Disclosure of Invention
The invention provides an intelligent control system for parameters of a livestock and poultry house breeding environment, which effectively solves the problems that the existing livestock and poultry breeding environment does not have the influence on the production of the livestock and poultry breeding environment according to the change of growth parameters of the livestock and poultry breeding environment, has the characteristics of time variation, nonlinearity, multivariable coupling and the like, and does not carry out accurate detection and decoupling control on the growth parameters of the livestock and poultry breeding environment, so that the production benefit and the production management of the livestock and poultry breeding environment are greatly influenced.
The invention is realized by the following technical scheme:
the utility model provides a beasts and birds house farming environment parameter intelligence control system which characterized in that: the system comprises an animal house environment parameter acquisition and control platform and a growth parameter intelligent decoupling control subsystem, and realizes detection and growth parameter intelligent adjustment of animal environment parameters.
The invention further adopts the technical improvement scheme that:
the livestock and poultry house environment parameter acquisition and control platform consists of a detection node, a control node, a gateway node, a field monitoring end, a cloud platform and a mobile phone APP, wherein the detection node acquires livestock and poultry house environment parameters and uploads the livestock and poultry house environment parameters to the cloud platform through the gateway node, and data and release information are stored and issued at the cloud platform end; the mobile phone APP can monitor the environment parameters of the livestock and poultry house in real time through the environment information of the livestock and poultry house provided by the cloud platform; the detection nodes and the control nodes are responsible for collecting the environment parameters of the livestock and poultry house and controlling the environment equipment of the livestock and poultry house, the bidirectional communication among the detection nodes, the control nodes, the field monitoring end, the cloud platform and the mobile phone APP is realized through the gateway node, and the collection of the environment parameters of the livestock and poultry house and the control of the equipment of the livestock and poultry house are realized; the structure of the livestock and poultry house environment parameter acquisition and control platform is shown in figure 1.
The invention further adopts the technical improvement scheme that:
the growth parameter intelligent decoupling control subsystem consists of an ESN neural network model, a noise reduction self-encoder A, a time delay neural network model, a PID controller, an integral loop, a fuzzy recursive neural network compensation controller, an FLNN neural network model and a parameter decoupling control module, 2 integral operators S are connected in series to form 1 integral loop, and 2 integral operator connecting ends of each integral loop and the output of the integral loop are respectively used as 2 corresponding inputs of the FLNN neural network model; the expected values of temperature, humidity, wind speed and illuminance are used as corresponding inputs of an ESN (electronic stability network) neural network model, the outputs of a plurality of groups of temperature, humidity, wind speed and illuminance sensors are respectively used as the inputs of a plurality of corresponding time delay neural network models, the outputs of the plurality of time delay neural network models are used as the inputs of a noise reduction self-encoder A, and the output of the noise reduction self-encoder A is used as the corresponding input of the ESN neural network model; the difference value of the ESN neural network model output and the noise reduction self-encoder A output of the parameter decoupling control module is used as a growing environment grade difference value, the growing environment grade difference value and the change rate of the growing environment grade difference value are respectively used as the input of a PID controller and the input of a fuzzy recurrent neural network compensation controller, the PID controller output is used as the input of an integral loop and the corresponding input of an FLNN neural network model, and the output of the FLNN neural network model and the output of the fuzzy recurrent neural network compensation controller are respectively used as 2 corresponding inputs of an NARX neural network control decoupling controller of the parameter decoupling control module; the intelligent decoupling control subsystem for the growth parameters is shown in figure 2.
The invention further adopts the technical improvement scheme that:
the parameter decoupling control module consists of an LSTM neural network disturbance controller, an NARX neural network control decoupler and a noise reduction self-encoder B; the output of a plurality of temperature, humidity, wind speed and illuminance sensors is used as a plurality of corresponding inputs of a noise reduction self-encoder B, the output of the noise reduction self-encoder B is used as a corresponding input of an LSTM neural network disturbance controller, the output of the LSTM neural network disturbance controller is used as a corresponding input of an NARX neural network control decoupling controller, and a temperature control value, a humidity control value, a illuminance control value and a wind speed control value output by the NARX neural network control decoupling controller are respectively 4 corresponding inputs of the LSTM neural network disturbance controller and corresponding inputs of a temperature regulation device, a humidity regulation device, an illuminance regulation device and a wind speed regulation device. The parameter decoupling control module is shown in fig. 2.
