Disclosure of Invention
The invention aims to provide a big data detection system for livestock and poultry activity information, which monitors the temperature and activity information of livestock and poultry bodies in real time, thereby providing data and early warning for preventing livestock and poultry diseases.
The invention is realized by the following technical scheme:
the big data detection system for the livestock and poultry activity information realizes detection of livestock and poultry body temperature and activity information parameters and intelligent prediction of livestock and poultry postures, and comprises a cloud platform-based livestock and poultry sign parameter acquisition and intelligent prediction platform and a livestock and poultry activity big data prediction subsystem.
The invention further adopts the technical improvement scheme that:
the livestock and poultry physical sign parameter acquisition and intelligent prediction platform based on the cloud platform is composed of a detection node, a gateway node, an on-site monitoring end, the cloud platform and a mobile phone APP, wherein the detection node acquires the body temperature and activity information parameters of livestock and poultry and uploads the body temperature and activity information parameters to the cloud platform through the gateway node, the cloud platform end stores data and releases information, the mobile phone APP can monitor the body temperature and activity parameters of the livestock and poultry in real time through the livestock and poultry body temperature and activity information parameters provided by the cloud platform, the detection node is responsible for acquiring the body temperature and activity information parameters of the livestock and poultry, and the gateway node is used for realizing the bidirectional communication among the detection node, the on-site monitoring end, the cloud platform and the mobile phone APP and realizing the acquisition and prediction of the body temperature and activity information parameters of the livestock and poultry; the structure diagram of the cloud platform-based livestock and poultry sign parameter acquisition and intelligent prediction platform is shown in figure 1.
The invention further adopts the technical improvement scheme that:
the livestock and poultry activity big data prediction subsystem consists of 3 parameter detection modules, a BAM neural network model, 2 beat delay lines TDL, 2 ARIMA prediction models and an ART2 neural network model part, wherein a triaxial acceleration sensor ADXL362 senses X, Y and the acceleration of a detected livestock and poultry in a Z-axis direction as the input of the corresponding parameter detection modules respectively, 3 coupling coefficients output by the 3 parameter detection modules are used as the input of the BAM neural network model, a binary coupling coefficient output by the BAM neural network model is used as the corresponding input of the ART2 neural network model, a determined value and a fluctuation value output by the ART2 neural network model are used as the input of the 2 beat delay lines TDL and the corresponding input of the ART2 neural network model, the 2 beat delay line TDL outputs are respectively used as the corresponding input of the 2 ARIMA prediction models, and the 2 ARIMA prediction models are output as the corresponding input of the ART2 neural network model, the determined value c and the fluctuation value d output by the ART2 neural network model form a binary coefficient of c + di, and the binary coefficient output by the ART2 neural network model respectively corresponds to 5 different states of the livestock and poultry in horizontal climbing, walking, standing, horizontal lying and lateral lying; the structure diagram of the big data prediction subsystem of the livestock and poultry activities is shown in figure 2.
The invention further adopts the technical improvement scheme that:
the parameter detection module consists of an Adaline neural network model with a time-lag unit, an EMD empirical mode decomposition model, a GM (1,1) gray prediction model, a plurality of NARX neural network prediction models, 2 beat-to-beat delay lines TDL, 2 ARIMA prediction models and a wavelet neural network model of a binary coefficient; the output of the parameter detection sensor is used as the input of an Adaline neural network model with a time delay unit, the output of the Adaline neural network model with the time delay unit is used as the input of an EMD empirical mode decomposition model, the low-frequency trend value of a measured parameter output by the EMD empirical mode decomposition model is used as the input of a GM (1,1) gray prediction model, a plurality of different high-frequency fluctuation values of the measured parameter output by the EMD empirical mode decomposition model are respectively used as the input of a plurality of corresponding NARX neural network prediction models, the output of the GM (1,1) gray prediction model and the outputs of the NARX neural network prediction models are respectively used as the corresponding input of a wavelet neural network model of a dyadic coefficient, the determined value a and the fluctuation value b of the measured parameter output by the wavelet neural network model of the dyadic coefficient form a + bi of the size of the measured parameter, and the determined value a and the fluctuation value b of the measured parameter are respectively used as the input of a delay line TDL and the minbi coefficient of the dyadic coefficient Through 2 corresponding inputs of the network model, 2 outputs of the TDL are respectively used as the inputs of corresponding ARIMA prediction models, 2 outputs of the ARIMA prediction models are respectively used as the corresponding inputs of a wavelet neural network model of a binary coefficient, and the binary coefficient output by the wavelet neural network model of the binary coefficient is used as a measured parameter prediction value. The structure of the parameter detection module is shown in FIG. 3.
