Elastic network personnel health monitoring model based on differential evolution optimization

Abstract

The invention relates to the technical field of hot-line work, in particular to an elastic network personnel health monitoring model based on differential evolution optimization, which comprises data collection and preprocessing, and the data collection comprises collection of body temperature, heart rate and blood pressure data of each individual of a helmet wearer; a cost function of an elastic network regression model is defined by combining a regularization method of Lasso regression and ridge regression, hyper-parameters in the elastic network regression model are optimized by using differential evolution, an optimal model is obtained through continuous iteration, an input feature vector is introduced into the model, the model outputs a health state label, and the health of elastic network personnel is predicted. According to the method, the hyper-parameters of the elastic network regression model are optimized by adopting the differential evolution algorithm, and the optimal model is obtained through continuous iteration, so that the health state of a person is judged more accurately, and new data are continuously collected to update and optimize the model so as to adapt to different conditions and changes.

Description

Technical Field

[0001] The present invention relates to the technical field of live working, and in particular to a health monitoring model for elastic network personnel based on differential evolution optimization. Background Art

[0002] The present invention mainly studies the intelligent monitoring insulating helmet with communication function, which is applied to various work sites, especially the live power distribution maintenance work site. In traditional work, the traditional helmet only provides a single function of physical protection, and provides communication, lighting, positioning, data recording and other functions for the workers. There are deficiencies in the safety management, communication and collaboration, and information recording of the workers. Even if there is a helmet with a monitoring function, it only relies on a single sensor for simple data transmission, which cannot monitor and accurately analyze the health status of the workers in real time, and it is difficult to respond in time when an emergency occurs. For the above situation, the vital signs monitoring module in the intelligent monitoring insulating helmet with communication function can efficiently monitor people's blood pressure, body temperature, and heart rate. Summary of the invention

[0003] The present invention provides an elastic network personnel health monitoring model based on differential evolution optimization. The differential evolution algorithm is used to optimize the hyperparameters of the elastic network regression model. The optimal model is obtained through continuous iteration, so that the health status of personnel can be judged more accurately. New data is continuously collected to update and optimize the model to adapt to different situations and changes.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A health monitoring model for elastic network personnel based on differential evolution optimization includes the following steps:

[0006] S1. Data collection and preprocessing: Data collection includes collecting body temperature, heart rate and blood pressure data of each individual helmet wearer;

[0007] S2, define the cost function of the elastic network regression model by combining the regularization method of Lasso regression and ridge regression;

[0008] S3. Use differential evolution to optimize the hyperparameters in the elastic network regression model and obtain the optimal model through continuous iteration;

[0009] S4. The input feature vector is passed into the model, and the model outputs the health status label to predict the health of the elastic network personnel.

[0010] Furthermore, the preprocessing includes cleaning the data, processing missing values ​​and outliers, and standardizing body temperature, heart rate, and blood pressure.

[0011] Furthermore, the cost function is:

[0012]

[0013] Among them, y i is the true value of the i-th sample, i.e., the health status label, w is the weight coefficient, N is the number of samples, x i is the feature vector of the i-th sample, including vital signs data such as body temperature, heart rate, and blood pressure, T is the transpose operation of the matrix, λ is the regularization strength, and ρ is the mixing ratio of the L1 regularization term and the L2 regularization term in the elastic network regression;

[0014] The weight coefficient w when the cost function is minimized is:

[0015]

[0016] When ρ = 0, its cost function is equivalent to the cost function of ridge regression, and when ρ = 1, its cost function is equivalent to the cost function of lasso regression.

[0017] Furthermore, the weight coefficient w is solved by coordinate descent method, including iterating the weight coefficient, wherein the weight coefficient w is initialized first, and then all weight coefficients are traversed, and one of the weight coefficients is taken as a variable in turn, and the other weight coefficients are fixed to the result of the previous calculation as a constant, and the optimal solution is found when there is only one weight coefficient variable under the current conditions. When the changes of all weight coefficients remain unchanged or the maximum number of iterations is reached, the iteration is terminated.