Compared with the prior art, the invention has the following obvious advantages:
firstly, the temperature, the humidity and the wind speed of the environment of the livestock and poultry house have the characteristics of nonlinearity, large hysteresis, complex dynamic change and the like, and a sensor for measuring the environment parameters of the livestock and poultry house is easily interfered, so that the environment parameter measurement of the livestock and poultry house often contains large noise. On the other hand, the measured parameters of the poultry house environment are more than the number of independent variables thereof, i.e. there is redundant information among these measured parameters. The time delay neural networks and the noise reduction self-encoders can utilize redundant information to suppress measurement noise of the time delay neural networks and the noise reduction self-encoders through compression and decompression processes of environment temperature, humidity, wind speed and illuminance information of the livestock and poultry houses, and predict and fuse environment measurement parameters of the livestock and poultry houses by applying the time delay neural networks and the noise reduction self-encoders in the process of processing big data of the environment of the livestock and poultry houses, so that the accuracy of the environment parameters of the livestock and poultry houses can be greatly improved.
The invention relates to a NARX neural network control decoupling controller, which is a dynamic recursive network for establishing the NARX neural network control decoupling controller by introducing the output of an LSTM neural network disturbance controller, the output and feedback realization of an FLNN neural network model and a fuzzy recursive neural network compensation controller, and the NARX neural network control decoupling controller is a data association modeling idea of function simulation function realized by a sequence of a plurality of time environment parameter control device control quantity state characteristic parameters expanded along the time axis direction of input control quantity state characteristic parameters of a livestock and poultry house environment parameter control device And the calculation accuracy is calculated, and the continuous and dynamic output of the control quantity state of the livestock and poultry house environment regulation and control device is realized.
Three, the LSTM neural network perturbation controller is a recurrent neural network with 4 interaction layers in a repeating network. The method not only can extract information from the output of the NARX neural network controlled decoupler and sequence data of the livestock and poultry house environment growth parameters like a standard recurrent neural network, but also can retain information of long-term correlation of the output of the NARX neural network controlled decoupler and the livestock and poultry house environment growth parameters from previous remote steps. In addition, because the sampling interval of the input control quantity of the environment parameter regulating and controlling device is relatively small, the input control quantity of the environment parameter regulating and controlling device has long-term spatial correlation, and the LSTM neural network disturbance controller has enough long-term memory to solve the problem, the accuracy of preventing the environment parameter grade of the livestock and poultry house from being disturbed when the output of the LSTM neural network disturbance controller is used as the input of the livestock and poultry environment regulating and controlling device is improved, and the accuracy and the robustness of controlling the environment parameter grade of the livestock and poultry house are improved.
And the FLNN function connection neural network model is composed of an input layer and an output layer, and has no hidden layer, so that compared with the traditional neural network, the FLNN function connection neural network model has smaller network calculation amount and higher training speed. The method can avoid updating the weight of the hidden layer, only needs to adjust the weight of the output layer, has higher convergence speed and less on-line calculation amount, simultaneously expands the input variable of the FLNN function connection neural network model, can improve the network resolution capability of the FLNN function connection neural network model, and improves the accuracy of the output control amount of the FLNN function connection neural network model.
The controller of the invention is composed of an ESN neural network model, a PID controller, an interference controller and a decoupling control module to realize combined control on the level of the livestock and poultry house breeding environment parameters, and the influence of the forecast values of the livestock and poultry house environment temperature, humidity, illuminance and wind speed on the control of the livestock and poultry house environment parameter level is considered in the regulation of the ESN neural network model, so that the pre-regulation of the livestock and poultry house environment parameter level is realized; the PID controller realizes dynamic and static adjustment of the level change of the environmental parameters of the livestock and poultry house, the decoupling control module decouples and disturbs the adjustment of the level of the environmental parameters of the livestock and poultry house, and the four adjustments act together to improve the accuracy and robustness of the level control of the environmental parameters of the livestock and poultry house.