Compared with the prior art, the invention has the following obvious advantages:
firstly, the invention decomposes the livestock activity information sequence output by the original Adaline neural network model with a time delay unit into components of different frequency bands through an EMD empirical mode decomposition model, and each component displays different characteristic information hidden in the original sequence. To reduce the non-stationarity of the sequence. The data relevance of the high-frequency fluctuation part of the livestock and poultry activity process is not strong, the frequency is higher, the high-frequency fluctuation part represents the fluctuation component of the original sequence, and the high-frequency fluctuation part has certain periodicity and randomness, and the periodicity and the randomness are consistent with the periodicity change of the livestock and poultry activity process; the low-frequency components represent the variation trend of the original sequence in the activity process of the livestock and poultry. Therefore, the EMD can gradually decompose fluctuation components, period components and trend components in the livestock and poultry activity process, each decomposed component contains the same deformation information, mutual interference among different characteristic information is reduced to a certain extent, and the decomposed component change curves are smoother than original livestock and poultry activity deformation sequence curves. Therefore, EMD can effectively analyze deformation data of the livestock and poultry in the activity process under the multi-factor combined action, and each component obtained through decomposition is output by a GM (1,1) gray prediction model and establishment and better prediction of a plurality of NARX neural network prediction models. And finally, superposing the component prediction results to obtain a final fusion prediction result. Example researches show that the provided fusion prediction result has higher prediction precision.
Secondly, the time span of parameter measurement low-frequency trend in the process of predicting animal and poultry activities by adopting the GM (1,1) gray prediction model is long. The GM (1,1) grey prediction model can predict the parameter measurement low-frequency trend value at the future time according to the parameter measurement low-frequency trend value, after the parameter measurement low-frequency trend predicted by the method is measured, the parameter measurement low-frequency trend value is added into the original number series of the parameter measurement low-frequency trend respectively, a data model at the beginning of the number series is correspondingly removed, and then the prediction of the parameter measurement low-frequency trend is predicted. And by analogy, predicting a parameter measurement low-frequency trend value. The method is called an equal-dimensional gray number successive compensation model, and can realize long-time prediction. The driver can more accurately master the variation trend of the low-frequency trend of the parameter measurement, and preparation is made for effectively avoiding the fluctuation of the low-frequency trend of the parameter measurement.
The ARIMA prediction model is adopted to obey time sequence distribution on the basis of the original data of the determined value and the fluctuation value of the parameter measurement, the principle that the determined value and the fluctuation value of the parameter measurement have certain inertial trends is utilized, the determined value and the original time sequence variable of the fluctuation value of the parameter measurement of factors such as trend factors, periodic factors, random errors and the like are integrated, the non-stationary sequence is converted into a stationary random sequence with a zero mean value through methods such as differential data conversion and the like, and the number fitting and prediction of the determined value and the fluctuation value of the parameter measurement are carried out through repeated recognition, model diagnosis and comparison and selection of an ideal model. The method combines the advantages of autoregressive and moving average methods, has the characteristics of no data type constraint and strong applicability, and is a model for predicting the determined value and the fluctuation value of parameter measurement in a short term.
The BAM neural network with the binary association coefficient is a double-layer feedback neural network, and can realize the function of different association memory; which when an input signal is added to one of the layers, the other layer gets an output. There is no explicit input layer or output layer, since the initial mode can act on any layer of the network, and the information can also be propagated in both directions. The learning speed of the BAM neural network model is high, the convergence speed is low during BP learning, the final convergence can possibly reach a local minimum point instead of a global minimum point, and the BAM reaches an energy minimum point; the BAM neural network model is provided with a feedback network, and when an input has an error, the BAM neural network model not only can output an accurate fault reason, but also can correct the error of the original input. The BAM neural network model is suitable for systems that require correction of symptoms of erroneous inputs. The BAM neural network model improves the uncertain information processing capability of the material parameter sensor predicted value in the reasoning process by utilizing the characteristic of bidirectional association storage of the BAM neural network.
