Intelligent calculation and dynamic adjustment method and system for breast feeding amount

Abstract

The invention provides a breast milk feeding amount intelligent calculation and dynamic adjustment method and system, and relates to the technical field of intelligent feeding, and the method comprises the steps: collecting real-time physiological parameters of an infant, constructing an infant digital model through a multi-scale morphological neural network, calculating a basic energy consumption value through combining with a biological rhythm prediction model, and calculating the feeding amount of the infant. The standard breast milk feeding amount is calculated through the self-adaptive fuzzy neural network; the method comprises the following steps: collecting mother physiological state data and breast milk component data, calculating a breast milk nutrition density coefficient by using a deep mixing cognitive network and an immune evolution self-organizing network, correcting a breast milk reference feeding amount, and finally calculating an actual breast milk feeding amount through a self-adaptive resonance theory network; growth and development data of an infant after breast milk intake are collected, a multi-layer spiral pulse neural network is used for extracting development feature vectors, a development deviation value is calculated based on a fractal dynamics model, network parameters are dynamically adjusted through a biological group intelligent optimization system, and dynamic optimization of breast milk feeding amount is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent feeding, and particularly to an intelligent calculation and dynamic adjustment method and system for breast milk feeding volume. Background Art

[0002] Scientific and reasonable breast milk feeding volume directly affects the healthy growth of infants and is of great significance to the physical and neurological development of infants. At present, the determination of breast milk feeding volume mainly relies on empirical judgment or fixed calculation formulas, and this method is difficult to adapt to the individual differences and dynamic development needs of infants. With the development of artificial intelligence technology, it has become possible to apply intelligent calculation methods to the accurate calculation and dynamic adjustment of breast milk feeding volume.

[0003] However, there are still deficiencies in the calculation and adjustment of breast milk feeding volume in the existing technology. The traditional calculation methods do not fully consider the individual characteristics of infants, and it is difficult to meet the actual needs of different infants by adopting a unified calculation standard; the existing methods lack real-time monitoring and analysis of breast milk components and the mother's physiological state, and it is impossible to accurately evaluate the nutritional density of breast milk and its suitability for infants; the existing technology lacks a dynamic feedback mechanism for the growth and development status of infants, and it is difficult to adjust the feeding volume in a timely manner according to the actual development of infants, which is likely to cause a mismatch between nutritional supply and demand.

[0004] In summary, it is urgent to solve the technical problem of personalized calculation and dynamic adjustment of breast milk feeding volume. By constructing a multi-level intelligent calculation framework, realizing real-time monitoring and analysis of infant physiological characteristics, establishing an accurate evaluation mechanism for breast milk nutritional density, and achieving adaptive adjustment of feeding volume based on dynamic feedback of infant development status; according to the individual characteristics and development needs of infants, combined with the nutritional characteristics of breast milk, realizing intelligent calculation and dynamic optimization of breast milk feeding volume, and providing more scientific nutritional support for the healthy growth of infants. The present invention can solve the problems in the existing technology. Summary of the Invention

[0005] An embodiment of the present invention provides an intelligent calculation and dynamic adjustment method and system for breast milk feeding volume, which can solve the problems in the existing technology.

[0006] In the first aspect of the embodiment of the present invention,

[0007] Provided is an intelligent calculation and dynamic adjustment method for breast milk feeding volume, including:

[0008] Collect the real-time physiological parameters of the infant to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to construct an infant digital model, perform feature dimensionality reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set; calculate the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, input it into a dynamic fuzzy inference system, combine with a biological rhythm prediction model to calculate the basal energy consumption value, and calculate the benchmark breast milk feeding amount through an adaptive fuzzy neural network;

[0009] Collect the physiological state data and breast milk composition data of the mother, input them into a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain time-series features through chaotic memory mapping; input the feature vectors and time-series features into an immune evolutionary self-organizing network to calculate the breast milk nutrition density coefficient, correct the benchmark breast milk feeding amount according to the breast milk nutrition density coefficient, and calculate the actual breast milk feeding amount through an adaptive resonance theory network;

[0010] Collect the growth and development data of the infant after ingesting the actual breast milk feeding amount, input it into a multi-layer spiral pulse neural network to extract development feature vectors; use a fractal dynamics model to map with standard growth and development parameters to calculate the development deviation value; a biotic swarm intelligence optimization system constructed according to the development deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on a mutation algorithm to realize the dynamic optimization of the breast milk feeding amount.

[0011] In an alternative embodiment,

[0012] Collect the real-time physiological parameters of the infant to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to construct an infant digital model, and perform feature dimensionality reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set including:

[0013] Collect the physiological parameters of the infant, including body temperature parameters, heart rate parameters, respiratory rate parameters, blood oxygen saturation parameters, and weight parameters; classify the physiological parameters according to the change characteristics of the physiological parameters to determine fast-changing parameters and slow-changing parameters; determine the sampling frequency of the fast-changing parameters according to the ratio of the standard deviation to the mean of the fast-changing parameters, and determine the sampling frequency of the slow-changing parameters according to the ratio of the change amount of the slow-changing parameters to the time interval; sample the physiological parameters through the sampling frequency to obtain raw data, and use wavelet threshold filtering to denoise the raw data to obtain a dynamic data stream;

[0014] Input the dynamic data stream into a multi-scale morphological neural network. Determine the scale factor according to the ratio of the characteristic period of the dynamic data stream to the basic sampling interval. Use the scale factor to perform multi-scale decomposition on the dynamic data stream to obtain multi-scale data; calculate the variance of the multi-scale data within a local window, determine the size of the structural element of the morphological operator according to the variance, construct a dilation operator and an erosion operator using the size of the structural element, and process the multi-scale data through the dilation operator and the erosion operator to construct a baby digital model;

[0015] Determine the initial number of nodes of the self-organizing competitive mapping network according to the number of samples and the input dimension of the baby digital model. The self-organizing competitive mapping network adopts an adaptive growth structure; input the baby digital model into the self-organizing competitive mapping network, calculate the Euclidean distance and Manhattan distance between samples, construct an improved distance metric according to the Euclidean distance and the Manhattan distance, and use the improved distance metric to perform local response normalization processing on the baby digital model to obtain normalized features; construct a timing constraint function, calculate the timing constraint value of the normalized features according to the timing constraint function, and fuse the timing constraint value with the normalized features to obtain a physiological feature set.

