CN109886350A - A method for predicting the digestible energy of dairy cows' diets based on the kernel extreme learning machine - Google Patents

Abstract

The invention provides a method for predicting the digestive energy of dairy cows' ration based on a nuclear extreme learning machine, which belongs to the field of nutritional value evaluation of livestock and poultry rations. Generate the dairy cow's dietary digestible energy prediction sample and divide it into training sample set and test sample set; (2) For the established training sample set, construct the output of the extreme learning machine network and express it in the form of a matrix; (3) Select the Gaussian kernel function Solve, determine the parameter set of the kernel function, and obtain the output function based on the KELM prediction model; (4) Compare the test sample with the prediction results of the KELM model, and calculate the mean absolute error, mean absolute percentage error and mean absolute percentage error between the predicted digestibility and the true value. The square root error is used to evaluate the effectiveness of the prediction method. The prediction method provided by the present invention belongs to a non-parametric machine learning model, which can perform effective prediction only by learning the training samples, and can obtain higher prediction accuracy.

Description

A method for predicting the digestible energy of dairy cows' diets based on the kernel extreme learning machine

Technical field:

The invention belongs to the field of nutritional value evaluation of livestock and poultry rations, and in particular relates to a method for predicting the digestibility of dairy cow rations based on a nuclear extreme learning machine.

Background technique:

Comprehensive and accurate evaluation of dietary nutrition and feeding value is a long-term concern of livestock breeders, feed suppliers and animal nutrition experts. Among them, the prediction and evaluation of dietary energy digestion of dairy cows is an important measure of feed nutrition and feeding value. aspect. By establishing a prediction model for the digestibility of dairy cows' dietary energy, it is possible to accurately grasp the dietary energy digestion of dairy cows in advance, so as to optimize the feed mix and management, thereby improving the breeding efficiency, and also meeting the development needs of modern animal husbandry.

The traditional mathematical model for predicting the dietary energy digestion index of dairy cows is mainly based on the linear regression (LR) method. The LR-based model is a typical parameter model, which is characterized by simple and fast modeling, and can reflect the predicted value within a certain range. The change trend of the data, but the parametric model usually assumes the form of the objective function followed by the predicted data, and estimates the parameters of the objective function during the training process, and then determines the previously proposed hypothetical model. However, since the animal body itself, including the cow, is a complex system, this assumption about the predictive model does not always hold, and when it does not hold, there will be large errors.

Extreme learning machine (ELM) is a machine learning algorithm based on non-parametric model. It usually does not need to make any assumptions about the objective function in advance during modeling. Compared with other non-parametric models, it has fast learning ability and strong generalization characteristics, and currently has a wide range of prediction fields in various fields. Application, the patent of the present invention is to use KELM technology to predict the digestibility of dairy cows' diets.

Invention content:

The purpose of the present invention is to provide a method for predicting the digestibility of dairy cows' ration based on a nuclear extreme learning machine.

The above purpose can be achieved through the following technical solutions:

1. A method for predicting the digestibility of dairy cow ration based on nuclear extreme learning machine, characterized in that the method comprises the following steps:

(1) Measure the dietary nutrient intake and digestible energy data of dairy cows, and generate the dairy cow's dietary digestible energy prediction sample, which is divided into two parts: training sample set and test sample set;

(2) Constructing a training sample set for the prediction of dietary digestibility of dairy cows based on the dairy cow's dietary digestible energy prediction samples generated in step (1). Among them, x _i ∈ R ⁿ represents the ith n-dimensional input vector, t _i ∈ R ^m represents the corresponding output vector, and N represents the number of training samples;

(3) For the training sample set established in step (2), construct the output of the extreme learning machine network, and the goal is to minimize the error of the output, that is: in, is the number of hidden layer nodes, w _i =[w _i1 ,w _i2 ,...,w _in ] ^T is the weight vector connecting the ith hidden layer node and the input layer node, β _i =[β _i1 ,β _i2 ,…,β _im ] ^T is the weight vector connecting the ith hidden layer node and the output layer node, g( ) is the activation function, and b _i is the bias of the ith hidden layer node;

(4) Rewrite the equation in step (3) into matrix form, that is: HB=T, and have,

Among them, H is the output matrix of the hidden layer node, and the jth row of the hidden layer feature map B is the output weight matrix, T is the expected output matrix;

(5) Solve the matrix in step (4), define the kernel function k(x _i , x _j )=h(x _i )·h(x _j ), and at the same time, define the kernel matrix The kernel extreme learning machine ( ^KELM ) is formed. In addition, the parameter I/C is chosen to be added to HHT so that its eigenvalue is not zero, which is used to improve the stability and generalization ability of KELM, where I is the identity matrix, C is the penalty factor, then the network output weight matrix B of KELM is: B=H ^T (I/C+HH ^T ) ^-1 T=H ^T (I/C+Ω) ^-1 T;