According to the scientificity and reliability of the classification of the livestock activity state grades, the ART2 neural network classifier with the binary coupling coefficient outputs 5 different predicted values through the ART2 neural network of the binary coupling coefficient according to the engineering practice experience of the livestock activity state, so that the dynamic degree of the livestock activity state is quantized into 5 different states of the livestock in the states of horizontal crawling, walking, standing, horizontal lying and lateral lying, and the classification of the livestock activity state dynamic performance and scientific classification are realized.
According to the invention, aiming at the uncertainty and randomness of the problems of sensor precision error, interference, measured value abnormality and the like in the parameter measurement process, the parameter value measured by the parameter sensor is converted into a binary coefficient form through the parameter detection module to be expressed, the ambiguity, the dynamic property and the uncertainty of the parameter measured by the parameter sensor are effectively processed, and the objectivity and the reliability of the parameter detected by the parameter sensor are improved.
According to the invention, aiming at the uncertainty and randomness of the problems of sensor precision error, interference, measured value abnormality and the like in the parameter measurement process, the parameter value measured by the parameter sensor is converted into a binary coefficient form through the parameter detection module to be expressed, the ambiguity, the dynamic property and the uncertainty of the parameter measured by the parameter sensor are effectively processed, and the objectivity and the reliability of the parameter value detection of the parameter sensor are improved.
Detailed Description
The technical scheme of the invention is further described by combining the attached drawings 1-6:
design of overall system function
The detection system provided by the invention can be used for detecting the body temperature and activity information parameters of livestock and poultry and predicting the postures of the livestock and poultry, and comprises a cloud platform-based livestock and poultry sign parameter acquisition and intelligent prediction platform and a livestock and poultry activity big data prediction subsystem. The livestock and poultry sign parameter acquisition and intelligent prediction platform based on the cloud platform comprises detection nodes of livestock and poultry sign parameters, gateway nodes, a field monitoring terminal, the cloud platform and a mobile phone App, and communication among the detection nodes and between the detection nodes and the gateway nodes is realized through a ZiGBee technology; the detection nodes send the detected livestock body temperature and activity parameters to the field monitoring end and the cloud platform through the gateway nodes, and bidirectional transmission of the livestock body temperature and activity information parameters is realized among the gateway nodes, the cloud platform, the field monitoring end and the mobile phone App. The cloud platform based livestock and poultry sign parameter acquisition and intelligent prediction 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 sensing terminals of the temperature and activity information parameter of the livestock, and the detection nodes realize mutual information interaction with gateway nodes through the self-organizing communication network. The detection node comprises a sensor for acquiring temperature and activity information parameters of livestock and poultry, a corresponding signal conditioning circuit, an STM32 singlechip and a CC2530 module; the software of the detection node mainly realizes the self-organizing network communication and the acquisition and the pretreatment of the temperature and activity information parameter parameters of the livestock and poultry. 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 4.
Third, gateway node design
The gateway node comprises a CC2530 module, an NB-IoT module, an STM32 single chip microcomputer and an RS232 interface, the gateway node realizes communication with the detection node through the CC2530 module, the NB-IoT module realizes data bidirectional interaction between the gateway and the cloud platform, and the RS232 interface is connected with the field monitoring end to realize information interaction between the gateway and the field monitoring end. The gateway node structure is shown in figure 5.