[0016] In an alternative embodiment,

[0017] Calculate the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, input it into a dynamic fuzzy inference system, and calculate the basal energy consumption value in combination with a biological rhythm prediction model. Calculating the breast milk benchmark feeding amount through an adaptive fuzzy neural network includes:

[0018] Perform pattern segmentation on the feature sequences in the physiological feature set, and determine the similarity threshold according to the standard deviation of the feature sequences; count the number of matching patterns in adjacent dimensions to construct a conditional probability matrix; obtain a single-feature entropy value through local sensitive hashing mapping; calculate the mutual information between different features in the physiological feature set to obtain an interaction entropy matrix; fuse the single-feature entropy value with the interaction entropy matrix to obtain an energy metabolism feature vector;

[0019] Construct a fuzzy rule base for the dynamic fuzzy inference system. The rule weights of the fuzzy rule base include historical data weights, current state weights, and future trend weights; input the energy metabolism feature vector into the dynamic fuzzy inference system, divide the time window to construct a biological rhythm prediction model to predict the energy metabolism pattern for each time window to obtain a timing energy feature; use a second-order filter to process the non-linear time-varying features of the timing energy feature, and calculate the basal energy consumption value in combination with the output result of the dynamic fuzzy inference system;

[0020] Construct a five - layer adaptive fuzzy neural network including an input layer, a fuzzy layer, a rule layer, a normalization layer and an output layer; the input layer receives the basic energy consumption value, and dynamically adjusts the weight coefficients according to the input time and the importance degree of the basic energy consumption value to obtain weighted features; the fuzzy layer uses a bell - shaped membership function to fuzzify the weighted features to obtain fuzzy features, and optimizes the parameters of the bell - shaped membership function by the gradient descent method; the rule layer receives the fuzzy features and performs rule reasoning. When the description accuracy of the existing rules for the fuzzy features is lower than the rule threshold, rule splitting is triggered. When the number of rules exceeds the redundancy threshold, rule merging is triggered to obtain the rule reasoning result; the normalization layer uses the softmax function to normalize the rule reasoning result to obtain normalized features; the output layer uses a hybrid learning algorithm to process the normalized features. In the forward propagation process, the least - squares method is used to estimate the consequent parameters, and in the backward propagation process, the gradient descent method is used to optimize the antecedent parameters. The learning step size is dynamically adjusted according to the network output error, and finally the breast milk benchmark feeding amount is output.

[0021] In an alternative embodiment,

[0022] Collect the physiological state data and breast milk composition data of the mother, input them into a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain temporal features through chaotic memory mapping, including:

[0023] Collect the physiological state data and breast milk composition data of the mother and input them into the deep hybrid cognitive network;

[0024] Construct a dynamic morphological topology adaptive mechanism. The dynamic morphological topology adaptive mechanism is based on a deep hybrid cognitive network. It uses a morphological gradient operator for edge enhancement to obtain key point data, determines a data window through the key point data, calculates an adaptive threshold based on the weighted sum of variance and mean within the data window, and filters to obtain feature points; constructs a dynamic connection matrix according to the feature points to determine the topological relationship between features, and uses an annealing strategy to adjust the connection strength of the dynamic connection matrix. The temperature parameter of the annealing strategy decreases with the number of training rounds; determines the structural element size of the morphological operator according to the local distribution of the feature points, and processes the feature points to obtain the initial feature vector.

[0025] The initial feature vector is transformed through a chaotic mapping to obtain non-linear features; a multi-layer memory unit is constructed according to the total number of preset memory units, and each layer of the memory unit calculates the correlation degree between the current moment and the historical state through an attention unit to obtain attention weights; a forgetting gate unit controls the retention degree of historical information based on the time interval and state similarity to obtain memory features; forward propagation is performed on the memory features to extract causal relationship features and backward propagation is performed to extract long-term dependence features, and the chaotic dynamics of both the forward propagation and the backward propagation are calculated using piecewise linear approximation; a residual connection is constructed between adjacent two layers of the multi-layer memory unit to transfer gradient information; the causal relationship features and the long-term dependence features are fused to obtain time series features.

[0026] In an alternative embodiment,

[0027] The feature vector and the time series features are input into an immune evolutionary self-organizing network to calculate the breast milk nutrition density coefficient, and the breast milk benchmark feeding amount is corrected according to the breast milk nutrition density coefficient. Calculating the actual breast milk feeding amount through an adaptive resonance theory network includes:

[0028] The feature vector and the time series features are combined to form a feature to be optimized, which serves as an antigen in the immune evolutionary self-organizing network. An antibody population is initialized and constructed, where each antibody corresponds to a set of nutrition density coefficients and is represented by real number coding;

[0029] An affinity function is constructed based on the nutritional component indicators and dynamic change trends in the feature to be optimized, and the matching degree between the antibody and the antigen is calculated to obtain an affinity value. The antibody population is cloned and amplified according to the affinity value to obtain a cloned antibody population, and the cloning number of each antibody is proportional to the corresponding affinity value; a Gaussian mutation operation is performed on the cloned antibody population to obtain a mutant antibody population, and the mutation probability is inversely proportional to the corresponding affinity value; an evolutionary antibody population is screened according to a preset affinity threshold;

[0030] The optimal antibody with the highest affinity is selected to determine the optimal nutrition density coefficient, which is multiplied by the breast milk benchmark feeding amount to obtain a preliminary corrected feeding amount;

[0031] An adaptive resonance theory network including a comparison layer and a recognition layer is constructed. The comparison layer receives the preliminary corrected feeding amount, the feature to be optimized, and the feedback signal from the recognition layer; the matching degree is calculated based on an adaptive threshold. When the matching degree exceeds a preset vigilance parameter, it is classified into an existing category and the feature template is updated, otherwise a new category is created; a lateral inhibition mechanism is constructed between the neurons in the recognition layer for competitive learning to ensure a single winning neuron, and the weight vector of the winning neuron is updated. The learning rate of the weight vector decreases with the number of training rounds; the actual breast milk feeding amount is calculated based on the weight vector of the winning neuron and the feature template.

[0032] In an alternative embodiment,

[0033] Collect the growth and development data of infants after ingesting the actual breast milk intake, and input it into a multi-layer spiral pulse neural network to extract the development feature vector; use the fractal dynamics model to map with the standard growth and development parameters to calculate the development deviation value, including:

[0034] Collect the growth and development data of infants after ingesting the actual breast milk intake, including physical development data and neurodevelopment data;

[0035] Construct a multi-layer spiral pulse neural network to convert the growth and development data into a pulse sequence, where the pulse frequency is proportional to the numerical size; construct an ion channel model through the hidden layer neurons, receive the pulse sequence through the synaptic weights and integrate it into the postsynaptic potential, and trigger an action potential when the postsynaptic potential exceeds the preset potential threshold to form an action potential sequence;

[0036] Based on the action potential sequence, construct a spiral synaptic transmission mechanism, including a short-term regulation module and a long-term regulation module; the short-term regulation module calculates the short-term intensity regulation factor based on the amplitude, and the long-term regulation module calculates the long-term intensity regulation factor based on the frequency and duration, and combines them to form the total regulation coefficient;

[0037] Based on the total regulation coefficient, construct a recursive loop to realize the reciprocating transmission of information between layers, dynamically adjust the synaptic weights to form a weight matrix, and combine the firing pattern of the output layer neurons with the weight matrix to construct a development feature vector;

[0038] Construct a fractal dynamics model, calculate the mapping trajectory of the development feature vector in the fractal phase space through an iterative function to obtain the actual development trajectory; construct a reference fractal trajectory with the same fractal dimension distribution and iterative mapping rule based on the standard growth and development parameters; calculate the Euclidean distance between the actual development trajectory and the reference fractal trajectory in the fractal phase space to obtain the development deviation value.

