Disclosure of Invention
Aiming at the problems that the traditional data inspection method is inaccurate in detecting the data related to the back-turning requirement of the elderly and difficult to effectively identify abnormal data, the method eliminates the abnormal data by constructing a feature vector, defining an enhanced weight, constructing an enhanced mixed feature model, calculating probability and updating parameters, describes the distribution characteristics of the data more accurately, improves the fitting capacity and generalization capacity of the model to the data, reduces the influence of noise data on the model, improves the quality and reliability of the data, and aims at solving the problems that the traditional back-turning requirement prediction model is low in prediction precision, difficult to effectively capture complex relations and dynamic changes among the data and incorrect in setting of model parameters.
The invention adopts the technical scheme that the old people turning over back beating auxiliary system based on artificial intelligence comprises a data acquisition module, a data inspection module, a turning over back beating demand prediction model building module and a turning over back beating auxiliary module;
The data acquisition module acquires historical old person physiological data, environment data and turnover back-beating demand levels, wherein the physiological data comprise heart rate, respiratory rate, body temperature, blood pressure and bedridden time, the environment data refer to temperature, humidity and noise intensity of the turnover back-beating environment, and the turnover back-beating demand levels comprise two levels which are not required and have requirements;
The data checking module clears abnormal data by constructing feature vectors, defining enhanced weights, constructing an enhanced hybrid feature model, calculating probability and updating parameters;
The turnover back beating demand prediction model building module builds a turnover back beating demand prediction model by setting tag data, defining difference entropy, defining oscillation factors, designing interleaving kernels, determining targets, calculating optimal hyperplane weights, calculating optimal hyperplane biases, preparing parameter optimization, generating jump factors, generating initial parameter points, defining kinetic energy control factors, designing parameter search functions, introducing a global reset mechanism and customizing a start-stop strategy;
the turnover back-beating auxiliary module predicts the turnover back-beating demand level of the old through the turnover back-beating demand prediction model and assists the old to turn over the back-beating.
Further, the data checking module specifically includes the following:
Constructing a characteristic vector, and forming characteristic data such as heart rate, respiratory rate, body temperature, blood pressure, bedridden time, temperature, humidity, noise intensity and turnover back beating demand level into the characteristic vector;
enhancement weights are defined, expressed as follows:
;
Wherein,Representing the input eigenvector, u andThe index of the gaussian component is shown,Representing input feature vectorsE represents a natural constant, U represents the total number of gaussian components,Representing the mean value of the u-th gaussian component,Represent the firstThe average value of the individual gaussian components,Representing the modulus length;
an enhanced hybrid feature model is constructed, expressed as follows:
;
Wherein,A parameter set representing an enhanced hybrid feature model includes hybrid weightsAverage value vectorSum covariance matrix,Feature vectors representing inputsAt the parameters ofThe probability density of the enhanced hybrid feature model under the condition of (1),Feature vectors representing inputsFor a mean value ofVariance isProbability density of gaussian distribution of (c);
Probability calculations are expressed as follows:
;
Where g represents the index of the feature vector,Representing the probability that the g-th feature vector comes from the u-th gaussian component,Representing a g-th feature vector;
Parameter updating, namely updating a mixing weight, a mean vector and a covariance matrix, wherein the mixing weight, the mean vector and the covariance matrix are represented as follows:
;
Where G represents the total number of feature vectors,Representing a transpose operation;
and (3) removing the abnormal data, repeating probability calculation and parameter updating until the parameters are converged, setting a probability density threshold, calculating the probability density of the enhanced hybrid feature model of each feature vector, setting the feature vector with the probability density lower than the probability density threshold as the abnormal vector, and removing.