Fourth, software design of on-site monitoring terminal
The on-site monitoring end is an industrial control computer, the on-site monitoring end mainly collects the body temperature and activity information parameters of livestock and poultry and predicts the postures of the livestock and poultry, information interaction with the detection nodes is realized through gateway nodes, the on-site monitoring end mainly has the functions of communication parameter setting, data analysis and data management and intelligent prediction of the postures of the livestock and poultry through a livestock and poultry activity big data prediction subsystem, Microsoft Visual + +6.0 is selected as a development tool by the management software, a communication program is designed by calling an Mscomm communication control of the system, and the software function of the on-site monitoring end is shown in figure 6. The structure of the big data prediction subsystem of the livestock and poultry activities is shown in figure 2. The big data prediction subsystem of livestock and poultry activity is composed of 3 parameter detection modules, a BAM neural network model, 2 TDL (time domain delay line) according to beat, 2 ARIMA prediction models and an ART2 neural network model, the functional diagram of the big data prediction subsystem of livestock and poultry activity is shown in figure 2, and the design process is as follows:
1. design of parameter detection module
The three-axis acceleration sensor ADXL362 senses X, Y of the detected livestock and poultry and acceleration in the Z-axis direction as input of corresponding parameter detection modules respectively, 3 coupling coefficients output by the 3 parameter detection modules are used as input of a BAM neural network model, and output of the BAM neural network model is used as corresponding input of an ART2 neural network model; the parameter detection module consists of an Adaline neural network model with a time-lag unit, an EMD empirical mode decomposition model, a GM (1,1) gray prediction model, a plurality of NARX neural network prediction models, 2 beat-to-beat delay lines TDL, 2 ARIMA prediction models and a wavelet neural network model of a binary coefficient; the functional diagram of the parameter detection module is shown in FIG. 3;
(1) adaline neural network model design with time delay unit
The output of the parameter sensor is used as the input of an Adaline neural network model with a time delay unit, and the output of the Adaline neural network model with the time delay unit is used as the input of an EMD empirical mode decomposition model; the Adaline neural network model with the time delay unit consists of 2 beat delay lines TDL and Adaline neural networks, the output of a parameter sensor is used as the input of the corresponding beat delay line TDL, the output of the beat delay line TDL is used as the input of the Adaline neural network, the output of the Adaline neural network is used as the input of the corresponding beat delay line TDL, and the output of the beat delay line TDL is the output of the Adaline neural network model with the time delay unit; the Adaptive Linear Element (Adaptive Linear Element) of the Adaline neural network model is one of the early neural network models, and the input signal of the model can be written in the form of vector, x (k) ═ x0(K),x1(K),…xn(K)]TEach set of input signals corresponds to a set of weight vectors expressed as W (K) ═ k0(K),k1(K),…k(K)],x0(K) When the bias value of the Adaline neural network model is equal to minus 1, the excitation or inhibition state of the neuron is determined, and the network output can be defined as follows according to the input vector and the weight vector of the Adaline neural network model:
in the Adaline neural network model, a special input, namely an ideal response output d (K), is sent into the Adaline neural network model, then the output y (K) of the network is compared, a difference value is sent into a learning algorithm mechanism to adjust a weight vector until an optimal weight vector is obtained, the y (K) and the d (K) tend to be consistent, the adjusting process of the weight vector is the learning process of the network, the learning algorithm is a core part of the learning process, and the weight optimization searching algorithm of the Adaline neural network model adopts a least square method of an LMS algorithm.
(2) EMD empirical mode decomposition model design
The output of the parameter sensor is used as the input of an Adaline neural network model with a time delay unit, the output of the Adaline neural network model with the time delay unit is used as the input of an EMD empirical mode decomposition model, the measured low-frequency trend value of the parameters output by the EMD empirical mode decomposition model is used as the input of a GM (1,1) gray prediction model, the measured high-frequency fluctuation values of a plurality of parameters output by the EMD empirical mode decomposition model are respectively used as the input of a plurality of corresponding NARX neural network prediction models, and the EMD empirical mode decomposition is an adaptive signal screening method and has the characteristics of simplicity and intuition in calculation, and based on experience and self adaptation. It can screen the trends of different characteristics existing in the parameter measurement signal step by step to obtain a plurality of high frequency fluctuation parts (IMF) and low frequency trend parts. The IMF component decomposed by EMD empirical mode contains components of different frequency bands of parameter measurement signals from high to low, and the frequency resolution contained in each frequency band changes along with the signals, so that the self-adaptive multi-resolution analysis characteristic is realized. The purpose of using EMD empirical mode decomposition is to extract parameter measurement information more accurately. The IMF component must satisfy two conditions simultaneously: in a parameter measurement signal to be decomposed, the number of extreme points is equal to the number of zero-crossing points, or the difference is one at most; the envelope mean defined by the local maxima and local minima is zero at any one time. The EMD empirical mode decomposition method aims at the screening process steps of the Adaline neural network model output value signals with the time delay units as follows:
(a) all local extreme points of the output value signals of the Adaline neural network model with the time delay unit are connected by three sample lines to form an upper envelope line.
(b) Local minimum value points of Adaline neural network model output values with time delay units are connected by three spline lines to form a lower envelope line, and the upper envelope line and the lower envelope line should envelop all data points.