[0039] In an alternative embodiment,

[0040] The biotic group intelligent optimization system constructed according to the development deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to realize the dynamic optimization of the breast milk intake, including:

[0041] Construct an optimization objective function by weighted summation of the physical development deviation value and the neurodevelopment deviation value, and the weight coefficients of the optimization objective function are allocated based on the importance of the development indicators to obtain the objective function value;

[0042] Convert the scale factor, morphological operator parameters of the multi-scale morphological neural network and the network weights and network bias values of the deep hybrid cognitive network into an optimization variable sequence, construct a parameter search space, and initialize the optimization variable sequence based on the objective function value;

[0043] Construct a dual-mode mutation operator, which includes a Gaussian mutation module and a Cauchy mutation module; the Gaussian mutation module performs local parameter search according to the first mutation step size to obtain a local search sequence, and the Cauchy mutation module performs global parameter exploration according to the second mutation step size to obtain a global search sequence, and combines the local search sequence and the global search sequence to form a mutation parameter sequence;

[0044] Input the mutation parameter sequence into the multi-scale morphological neural network and the deep hybrid cognitive network. The multi-scale morphological neural network outputs a feature extraction result, and the deep hybrid cognitive network outputs a development prediction value based on the feature extraction result; substitute the development prediction value into the optimization objective function to calculate the current objective function value, compare the current objective function value with the historical objective function value, and store them in the elite solution set in descending order;

[0045] When the number of iterations reaches a preset value and the difference between adjacent objective function values is less than the first preset threshold, re-execute the dual-mode mutation operator by increasing the mutation step size according to the escape index;

[0046] Select the parameter sequence with the optimal objective function value from the elite solution set as the optimal parameter sequence, drive the multi-scale morphological neural network and the deep hybrid cognitive network, extract the breast milk feature sequence and predict the final development prediction value; calculate the evaluation deviation value between the final development prediction value and the actual development data. When it is greater than the second preset threshold, return to execute the dual-mode mutation operator. When it is less than the second preset threshold, output the optimal breast milk feeding amount parameter.

[0047] In the second aspect of the embodiments of the present invention,

[0048] Provide an intelligent calculation and dynamic adjustment system for breast milk feeding amount, including:

[0049] The first unit is used to collect the real-time physiological parameters of the baby to form a dynamic data stream, input the data stream into the multi-scale morphological neural network to construct a baby digital model, perform feature dimensionality reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set; calculate the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, input it into the dynamic fuzzy inference system, calculate the basic energy consumption value in combination with the biological rhythm prediction model, and calculate the benchmark breast milk feeding amount through the adaptive fuzzy neural network;

[0050] The second unit is used to collect the physiological state data and breast milk composition data of the mother, input them into the deep hybrid cognitive network adopting the dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain time series features through chaotic memory mapping; input the feature vectors and time series features into the immune evolutionary self-organizing network to calculate the breast milk nutrition density coefficient, correct the benchmark breast milk feeding amount according to the breast milk nutrition density coefficient, and calculate the actual breast milk feeding amount through the adaptive resonance theory network;

[0051] The third unit is used to collect the growth and development data after the baby ingests the actual breast milk feeding amount, input the multi-layer spiral pulse neural network to extract the development feature vector; use the fractal dynamics model to map with the standard growth and development parameters to calculate the development deviation value; the biota intelligent optimization system constructed according to the development deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to realize the dynamic optimization of the breast milk feeding amount.

[0052] In the third aspect of the embodiments of the present invention,

[0053] A kind of electronic device is provided, including:

[0054] A processor;

[0055] A memory for storing instructions executable by the processor;

[0056] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0057] In the fourth aspect of the embodiments of the present invention,

[0058] A computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0059] In the embodiments of the present invention, by constructing a baby digital model and considering the mother's physiological state and breast milk composition, the breast milk feeding amount that more meets the individual needs of the baby can be calculated, avoiding the empiricism and subjectivity of the traditional method, and improving the scientificity and effectiveness of feeding; it can be dynamically adjusted according to the baby's real-time physiological parameters, growth and development data, and the nutritional density coefficient of breast milk, and the model parameters are continuously optimized through the biota intelligent optimization system, so that the breast milk feeding amount is always kept in the best state to better meet the needs of the baby's growth and development; using artificial intelligence and machine learning technologies, the intelligent calculation and dynamic adjustment of the breast milk feeding amount are realized, reducing the manual intervention and calculation complexity, and providing more convenient and efficient guidance for breast milk feeding. Description of the Drawings

[0060] Figure 1 It is a schematic flow chart of the method for intelligent calculation and dynamic adjustment of the breast milk feeding amount in the embodiments of the present invention;

[0061] Figure 2 It is a comparison curve of the feature extraction accuracy;

[0062] Figure 3 It is an affinity convergence curve of the immune algorithm;

[0063] Figure 4 It is a bar chart of the feeding amount prediction error distribution;

[0064] Figure 5 It is a scatter plot for comparing feature extraction and prediction performance;

[0065] Figure 6 It is a schematic structural diagram of the intelligent calculation and dynamic adjustment system for breast milk feeding volume in the embodiment of the present invention. Specific embodiments

[0066] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0068] Figure 1 It is a schematic flowchart of the intelligent calculation and dynamic adjustment method for breast milk feeding volume in the embodiment of the present invention. As Figure 1 shown, the method includes:

[0069] S101. Collect real-time physiological parameters of the infant to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to construct an infant digital model, perform feature dimensionality reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set; calculate the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, input it into a dynamic fuzzy inference system, calculate the basal energy consumption value in combination with a biological rhythm prediction model, and calculate the benchmark breast milk feeding volume through an adaptive fuzzy neural network;

[0070] S102. Collect the physiological state data and breast milk composition data of the mother, input them into a deep hybrid cognitive network using a dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain temporal features through a chaotic memory mapping; input the feature vectors and temporal features into an immune evolutionary self-organizing network to calculate the breast milk nutritional density coefficient, correct the benchmark breast milk feeding volume according to the breast milk nutritional density coefficient, and calculate the actual breast milk feeding volume through an adaptive resonance theory network;

[0071] S103. Collect the growth and development data of the infant after ingesting the actual breast milk intake, input the data into a multi-layer spiral pulse neural network to extract the development feature vector; use the fractal dynamics model to map with the standard growth and development parameters to calculate the development deviation value; the bio-group intelligent optimization system constructed based on the development deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to realize the dynamic optimization of the breast milk intake.