[0018] Furthermore, the hyperparameters in the elastic network regression model are optimized by differential evolution, and the differential evolution optimization includes the following steps:

[0019] S3.1. Initialize the model hyperparameters corresponding to the health monitoring data to provide a search starting point for subsequent optimization of the health status prediction model;

[0020] S3.2, after the initial population is generated, mutation operation is performed;

[0021] S3.3, perform overall crossover of the mutation vector and the target vector;

[0022] S3.4, after the above mutation and crossover operations, the differential evolution algorithm compares the test vector with the target vector in the current population according to the greedy criterion. In the next generation, if the target vector is better, the target vector is selected, and if the test vector is better, the test vector is selected. The test vector is only compared with the target vector for individuals, not all individuals in the existing population;

[0023] S3.5. During the mutation process, if a solution outside the feasible domain is compiled, the dimension value that exceeds the boundary of the parameter definition domain will be reset to a uniformly distributed random number between the upper and lower limits of the corresponding parameter to ensure that the solution is always within the feasible domain.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1) The differential evolution algorithm is used to optimize the hyperparameters of the elastic network regression model, and the optimal model is obtained through continuous iteration, so that it can more accurately judge the health status of people wearing helmets;

[0026] 2) It can continuously collect new data to update and optimize the model to adapt to different situations and changes. DETAILED DESCRIPTION

[0027] The specific embodiments of the present invention are further described below:

[0028] The present invention provides an elastic network personnel health monitoring model based on differential evolution optimization, comprising the following steps:

[0029] S1. Data preparation;

[0030] S1.1. Data collection: Data collection includes collecting the body temperature, heart rate and blood pressure data of each individual helmet wearer as the input features of the model, and collecting the corresponding health status labels, such as 0 or 1 representing different health states, as the output of the model;

[0031] S1.2, Data preprocessing:

[0032] (1) Cleaning data: dealing with missing values ​​and outliers;

[0033] (2) Standardization or normalization: The body temperature, heart rate, and blood pressure of the helmet wearer are standardized to ensure that all features are trained on the same scale.

[0034] S2, define the cost function of the elastic network regression model by combining the regularization method of Lasso regression and ridge regression;

[0035] Ridge regression is a biased estimation method for dealing with multicollinearity problems in linear regression. Lasso, the least absolute shrinkage and selection operator, is a regularization method for linear regression. Lasso can automatically perform feature selection and shrink unimportant feature coefficients to zero, which is very useful for processing high-dimensional data. It can reduce the complexity of the model and improve the interpretability of the model. When multiple features are related, Lasso regression may only randomly select one of them, while ridge regression will select all features. Combining these two regularization methods, this regularized algorithm is called elastic network regression. The model is introduced as follows:

[0036] The cost function of the elastic network regression algorithm combines the regularization methods of Lasso regression and ridge regression, and controls the size of the penalty term through two parameters λ and ρ.

[0037]

[0038] Among them, y i is the true value of the i-th sample, i.e., the health status label, w is the weight coefficient, N is the number of samples, x i is the feature vector of the i-th sample, including vital signs data such as body temperature, heart rate, and blood pressure, T is the transpose operation of the matrix, λ is the regularization strength, and ρ is the mixing ratio of the L1 regularization term and the L2 regularization term in the elastic network regression;

[0039] The same is to find the size of w when the cost function is minimized:

[0040]

[0041] When ρ = 0, its cost function is equivalent to the cost function of ridge regression. When ρ = 1, its cost function is equivalent to the cost function of Lasso regression. Like Lasso regression, there is an absolute value in the cost function, which is not differentiable everywhere. Therefore, it is impossible to directly obtain the analytical solution of w by direct derivation. However, the coordinate descent method can still be used to solve w.

[0042] The iterative steps belong to the coordinate descent solution process of the elastic network regression algorithm. Specifically: the algorithm steps describe the coordinate descent implementation steps used in the training process of the elastic network regression model:

[0043] (1) Initializing the weight coefficient w, for example, to a zero vector;

[0044] (2) Traverse all weight coefficients, take one of the weight coefficients as a variable in turn, and fix the other weight coefficients to the result of the previous calculation as constants, and find the optimal solution under the current conditions when there is only one weight coefficient variable. At the kth iteration, the method for updating the weight coefficient is as follows:

[0045] represents the kth iteration, the mth weight coefficient,

[0046]

[0047] (3) The above step (2) is a complete iteration. When the changes of all weight coefficients are small or the maximum number of iterations is reached, the iteration ends;

[0048] Objective function: Use the average error of cross validation (such as mean squared error MSE or classification accuracy) as the objective function to evaluate the performance of the model for a given combination of λ and a, where a is the elastic network mixing ratio parameter.