(c) The average of the upper and lower envelope lines is denoted as m1(t), obtaining:
x(t)-m1(t)=h1(t) (2)
x (t) is the output value raw signal of Adaline neural network model with time-lag unit, such asFruit h1(t) is an IMF, then h1(t) is the first IMF component of x (t). Note c1(t)=h1k(t), then c1(t) is the first component of signal x (t) that satisfies the IMF condition.
(d) C is to1(t) separating from x (t) to obtain:
r1(t)=x(t)-c1(t) (3)
will r is1(t) repeating steps (a) to (c) as raw data to obtain the 2 nd component c satisfying IMF condition of x (t)2. The cycle is repeated n times to obtain n components of the signal x (t) satisfying the IMF condition. Thus, the output of the Adaline neural network model with the time-lag unit is decomposed into a low-frequency trend part and a plurality of high-frequency fluctuation parts through an EMD empirical mode decomposition model, and the EMD empirical mode decomposition model is shown in figure 2.
(3) GM (1,1) grey prediction model design
The parameter measurement low-frequency trend value output by the EMD empirical mode decomposition model is used as the input of a GM (1,1) gray prediction model, a plurality of parameter measurement high-frequency fluctuation values output by the EMD empirical mode decomposition model are respectively used as the input of a plurality of corresponding NARX neural network prediction models, and the GM (1,1) gray prediction model output and the NARX neural network prediction model outputs are respectively used as the corresponding input of a wavelet neural network model of a binary coefficient; compared with the traditional statistical prediction method, the GM (1,1) gray prediction method has more advantages that whether the prediction variable obeys normal distribution or not is not required to be determined, large sample statistics is not required, the prediction model is not required to be changed at any time according to the change of the parameter measurement low-frequency trend value input variable, a uniform differential equation model is established through an accumulation generation technology, the accumulation parameter measurement low-frequency trend original value is restored to obtain a prediction result, and the differential equation model has higher prediction precision. The essence of establishing the GM (1,1) gray prediction model is that the low-frequency trend value original data is subjected to once accumulation generation, so that a generated sequence presents a certain rule, and a fitted curve is obtained by establishing a differential equation model so as to predict the parameter measurement low-frequency trend value.
(4) NARX neural network prediction model design
The parameter measurement low-frequency trend value output by the EMD empirical mode decomposition model is used as the input of a GM (1,1) gray prediction model, a plurality of parameter measurement high-frequency fluctuation values output by the EMD empirical mode decomposition model are respectively used as the input of a plurality of corresponding NARX neural network prediction models, and the GM (1,1) gray prediction model output and the NARX neural network prediction model outputs are respectively used as the corresponding input of a wavelet neural network model of a binary coefficient; the NARX neural network prediction model is a dynamic recurrent neural network with output feedback connection, can be equivalent to a BP neural network with input time delay on a topological connection relation and is added with time delay feedback connection from output to input, and the structure of the NARX neural network prediction model is composed 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 uses an activation function to perform nonlinear operation on the delayed signals, and an 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 prediction modeliComprises the following steps:
output o of j output layer node of NARX neural networkjComprises the following steps:
the input layer, the time delay layer, the hidden layer and the output layer of the NARX neural network are respectively 2-19-10-1 nodes.