[0072] In an alternative embodiment, collect the real-time physiological parameters of the infant to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to construct an infant digital model, and perform feature dimension reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set including:

[0073] Collect the physiological parameters of the infant, including body temperature parameters, heart rate parameters, respiratory rate parameters, blood oxygen saturation parameters, and weight parameters; classify the physiological parameters according to the change characteristics of the physiological parameters to determine the fast-changing parameters and slow-changing parameters; determine the sampling frequency of the fast-changing parameters according to the ratio of the standard deviation to the mean of the fast-changing parameters, and determine the sampling frequency of the slow-changing parameters according to the ratio of the change amount of the slow-changing parameters to the time interval; sample the physiological parameters through the sampling frequency to obtain the original data, and use wavelet threshold filtering to perform noise reduction processing on the original data to obtain a dynamic data stream;

[0074] Input the dynamic data stream into a multi-scale morphological neural network, determine the scale factor according to the ratio of the characteristic period of the dynamic data stream to the basic sampling interval, and perform multi-scale decomposition on the dynamic data stream using the scale factor to obtain multi-scale data; calculate the variance of the multi-scale data within the local window, determine the structural element size of the morphological operator according to the variance, construct the dilation operator and erosion operator using the structural element size, and process the multi-scale data through the dilation operator and the erosion operator to construct an infant digital model;

[0075] Determine the initial number of nodes of the self-organizing competitive mapping network according to the number of samples and input dimensions of the infant digital model, and the self-organizing competitive mapping network adopts an adaptive growth structure; input the infant digital model into the self-organizing competitive mapping network, calculate the Euclidean distance and Manhattan distance between samples, construct an improved distance metric according to the Euclidean distance and the Manhattan distance, perform local response normalization processing on the infant digital model using the improved distance metric to obtain the normalized features; construct a timing constraint function, calculate the timing constraint value of the normalized features according to the timing constraint function, and fuse the timing constraint value with the normalized features to obtain a physiological feature set.

[0076] The specific function of the dilation operator is to expand or magnify the high-value regions. By taking the maximum value within a local range, it widens the peak part of the data, fills in small gaps or voids, thereby enhancing the overall structure.

[0077] The specific function of the erosion operator is to contract or weaken the high-value regions. By taking the minimum value within a local range, it narrows the peak part of the data, removes isolated high-value points, thereby smoothing or reducing noise.

[0078] In a specific implementation, first, various physiological parameters of the infant are collected. These parameters include body temperature, heart rate, respiratory rate, blood oxygen saturation, and body weight. For the convenience of subsequent processing, these parameters are classified according to their change rates. The heart rate, respiratory rate, and blood oxygen saturation change relatively quickly and are classified as fast-changing parameters; the body temperature and body weight change relatively slowly and are classified as slow-changing parameters.

[0079] Next, the sampling frequencies of the various physiological parameters are determined. The sampling frequency of the fast-changing parameters is determined by the ratio of their standard deviation to the mean. The larger the ratio, the greater the data fluctuation, and a faster sampling frequency is required to capture these fluctuations. For example, if the heart rate of a certain infant fluctuates greatly and the ratio of the standard deviation to the mean is 0.2, a higher sampling frequency is required, such as once per second; while if the heart rate of another infant is relatively stable and the ratio of the standard deviation to the mean is only 0.05, the sampling frequency can be reduced, such as once every 2 seconds. The sampling frequency of the slow-changing parameters is determined according to the ratio of their change amount to the time interval. For example, if the body temperature of a certain infant changes by 0.5 degrees Celsius within 1 hour, the body temperature data can be collected once every 10 minutes; while if the body temperature of another infant hardly changes within 1 hour, the sampling interval can be extended, such as once every 30 minutes.

[0080] After the original data is collected through the determined sampling frequencies, the data needs to be denoised. Here, the wavelet threshold filtering method is used to remove the noise in the original data. For example, the collected infant heart rate data may contain some noise caused by measurement errors or other interferences. Wavelet threshold filtering can effectively remove these noises, retain the true physiological signals, and obtain a smooth dynamic data stream.

[0081] Input the denoised dynamic data stream into a multi-scale morphological neural network to construct a baby digital model. First, determine the scale factor according to the ratio of the characteristic period of the dynamic data stream to the basic sampling interval. For example, if the characteristic period of a baby's breathing is 1 second and the basic sampling interval is 0.1 second, then the scale factor is 10. Then, perform multi-scale decomposition on the dynamic data stream using the scale factor. For example, the heart rate data can be decomposed into components at different time scales, representing the long-term trend, medium-term fluctuations, and short-term variations of the heart rate respectively. Next, calculate the variance of the multi-scale data within a local window. For example, the variance within a window composed of every 10 data points can be calculated. Determine the size of the structural element of the morphological operator according to the variance. The larger the variance, the greater the data fluctuation and the larger the structural element required. For example, if the variance of the heart rate data within a certain window is large, a larger structural element such as a 5x5 matrix can be used; while if the variance is small, a smaller structural element such as a 3x3 matrix can be used. Construct dilation and erosion operators using the determined size of the structural element, process the multi-scale data, and finally construct a baby digital model.

[0082] Finally, use a self-organizing competitive mapping network to perform feature dimensionality reduction and clustering on the baby digital model to obtain a physiological feature set. First, determine the initial number of nodes of the self-organizing competitive mapping network according to the number of samples and input dimensions of the baby digital model. The network adopts an adaptive growth structure and can dynamically adjust the number of nodes according to the characteristics of the data. For example, if the number of samples of the baby digital model is large, more initial nodes are required. Input the baby digital model into the network and calculate the Euclidean distance and Manhattan distance between samples. For example, the square root of the sum of the squares of the differences between two samples in each dimension can be calculated as the Euclidean distance, and the sum of the absolute values of the differences in each dimension can be used as the Manhattan distance. Construct an improved distance metric based on the Euclidean distance and Manhattan distance. For example, the weighted sum of the Euclidean distance and Manhattan distance can be used as the improved distance metric. Perform local response normalization processing on the baby digital model using the improved distance metric to obtain normalized features. Construct a temporal constraint function and calculate the temporal constraint value of the normalized features according to the temporal constraint function. For example, the difference between features at adjacent time points can be calculated as the temporal constraint value. Fuse the temporal constraint value with the normalized features to obtain the final physiological feature set. For example, the temporal constraint value can be added as a new dimension to the normalized features.

[0083] In this embodiment, through multi-scale decomposition and morphological processing, key information in the dynamic data stream can be effectively extracted, the data dimension can be reduced, and the efficiency of subsequent processing can be improved; by using wavelet threshold filtering and local response normalization processing, the influence of noise and outliers can be effectively removed, and the robustness of the model can be enhanced; by constructing a baby digital model and extracting a physiological feature set, personalized monitoring of the baby's physiological state can be achieved, providing a more accurate basis for baby health management.

