Intelligent weight management method and system

Abstract

The invention provides an intelligent weight management method and system, and the method comprises the steps: presetting a periodic time sequence, configuring a corresponding weight management decision which comprises different types of decision data sub-elements, and binding the sub-elements according to the corresponding time sequence; acquiring endocrine related detection data of the target weight management personnel according to the periodic time sequence and the corresponding decision data sub-elements, and constructing corresponding target weight management personnel detection data sub-elements according to endocrine related detection data types; taking a set of multiple types of sub-elements of the same period time sequence as elements of a previous level, and constructing a sample set; training a deep learning model by using the sample set, and optimizing the deep learning model by using a reinforcement learning method to obtain a strong correlation relation prediction model of the detection data sub-elements and the weight management decision data sub-elements in the same time period; and detecting the data sub-elements to construct a fitness index of weight management safety, and outputting a corresponding weight management decision according to the fitness index.

Description

Technical Field

[0001] The present invention relates to the technical field of weight management, and in particular to an intelligent weight management method and system. Background Art

[0002] At present, the existing weight management technology is still based on the measurement of traditional weight measurement, BIM (body mass data) and body fat rate and other basic data to predict and manage weight. However, the above basic data is only a phased result in the weight management process. The above phased results cannot reflect whether the weight management process is scientific and accurate. For example, the training intensity of weight management, the changes in relevant data of the endocrine system during training, and other data have not been fully utilized scientifically. Due to the fact that different ages, basic weights, metabolic conditions, and illnesses have an important impact on weight management, different dietary conditions, training methods, and training intensities need to be matched accordingly. If the corresponding weight management method does not match the body's own physical condition, it is easy to cause a series of health-damaging problems including but not limited to malnutrition, anemia, hypoglycemia, and osteoporosis. Even some excessive high-intensity exercise can cause sudden muscle dissolution, further causing acute renal failure. Therefore, maintaining the scientific nature of the weight management process is crucial to physical health. In other words, the prior art does not use the detection data related to the endocrine system as the process data of weight management as the guiding data for weight management. Summary of the invention

[0003] One of the inventive purposes of the present invention is to provide an intelligent weight management method and system, wherein the method and system use the relevant detection data of the endocrine system in the process of weight management as the basic data for weight management, and monitor the numerical value and dynamically changing numerical value of the basic data to construct feature data, and use the constructed feature data as elements to construct a sample set, wherein the elements in the sample set also include corresponding exercise methods, exercise intensity, diet information, etc. to construct strategy data for weight management. Therefore, the present invention can better control the process of weight management based on the numerical value and dynamically changing numerical value of the relevant detection data of the endocrine system, and avoid certain harm to the body caused by overly strong weight management control strategies. The present invention uses a reinforcement learning model to perform reward-based reasoning learning based on the numerical value and dynamically changing numerical value of the relevant detection data of the endocrine system, thereby improving the scientificity and safety of the weight management process.

[0004] Another inventive object of the present invention is to provide an intelligent weight management method and system, wherein the method and system construct the numerical value and dynamically changing numerical value of the relevant detection data of the endocrine system with a periodic time series, wherein the time series uses the time period of the strategy of the exercise mode, exercise intensity, and diet information as a matching time series, that is to say, the present invention obtains the numerical value and dynamically changing numerical actual detection data of the relevant detection data of the endocrine system within the periodic time series based on the weight management decision of the exercise mode, exercise intensity, and diet information. Each element in the sample set described in the present invention includes sub-element data matched with a periodic time series, wherein the sub-element data includes the weight management decision of the exercise mode, exercise intensity, and diet information of the corresponding time series and the weight management strategy matching the corresponding time series within the periodic time series. The sample set constructed based on multi-level elements can effectively improve the sample quality and improve the training effect of the reinforcement learning model.

[0005] Another inventive object of the present invention is to provide an intelligent weight management method and system. After using the reinforcement learning to obtain a relationship explanation model between endocrine-related test data and corresponding weight management decisions, the method and system use the predictive reasoning ability of the relationship explanation model to automatically generate and adjust the weight management strategy of the corresponding target person, so that the corresponding target person can reduce the physical harm of the target person caused by mismatched weight management strategies on the basis of meeting the weight management requirements.