Further, the module for constructing the turnover back beating demand prediction model specifically comprises the following contents:
Setting tag data, and setting a turnover back-beating demand level as tag data of a turnover back-beating demand prediction model;
The differential entropy is defined as follows:
;
Where x1 and x2 represent input feature vectors,Representing the difference entropy between x1 and x2, i representing the dimension index of the feature vector,AndRepresenting the probability that the difference between the value of the feature vector x1 and x2 in the ith dimension and the mean is mapped to interval 0,1,Representing a logarithmic function;
The oscillation factor is defined as follows:
;
Wherein,Representing the oscillation factor between the eigenvectors x1 and x2,AndRepresenting the second derivatives of the feature vectors x2 and x2 in the ith dimension respectively,AndRepresenting the first derivatives of the feature vectors x2 and x2 in the ith dimension respectively,The representation takes the absolute value of the value,Representing a nulling factor;
the interleaving core is designed as follows:
;
Wherein,Representing the interleaving kernel between feature vectors x1 and x2, D representing the largest dimension of the feature vector,Representing the eigenvalue of the eigenvector x1 in the ith dimension,Representing the eigenvalue of the eigenvector x2 in the ith dimension,A width parameter representing the i-th dimension,AndRespectively representing the cross weight and the oscillation weight;
the target is determined as follows:
;
where w is the hyperplane weight vector,A discrete weight vector representing the i-th dimension,Indicating that the maximum value is taken,Representing the square of the L2 norm, j and k representing the index of the feature vector,Representing the adjustment of the weight(s),AndThe lagrangian multipliers for the jth and kth eigenvectors respectively,AndLabels respectively representing the jth and kth feature vectors;
The optimal hyperplane weights are calculated as follows:
;
Wherein,The weights representing the optimal hyperplane are represented,Represents the optimal solution of the lagrangian multiplier for the j-th feature vector,Representing a feature map of the interleaving kernel for the jth feature vector;
The optimal hyperplane bias is calculated as follows:
;
Wherein,Representing the offset of the optimal hyperplane,Representing the total number of support vectors,Representing the feature vector index belonging to the support vector,An optimal solution of the lagrangian multiplier representing the kth eigenvector;
Parameter optimization preparation, namely setting the accuracy of a turn-over back-beat demand prediction model as a performance value of a parameter individual, and determining optimization parameters including width parameters, cross weights, oscillation weights and adjustment weights;
the jump factor is generated as follows:
;
where q represents the number of times the initial parameter point is generated,Representing the jump factor at the q +1 th generation of the initial parameter point,Representing the skip factor at the q-th generation of the initial parameter point, wherein the initial skip factorIs a random number with a value range of 0 to 1;
Generating initial parameter points, expressed as follows:
;
Wherein,Represents the position of the initial parameter point generated at the q+1st time,Representing the upper bound of the parameter space,Representing a lower bound of the parameter space;
Kinetic energy control factors are defined and expressed as follows:
;
where t represents the number of searches for the current parameter,Represents the kinetic energy control factor at the time of the t-th parameter search,Indicating the maximum number of parameter searches,Representing a random number between 0 and 1,Representing a random number between 0 and 2,The sign function is represented by a sign function,Representing the average position of the generated initial parameter search points;
the design parameter search function is expressed as follows:
;
Wherein,Representing the parameter position obtained by the t+1st parameter search,Representing the parameter position obtained by the t-th parameter search,Represents the position with highest global parameter performance in the t-th parameter search, r3 represents a random number with a value range of 0 to 1,Representing the position with the lowest global parameter performance in the t-th parameter search;
A global reset mechanism is introduced, expressed as follows:
;
Wherein,Representing the position of the parameter during the parameter search,Indicating the parameter position after the reset,Representing the position of the parameter closest to the acquired parameter;
Setting a parameter performance threshold and the maximum parameter searching times, carrying out parameter searching on initial parameter searching points by using a parameter searching function, stopping searching if the parameter performance of the parameter position is greater than the parameter performance threshold in the searching process, setting the parameter of the position with the highest global parameter performance at the moment as a model parameter, carrying out searching again if the parameter searching times reach the maximum parameter searching times, and otherwise, continuing searching.
Further, the turning-over back-beating auxiliary module is used for inputting data into a turning-over back-beating demand prediction model through collecting physiological data and environmental data of the old people in real time, predicting a turning-over back-beating demand level of the old people by the model, and assisting the old people in turning over back-beating in real time.
By adopting the scheme, the beneficial effects obtained by the invention are as follows:
(1) Aiming at the problems that the traditional data inspection method is inaccurate in data inspection related to the turn-over back clapping requirement of the old and difficult to effectively identify abnormal data, the method eliminates the abnormal data by constructing a feature vector, defining an enhanced weight, constructing an enhanced mixed feature model, calculating probability and updating parameters, describes the distribution characteristics of the data more accurately, improves the fitting capacity and generalization capacity of the model to the data, reduces the influence of noise data on the model, and improves the quality and reliability of the data.
(2) Aiming at the problems that the traditional turn-over back-beating demand prediction model is low in prediction precision, difficult to effectively capture complex relation and dynamic change among data and improper in model parameter setting, the method and the device have the advantages that by introducing difference entropy and oscillation factors, designing an interleaving kernel, optimizing an objective function, carrying out parameter searching and optimizing, the complex relation among data is more comprehensively described, so that the model can capture the distribution characteristics and rules of the data more accurately, the parameter searching efficiency and accuracy are improved, the model parameters are set more properly, and the model precision is improved.