Detailed Description
The technical scheme of the invention is further described by combining the attached drawings 1-6:
design of overall system function
The invention discloses an intelligent control system for livestock and poultry house breeding environment parameters, which realizes detection and intelligent decoupling control of the livestock and poultry house breeding environment parameters and consists of an livestock and poultry house environment parameter acquisition and control platform and a growth parameter intelligent decoupling control subsystem. The livestock and poultry house environment parameter acquisition and control platform comprises detection nodes, control nodes, gateway nodes, a field monitoring end, a cloud platform, a remote monitoring end and a mobile phone App of livestock and poultry house breeding environment parameters, wherein the detection nodes and the control nodes construct a CC2530 self-organizing communication network to realize self-organizing network communication among the detection nodes, the control nodes and the gateway nodes; the detection node sends the livestock and poultry house breeding environment parameters to be detected to the field monitoring end and the cloud platform through the gateway node, bidirectional transmission of the livestock and poultry house breeding environment parameters and relevant control information is achieved among the gateway node, the cloud platform, the field monitoring end and the mobile phone App, and the mobile phone APP can monitor the livestock and poultry house environment parameters in real time through the livestock and poultry house environment information provided by the cloud platform. The livestock and poultry house environment parameter acquisition and control platform is shown in figure 1.
Design of detection node
A large number of detection nodes of the CC 2530-based self-organizing communication network are used as livestock and poultry breeding environment parameter sensing terminals, and the mutual information interaction between the detection nodes and gateway nodes is realized through the self-organizing communication network. The detection node comprises a sensor for acquiring the temperature, humidity, wind speed and illuminance of the livestock and poultry house culture environment, a corresponding signal conditioning circuit, an MSP430 microprocessor and a CC2530 module; the software of the detection node mainly realizes the self-organizing network communication and the collection and pretreatment of the environmental parameters of the livestock and poultry house. The software is designed by adopting a C language program, the compatibility degree is high, the working efficiency of software design and development is greatly improved, the reliability, readability and transportability of program codes are enhanced, and the structure of the detection node is shown in figure 3.
Design of control node
The control node realizes mutual information interaction with the gateway node through a self-organizing communication network of the CC2530, and comprises 4 digital-to-analog conversion circuits corresponding to control external equipment, an MSP430 microprocessor, a CC2530 module and 4 external equipment controllers; the 4 external equipment controllers are respectively a temperature regulating device, a humidity regulating device, a wind speed regulating device and a light intensity regulating device; the control node is shown in figure 4.
Fourth, gateway node design
The gateway node comprises a CC2530 module, an NB-IoT module, an MSP430 microprocessor and an RS232 interface, the gateway node comprises a CC2530 module and is used for realizing a self-organizing communication network communicated with the detection node and the control node, the NB-IoT module is used for realizing data bidirectional interaction between the gateway and the cloud platform, and the RS232 interface is connected with the field monitoring terminal to realize information interaction between the gateway and the field monitoring terminal. The gateway node is shown in figure 5.
Fifthly, field monitoring terminal software design
The on-site monitoring end is an industrial control computer, mainly realizes the collection of the parameters of the livestock and poultry house and the processing of the parameters of the livestock and poultry house, realizes the information interaction with the gateway node, and has the main functions of communication parameter setting, data analysis, data management and growth parameter intelligent decoupling control subsystem. The structure of the intelligent decoupling control subsystem of the growth parameters is shown in figure 2. The management software selects Microsoft Visual + +6.0 as a development tool, calls the Mscomm communication control of the system to design a communication program, and the functions of the field monitoring end software are shown in FIG. 6. The growth parameter intelligent decoupling control subsystem is composed of an ESN neural network model, a noise reduction self-encoder A, a time delay neural network model, a PID controller, an integral loop, a fuzzy recursive neural network compensation controller, an FLNN neural network model and a parameter decoupling control module. The design process of each model is as follows:
1. ESN neural network model design
The expected values of temperature, humidity, wind speed and illuminance are used as corresponding inputs of an ESN (environmental health protection network) neural network model, the output of the noise reduction self-encoder A is used as a corresponding input of the ESN neural network model, and the difference value between the output of the ESN neural network model and the output of the noise reduction self-encoder A of the parameter decoupling control module is used as a growing environment grade difference value; an ESN (Echo state network) is a novel dynamic neural network, has all the advantages of the dynamic neural network, and can better adapt to nonlinear system identification compared with a common dynamic neural network because the Echo state network introduces a reserve pool concept. The reserve pool is a randomly connected reserve pool which is formed by converting a part connected among traditional dynamic neural networks, and the whole learning process is a process of learning how to connect the reserve pool. The "pool" is actually a randomly generated large-scale recursive structure in which the interconnection of neurons is sparse, usually denoted SD as the percentage of interconnected neurons in the total number of neurons N. The state equation of the ESN neural network model is as follows:
wherein W is the state variable of the neural network, W
inIs an input variable of the neural network; w
backConnecting a weight matrix for the output state variables of the neural network; x (n) represents the internal state of the neural network; w
outA connection weight matrix among a nuclear reserve pool of the ESN neural network model, the input of the neural network and the output of the neural network;
is the output deviation of the neural network or may represent noise; f ═ f [ f
1,f
2,…,f
n]N activation functions for neurons within the "pool of stores"; f. of
iIs a hyperbolic tangent function; f. of
outIs the epsilon output functions of the ESN neural network model. And the ESN neural network model outputs the actual control value of the growth parameter grade of the environment of the livestock and poultry house.