[0006] In order to achieve at least one of the above-mentioned invention objects, the present invention further provides an intelligent weight management method, the method comprising:

[0007] Preset a periodic time series, and configure corresponding weight management decisions in the time series, wherein the weight management decisions include different types of decision data sub-elements, and the decision data sub-elements are bound according to the corresponding time series;

[0008] Acquire endocrine-related test data of the target weight management personnel according to the periodic time series and the corresponding decision data sub-element, and construct the corresponding target weight management personnel test data sub-element with the endocrine-related test data type;

[0009] The set of multi-type sub-elements of the same periodic time series is used as the element of the previous level to construct a sample set of multi-element high-dimensional input;

[0010] The deep learning model is trained using the sample set, and the deep learning model is optimized using a reinforcement learning method to obtain a relationship prediction model in which the detection data sub-elements and the weight management decision data sub-elements are strongly correlated in the same time period;

[0011] A fitness index for weight management safety is constructed according to the detection data sub-elements, and a corresponding weight management decision is output according to the fitness index.

[0012] According to one of the preferred embodiments of the present invention, the sample element also includes a basic data sub-element of the target weight management person, wherein the basic data sub-element includes the age, height, weight, gender, and underlying diseases of the target weight management person, and a binding relationship is established between the basic data sub-element and the detection data sub-element of the target weight management person, and the basic data sub-element, the detection data sub-element, and the decision data sub-element are constructed into an upper-level element in units of corresponding time series for constructing the sample set.

[0013] According to another preferred embodiment of the present invention, the basic data sub-elements, detection data sub-elements and decision data sub-elements are input into a deep neural network model for initialization training, and the hyperparameters of the deep neural model are adjusted through a loss function and a back-propagation method to obtain a first deep neural network model for initialization training. Further, reinforcement learning of Markov decision making is performed based on relevant hyperparameters of the first deep neural network model and actual corresponding sub-elements to obtain a relationship prediction model with strong correlation between the detection data sub-elements and the weight management decision data sub-elements. The relationship prediction model is used to output endocrine-related detection data and weight management decisions that conform to the real relationship.

[0014] According to another preferred embodiment of the present invention, the reinforcement learning method includes: obtaining the pre-trained hyperparameter W of the first deep neural network model n , wherein n represents a hyperparameter type, and the hyperparameter type includes a learning rate, network weights of different network layers of the first deep neural network model, a preset number of network layers, a regularization parameter, and an iteration number; using the hyperparameters of different types and the corresponding weight management decision data as an action value set e for reinforcement learning, and using the predicted value of the endocrine-related detection data of the first deep neural network model as a state value v for reinforcement learning, and constructing the action value function E(e, v) of the corresponding action value type, and automatically updating the action value function E(e, v) according to the predicted state value v, the action value function E(e, v) and the reward function r g (e,v), and according to the updated action-value function E g (e, v) automatically updates the action value set e until the state value v satisfies the preset iteration termination condition.

[0015] According to another preferred embodiment of the present invention, based on the network hyperparameter W of the first deep neural network model nThe action function E(e,v) is a linear relationship constructed with the state function v of the endocrine-related detection data prediction value, and the action function is , where in the action function e m Indicates the current corresponding action type, v l Represents the predicted values ​​of different status types of different endocrine-related test data, y v Represents the relationship coefficient, θ represents the regularization term, and the corresponding action value set e is dynamically updated through the linearly related action function E(e,v) and the reward function to obtain a relationship prediction model between the action value set e and the corresponding state value v.