Detailed Description
The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments, and all other embodiments obtained by those skilled in the art without making any inventive effort based on the embodiments of the present invention are within the scope of protection of the present invention.
In the description of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," "outer," and the like indicate orientation or positional relationships based on those shown in the drawings, merely to facilitate description of the invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus should not be construed as limiting the invention.
Referring to fig. 1, the artificial intelligence-based old people turning over back beating auxiliary system provided by the invention comprises a data acquisition module, a data inspection module, a turning over back beating demand prediction model building module and a turning over back beating auxiliary module;
the data acquisition module acquires historical old person physiological data, environmental data and turnover back-beat requirement levels, and sends the data to the data inspection module;
the data checking module clears abnormal data by constructing feature vectors, defining enhanced weights, constructing an enhanced mixed feature model, calculating probability and updating parameters, and sends the data to the model module for constructing a turnover back-beat demand prediction model;
The turnover back beating demand prediction model building module builds a turnover back beating demand prediction model by setting tag data, defining difference entropy, defining oscillation factors, designing an interleaving kernel, determining a target, calculating optimal hyperplane weights, calculating optimal hyperplane biases, preparing parameter optimization, generating jump factors, generating initial parameter points, defining kinetic energy control factors, designing parameter search functions, introducing a global reset mechanism and customizing a start-stop strategy, and sending the data to the turnover back beating auxiliary module;
the turnover back-beating auxiliary module predicts the turnover back-beating demand level of the old through the turnover back-beating demand prediction model and assists the old to turn over the back-beating.
Referring to fig. 1 and 2, the second embodiment is based on the above embodiment, and the data checking module specifically includes the following:
Constructing a characteristic vector, and forming characteristic data such as heart rate, respiratory rate, body temperature, blood pressure, bedridden time, temperature, humidity, noise intensity and turnover back beating demand level into the characteristic vector;
enhancement weights are defined, expressed as follows:
;
Wherein,Representing the input eigenvector, u andThe index of the gaussian component is shown,Representing input feature vectorsE represents a natural constant, U represents the total number of gaussian components,Representing the mean value of the u-th gaussian component,Represent the firstThe average value of the individual gaussian components,Representing the modulus length;
an enhanced hybrid feature model is constructed, expressed as follows:
;
Wherein,A parameter set representing an enhanced hybrid feature model includes hybrid weightsAverage value vectorSum covariance matrix,Feature vectors representing inputsAt the parameters ofThe probability density of the enhanced hybrid feature model under the condition of (1),Feature vectors representing inputsFor a mean value ofVariance isProbability density of gaussian distribution of (c);
Probability calculations are expressed as follows:
;
Where g represents the index of the feature vector,Representing the probability that the g-th feature vector comes from the u-th gaussian component,Representing a g-th feature vector;
Parameter updating, namely updating a mixing weight, a mean vector and a covariance matrix, wherein the mixing weight, the mean vector and the covariance matrix are represented as follows:
;
Where G represents the total number of feature vectors,Representing a transpose operation;
and (3) removing the abnormal data, repeating probability calculation and parameter updating until the parameters are converged, setting a probability density threshold, calculating the probability density of the enhanced hybrid feature model of each feature vector, setting the feature vector with the probability density lower than the probability density threshold as the abnormal vector, and removing.
By executing the operations, the problem that the traditional data inspection method is inaccurate in data inspection related to the needs of the old for turning over and clapping back and difficult to effectively identify abnormal data is solved, the abnormal data is eliminated by constructing the feature vector, defining the enhanced weight, constructing the enhanced mixed feature model, calculating probability and updating parameters, the distribution characteristics of the data are more accurately described, the fitting capacity and generalization capacity of the model to the data are improved, the influence of noise data on the model is reduced, and the quality and reliability of the data are improved.