2. Noise reduction autoencoder A design
The output of the time delay neural network models is used as the input of a noise reduction self-encoder A, and the output of the noise reduction self-encoder A is used as the corresponding input of the ESN neural network model; a noise-reducing self-encoder (DAE) is a dimension-reducing method that converts high-dimensional data into low-dimensional data by training a multi-layer neural network having a small center layer. A noise-reducing self-encoder (DAE) is a typical three-layer neural network with an encoding process between a hidden layer and an input layer and a decoding process between an output layer and the hidden layer. The auto-encoder obtains the encoded representation (encoder) by an encoding operation on the input data and the reconstructed input data (decoder) by an output decoding operation on the hidden layer, the data of the hidden layer being the dimension-reduced data. A reconstruction error function is then defined to measure the learning effect of the auto-encoder. Constraints may be added based on the error function to generate various types of autoencoders. The encoder and decoder and the loss function are as follows:
an encoder: h ═ delta (Wx + b) (2)
the training process of AE is similar to BP neural network, W and W 'are weight matrix, b and b' are offset, h is output value of hidden layer, x is input vector,
![Figure BDA0002981221510000075](https://patentimages.storage.googleapis.com/79/25/52/30185803bde17c/BDA0002981221510000075.png)
to output the vector, δ is the excitation function, typically using a Sigmoid function or a tanh function. The noise reduction self-encoder A isThe sparse self-coding network is trained by adding noise data into input data, the data characteristics learned by the self-coding network are more robust due to the action of the noise data, the self-coding network is divided into a coding process and a decoding process, the coding process is from an input layer to a hidden layer, and the decoding process is from the hidden layer to an output layer. The self-coding network aims to make input and output as close as possible by utilizing an error function, obtain the optimal weight and bias of the self-coding network by reversely propagating the minimized error function and prepare for establishing a deep self-coding network model. In the noise reduction self-coding network process, random probability is used for setting certain values in the original data of the livestock and poultry house environment to be 0 to obtain data containing noise, according to the self-coding network coding and decoding principle, the data containing the noise of the livestock and poultry house environment are used for obtaining coded data and decoded data, finally, an error function is constructed through the decoded data and the original data, and the optimal network weight and bias are obtained through back propagation minimizing the error function. Original data of the environment of the livestock and poultry house are destroyed by adding noise, and then the destroyed data are input into a neural network as an input layer. The reconstruction result of the neural network of the noise reduction self-encoder A is similar to the original data of the environment of the livestock and poultry house, and by the method, the disturbance of the environment of the livestock and poultry house can be eliminated, and a stable structure can be obtained. The original livestock and poultry house environment input data is subjected to interference input by adding noise, then is input into an encoder to obtain characteristic expression, and is mapped to an output layer through a decoder.