[0016] According to another preferred embodiment of the present invention, the method for constructing the reward function r includes: inputting the weight management decision sub-element in the current action set e into the first deep neural network model, and using the hyperparameters of the first deep neural network model for initializing reinforcement learning in the current action set e to obtain the corresponding prediction values ​​v of different state types of endocrine-related detection data l , and at the same time obtain the actual value v of different state types based on the weight management decision sub-element actually detected in the current action set e m , solve the predicted value v of the same state type l The corresponding true value v of the same state type as the actual detection m The square difference between , and solve the predicted value v of all types of states l Corresponding to the real value v of the same state type m The sum of squared differences between , according to the difference square sum Q, the reward function λ is configured to automatically adjust the relationship coefficient y in the action function E(e,v) v and a regularization term θ.

[0017] According to another preferred embodiment of the present invention, the method for updating the action function E(e,v) based on the Markov decision reinforcement learning comprises: , where α is the learning rate and r is the reward function value, where the reward function value is calculated as: , is the reward optimization coefficient; and δ is the corresponding discount factor, and the updated action function is E g (e,v); when the action function E is completed g After (e,v) is updated, the updated action function E is used g (e, v) obtains the corresponding action set e, and at this time adjusts the first deep neural network model hyperparameters and / or the corresponding weight management decision data according to the updated action set, and sets the upper limit threshold Q of the sum of squared differences Qs , the action function E(e,v) is iterated multiple times until the sum of squared differences Q is less than the upper threshold Q s As a termination condition of reinforcement learning, the relationship prediction model is obtained.

[0018] According to another preferred embodiment of the present invention, the method for calculating the fitness index of weight management safety includes: inputting the corresponding weight management decision into the intensively trained relationship prediction model, and obtaining the predicted value v of endocrine-related test data, setting the safety upper limit value, safety lower limit value and safety change limit value of the endocrine test data according to the medical knowledge base, obtaining the predicted value v of the endocrine-related test data through the relationship prediction model and comparing the preset safety upper limit value, safety lower limit value and safety change limit value of the endocrine test data, calculating the gap value between the predicted value v of the endocrine-related test data and the safety limit value, and weighted summing the gap value to obtain the fitness index P of weight management safety.

[0019] In order to achieve at least one of the above-mentioned objects of the invention, the present invention further provides an intelligent weight management system, which executes the above-mentioned intelligent weight management method.

[0020] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the above-mentioned intelligent weight management method. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Shown is a flow chart of an intelligent weight management method of the present invention. DETAILED DESCRIPTION

[0022] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0023] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number. Figure 1The present invention discloses an intelligent weight management method and system. The method and system construct a relationship prediction model including a strong correlation between weight management strategy and endocrine-related test data through a reinforcement learning method based on a deep learning model, and improve the correlation between the weight management strategy and endocrine test data through the relationship prediction model. And further output the corresponding weight management strategy based on the prediction result of the relationship prediction model, and construct the safety fitness index of the weight management strategy according to the limit value of the endocrine test data that may exceed the tolerance or have obvious damage to the body in the existing medical knowledge base, and use the safety fitness index as the guidance data of the corresponding weight management strategy to avoid the harm to the body caused by the weight management strategy of the mismatched individual.

[0024] Specifically, in order to maintain the safety and scientific nature of weight management strategies, weight management strategies are often affected by obvious personal factors, such as the age, gender, height, basal weight, basal metabolism, and diseases of the target weight management personnel; the above different personal factors will greatly affect the safety of the weight management strategy and the actual effect of weight management. For example, for target weight management personnel with hypoglycemia or anemia or basal metabolism, they need to increase their intake of carbohydrates or meat, and in the weight management strategy based on exercise, they need to choose a weight management strategy with lower exercise intensity, otherwise it may aggravate the disease symptoms and metabolic conditions of the target weight management personnel themselves. Therefore, the present invention constructs the basic data element C=[c1, c2, c3, c4, c5] based on the basic physical data of different individuals. 5.... c n ], wherein the basic data element C contains the corresponding basic data sub-element c n , c n The data type of the basic data sub-element in the representation, the basic data sub-element c n It may include but is not limited to the above-mentioned age, gender, height, basic weight, basic metabolism, and diseases. Since different types of basic data sub-elements may be numerical data or state data, the present invention provides the basic data sub-element c n The state data in the data set is assigned a value, and the numerical data is preprocessed, wherein the data preprocessing method includes but is not limited to normalization, removal of irrelevant data, and interpolation of missing data by linear interpolation, so as to obtain a relatively clean basic data sub-element c n It is worth mentioning that in one of the preferred embodiments of the present invention, since some basic data may change with time, the present invention can further set the detection time and bind the basic data element C to the corresponding detection time point t. At this time, the corresponding basic data element can be expressed as C t =[c1,t , c 2,t , c 3,t , c 4,t , c 5,t.... c n,t ].