(6) Bring in the kernel function and parameter I/C defined in step (5) to obtain the output function based on the KELM prediction model:

(7) For the KELM prediction model in step (6), take the dairy cow's dietary digestible energy prediction test sample as the input x, calculate the output function f(x) of the prediction model, and calculate the average absolute error between the prediction result and the real value MAE, mean absolute percentage error MAPE and root mean square error RMSE to evaluate the effectiveness of the method for predicting dietary digestibility of dairy cows based on kernel extreme learning machine.

2. The method for predicting the dietary digestibility of dairy cows based on a nuclear extreme learning machine according to claim 1, characterized in that, for the training sample set data for predicting the dietary digestibility of dairy cows in step (2), according to the requirements of the CNCPS standard, Taking the intake of dairy cows' dietary CNCPS components (PA, PB1, PB2, PB3, PC, CA, CB1, CB2, CC) as the input x _i , and taking the dietary digestible energy as the corresponding output t _i , i =1,...,N, where N is the number of training samples.

3. The method according to claim 1, wherein the kernel function selected in step (5) is a Gaussian kernel function: k(x _i , x _j )=exp (-‖x _i -x _j ‖ ² /2σ ² ).

4. The method for predicting the dietary digestibility of dairy cows based on a kernel extreme learning machine according to claim 1, characterized in that, for the kernel function selected in step (5), a 5-fold cross-validation network method is used to determine the parameter set of the kernel function {C,σ}.

Beneficial effects of the present invention:

By replacing the traditional linear regression method based on the parametric model, the present invention based on the non-parametric model kernel extreme learning machine method replaces the traditional linear regression method based on the parametric model to predict the dietary digestibility of dairy cows, without making any assumptions about the prediction model in advance, and only by learning the training samples. Compared with traditional artificial neural network and support vector machine prediction models, it can obtain higher prediction accuracy, and is especially suitable for complex system prediction problems such as dairy cow ration energy digestion.

Description of drawings:

Figure 1 is a schematic flowchart of the method for predicting the digestibility of a dairy cow's ration based on the nuclear extreme learning machine of the present invention.

Accompanying drawing 2 is a comparison chart of different methods for predicting the digestibility of dairy cows' diets.

Figure 3 is a graph showing the comparison of the performance of different methods of training samples for predicting the digestibility of diets.

Figure 4 is a comparison chart of the performance of different methods for predicting the digestibility of dietary digestibility of the test samples.

Detailed ways:

The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are intended to illustrate the present invention, but not to limit the scope of the present invention.

The invention provides a method for predicting the dietary digestibility of dairy cows based on a nuclear extreme learning machine, which mainly includes the following steps:

1. Measure the dietary nutrient intake and digestible energy data of dairy cows, and generate the prediction sample of dairy cow's dietary digestible energy, which is divided into two parts: training sample set and test sample set.

2. Construct a training sample set for prediction of dietary digestibility of dairy cows At the same time, according to the requirements of CNCPS standards, the intake of dairy cows' CNCPS components (PA, PB1, PB2, PB3, PC, CA, CB1, CB2, CC) was taken as the input amount _xi , and the digestible energy of the diet was taken as the The corresponding output quantities t _i , i=1,...,N, where x _i ∈ R ⁿ represents the i-th n-dimensional input vector, t _i ∈ R ^m represents the corresponding output vector, and N represents the number of training samples.

3. Construct the output of the extreme learning machine network, the goal is to minimize the error of the output, that is: in, is the number of hidden layer nodes, w _i =[w _i1 ,w _i2 ,...,w _in ] ^T is the weight vector connecting the ith hidden layer node and the input layer node, β _i =[β _i1 ,β _i2 ,...,β _im ] ^T is the weight vector connecting the ith hidden layer node and the output layer node, g( ) is the activation function, and b _i is the bias of the ith hidden layer node. The above equation can be expressed in matrix form as: HB=T, and there is,

Among them, H is the output matrix of the hidden layer node, and the jth row of the hidden layer feature map B is the output weight matrix, and T is the expected output matrix.

4. Solve the above matrix and define the kernel function k(x _i ,x _j )=h(x _i )·h(x _j ), where the selected kernel function is a Gaussian kernel function: k(x _i ,x _j ) =exp(-‖x _i -x _j ‖ ² /2σ ² ), meanwhile, define the kernel matrix The kernel extreme learning machine ( ^KELM ) is formed. In addition, the parameter I/C is chosen to be added to HHT so that its eigenvalue is not zero, which is used to improve the stability and generalization ability of KELM, where I is the identity matrix, C is the penalty factor, then the network output weight matrix B of KELM is: B= ^HT (I/C+HHT ) ^{-1 T} = ^HT (I/C+Ω) ^-1 T.