3. PID controller design
The difference value of the ESN neural network model output and the noise reduction self-encoder A output of the parameter decoupling control module is used as a growth environment grade difference value, the change rates of the growth environment grade difference value and the growth environment grade difference value are respectively used as the input of a PID controller and the input of a fuzzy recurrent neural network compensation controller, and the PID controller output is used as the input of an integral loop and the corresponding input of an FLNN neural network model; the PID closed-loop controller is mainly composed of a proportion P and an integral I and a derivative D, and is mainly used for calculating a control quantity through the proportion, the integral and the derivative based on a growth environment grade difference value so as to realize effective control. The important basis of the PID closed-loop control is proportional control, while integral control can effectively reduce steady-state errors, but is most likely to cause overshoot to increase, differential control can promote the response speed of a large inertia system to be accelerated, and overshoot is effectively reduced, and the relation between PID input and output u (t) is as follows:
wherein e (t) represents an input; u (t) represents the output; kPRepresents a proportionality coefficient; kIRepresents an integral coefficient; kDRepresents a differential coefficient;
4. time delay neural network model design
The outputs of a plurality of groups of temperature, humidity, wind speed and illumination sensors are respectively used as the inputs of a plurality of corresponding time delay neural network models, and the outputs of the plurality of time delay neural network models are used as the inputs of a noise reduction self-encoder A; the Time Delay Neural Network (TDNN) is a self-adaptive linear network, wherein multiple groups of temperature, humidity, wind speed and illumination sensor outputs enter from the left side of the network, and become the input of a D + 1-dimensional vector after D-step Delay under the action of a single-step Delay line D, the vector is formed by combining signals output by multiple groups of temperature, humidity, wind speed and illumination sensors at the current K moments and signals output by multiple groups of temperature, humidity, wind speed and illumination sensors at D-1 times before K, the neuron adopts a linear activation function, and the Time Delay Neural network belongs to the variation of the traditional artificial Neural network. The time delay neural network structure consists of an input layer, an output layer and one or a plurality of hidden layers, and the neural network establishes a mapping relation between input and output. Different from the traditional neural network, the time delay neural network realizes the memory of the preamble input by delaying the input at the input layer, and delays the output values of a plurality of groups of temperature, humidity, wind speed and illumination sensors at the input layer, so that the network can jointly predict the level value output of the environment parameters of the livestock and poultry house at the current time point by using the output of the plurality of groups of temperature, humidity, wind speed and illumination sensors at the previous d steps and the output of the plurality of groups of temperature, humidity, wind speed and illumination sensors at the current time point, and for the time delay neural network with the delay step number of d at the input layer, R is a forward propagation operator of the time delay neural network, the relation between an input sequence X and an output sequence Y can be simply expressed as follows:
Y(t)=R(X(t),X(t-1),…,X(t-d)) (6)
5. FLNN neural network model design
The method comprises the following steps that 1 integral loop is formed by connecting 2 integral operators S in series, 2 integral operator connecting ends of each integral loop and the output of each integral loop are respectively used as 2 corresponding inputs of an FLNN neural network model, the output of a PID controller is used as the input of the integral loop and the corresponding input of the FLNN neural network model, and the outputs of the FLNN neural network model and a fuzzy recurrent neural network compensation controller are respectively used as 2 corresponding inputs of an NARX neural network control decoupler of a parameter decoupling control module; the FLNN functionally connected neural network is a functional neural network model in which the functional connection functions to multiply each component of the input pattern of the FLNN neural network model by the entire pattern vector, resulting in a product of the original pattern vectors. The FLNN function connection type neural network performs nonlinear expansion on the input mode of the FLNN function connection type neural network model in advance, introduces a high-order item into the FLNN function connection type neural network, and maps the input mode of the FLNN function connection type neural network model to a larger mode space through the nonlinear expansion on the input mode of the FLNN function connection type neural network model, so that the mode expression of the input signal of the FLNN function connection type neural network model is enhanced, and the network structure of the FLNN function connection type neural network model is greatly simplified. Although the input information of the FLNN neural network model input by the FLNN function connection type neural network model is not increased, the enhancement of the FLNN function connection type neural network model brings simplification of the network structure of the FLNN function connection type neural network model and improvement of the learning speed, the supervised learning can be realized by using a single-layer network, and the method has great advantages compared with a multi-layer forward neural network. The FLNN function connection type neural network model realizes supervised learning by using a single-layer network, and the solving process can be completed by the following self-adaptive supervised learning algorithm. The learning algorithm of the FLNN functional type connected neural network model may be represented by the following equation:
weight adjustment:
wherein: f
i(k)、
e
i(k) And w
n(k) Respectively the expected output, the estimated output, the error and the nth connection weight of the functional network in the kth step of the ith input mode; alpha is a learning factor and affects stability and convergence rate. The FLNN function connection neural network model adopts a function expansion mode to expand the input of the FLNN neural network model, so that the input of the FLNN neural network model is converted into another space, and the enhanced mode is used as the input of a network input layer of the FLNN function connection neural network model, and the method is used for better solving the nonlinear problem of the level control output control quantity of the environment parameters of the livestock and poultry house; the FLNN function connection neural network model is composed of an input layer and an output layer, and has no hidden layer, so that compared with the traditional neural network, the FLNN function connection neural network model has smaller network calculation amount and higher training speed. The method can avoid updating the weight of the hidden layer, only needs to adjust the weight of the output layer, has higher convergence speed and less on-line calculation amount, simultaneously expands the input variable of the FLNN neural network model, and can improve the network resolution capability of the FLNN function connection neural network model.