[0084] In an alternative embodiment, calculating the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, inputting it into a dynamic fuzzy inference system, and combining with a biological rhythm prediction model to calculate the basal energy expenditure value. Calculating the breast milk benchmark feeding amount through an adaptive fuzzy neural network includes:

[0085] Performing pattern segmentation on the feature sequences in the physiological feature set, determining the similarity threshold according to the standard deviation of the feature sequences; counting the number of matching patterns in adjacent dimensions to construct a conditional probability matrix; obtaining the single-feature entropy value through local sensitive hashing mapping; calculating the mutual information between different features in the physiological feature set to obtain an interaction entropy matrix; fusing the single-feature entropy value with the interaction entropy matrix to obtain an energy metabolism feature vector;

[0086] Constructing a fuzzy rule base for the dynamic fuzzy inference system, where the rule weights of the fuzzy rule base include historical data weights, current state weights, and future trend weights; inputting the energy metabolism feature vector into the dynamic fuzzy inference system, dividing time windows to construct a biological rhythm prediction model to predict the energy metabolism pattern for each time window, obtaining time-series energy features; processing the non-linear time-varying features of the time-series energy features using a second-order filter, and combining with the output result of the dynamic fuzzy inference system to calculate the basal energy expenditure value;

[0087] Construct a five - layer adaptive fuzzy neural network including an input layer, a fuzzy layer, a rule layer, a normalization layer, and an output layer; the input layer receives the basic energy consumption value, and dynamically adjusts the weight coefficients according to the input time and the importance degree of the basic energy consumption value to obtain weighted features; the fuzzy layer uses a bell - shaped membership function to fuzzify the weighted features to obtain fuzzy features, and optimizes the parameters of the bell - shaped membership function by the gradient descent method; the rule layer receives the fuzzy features and performs rule inference, triggers rule splitting when the description accuracy of the existing rules for the fuzzy features is lower than the rule threshold, and triggers rule merging when the number of rules exceeds the redundancy threshold to obtain a rule inference result; the normalization layer uses a softmax function to normalize the rule inference result to obtain normalized features; the output layer uses a hybrid learning algorithm to process the normalized features, estimates the consequent parameters by the least - squares method in the forward propagation process, optimizes the antecedent parameters by the gradient descent method in the reverse propagation process, dynamically adjusts the learning step size according to the network output error, and finally outputs the breast milk benchmark feeding amount.

[0088] The antecedent parameters specifically refer to the description parameters of the input conditions in the fuzzy inference rules, which determine how the input data is classified into different fuzzy sets. For example, when judging the baby's energy consumption level, the antecedent parameters can be used to define the ranges of "high energy consumption" or "low energy consumption", including the central values, variation ranges, and smoothness of transitions of these categories. These parameters affect how the input data is fuzzified, thus determining the applicability of the rules.

[0089] The consequent parameters specifically refer to the calculation parameters of the output results in the fuzzy inference rules, which determine how the rules affect the final decision or prediction after being triggered. For example, when inferring the breast milk feeding amount, the consequent parameters are used to calculate the specific adjustment amplitude of the feeding amount, usually represented by a mathematical formula, and it may have a linear or non - linear relationship with the input variables. These parameters directly affect the final numerical output, enabling the fuzzy rules to be converted into specific operation suggestions.

[0090] In a specific embodiment, first, collect the physiological feature set of the baby, such as heart rate, respiratory rate, body temperature, crying duration, sleep duration, etc. Suppose the sleep duration data of a baby for one week is collected, with the unit of hours: [7, 8, 7.5, 6.5, 7, 8.5, 7].

[0091] Next, perform pattern segmentation on the collected feature sequence. Taking the sleep duration as an example, the data for one week can be divided into multiple time periods according to the change of the baby's sleep duration. For example, [7, 8, 7.5], [6.5, 7], and [8.5, 7] can be divided into three time periods respectively, representing different patterns of the baby's sleep duration.

[0092] Then, determine the similarity threshold according to the standard deviation of the feature sequence. Calculate the standard deviation of the one-week sleep duration data, and assume the obtained value is 0.7. Set a scaling factor, for example, 0.5, then the similarity threshold is 0.7×0.5 = 0.35.

[0093] Count the number of matching patterns of adjacent dimensions in the feature sequence to construct a conditional probability matrix. For example, count the probability of the change in sleep duration between the aforementioned three time periods to construct a 3×3 conditional probability matrix.

[0094] Use locality-sensitive hashing to map the conditional probability matrix to a low-dimensional space to obtain a single-feature entropy value. Map the 3×3 conditional probability matrix to a three-dimensional vector, for example, [0.2, 0.3, 0.5], representing the entropy value of the sleep pattern in each time period.

[0095] Calculate the mutual information between different features in the physiological feature set to obtain an interaction entropy matrix. For example, calculate the mutual information between sleep duration and heart rate to construct an interaction entropy matrix. Assume only these two features are considered, then a 2×2 matrix is obtained.

[0096] Fuse the single-feature entropy value with the interaction entropy matrix to obtain an energy metabolism feature vector. Fuse the single-feature entropy value [0.2, 0.3, 0.5] of sleep duration with the interaction entropy matrix of sleep duration and heart rate to obtain a new vector as the energy metabolism feature vector.

[0097] Construct a fuzzy rule base for the dynamic fuzzy inference system. The rule weights of the fuzzy rule base include historical data weights, current state weights, and future trend weights. For example, a rule can be: If the baby has a long sleep duration and a low heart rate, then the energy metabolism level is low. The rule weights can be dynamically adjusted according to historical data, current state, and future trend.

[0098] Input the energy metabolism feature vector into the dynamic fuzzy inference system. Divide the 24-hour cycle into multiple time windows, for example, one time window per hour. Use the biological rhythm prediction model to predict the energy metabolism pattern of each time window to obtain the time-series energy features.

[0099] Adopt a second-order filter to process the non-linear time-varying features of the time-series energy features. Combine the output results of the dynamic fuzzy inference system to calculate the basic energy consumption value.

[0100] Construct a five-layer adaptive fuzzy neural network, including an input layer, a fuzzy layer, a rule layer, a normalization layer, and an output layer.

[0101] The input layer receives the basic energy consumption value and dynamically adjusts the weight coefficients according to the importance of the input time and the basic energy consumption value to obtain the weighted features.

[0102] The fuzzy layer uses a bell-shaped membership function to fuzzify the weighted features and obtain fuzzy features. The parameters of the bell-shaped membership function are optimized by the gradient descent method.

[0103] The rule layer receives the fuzzy features and performs rule inference. When the description accuracy of the existing rules for the fuzzy features is lower than the rule threshold, rule splitting is triggered. When the number of rules exceeds the redundancy threshold, rule merging is triggered, and the rule inference result is obtained.

[0104] The normalization layer uses the softmax function to normalize the rule inference result and obtain the normalized features.

[0105] The output layer processes the normalized features using a hybrid learning algorithm. In the forward propagation process, the least squares method is used to estimate the consequent parameters. In the backward propagation process, the gradient descent method is used to optimize the antecedent parameters. The learning step size is dynamically adjusted according to the network output error, and the breast milk benchmark feeding amount is finally output.

[0106] In this embodiment, calculating the breast milk feeding amount based on the individual physiological characteristics of the infant avoids the "one-size-fits-all" problem of traditional methods, provides more accurate personalized feeding guidance, and helps the healthy growth of the infant; it takes into account the dynamic changes of the infant's biological rhythm and energy metabolism, can adjust the feeding amount according to the real-time state of the infant, and better meets the actual needs of the infant; using intelligent algorithms such as fuzzy inference and neural networks can automatically learn and adapt to the growth and development laws of the infant, realize the intelligent calculation of the breast milk feeding amount, and reduce the feeding burden on parents.