[0025] It is worth mentioning that the present invention needs to pre-set a periodic time series T. In the present invention, a time series with a period of 24 hours is preferred, and a time series can be set every 1 hour, that is, T=[t1, t2, t3...t 24 ]. In some other preferred embodiments of the present invention, a periodic time series of one week or longer can be set, but the present invention is not limited thereto. Based on the preset periodic time series T, the endocrine system-related data detection of the target weight management person is performed within the corresponding time series, wherein the data types of the endocrine system-related data detection include but are not limited to: blood sugar, blood lipids, blood creatinine, blood cell values, different types of hormone content values, and urine-related detection data caused by diseases or endocrine dysfunction. The above-mentioned different types of detection data can be obtained by detection through relevant biosensors, and the detection method of the above-mentioned biosensor can be implemented with reference to the prior art, and the present invention will not elaborate on this.

[0026] Since the change value or change rate of some detection data will affect the health of the target weight management personnel, the present invention also calculates the change value or change rate of the corresponding endocrine system related data based on the detection data value of the biosensor, and uses the change value or change rate as the corresponding detection data sub-element. For example, by monitoring the blood sugar change value or change rate in the corresponding periodic time series through a blood sugar monitor, it is possible to determine whether the relevant blood sugar changes may cause hypoglycemia caused by rapid and high-intensity exercise and affect physical health.

[0027] Furthermore, the present invention defines the endocrine system-related detection data element as F t =[f 1,t , f 2,t , f 3,t , f 4,t.... f n,t ], where f n,t The subscript n indicates the different types of test data corresponding to the endocrine system. n is a detection data sub-element related to the endocrine system. n,t The time series binding t of the corresponding detection period T. At this time, the detection data sub-element f related to the endocrine system is n,tSample sub-element of the deep neural network model. For example, the target weight management personnel in all time series of a time period have performed the above-mentioned endocrine system-related test data sub-element f, including but not limited to blood sugar, blood lipids, blood creatinine, blood cell values, different types of hormone content values, and urine-related test data caused by diseases or endocrine dysfunction. n,t If there is no missed detection, n*t detection data sub-elements will be generated. Since the degree of endocrine response of the human body based on the corresponding weight control strategy in different time periods is different, the above time series has a better process effect of weight management. Compared with results such as body fat percentage and target weight, process weight management has a safer, more scientific and more reasonable management effect.

[0028] In another preferred embodiment of the present invention, since the standard values ​​of the above-mentioned endocrine system related test data may have a large order of magnitude difference due to the test unit or the actual test content, the present invention further adopts a data processing method of decimal calibration standardization for different types of test data sub-elements, and the calculation steps of the decimal calibration standardization processing method include: = , where j is the minimum integer value that satisfies the condition, is the value after normalization, f n,t is the original data.

[0029] It should be noted that the weight management strategy in the present invention is not limited to exercise monitoring, but also includes diet monitoring, wherein exercise monitoring includes exercise mode and exercise intensity, for example, the exercise mode may include but not be limited to aerobic exercise, strength training and functional training, and the intensity of different exercise modes may be further pre-classified; wherein the exercise intensity may be comprehensively set in combination with the corresponding exercise mode and exercise frequency, etc., and the present invention does not specifically limit how to set the exercise mode and exercise intensity. In addition, the diet monitoring described in the present invention also needs to be pre-classified, for example, the dietary nutrient types include carbon water, protein, fat, vitamins, minerals and water, etc., and the grading of the supplementary amount per unit body weight of each nutrient in the diet needs to be pre-configured, such as the protein type and intake per meal corresponding to a high-protein diet, and the intake of cereal carbon water per meal corresponding to a low-carbohydrate diet. The grading setting belongs to the technical personnel in this field and can be set according to actual needs, and the present invention will not go into details. Define the weight management strategy element described in the present invention as B t =[b 1,t , b 2,t , b 3,t , b 4, t.... b n,t ], where b n,tIndicates the type and method of implementing weight management decision under the time series t of period T. In the present invention, the numerical data in the weight management strategy is processed by normalization, non-numerical data assignment conversion, difference processing and other data processing methods to obtain a clean and standardized weight management decision data sub-element b n,t .