6. Fuzzy recurrent neural network compensation controller design
The difference value of ESN neural network model output and noise reduction self-encoder A output of the parameter decoupling control module is used as a growth environment grade difference value, the change rates of the growth environment grade difference value and the growth environment grade difference value are respectively used as the input of a PID controller and a fuzzy recurrent neural network compensation controller, and the FLNN neural networkThe outputs of the network model and the fuzzy recurrent neural network compensation controller are respectively used as 2 corresponding inputs of an NARX neural network control decoupling controller of the parameter decoupling control module; the fuzzy recurrent neural network compensation controller (HRFNN) is a 2-input 1-output network topology structure, and a network consists of 4 layers: an input layer, a membership function layer, a rule layer, and an output layer. The fuzzy recurrent neural network compensation controller comprises 2 input nodes, wherein each input node corresponds to m condition nodes, m represents a rule number, nm rule nodes and 1 output node. Layer I in the figure introduces the input into the network; the second layer fuzzifies the input, and the adopted membership function is a Gaussian function; the third layer corresponds to fuzzy reasoning; layer iv corresponds to the defuzzification operation. By using
![Figure BDA0002981221510000111](https://patentimages.storage.googleapis.com/4b/57/7f/da7e3c0c7abbc7/BDA0002981221510000111.png)
Representing the input and output of the ith node of the kth layer, respectively, the signal transfer process inside the network and the input-output relationship between the layers can be described as follows. Layer I: an input layer, each input node of the layer being directly connected to an input variable, the input and output of the network being represented as:
in the formula
And
for the input and output of the ith node of the network input layer, N represents the number of iterations.
Layer II: and in the membership function layer, nodes of the membership function layer fuzzify input variables, each node represents a membership function, and a Gaussian function is adopted as the membership function. The inputs and outputs of the network are represented as:
where mij and σijRespectively representing the mean center and width value of the j term Gaussian function of the ith linguistic variable of the II layer, wherein m is the number of all linguistic variables corresponding to the input node.
Layer III: the fuzzy inference layer, namely the rule layer, adds dynamic feedback to ensure that the network has better learning efficiency, and the feedback link introduces an internal variable hkAnd selecting a sigmoid function as an activation function of the internal variable of the feedback link. The inputs and outputs of the network are represented as:
in the formula of omega
jkIs the connecting weight value of the recursion part, the neuron of the layer represents the front-piece part of the fuzzy logic rule, the node of the layer performs pi operation on the output quantity of the second layer and the feedback quantity of the third layer,
![Figure BDA0002981221510000122](https://patentimages.storage.googleapis.com/e4/8a/af/3680b999e0f3f1/BDA0002981221510000122.png)
is the output of the third layer, and m represents the number of rules in a full connection. The feedback link mainly calculates the value of the internal variable and the activation strength of the corresponding membership function of the internal variable. The activation strength is related to the rule node matching degree of the layer 3. The internal variables introduced by the feedback link comprise two types of nodes: and the receiving node and the feedback node. The carrying node calculates an internal variable by using weighted summation to realize the defuzzification function; the result of fuzzy inference of hidden rules represented by internal variables. And the feedback node adopts a sigmoid function as a fuzzy membership function to realize the fuzzification of the internal variable. The membership function layer of the HRFNN network uses a local membership function, which is different from the local membership function: the feedback part adopts a global membership function on the domain of an internal variable to simplify the network structure and realize the feedback of global historical information. The number of the receiving nodes is equal to the number of the feedback nodes; the number of the bearing nodes is equal to the number of the nodes of the rule layer. The feedback quantity is connected to the 3 rd layer and is used as the input quantity of the fuzzy rule layer, and the output of the feedback node contains the fuzzy ruleThe fuzzy recurrent neural network of the activation intensity compensates the historical information output by the controller.