[0107] In an alternative embodiment, physiological state data and breast milk composition data of the mother are collected, and a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism is input to extract feature vectors, and time series features are obtained through chaotic memory mapping, including:

[0108] Collect the physiological state data and breast milk composition data of the mother and input them into the deep hybrid cognitive network;

[0109] Construct a dynamic morphological topology adaptive mechanism. The dynamic morphological topology adaptive mechanism is based on the deep hybrid cognitive network. The morphological gradient operator is used for edge enhancement to obtain key point data. The data window is determined through the key point data. Based on the weighted sum of the variance and mean within the data window, the adaptive threshold is calculated, and the feature points are screened. The dynamic connection matrix is constructed according to the feature points to determine the topological relationship between features. The connection strength of the dynamic connection matrix is adjusted using the annealing strategy, and the temperature parameter of the annealing strategy decreases with the number of training rounds. The size of the structural element of the morphological operator is determined according to the local distribution of the feature points, and the feature points are processed to obtain the initial feature vector;

[0110] The initial feature vector is transformed through a chaotic mapping to obtain non-linear features; a multi-layer memory unit is constructed according to the total number of preset memory units, and each layer of the memory unit calculates the correlation degree between the current moment and the historical state through an attention unit to obtain attention weights; a forgetting gate unit controls the retention degree of historical information based on the time interval and state similarity to obtain memory features; forward propagation is performed on the memory features to extract causal relationship features and backward propagation is performed to extract long-term dependence features, and the chaotic dynamics of both the forward propagation and the backward propagation are calculated using piecewise linear approximation; a residual connection is constructed between adjacent two layers of the multi-layer memory unit to transfer gradient information; the causal relationship features and the long-term dependence features are fused to obtain temporal features.

[0111] In a specific embodiment, first, physiological state data of the mother is collected, such as heart rate, blood pressure, body temperature, sleep duration, etc., and breast milk composition data, such as the contents of protein, fat, lactose, vitamins, minerals, etc. These data can be obtained by using wearable devices, household medical devices, and laboratory tests. For example, a smart bracelet is used to record the mother's heart rate and sleep duration, a household sphygmomanometer is used to measure blood pressure, a thermometer is used to measure body temperature, and the collected breast milk samples are sent to a laboratory for component analysis.

[0112] Next, the collected physiological state data and breast milk composition data are input into a deep hybrid cognitive network. The structure of this network can be adjusted according to the actual situation. For example, a convolutional neural network, a recurrent neural network, or a combination of them can be used.

[0113] In the deep hybrid cognitive network, a dynamic morphological topology adaptive mechanism is constructed. This mechanism first uses a morphological gradient operator to enhance the edges of the input data. For example, dilation and erosion operations are used to highlight the edge information of the data to obtain key point data. Suppose a certain physiological index of a mother fluctuates greatly within a period of time, and these points with obvious fluctuations can be identified as key points through the morphological gradient operator.

[0114] Then, based on these key point data, a data window is determined. For example, taking the data within a certain time range before and after each key point as a window. Suppose the time corresponding to a certain key point is the 10th day, then the data between the 8th day and the 12th day can be taken as a window.

[0115] According to the weighted sum of the data variance and the data mean within the data window, an adaptive threshold is calculated. For example, the variance and the mean are multiplied by different weights and then added to obtain the threshold. Suppose the variance of the data within a certain window is 10, the mean is 5, and the weights are 0.8 and 0.2 respectively, then the threshold is 10×0.8 + 5×0.2 = 9.

[0116] According to this adaptive threshold, feature points are screened from the key point data. For example, the points in the key point data that exceed the threshold are retained as feature points. Suppose the value of a certain key point is 12, which exceeds the threshold of 9, then this key point is retained as a feature point.

[0117] Next, based on these feature points, a dynamic connection matrix is constructed to determine the topological relationship between features. For example, the K-nearest neighbor algorithm can be used to connect the feature points that are relatively close.

[0118] The annealing strategy is adopted to adjust the connection strength of the dynamic connection matrix. The temperature parameter of the annealing strategy gradually decreases as the number of training rounds increases. For example, the initial temperature is set to 100, and the temperature decreases by 10% after each round of training.

[0119] The size of the structuring element of the morphological operator is determined according to the local data distribution of the feature points. For example, if the local data distribution of the feature points is relatively concentrated, a smaller structuring element is used; if the data distribution is relatively dispersed, a larger structuring element is used.

[0120] The morphological operator is used to process the feature points to obtain the initial feature vector. For example, morphological opening or closing operations are applied to each feature point to obtain new feature values, and these feature values are combined into the initial feature vector.

[0121] The initial feature vector is transformed into the chaotic space through a chaotic mapping to obtain non-linear features. For example, the Logistic mapping or Tent mapping is used to map each value in the feature vector to a new value.

[0122] Based on the non-linear features, multi-layer memory units are constructed. Each layer of memory units calculates the correlation degree between the current moment and the historical state through the attention unit to obtain the attention weights. For example, the dot product attention mechanism is used to calculate the similarity between the feature vector at the current moment and the feature vector at the historical moment to obtain the attention weights.

[0123] The attention weights are input into the forget gate unit. The forget gate unit calculates the time interval and state similarity, and based on this information, controls the retention degree of historical information to obtain the memory features. For example, the longer the time interval and the lower the state similarity, the smaller the value output by the forget gate unit and the less historical information is retained.

[0124] Forward propagation and backward propagation processing are performed on the memory features. Causal relationship features are extracted through forward propagation, and long-term dependence features are extracted through backward propagation. The chaotic dynamics of both forward propagation and backward propagation are calculated using the piecewise linear approximation method.

[0125] Residual connections are constructed between adjacent layers of the multi-layer memory units to transmit gradient information.

[0126] Fuse the causal relationship features and long-term dependence features to obtain time series features. For example, concatenate the two feature vectors together, or add them after multiplying them by different weights respectively.

[0127] In Figure 2 , the comparative analysis results of the feature extraction accuracy of this method with traditional CNN and traditional RNN within a 24-hour cycle are shown. From the perspective of time series performance, this method maintains a significant performance advantage throughout the observation period. Specifically:

[0128] At the initial 0h moment, the accuracy of this method reaches 95.3%, while the accuracies of traditional CNN and RNN are 85.7% and 88.4% respectively, and this method is 9.6 and 6.9 percentage points higher respectively. At the 6h moment, although the accuracies of each method show a slight decline, this method still maintains a high accuracy of 93.8%, while traditional CNN and RNN drop to 83.2% and 86.9% respectively. This indicates that this method has stronger robustness during periods of large data fluctuations.

[0129] By the 12h moment, the accuracy of this method rebounds to 94.7%, while traditional CNN and RNN are 84.5% and 87.8% respectively. At the 18h moment, this method reaches the highest value of 96.1% for the whole day, while the traditional methods are 86.3% and 89.2% respectively. Finally, at the 24h moment, the accuracies of the three methods are 95.8%, 85.9% and 88.7% respectively. From the overall trend, the accuracy of this method always remains within the high level range of 93.8% - 96.1%, with a fluctuation range of only 2.3 percentage points; while the accuracies of traditional CNN and RNN fluctuate between 83.2% - 86.3% and 86.9% - 89.2% respectively, and the performance fluctuations are more obvious. These data fully illustrate that this method has significant advantages in both the accuracy and stability of feature extraction, especially showing stronger adaptability when dealing with time series data fluctuations. This method has an average accuracy improvement of 8 - 10 percentage points, providing a more reliable feature basis for subsequent analysis.