[0030] The weight management strategy elements B after the standardized processing are obtained respectively t =[b 1,t , b 2,t , b 3,t , b 4, t.... b n,t ]、Basic data element C t =[c 1,t , c 2,t , c 3,t , c 4,t , c 5,t.... c n,t ] and endocrine system related test data element F t =[f 1,t , f 2,t , f 3,t , f 4,t.... f n,t ], the weight management strategy element B is simultaneously converted into t , Basic data element C t Test data element F related to the endocrine system t The input data is input into a deep neural network for training, wherein the present invention can use a recurrent neural network model RNN model to initialize the architecture of the deep neural network model to perform the detection data element F related to the endocrine system in the present invention. t and the weight management strategy element B t Solution to the regression prediction problem between .

[0031] The training method of the deep neural network model includes but is not limited to: 1. Setting initialization hyperparameters and setting corresponding loss functions. Using the back propagation mechanism to update the hyperparameters of the deep neural network model according to the loss function. The hyperparameter types of the deep neural network model include but are not limited to: learning rate, network weights of different network layers, preset number of network layers, regularization parameters, and number of iterations. The first deep neural network model is obtained by initializing and training the deep neural network model.

[0032] Furthermore, the present invention obtains the pre-trained hyperparameters of the first deep neural network model, including but not limited to learning rate, network weights of different network layers, preset number of network layers, regularization parameters, number of iterations, etc., and performs reinforcement learning based on the hyperparameters of the first deep neural network model. The specific method of reinforcement learning includes: obtaining the pre-trained hyperparameters W of the first deep neural network model n , where n represents the type of hyperparameter; using the hyperparameters of different types and the corresponding weight management decision data as the action value set e of reinforcement learning, and using the predicted value of the endocrine-related detection data of the first deep neural network model as the state value v of the reinforcement learning, and constructing the action value function E(e,v) of the corresponding action value type, and automatically updating the action value function E(e,v) according to the predicted state value v, the action value function E(e,v) and the reward function r g (e,v), and according to the updated action value function E g (e, v) automatically updates the action value set e until the state value v satisfies a preset iteration termination condition. The iteration termination condition of the reinforcement learning may include but is not limited to a setting based on the convergence state of accuracy and reward value.

[0033] In one of the preferred embodiments of the present invention, the reinforcement learning needs to set a reward function, and the method for constructing the reward function r includes: inputting the weight management decision sub-element in the current action set e into the first deep neural network model, and using the hyperparameters of the first deep neural network model for initializing reinforcement learning in the current action set e to obtain the corresponding prediction values ​​v of different state types of endocrine-related detection data l , and at the same time obtain the actual value v of different state types based on the weight management decision sub-element actually detected in the current action set e m Since there is a certain deviation between the relevant hyperparameters of the first deep neural network model and the actual detected true value, the deviation problem can be better solved by the reward function of reinforcement learning, and the predicted value v of the same state type is solved. l The corresponding true value v of the same state type as the actual detection m The square difference between , and solve the predicted value v of all types of states l Corresponding to the real value v of the same state type m The sum of squared differences between , according to the difference square sum Q, the reward function λ is configured to automatically adjust the relationship coefficient y in the action function E(e,v) v and a regularization term θ.