Referring to fig. 1, 3 and 4, the third embodiment is based on the above embodiment, and the building of the turnover back-beat demand prediction model module specifically includes the following:
Setting tag data, and setting a turnover back-beating demand level as tag data of a turnover back-beating demand prediction model;
The differential entropy is defined as follows:
;
Where x1 and x2 represent input feature vectors,Representing the difference entropy between x1 and x2, i representing the dimension index of the feature vector,AndRepresenting the probability that the difference between the value of the feature vector x1 and x2 in the ith dimension and the mean is mapped to interval 0,1,Representing a logarithmic function;
The oscillation factor is defined as follows:
;
Wherein,Representing the oscillation factor between the eigenvectors x1 and x2,AndRepresenting the second derivatives of the feature vectors x2 and x2 in the ith dimension respectively,AndRepresenting the first derivatives of the feature vectors x2 and x2 in the ith dimension respectively,The representation takes the absolute value of the value,Representing a nulling factor;
the interleaving core is designed as follows:
;
Wherein,Representing the interleaving kernel between feature vectors x1 and x2, D representing the largest dimension of the feature vector,Representing the eigenvalue of the eigenvector x1 in the ith dimension,Representing the eigenvalue of the eigenvector x2 in the ith dimension,A width parameter representing the i-th dimension,AndRespectively representing the cross weight and the oscillation weight;
the target is determined as follows:
;
where w is the hyperplane weight vector,A discrete weight vector representing the i-th dimension,Indicating that the maximum value is taken,Representing the square of the L2 norm, j and k representing the index of the feature vector,Representing the adjustment of the weight(s),AndThe lagrangian multipliers for the jth and kth eigenvectors respectively,AndLabels respectively representing the jth and kth feature vectors;
The optimal hyperplane weights are calculated as follows:
;
Wherein,The weights representing the optimal hyperplane are represented,Represents the optimal solution of the lagrangian multiplier for the j-th feature vector,Representing a feature map of the interleaving kernel for the jth feature vector;
The optimal hyperplane bias is calculated as follows:
;
Wherein,Representing the offset of the optimal hyperplane,Representing the total number of support vectors,Representing the feature vector index belonging to the support vector,An optimal solution of the lagrangian multiplier representing the kth eigenvector;
Parameter optimization preparation, namely setting the accuracy of a turn-over back-beat demand prediction model as a performance value of a parameter individual, and determining optimization parameters including width parameters, cross weights, oscillation weights and adjustment weights;
the jump factor is generated as follows:
;
where q represents the number of times the initial parameter point is generated,Representing the jump factor at the q +1 th generation of the initial parameter point,Representing the skip factor at the q-th generation of the initial parameter point, wherein the initial skip factorIs a random number with a value range of 0 to 1;
Generating initial parameter points, expressed as follows:
;
Wherein,Represents the position of the initial parameter point generated at the q+1st time,Representing the upper bound of the parameter space,Representing a lower bound of the parameter space;
Kinetic energy control factors are defined and expressed as follows:
;
where t represents the number of searches for the current parameter,Represents the kinetic energy control factor at the time of the t-th parameter search,Indicating the maximum number of parameter searches,Representing a random number between 0 and 1,Representing a random number between 0 and 2,The sign function is represented by a sign function,Representing the average position of the generated initial parameter search points;
the design parameter search function is expressed as follows:
;
Wherein,Representing the parameter position obtained by the t+1st parameter search,Representing the parameter position obtained by the t-th parameter search,Represents the position with highest global parameter performance in the t-th parameter search, r3 represents a random number with a value range of 0 to 1,Representing the position with the lowest global parameter performance in the t-th parameter search;
A global reset mechanism is introduced, expressed as follows:
;
Wherein,Representing the position of the parameter during the parameter search,Indicating the parameter position after the reset,Representing the position of the parameter closest to the acquired parameter;
Setting a parameter performance threshold and the maximum parameter searching times, carrying out parameter searching on initial parameter searching points by using a parameter searching function, stopping searching if the parameter performance of the parameter position is greater than the parameter performance threshold in the searching process, setting the parameter of the position with the highest global parameter performance at the moment as a model parameter, carrying out searching again if the parameter searching times reach the maximum parameter searching times, and otherwise, continuing searching.
By executing the operation, the scheme aims at the problems that the traditional turn-over back-beating demand prediction model has low prediction precision, is difficult to effectively capture complex relations and dynamic changes among data and has improper model parameter setting, and by introducing differential entropy and oscillation factors, designing interleaving kernels, optimizing objective functions, carrying out parameter searching and optimizing, the complex relation between the data is more comprehensively described, so that the model can more accurately capture the distribution characteristics and rules of the data, the efficiency and accuracy of parameter searching are improved, the model parameters are more properly set, and the accuracy of the model is improved.
Referring to fig. 1, the embodiment is based on the above embodiment, and the turning-over back-beating auxiliary module inputs the data into the turning-over back-beating demand prediction model by collecting the physiological data and the environmental data of the elderly, and the model predicts the turning-over back-beating demand level of the elderly, and assists the elderly in turning over back-beating in real time.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made hereto without departing from the spirit and principles of the present invention.
The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.