5. Use the 5-fold cross-validation network method to determine the parameter set {C,σ} of the kernel function. According to the determined kernel function and parameter I/C, the output function based on the KELM prediction model is finally obtained:

6. Taking the input component of the dairy cow's dietary digestible energy prediction test sample as the input x, calculate the output function f(x) of the prediction model, and calculate the average absolute error MAE, average absolute error between the prediction result and the output component (true value) of the test sample. The absolute percentage error MAPE and the root mean square error RMSE were used to evaluate the effectiveness of the method for predicting the dietary digestibility of dairy cows based on the kernel extreme learning machine.

The overall flow chart of the above steps is shown in Figure 1. The training and testing are completed based on the MATLAB R2010b platform. The prediction comparison of different methods is shown in Figure 2. The prediction performance comparison of different methods for training samples and test samples is shown in Figure 3 and Figure 4.

Claims (4)

1. A method for predicting the daily ration digestion energy of dairy cows based on a nuclear extreme learning machine is characterized by comprising the following steps:

(1) actually measuring the nutrient intake and the digestion energy data of the daily ration of the dairy cow to generate a daily ration digestion energy prediction sample of the dairy cow, and dividing the daily ration digestion energy prediction sample into a training sample set and a test sample set;

(2) constructing a milk cow daily ration digestion energy prediction training sample set for the milk cow daily ration digestion energy prediction sample generated in the step (1)Wherein x is_i∈RⁿRepresenting the ith n-dimensional input vector, t_i∈R^mRepresenting the corresponding output vector, and N representing the number of training samples;

(3) constructing the output of the extreme learning machine network for the training sample set established in the step (2), wherein the aim is to minimize the output error, namely:wherein,to imply the number of layer nodes, w_i＝[w_i1,w_i2,…,w_in]^TFor the weight vector connecting the ith hidden layer node and the input layer node, β_i＝[β_i1,β_i2,…,β_im]^TFor the weight vector connecting the i-th hidden layer node with the output layer node, g (-) is the activation function, b_iA bias for the ith hidden layer node;

(4) rewriting the equation in step (3) into a matrix form, namely: HB ═ T, and has,

wherein H is the output matrix of hidden layer node, and the jth row of hidden layer feature mapping B is an output weight matrix, and T is an expected output matrix;

(5) solving the matrix in the step (4) to define a kernel function k (x)_i,x_j)＝h(x_i)·h(x_j) While defining a kernel matrixForm a Kernel Extreme Learning Machine (KELM), and choose to add the parameter I/C to HH^TAnd the characteristic root is not zero, and the method is used for improving the stability and generalization capability of the KELM, wherein I is a unit matrix, C is a penalty factor, and the network output weight matrix B of the KELM is as follows: b ═ H^T(I/C+HH^T)^-1T＝H^T(I/C+Ω)^-1T；

(6) Substituting the kernel function and the parameter I/C defined in the step (5) to obtain an output function based on a KELM prediction model:

(7) and (4) regarding the KELM prediction model in the step (6), taking the cow daily ration digestion energy prediction test sample as an input x, calculating an output function f (x) of the prediction model, and evaluating the effectiveness of the cow daily ration digestion energy prediction method based on the kernel-limit learning machine by calculating the average absolute error MAE, the average absolute percentage error MAPE and the root mean square error RMSE of the prediction result and the true value.

2. The method for predicting the daily ration digestion energy of dairy cows based on the nuclear limit learning machine as claimed in claim 1, wherein the method for predicting the daily ration digestion energy of dairy cows in the step (2) is characterized in that the intake of the CNCPS component (PA, PB1, PB2, PB3, PC, CA, CB1, CB2, CC) of the daily ration of dairy cows is used as the input x according to the requirements of the CNCPS standard_iTaking the digestion energy of daily ration as the corresponding output t_iI is 1, …, and N is the number of training samples.

3. The method for predicting the daily ration digestion energy of the dairy cow based on the kernel-based extreme learning machine as claimed in claim 1, wherein the kernel function selected in the step (5) is a gaussian kernel function: k (x)_i,x_j)＝exp(-‖x_i-x_j‖²/2σ²)。

4. The method for predicting the daily ration digestion energy of dairy cows based on the kernel-based extreme learning machine as claimed in claim 1, wherein the parameter set { C, σ } of the kernel function is determined by a 5-fold cross validation network method for the kernel function selected in the step (5).