[0034] The method for updating the action function E(e,v) based on the Markov decision reinforcement learning in the present invention includes: , where α is the learning rate and r is the reward function value, where the reward function value is calculated as: , is the reward optimization coefficient; and δ is the corresponding discount factor, and the updated action function is E g (e,v); when the action function E is completed g After (e,v) is updated, the updated action function E is used g (e, v) obtains the corresponding action set e, and at this time adjusts the first deep neural network model hyperparameters and / or the corresponding weight management decision data according to the updated action set, and sets the upper limit threshold Q of the sum of squared differences Q s , the action function E(e,v) is iterated multiple times until the sum of squared differences Q is less than the upper threshold Q s As the termination condition of reinforcement learning, the relationship prediction model is obtained. The deviation of the reinforcement learning can be effectively obtained by calculating the sum of squared differences. When the predicted value of the endocrine-related test data predicted by actual regression prediction and the actual test value meet the deviation range condition, it can be considered that the requirements of reinforcement training are met.

[0035] In another preferred embodiment of the present invention, the present invention also needs to calculate the fitness index of weight management safety, and the fitness index calculation method of weight management safety includes: randomly inputting the corresponding weight management decision into the intensively trained relationship prediction model, and obtaining the predicted value v of the endocrine-related test data, setting the upper safety limit, lower safety limit and safe change limit of the endocrine test data according to the medical knowledge base, and obtaining the predicted value v of the endocrine-related test data and the preset upper safety limit, lower safety limit and safe change limit of the endocrine test data through the relationship prediction model for comparison and calculation, calculating the gap between the predicted value v of the endocrine-related test data and the safety limit, and weighted summing the gap to obtain the fitness index P of weight management safety. For example: define a total of m predicted values ​​h of endocrine-related test data m , if there are n endocrine-related test data prediction values ​​h n If the safety upper limit, safety lower limit and safety change limit of the corresponding endocrine test data are exceeded, the fitness index P of the weight management safety is obtained by weighted summation according to the size and type of the excess. ,in is the weight coefficient corresponding to different detection data types.

[0036] The embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the functions defined in the method of the present application are executed. It should be noted that the computer-readable medium described in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may be of various types, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless segments, wire segments, optical cables, RF, etc., or any suitable combination thereof.

[0037] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0038] Those skilled in the art should understand that the embodiments of the present invention described and shown in the drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be deformed or modified in any way.

Claims

1. An intelligent weight management method, characterized in that: The method comprises: Preset a periodic time series, and configure corresponding weight management decisions in the time series, wherein the weight management decisions include different types of decision data sub-elements, and the decision data sub-elements are bound according to the corresponding time series; Acquire endocrine-related test data of the target weight management personnel according to the periodic time series and the corresponding decision data sub-element, and construct the corresponding target weight management personnel test data sub-element with the endocrine-related test data type; The set of multi-type sub-elements of the same periodic time series is used as the element of the previous level to construct a sample set of multi-element high-dimensional input; The deep learning model is trained using the sample set, and the deep learning model is optimized using a reinforcement learning method to obtain a relationship prediction model in which the detection data sub-elements and the weight management decision data sub-elements are strongly correlated in the same time period; A fitness index for weight management safety is constructed according to the detection data sub-elements, and a corresponding weight management decision is output according to the fitness index.

2. The intelligent weight management method according to claim 1, characterized in that: The sample element also includes a basic data sub-element of the target weight management person, wherein the basic data sub-element includes the age, height, weight, gender, and underlying diseases of the target weight management person. The basic data sub-element and the detection data sub-element of the target weight management person establish a binding relationship, and the basic data sub-element, detection data sub-element and decision data sub-element are constructed into the upper-level element in units of corresponding time series for constructing the sample set.

3. The intelligent weight management method according to claim 2, characterized in that: The basic data sub-elements, detection data sub-elements and decision data sub-elements are input into a deep neural network model for initialization training, and the hyperparameters of the deep neural model are adjusted through a loss function and a back-propagation method to obtain a first deep neural network model of initialization training. Further, reinforcement learning of Markov decision making is performed based on relevant hyperparameters of the first deep neural network model and actual corresponding sub-elements to obtain a relationship prediction model with strong correlation between the detection data sub-elements and the weight management decision data sub-elements. The relationship prediction model is used to output endocrine-related detection data and weight management decisions that conform to the real relationship.