[0130] In this embodiment, through the deep hybrid cognitive network and the dynamic morphological topology adaptive mechanism, key features in the mother's physiological state and breast milk component data can be extracted more effectively, thereby improving the accuracy of breast milk component prediction. For example, the contents of key nutritional components such as protein, fat, and lactose in breast milk can be predicted more accurately, providing a more precise reference for guiding infant feeding; through the chaotic memory mapping module, this technical solution can capture the dynamic laws of breast milk components changing over time and reveal the complex relationship between them and the mother's physiological state. For example, the effects of factors such as the mother's sleep quality and stress level on breast milk components can be analyzed, providing a scientific basis for deeply understanding the dynamic change mechanism of breast milk components; it can perform personalized breast milk component analysis and prediction for the individual differences of different mothers. For example, according to factors such as the mother's age, physical condition, and living habits, a personalized prediction model can be established to provide more precise breast milk component analysis and guidance for each mother.

[0131] In an alternative embodiment, the feature vector and the time series feature are input into the immune evolutionary self-organizing network to calculate the breast milk nutrition density coefficient, and the breast milk benchmark feeding amount is corrected according to the breast milk nutrition density coefficient. Calculating the actual breast feeding amount through the adaptive resonance theory network includes:

[0132] The feature vector and the time series feature are combined to form a feature to be optimized, which is used as an antigen in the immune evolutionary self-organizing network. An antibody population is initialized and constructed, where each antibody corresponds to a set of nutrition density coefficients and is represented by real number coding;

[0133] Based on the nutritional component indicators and the dynamic change trends in the feature to be optimized, an affinity function is constructed, and the matching degree between the antibody and the antigen is calculated to obtain an affinity value. According to the affinity value, the antibody population is cloned and amplified to obtain a cloned antibody population, and the cloning number of each antibody is proportional to the corresponding affinity value; a Gaussian mutation operation is performed on the cloned antibody population to obtain a mutant antibody population, and the mutation probability is inversely proportional to the corresponding affinity value; an evolutionary antibody population is screened according to a preset affinity threshold;

[0134] The optimal antibody with the highest affinity is selected to determine the optimal nutrition density coefficient, which is multiplied by the breast milk benchmark feeding amount to obtain a preliminary corrected feeding amount;

[0135] Construct an Adaptive Resonance Theory network including a comparison layer and a recognition layer. The comparison layer receives the preliminarily corrected feeding amount, the features to be optimized, and the feedback signal from the recognition layer. Calculate the matching degree based on an adaptive threshold. When the matching degree exceeds the preset vigilance parameter, classify it into the existing category and update the feature template; otherwise, create a new category. Build a lateral inhibition mechanism among the neurons in the recognition layer for competitive learning to ensure a single winning neuron, and update the weight vector of the winning neuron. The learning rate of the weight vector decreases with the number of training rounds. Calculate the actual breastfeeding amount based on the weight vector of the winning neuron and the feature template.

[0136] In a specific embodiment, first, collect the feature data of the infant, including but not limited to weight, height, age, gender, health status, etc., and the time-series features reflecting the dynamic change trend of the infant, such as appetite change, sleep duration, defecation situation, etc. These data will be quantified and combined into a feature vector as the input for subsequent calculations. For example, a healthy baby girl at 6 months old with a weight of 7 kg, her feature vector can be expressed as [6, female, 7, healthy,...], and at the same time record her recent appetite change, sleep duration, defecation situation, etc., which are quantified and added to the feature vector after quantification.

[0137] Combine the above feature vector and time-series features to form the features to be optimized. Next, use the immune evolutionary self-organizing network to calculate the nutritional density coefficient of breast milk. Initialize an antibody population, where each antibody represents a set of nutritional density coefficients of breast milk, such as the contents of protein, fat, carbohydrates, vitamins, minerals, etc. Each antibody uses real-number coding to represent these nutritional density coefficients, such as [protein content, fat content, carbohydrate content,...].

[0138] Then, construct an affinity function, which is used to evaluate the matching degree between the antibody and the antigen (i.e., the features to be optimized). The design of the affinity function needs to consider the nutritional needs of the infant and its dynamic change trend. For example, if the infant's appetite increases, the affinity function should be more inclined to select antibodies with higher nutritional density. Calculate the affinity value between each antibody and the features to be optimized through the affinity function.

[0139] Clone and amplify the antibody population according to the affinity value. The higher the affinity value, the more clones of the antibody. For example, the antibody with an affinity value of 0.9 is cloned 10 times, and the antibody with an affinity value of 0.6 is cloned 6 times. Then, perform Gaussian mutation operation on the cloned antibody population. The mutation probability is inversely proportional to the affinity value, that is, the higher the affinity value, the lower the mutation probability of the antibody, to ensure the genetic stability of excellent antibodies. For example, the mutation probability of the antibody with an affinity value of 0.9 is 0.1, and the mutation probability of the antibody with an affinity value of 0.6 is 0.4. Screen the mutated antibody population and the original antibody population, and select antibodies according to the preset affinity threshold to form an evolutionary antibody population.

[0140] Select the antibody with the highest affinity value from the evolved antibody population as the optimal antibody, and the corresponding nutrient density coefficient is the optimal nutrient density coefficient. Multiply the optimal nutrient density coefficient by the pre-set breast milk benchmark feeding amount to obtain the preliminary corrected feeding amount. For example, if the benchmark feeding amount is 200 ml and the optimal nutrient density coefficient is 1.2, then the preliminary corrected feeding amount is 240 ml.

[0141] Finally, use the Adaptive Resonance Theory (ART) network to further correct the feeding amount. Input the preliminary corrected feeding amount and the features to be optimized as input vectors into the comparison layer of the ART network. The recognition layer of the ART network contains multiple neurons, and each neuron represents a feeding pattern. The comparison layer compares the input vector with the weight vectors of each neuron in the recognition layer and calculates the matching degree.

[0142] If the matching degree exceeds the pre-set vigilance parameter, classify the input vector into the existing category and update the feature template of the corresponding category; if the matching degree does not exceed the vigilance parameter, create a new recognition neuron as a new category. Construct a lateral inhibition mechanism between the neurons in the recognition layer to ensure that only one neuron wins each time. Update the weight vector of the winning neuron, and the learning rate decreases with the number of training rounds. Finally, calculate the actual breast feeding amount based on the weight vector and feature template of the winning neuron.

[0143] In Figure 3 shows the comparison of the affinity convergence performance of the three algorithms during the feeding amount optimization process. From the iterative process, the initial affinity values of the three methods are all 0.45, but significant performance differences emerge in the subsequent iterations. This method rapidly increases to 0.68 in the first 5 iterations, while the traditional immune algorithm and genetic algorithm only reach 0.52 and 0.56 respectively. By the 15th iteration, the affinity of this method significantly increases to 0.89. In contrast, the traditional immune algorithm and genetic algorithm are 0.63 and 0.72 respectively, fully demonstrating the faster convergence speed of this method. In the later stage of iteration (20 - 30 times), this method continues to maintain its advantage and finally reaches a high affinity value of 0.96 at the 30th iteration, while the traditional immune algorithm and genetic algorithm only reach 0.72 and 0.83 respectively. It is particularly noteworthy that the increase in affinity of this method remains stable in the 25 - 30 iteration interval (from 0.95 to 0.96), indicating that the algorithm has good detailed optimization ability, while the comparison methods almost stagnate in this stage.