(5) ARIMA prediction model design
The parameter measurement determination value a and fluctuation value b of wavelet neural network output of the binary coefficient are respectively used as the input of corresponding beat delay line TDL, 2 outputs of beat delay line TDL are respectively used as the input of corresponding ARIMA prediction model, 2 outputs of ARIMA prediction model are respectively used as the corresponding input of wavelet neural network model of the binary coefficient, the ARIMA (automatic regression Integrated Moving Average) prediction model is an Autoregressive integral sliding Average model, and the Autoregressive model (AR) and the sliding Average model (Moving Average, MA) are organically combined to form a comprehensive prediction method. As one of effective modern data processing methods, the method is known as the most complex and highest-level model in a time sequence prediction method, in practical application, because an input original data sequence often shows a certain trend or cycle characteristic, the requirement of an ARMA model on the stationarity of a time sequence is not met, and taking difference is a convenient and effective method for eliminating data trend. A model established based on the differentiated data sequence is called an ARIMA model and is marked as { Xt } -ARIMA (p, d, q), wherein p and q are called orders of the model, and d represents the difference times. Obviously, when d is 0, the ARIMA model is an ARMA model, which is defined as:
xt=b1xt-1+…+bpxt-p+εt+a1εt-1+…+aqεt-q (6)
{xtdata sequence of a determined value a and a fluctuating value b for the measurement of parameters of the wavelet neural network output for dyadic coefficients to be predicted, { epsilon }t}~WN(0,σ2). The ARIMA model building mainly comprises model identification, parameter estimation and model diagnosis. The model identification mainly comprises the preprocessing of a time sequence and the preliminary order determination of model parameters; after the order of the model is fixed, unknown parameters in the model are estimated by observing values through a time sequence and combining the values of p, d and q; the diagnosis of the model is mainly a significance test for the whole model and a significance test for parameters in the model. Generally, the establishment of the model is a continuous optimization process, and the model optimization is commonly used by AIC and BIC criteria, namely the smaller the value of the minimum information criterion is, the more suitable the model is, and the BIC criteria is an improvement on the deficiency of the AIC criterion on a large sample sequence. The time series can be fitted with an ARIMA (p, d, q) model the ARIMA (p, d, q) modeling steps are as follows:
A. and obtaining a parameter measurement determination value a and a fluctuation value b sequence output by the wavelet neural network of the binary coefficient.
B. And judging the stationarity of the sequence, if the sequence is not stationary, performing data preprocessing and differential operation on the data to stabilize the sequence, and determining the value of the differential order d.
C. When the post-differential sequence is a stationary non-white noise sequence, we can select an ARMA (p, q) model of the appropriate order to model the sequence.
D. And estimating unknown parameters in the model according to the identified model and the order thereof.
E. And (5) testing the residual sequence, and testing whether the preliminary model is effective by using a statistical test method.
F. And predicting the future development trend of the smoothed time series by using the obtained fitting model.
(6) Wavelet neural network model design of binary combined coefficient
The wavelet neural network model of the dyadic couple coefficient outputs a definite value a and a fluctuation value b of the parameter measurement value, which form a dyadic couple coefficient of the parameter measurement value as a + bi, the definite value a and the fluctuation value b of the parameter measurement value are respectively used as the input of a corresponding beat delay line TDL and 2 corresponding inputs of the wavelet neural network model of the dyadic couple coefficient, the TDL outputs of 2 beat delay lines are respectively used as the input of corresponding ARIMA prediction models, the 2 ARIMA prediction models are respectively used as the corresponding inputs of the wavelet neural network model of the dyadic couple coefficient, and the wavelet neural network model of the dyadic couple coefficient outputs a dyadic couple coefficient value of a parameter output by a parameter detection module; the structure of the parameter detection module is shown in fig. 2. The wavelet Neural network model WNN (wavelet Neural networks) is a feedforward network which is provided by combining an artificial Neural network on the basis of a wavelet theory. The method takes a wavelet function as an excitation function of a neuron, and the expansion, translation factors and connection weights of the wavelet are adaptively adjusted in the optimization process of an error energy function. An input signal of the wavelet neural network model can be represented as an input one-dimensional vector xi(i ═ 1,2, …, n), the output signal is denoted yk(k ═ 1,2, …, m), the calculation formula of the wavelet neural network model output layer output value is:
in the formula omega
ijInputting the connection weight between the i node of the layer and the j node of the hidden layer,
as wavelet basis functions, b
jIs a shift factor of the wavelet basis function, a
jScale factor, omega, of wavelet basis functions
jkThe connection weight between the node of the hidden layer j and the node of the output layer k. The correction algorithm of the weight and the threshold of the wavelet neural network model in the patent adopts a gradient correction method to update the network weight and the wavelet basis function parameters, so that the output of the wavelet neural network continuously approaches to the expected output. The output of the wavelet neural network model is a dynamic binary coefficient representing the value of the parameter measuring sensor in a period of time, the dynamic binary coefficient is a + bi, and the a + bi forms the dynamic binary coefficient value of the measured parameter output by the parameter measuring sensor in a period of time.