4. The intelligent weight management method according to claim 3, characterized in that: The reinforcement learning method includes: obtaining a pre-trained hyperparameter W of the first deep neural network model n , wherein n represents a hyperparameter type, and the hyperparameter type includes a learning rate, network weights of different network layers of the first deep neural network model, a preset number of network layers, a regularization parameter, and an iteration number; using the hyperparameters of different types and the corresponding weight management decision data as an action value set e for reinforcement learning, and using the predicted value of the endocrine-related detection data of the first deep neural network model as a state value v for reinforcement learning, and constructing the action value function E(e, v) of the corresponding action value type, and automatically updating the action value function E(e, v) according to the predicted state value v, the action value function E(e, v) and the reward function r g (e,v), and according to the updated action-value function E g (e, v) automatically updates the action value set e until the state value v satisfies the preset iteration termination condition.

5. The intelligent weight management method according to claim 4, characterized in that: Based on the network hyperparameter W of the first deep neural network model n The action function E(e,v) is a linear relationship constructed with the state function v of the predicted value of the endocrine-related detection data. The action function is , where in the action function e m Indicates the current corresponding action type, v l Represents the predicted values ​​of different status types of different endocrine-related test data, y v Represents the relationship coefficient, θ represents the regularization term, and the corresponding action value set e is dynamically updated through the linearly related action function E(e,v) and the reward function to obtain a relationship prediction model between the action value set e and the corresponding state value v.

6. The intelligent weight management method according to claim 5, characterized in that: The method for constructing the reward function r includes: inputting the weight management decision sub-element in the current action set e into the first deep neural network model, and obtaining the corresponding prediction value v of different state types of endocrine-related detection data using the hyperparameters of the first deep neural network model for initializing reinforcement learning in the current action set e l , and at the same time obtain the actual value v of different state types based on the weight management decision sub-element actually detected in the current action set e m , solve the predicted value v of the same state type l The corresponding true value v of the same state type as the actual detection m The square difference between , and solve the predicted value v of all types of states l Corresponding to the real value v of the same state type m The sum of squared differences between , according to the difference square sum Q, the reward function λ is configured to automatically adjust the relationship coefficient y in the action function E(e,v) v and a regularization term θ.

7. The intelligent weight management method according to claim 6, characterized in that: The method for updating the action function E(e,v) based on the Markov decision reinforcement learning includes: , where α is the learning rate and r is the reward function value, where the reward function value is calculated as: , is the reward optimization coefficient; and δ is the corresponding discount factor, and the updated action function is E g (e,v); when the action function E is completed g After (e,v) is updated, the updated action function E is used g (e, v) obtains the corresponding action set e, and at this time adjusts the first deep neural network model hyperparameters and / or the corresponding weight management decision data according to the updated action set, and sets the upper limit threshold Q of the sum of squared differences Q s , the action function E(e,v) is iterated multiple times until the sum of squared differences Q is less than the upper threshold Q s As a termination condition of reinforcement learning, the relationship prediction model is obtained.

8. The intelligent weight management method according to claim 1, characterized in that: The method for calculating the fitness index of weight management safety includes: inputting the corresponding weight management decision into the intensively trained relationship prediction model, and obtaining the predicted value v of endocrine-related detection data, setting the safety upper limit value, safety lower limit value and safety change limit value of the endocrine detection data according to the medical knowledge base, obtaining the predicted value v of the endocrine-related detection data through the relationship prediction model and comparing the preset safety upper limit value, safety lower limit value and safety change limit value of the endocrine detection data, calculating the gap value between the predicted value v of the endocrine-related detection data and the safety limit value, and weighted summing the gap value to obtain the fitness index P of weight management safety.

9. An intelligent weight management system, characterized in that: The system executes an intelligent weight management method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the intelligent weight management method described in any one of claims 1-8.

Images