in the wood outward, so that the moisture rapidly leaves the wood in the air with a certain flow speed, thereby achieving the purpose of drying.

The drying process is divided into a plurality of stages by taking the moisture content of the wood as a reference, and the temperature and the humidity corresponding to each stage are determined.

The wood drying process can be roughly divided into three stages of preheating, constant-speed drying and speed-reducing drying. The change of the moisture content of the wood along with time in the constant-speed drying stage is approximately linear, the control model is relatively simple, the existing deceleration drying stage control model is relatively complex, and the control precision of the moisture content of the wood in the deceleration drying stage is low, so that the simplification of the deceleration drying stage control model is very necessary for improving the moisture content control precision in the wood drying process.

Disclosure of Invention

The invention aims to solve the problem that the control precision of the moisture content of the wood is low in the deceleration drying stage by adopting the conventional method, and provides a method for controlling the moisture content of the wood based on a hidden Markov model and particle swarm optimization.

The technical scheme adopted by the invention for solving the technical problems is as follows: a method for controlling the moisture content of wood based on hidden Markov model and particle swarm optimization comprises the following steps:

step one, measuring the moisture content M of the wood at the time t by using a measuring instrument_tWood temperature T at time T_tWood moisture H at time t_tAnd the moisture content M of the wood at the time of t +1_t+1Wood temperature T at time T +1_t+1Wood moisture H at time t +1_t+1；

Wood moisture content M at time t_tWood temperature T at time T_tAnd wood moisture H at time t_tAs independent variable, the water content M of the wood at the t +1 moment_t+1Wood temperature T at time T +1_t+1And wood moisture H at time t +1_t+1Establishing a sample set consisting of independent variables and dependent variables for the dependent variables;

training an HMM prediction model by using the established sample set, and stopping training until the difference value of parameters of the HMM prediction model obtained by two continuous iterations is smaller than a threshold value Q to obtain a trained HMM prediction model;

step two, establishing a fitness function F;

step three, after measuring the moisture content, the temperature and the humidity of the wood at the current moment, inputting the moisture content, the temperature and the humidity data of the wood at the current moment into the HMM prediction model trained in the step one, and obtaining the moisture content, the temperature and the humidity data of the wood at the next moment output by the HMM prediction model;

substituting the measured wood temperature and humidity data at the current moment, the wood moisture content, temperature and humidity data at the next moment output by the model and the wood moisture content reference value at the next moment into a fitness function F, and calculating the optimal temperature and humidity required to be set by the wood drying system at the current moment according to the fitness function F;

step four, after the temperature and the humidity of the wood drying system at the current moment are respectively set to be the optimal temperature and the optimal humidity obtained in the step three, the moisture content, the temperature and the humidity data of the wood at the next moment are measured after the wood is dried;

adding the moisture content, temperature and humidity data of the wood at the current moment in the third step and the moisture content, temperature and humidity data of the wood at the next moment in the fourth step into the sample set, and continuously updating the sample set;

step six, if the difference value between the next moment wood water content data measured in the step four and the next moment wood water content data output by the HMM prediction model in the step three is larger than or equal to a threshold value, taking the parameters of the HMM prediction model trained in the step one as initialization input parameters, retraining the HMM prediction model by using the updated sample set, and executing the step seven by using the retrained HMM prediction model;

otherwise, continuing to use the HMM prediction model trained in the first step to execute a seventh step;

and step seven, repeatedly executing the process from the step three to the step six, and realizing the control of the water content in the wood drying process.

The invention has the beneficial effects that: the invention provides a method for controlling the moisture content of wood based on hidden Markov model and particle swarm optimization. The wood temperature and humidity data at the current moment, the wood moisture content, the temperature and humidity data at the next moment output by the model and the moisture content reference data are substituted into the fitness function, the fitness function is optimized to obtain the optimal temperature and humidity required to be set by the drying system at the current moment, the control of the wood moisture content in the drying process is realized, and the control precision of the wood moisture content in the deceleration drying stage can be obviously improved by the method.

Experiments show that the maximum difference value between the actual value and the standard value of the water content obtained by the method is about 1.5, the maximum difference value between the actual value and the standard value of the water content obtained by the conventional SVM prediction method is about 2.5, the difference value between the actual value and the standard value of the water content obtained by the method is obviously smaller than that of the SVM prediction method, and the control precision is greatly improved.

Drawings

FIG. 1 is a graph of water content versus time obtained by the method of the present invention;

FIG. 2 is a graph of water content versus time obtained using an SVM method;

FIG. 3 is a graph showing the relationship between the water content difference (difference between the actual value and the standard value) and time obtained by the method of the present invention;

FIG. 4 is a graph of water cut difference (difference between actual value and standard value) versus time obtained using SVM method.

Detailed Description