[0049] S3. Use differential evolution to optimize the hyperparameters in the elastic network regression model and obtain the optimal model through continuous iteration;

[0050] The differential evolution algorithm DE is used to solve the Chebyshev polynomial problem and is also an effective technology for solving complex optimization problems. The differential evolution algorithm is very similar to the genetic algorithm and is also an optimization algorithm based on swarm intelligence theory. It is a global search strategy generated by cooperation and competition among individuals in the group. It uses real number coding, simple mutation operations based on differences, and a "one-to-one" competitive survival strategy to reduce the complexity of evolutionary computing operations. At the same time, the differential evolution algorithm has a memory ability that allows it to dynamically track the current search situation to adjust its search strategy. It has strong global convergence ability and robustness, and does not require the use of feature information of the problem. It is suitable for solving some complex optimization problems that are difficult or even impossible to solve using conventional mathematical programming methods. The process of the differential optimization algorithm includes the following:

[0051] S3.1. Initialization

[0052] The model hyperparameters corresponding to the health monitoring data (such as body temperature, heart rate, blood pressure and other features), such as regularization strength λ and mixing ratio ρ, are initialized to provide a search starting point for the subsequent optimization of the health status prediction model. Each individual is represented as follows:

[0053] x i,G (i=1,2,...,NP) (8)

[0054] Among them, i represents the number of individuals in the population, G represents the evolutionary generation, and NP represents the population size;

[0055] In the differential evolution algorithm, it is generally assumed that all randomly initialized populations conform to a uniform distribution, and the bounds of the parameter variables are set to but

[0056] in, is the lower bound of the jth parameter variable, x j is the value range of the jth parameter variable, is the upper bound of the jth parameter variable, x ji,0is the initial value of the i-th individual in the j-th parameter dimension. rand[0,1] means generating uniform real numbers between [0,1]. Uniform generation of random numbers is only a possibility. If the probability distribution of the solution can be known in advance, uniform generation is not necessary. The distribution law can be used to generate solutions that cover more information, thereby improving the reconstruction effect.

[0057] S3.2 Variation

[0058] After generating the initial population, perform mutation operation. For each target x i,G (i=1,2,...,NP), the mutation vector of the basic differential evolution algorithm is generated as follows:

[0059]

[0060] Among them, v i,G+1 is the mutation vector of the i-th individual in the G+1-th generation;

[0061] is the first benchmark individual randomly selected from the G-th generation population; F is the scaling factor;

[0062] is the second individual randomly selected from the G-generation population;

[0063] is the third individual randomly selected from the G-generation population;

[0064] Requires the individual number r to be randomly selected 1 、r 2 、r 3 They are different from each other and cannot be the same as the target vector sequence number i, so NP ≥ 4 must be satisfied. The mutation operator F ∈ [0, 2] is a real constant factor that controls the scaling of the deviation variable.

[0065] S3.3 Crossover

[0066] Perform a general crossover of the mutation vector and the target vector. In order to increase the diversity of the interference parameter vector, a crossover operation is introduced, and the test vector becomes:

[0067]

[0068] Among them, u j,i,G+1 is the experimental value of the jth parameter of the ith individual in the G+1th generation, v j,i,G+1 represents the variation value of the jth parameter of the ith individual in the G+1th generation, rand b (j) represents a uniformly distributed random number in the range [0,1] generated for the jth parameter, CR represents the crossover probability, which is used to control the parameter replacement ratio, and r nbr(i) represents a parameter dimension index randomly selected by the algorithm for individual i, which is used to ensure that the crossover operation replaces at least one parameter;

[0069] S3.4 Selection

[0070] After the above mutation and crossover operations, the differential evolution algorithm compares the test vector with the target vector x in the current population according to the greedy criterion. i,G For comparison, in the next generation, if the target vector is better, the target vector is selected, and if the trial vector is better, the trial vector is selected. The trial vector is only compared with the target vector for individuals, not all individuals in the existing population.

[0071] S3.5. Boundary Condition Processing

[0072] During the mutation process, if a solution outside the feasible domain is compiled, the dimension value that exceeds the boundary of the parameter definition domain is reset to a uniformly distributed random number between the upper and lower limits of the corresponding parameter to ensure that the solution is always within the feasible domain, that is: or So

[0073] The differential algorithm has a simple structure, is easy to use, and has good reliability, efficiency, and robustness. For large-space, nonlinear, and non-differentiable continuous problems, its solution rate is better than other evolutionary methods;

[0074] The method of optimizing hyperparameters in the elastic network regression model by differential evolution comprises the following specific steps:

[0075] (1) Initialize the population: Create a population consisting of random hyperparameters. Each individual is a tuple (λ, a), where a is λ and can be initialized in a logarithmic scale, such as [10 -3 ,10 3 ];

[0076] (2) Evaluate the initial population: Use the defined objective function to evaluate the performance of each individual in the population;

[0077] (3) Iteratively update the population: In each generation, perform the following steps until the stopping condition is met, such as reaching the maximum number of iterations or error convergence.