2. BAM neural network model design
3 joint coefficients output by the 3 parameter detection modules are used as the input of the BAM neural network model, and a binary joint coefficient output by the BAM neural network model is used as the corresponding input of the ART2 neural network model; in the BAM neural network model topological structure, the initial mode of the network input end is x (t), and the initial mode is obtained by a weight matrix W1Weighted and then reaches the y end of the output end and passes through the transfer characteristic f of the output nodeyNon-linear transformation of (1) and (W)2The matrix is weighted and returns to the input end x, and then the transfer characteristic f of the output node at the x end is passedxThe nonlinear transformation of the BAM neural network model is changed into the output of the input terminal x, and the operation process is repeated, so that the state transition equation of the BAM neural network model is shown in an equation (8).
The output of the BAM neural network model is a dynamic binary coefficient representing the activity state of the livestock and poultry in a period of time, the dynamic binary coefficient is a + bi, and the a + bi forms a binary coefficient value of the activity state of the livestock and poultry in a period of time.
3. Design of ARIMA prediction model
The determined value and the fluctuation value of the ART2 neural network model output are used as the input of 2 beat-to-beat delay lines TDL and the corresponding input of the ART2 neural network model, the 2 beat-to-beat delay line TDL outputs are respectively used as the corresponding 2 ARIMA prediction model inputs, and the 2 ARIMA prediction model outputs are used as the corresponding input of the ART2 neural network model; the ARIMA prediction model design method refers to the ARIMA prediction model design process and method of the parameter detection module.
4. Design of ART2 neural network model
The binary joint coefficient output by the BAM neural network model is used as the corresponding input of the ART2 neural network model, the determined value and the fluctuation value output by the ART2 neural network model are used as the input of 2 beat-to-beat delay lines TDL and the corresponding input of the ART2 neural network model, the 2 beat-to-beat delay line TDL outputs are respectively used as the corresponding input of 2 ARIMA prediction models, the 2 ARIMA prediction models are output as the corresponding input of the ART2 neural network model, the determined value and the fluctuation value output by the ART2 neural network model form a binary joint coefficient, and 5 different binary joint coefficients respectively correspond to 5 different states of the livestock and poultry in horizontal climbing, walking, standing, horizontal lying and lateral lying; the ART2 neural network model structure comprises an F1 attention subsystem and an F2 orientation subsystem, wherein the F1 layer is divided into an upper layer,
The middle layer and the lower layer form two closed positive feedback loops respectively to realize the functions of feature enhancement and noise suppression; the short-term memory STM of the M-dimensional state vector representation network is formed by M neurons formed by long-term and short-term memories F1 and F2; the inner and outer star connection weight vectors of F1 and F2 form the self-adaptive long-term memory LTM of the network; the network comprises two functional neurons represented by hollow and solid, wherein the hollow neuron represents the superposition of input stimuli, and the solid neuron represents the mode of an input vector. The ART2 neural network model outputs a determined value c and a fluctuation value d, which form a binary coefficient of c + di, and 5 different binary coefficients respectively correspond to 5 different states of the livestock and poultry in the states of horizontal climbing, walking, standing, horizontal lying and lateral lying. The corresponding relation between the binary coefficient output by the ART2 neural network model and the livestock and poultry states is shown in table 1.
Table 15 corresponding relation table of different binary coefficients and animal state
5. Livestock and poultry sign parameter acquisition and intelligent prediction platform design based on cloud platform
The system comprises detection nodes of livestock and poultry physical sign parameters, gateway nodes, an on-site monitoring terminal, a cloud platform and a mobile phone App, wherein communication among the detection nodes and between the detection nodes and the gateway nodes is realized through a ZiGBee technology; the detection nodes send the detected livestock body temperature and activity parameters to the field monitoring end and the cloud platform through the gateway nodes, and bidirectional transmission of the livestock body temperature and activity information parameters is realized among the gateway nodes, the cloud platform, the field monitoring end and the mobile phone App; according to the distribution condition of the livestock and poultry parameters, the wearing mode is adopted to detect that the nodes are worn on the body surfaces of the livestock and poultry, the gateway nodes and the field monitoring end are placed in a livestock and poultry farm, the detection of the livestock and poultry temperature and activity parameter information is realized by the detection nodes, and the monitoring of the livestock and poultry activity information and the intelligent prediction of the livestock and poultry postures are realized through the system.
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 it would be apparent to those skilled in the art that several modifications and adaptations can be made without departing from the principles of the invention and are intended to be within the scope of the invention.