A fourth layer: the deblurring layer, i.e., the output layer. The layer node performs a summation operation on the input quantities. The inputs and outputs of the network are represented as:
in the formula lambdajIs the connection weight of the output layer. The fuzzy recurrent neural network has the performance approaching to a highly nonlinear dynamic system, the training error and the testing error of the fuzzy recurrent neural network added with the internal variable are respectively obviously reduced, the network prediction effect is superior to that of the fuzzy neural network with the self-feedback fuzzy recurrent neural network and the dynamic modeling, which shows that the learning capacity of the network is enhanced after the internal variable is added, and the dynamic characteristic of the environmental parameter control quantity of the livestock and poultry house is more fully reflected. The simulation result proves the effectiveness of the network. The fuzzy recurrent neural network HRFNN of the patent adopts a gradient descent algorithm added with cross validation to train the weight of the neural network. And performing error and error variable compensation control on the environmental parameters of the livestock and poultry house by using the HRFNN, introducing an internal variable into a feedback link by the HRFNN, performing weighted summation on the output quantity of the rule layer, performing defuzzification output as a feedback quantity, and using the feedback quantity and the output quantity of the membership function layer as the input of the rule layer at the next moment. The fuzzy recurrent neural network compensation controller outputs historical information containing the activation intensity of the regular layer and the output animal house environment parameter compensation control, and the fast tracking capability and the control accuracy of the HRFNN adaptive animal house environment parameter grade control nonlinear dynamic system are enhanced. Experiments show that the HRFNN can more accurately control the grade of the environmental growth parameters of the livestock and poultry house.
7. Design of parameter decoupling control module
The parameter decoupling control module consists of an LSTM neural network disturbance controller, an NARX neural network control decoupler and a noise reduction self-encoder B; the parameter decoupling control module is shown in fig. 2.
(1) LSTM neural network disturbance controller design
The output of the noise reduction self-encoder B is used as the corresponding input of an LSTM neural network disturbance controller, the temperature control value, the humidity control value, the illuminance control value and the wind speed control value output by an NARX neural network control decoupling controller are respectively 4 corresponding inputs of the LSTM neural network disturbance controller, and the output of the LSTM neural network disturbance controller is used as the corresponding input of the NARX neural network control decoupling controller; LSTM neural network perturbation controllers introduce mechanisms of Memory cells (Memory cells) and hidden layer states (Cell states) to control the transfer of information between hidden layers. The memory unit of an LSTM neural network has 3 Gates (Gates) as Input Gate, forgetting Gate and Output Gate. Wherein, the input gate can control the adding or filtering of new information; the forgetting door can forget the information to be lost and keep the useful information in the past; the output gate enables the memory unit to output only information related to the current time step. The 3 gate structures carry out operations such as matrix multiplication, nonlinear summation and the like in the memory unit, so that the memory still cannot be attenuated in continuous iteration. The long-short term memory unit (LSTM) structure unit is composed of a unit (Cell), an Input Gate (Input Gate), an Output Gate (Output Gate) and a forgetting Gate (Forget Gate). The LSTM neural network disturbance controller is a model which can last for long time and short term memory and is suitable for predicting time sequences to control the change of the disturbance of the livestock and poultry house breeding environment parameters, the LSTM neural network disturbance controller effectively prevents gradient disappearance during RNN training, and a Long Short Term Memory (LSTM) network is a special RNN. The LSTM neural network perturbation controller can learn long-term dependency information while avoiding the gradient vanishing problem. The LSTM adds a structure called a Memory Cell (Memory Cell) in a nerve node of a hidden layer of a neuron internal structure RNN for memorizing animal house environment parameters of past detection points and dynamic change information of a NARX neural network control decoupler, and adds three gate (Input, Forget, Output) structures for controlling the use of history information of the detection points. Setting the time sequence value of the input livestock and poultry house environment parameter and the output of the NARX neural network control decoupler as (x)1,x2,…,xT) The hidden layer state is (h)1,h2,…,hT) Then, time t has:
it=sigmoid(Whiht-1+WxiXt) (13)
ft=sigmoid(Whfht-1+WhfXt) (14)
ct=ft⊙ct-1+it⊙tanh(Whcht-1+WxcXt) (15)
ot=sigmoid(Whoht-1+WhxXt+Wcoct) (16)
ht=ot⊙tanh(ct) (17)
wherein it、ft、otRepresenting input, forget and output doors, ctRepresenting a cell, WhRepresenting the weight of the recursive connection, WxSigmoid and tanh represent the weights from the input layer to the hidden layer, and are two activation functions. The method comprises the steps of firstly establishing an LSTM neural network controller model, establishing a training set by utilizing preprocessed input quantity of the livestock house environment parameter device and time sequence value data of the livestock house environment parameters, and training the controller, wherein the LSTM neural network disturbance controller takes the time sequence and nonlinearity of the time sequence values output by the input livestock house environment parameters and the NARX neural network control decoupling controller into consideration to output control quantity for preventing livestock house breeding environment parameters, and has higher control precision of the time sequence value for outputting control quantity for controlling the livestock house environment parameter disturbance.
(2) Design of NARX neural network control decoupler
The output of the FLNN neural network model and the output of the fuzzy recurrent neural network compensation controller are respectively used as NARX neural network of the parameter decoupling control module2 corresponding inputs of a network controlled decoupler; the output of the LSTM neural network disturbance controller is used as the corresponding input of the NARX neural network control decoupling controller, and the temperature control value, the humidity control value, the illuminance control value and the wind speed control value output by the NARX neural network control decoupling controller are respectively 4 corresponding inputs of the LSTM neural network disturbance controller and the corresponding inputs of the temperature regulation device, the humidity regulation device, the illuminance regulation device and the wind speed regulation device; the NARX neural network controller is a dynamic recurrent neural network with output feedback connection, can be equivalent to a BP neural network with input time delay and is added with time delay feedback connection from output to input on a topological connection relation, and the structure of the NARX neural network controller consists of an input layer, a time delay layer, a hidden layer and an output layer, wherein an input layer node is used for signal input, a time delay layer node is used for time delay of input signals and output feedback signals, the hidden layer node performs nonlinear operation on the delayed signals by using an activation function, and the output layer node is used for performing linear weighting on hidden layer output to obtain final network output. Output h of ith hidden layer node of NARX neural network controlleriComprises the following steps:
output o of j output layer node of NARX neural networkjComprises the following steps:
the input layer, the time-extension layer, the hidden layer and the output layer of the NARX neural network controller are respectively 3-19-10-4 nodes.
(3) Noise reduction self-encoder B design
The outputs of a plurality of temperature, humidity, wind speed and illuminance sensors are used as a plurality of corresponding inputs of a noise reduction self-encoder B, and the output of the noise reduction self-encoder B is used as a corresponding input of an LSTM neural network disturbance controller; the design method of the noise reduction self-encoder B is similar to that of the patent; the design process of the noise reduction self-encoder B.
Design example of intelligent control system for breeding environmental parameters of livestock and poultry house
According to the actual condition of the intelligent control system for the livestock and poultry house breeding environment parameters, a plane layout installation diagram of detection nodes, control nodes and a gateway node field monitoring end of a livestock and poultry house breeding environment parameter detection and control platform is arranged in the system, sensors are evenly arranged in all directions of the livestock and poultry house according to the detection requirement, and the system is used for realizing the collection of the livestock and poultry house environment parameters and the intelligent decoupling control of growth parameters.
The technical means disclosed in the invention scheme are not limited to the technical means disclosed in the above embodiments, but also include the technical scheme formed by any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and such improvements and modifications are also considered to be within the scope of the present invention.