[0078] (4) Mutation: Select three different individuals for each individual Among them, r 1 、r 2 、r 3 All of them are different and not equal to the index of the current individual, generating mutant individuals: Where F is the scaling factor, between 0.5 and 1;

[0079] (5) Crossover: For variant v ij Perform crossover operation with the target individual to obtain the test individual u ij ,The method adopted is binomial crossover method;

[0080] First, determine the crossover probability CR, with a typical value of 0.7. This probability determines the probability that the gene of the variant will be selected in the crossover operation. Then, for each dimension j of the target individual, independently generate a uniformly distributed random number rand j ∈[0,1], assuming that the target individual and the variant are both n-dimensional vectors, perform the following operations: If the random number rand[0,1] is less than or equal to the crossover probability CR, or the index j of the current dimension is j rand , then the value of the experimental individual in this dimension is obtained from the variant, that is, u ij =v ij Otherwise, the value of the test individual in this dimension is obtained from the target individual, that is, u ij =x ij ;

[0081] (6) Selection: Evaluation of test individual u i The performance of the individuals is compared with that of the current individuals, and the individuals with better performance are selected to enter the next generation;

[0082] (7) Termination condition: reaching the maximum number of generations or the objective function change is less than the preset threshold.

[0083] S4. Pass the input feature vector into the model, and the model will output a prediction value. If the model outputs the probability distribution of health status, the category with the highest probability can be selected as the predicted health status. If the model outputs the probability of "healthy" as 0.7 and the probability of "abnormal" as 0.3, the health status of the person is judged to be "healthy".

[0084] The above embodiments are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given, but the protection scope of the present invention is not limited to the above embodiments. The methods used in the above embodiments are conventional methods unless otherwise specified.

Claims

1. A health monitoring model for elastic network personnel based on differential evolution optimization, characterized in that: The steps include: S1. Data collection and preprocessing: Data collection includes collecting body temperature, heart rate and blood pressure data of each individual helmet wearer; S2, define the cost function of the elastic network regression model by combining the regularization method of Lasso regression and ridge regression; S3. Use differential evolution to optimize the hyperparameters in the elastic network regression model and obtain the optimal model through continuous iteration; S4. The input feature vector is passed into the model, and the model outputs the health status label to predict the health of the elastic network personnel.

2. According to claim 1, a differential evolution optimization-based elastic network personnel health monitoring model is characterized in that: The preprocessing includes cleaning the data, processing missing values ​​and outliers, and standardizing body temperature, heart rate, and blood pressure.

3. According to the elastic network personnel health monitoring model based on differential evolution optimization according to claim 1, it is characterized in that: The cost function is: Among them, y i is the true value of the i-th sample, i.e., the health status label, w is the weight coefficient, M is the number of samples, and x i is the feature vector of the i-th sample, including vital signs data such as body temperature, heart rate, and blood pressure, T is the transpose operation of the matrix, λ is the regularization strength, and ρ is the mixing ratio of the L1 regularization term and the L2 regularization term in the elastic network regression; The weight coefficient w when the cost function is minimized is: When ρ = 0, its cost function is equivalent to the cost function of ridge regression, and when ρ = 1, its cost function is equivalent to the cost function of lasso regression.

4. The elastic network personnel health monitoring model based on differential evolution optimization according to claim 3 is characterized in that: The weight coefficient w is solved by the coordinate descent method, including iterating the weight coefficient, wherein the weight coefficient w is first initialized, and then all weight coefficients are traversed, one of the weight coefficients is taken as a variable in turn, and the other weight coefficients are fixed to the results of the previous calculation as constants, and the optimal solution is found when there is only one weight coefficient variable under the current conditions. When the changes of all weight coefficients remain unchanged or the maximum number of iterations is reached, the iteration is terminated.

5. The elastic network personnel health monitoring model based on differential evolution optimization according to claim 1 is characterized in that: The method of using differential evolution to optimize hyperparameters in the elastic network regression model includes the following steps: S3.

1. Initialize the model hyperparameters corresponding to the health monitoring data to provide a search starting point for subsequent optimization of the health status prediction model; S3.2, after the initial population is generated, mutation operation is performed; S3.3, perform overall crossover of the mutation vector and the target vector; S3.4, after the above mutation and crossover operations, the differential evolution algorithm compares the test vector with the target vector in the current population according to the greedy criterion. In the next generation, if the target vector is better, the target vector is selected, and if the test vector is better, the test vector is selected. The test vector is only compared with the target vector for individuals, not all individuals in the existing population; S3.

5. During the mutation process, if a solution outside the feasible domain is compiled, the dimension value that exceeds the boundary of the parameter definition domain will be reset to a uniformly distributed random number between the upper and lower limits of the corresponding parameter to ensure that the solution is always within the feasible domain.

Images