[0144] In Figure 4Among them, the prediction error sample distributions of the three methods in different feeding amount intervals are shown in the form of a bar chart. In the interval of 150 - 160 ml, the number of error samples of this method is 12, significantly lower than 28 of the traditional immune algorithm and 22 of the genetic algorithm. As the feeding amount increases, the prediction error of this method shows a significant downward trend, reaching the lowest value of 2 in the interval of 190 - 200 ml, while the traditional immune algorithm and the genetic algorithm still have 15 and 10 error samples in this interval. In the interval above 200 ml, the number of error samples of the three methods slightly rebounds, but this method always remains at a low level (3 - 4), while the traditional immune algorithm and the genetic algorithm rise to 17 - 19 and 13 - 16 respectively. Overall, the number of prediction error samples of this method in each feeding amount interval is significantly lower than that of the comparison methods. Especially in the common feeding amount interval of 170 - 200 ml, the advantage of prediction accuracy is more obvious, fully demonstrating the reliability of this method in practical applications.

[0145] In this embodiment, the individual characteristics and dynamic change trends of the infant are comprehensively considered, and the breast milk feeding amount can be calculated according to the actual needs of the infant to achieve personalized feeding and avoid overfeeding or underfeeding; it can be adjusted according to the growth and development of the infant and the dynamic change trends such as appetite and sleep to ensure that the feeding amount always meets the needs of the infant; based on the immune evolutionary algorithm and the adaptive resonance theory, it can calculate the breast milk feeding amount more scientifically and accurately, providing a guarantee for the healthy growth of the infant.

[0146] In an optional implementation manner, the growth and development data after the infant ingests the actual breast milk feeding amount are collected, and the development feature vectors are extracted by inputting them into a multi-layer spiral pulse neural network; the development deviation value is calculated by using the fractal dynamics model to map with the standard growth and development parameters, including:

[0147] The growth and development data after the infant ingests the actual breast milk feeding amount are collected, including physical development data and neurodevelopment data;

[0148] A multi-layer spiral pulse neural network is constructed to convert the growth and development data into a pulse sequence, and the pulse frequency is proportional to the numerical size; an ion channel model is constructed by the hidden layer neurons, and the pulse sequence is received through the synaptic weights and integrated into a postsynaptic potential. When the postsynaptic potential exceeds the preset potential threshold, an action potential is triggered to form an action potential sequence;

[0149] Based on the action potential sequence, a spiral synaptic transmission mechanism is constructed, including a short-term regulation module and a long-term regulation module; the short-term regulation module calculates the short-term intensity regulation factor based on the amplitude, and the long-term regulation module calculates the long-term intensity regulation factor based on the frequency and duration, and combines to form a total regulation coefficient;

[0150] Construct a recursive loop based on the total regulation coefficient to achieve the reciprocating transmission of interlayer information, dynamically adjust the synaptic weights to form a weight matrix, and combine the firing patterns of the output layer neurons and the weight matrix to construct a developmental feature vector;

[0151] Construct a fractal dynamics model, calculate the mapping trajectory of the developmental feature vector in the fractal phase space through an iterative function to obtain the actual developmental trajectory; construct a reference fractal trajectory with the same fractal dimension distribution and iterative mapping rule based on the standard growth and development parameters; calculate the Euclidean distance between the actual developmental trajectory and the reference fractal trajectory in the fractal phase space to obtain the developmental deviation value.

[0152] In a specific implementation, first, collect the growth and development data of the infant. These data include physical development data, such as height, weight, head circumference, etc., and neurodevelopment data, such as gross motor skills, fine motor skills, cognitive ability, language ability, and social adaptability. For example, record that the weight of an infant at 6 months is 8 kg, the height is 68 cm, the head circumference is 44 cm, can roll over, can grasp toys, and can make simple syllables. The breast milk intake also needs to be recorded. For example, this infant intakes 800 ml of breast milk per day.

[0153] Next, construct a multi-layer spiral spiking neural network. The input layer neurons convert the collected growth and development data into a pulse sequence. The pulse frequency of the pulse sequence is proportional to the numerical value of the growth and development data. For example, the pulse frequency corresponding to a weight of 8 kg is 8 pulses per second, and the pulse frequency corresponding to a height of 68 cm is 68 pulses per second.

[0154] In the hidden layer neurons, construct an ion channel model. The hidden layer neurons receive the pulse sequence from the input layer through synaptic weights and integrate it into a postsynaptic potential. When the postsynaptic potential exceeds a preset potential threshold, for example, the set threshold is -55 mV, if the postsynaptic potential reaches -50 mV, an action potential is triggered to form an action potential sequence.

[0155] Construct a spiral synaptic transmission mechanism based on the action potential sequence. This mechanism includes a short-term regulation module and a long-term regulation module. The short-term regulation module calculates the short-term intensity regulation factor according to the amplitude of the action potential sequence. For example, if the amplitude of the action potential is 100 mV, the short-term intensity regulation factor is 1.2. The long-term regulation module calculates the long-term intensity regulation factor according to the frequency and duration of the action potential sequence. For example, if the frequency of the action potential is 50 pulses per second and the duration is 100 milliseconds, the long-term intensity regulation factor is 1.5. Multiply the short-term intensity regulation factor and the long-term intensity regulation factor to obtain the total regulation coefficient. For example, 1.2×1.5 = 1.8.

[0156] In a multi-layer spiral spiking neural network, a recursive loop is constructed based on the total regulation coefficient to enable information to be reciprocally transmitted between different layers and dynamically adjust the synaptic weights to form a weight matrix. For example, the initial synaptic weight is 0.5, and after being adjusted by the recursive loop and the total regulation coefficient, the synaptic weight becomes 0.9. The firing patterns of the output layer neurons are combined with the weight matrix to construct a developmental feature vector. For example, if the firing pattern of the output layer neurons is [1, 0, 1, 1] and the weight matrix is [[0.1, 0.2], [0.3, 0.4]], then the developmental feature vector is [0.1, 0.2, 0.3, 0.4, 1, 0, 1, 1].

[0157] Then, a fractal dynamics model is constructed. The developmental feature vector is input into the fractal dynamics model, and an iterative function is used to calculate the mapping trajectory of the developmental feature vector in the fractal phase space to obtain the actual developmental trajectory. The actual developmental trajectory includes the fractal dimension and the fractal scaling exponent. For example, the calculated fractal dimension is 1.5 and the fractal scaling exponent is 0.8.

[0158] A reference fractal trajectory is constructed based on standard growth and development parameters. The reference fractal trajectory has the same fractal dimension distribution structure and iterative mapping rule as the actual developmental trajectory. For example, a reference fractal trajectory is constructed according to the child growth standards of the World Health Organization, and its fractal dimension is also 1.5. The Euclidean distance between the actual developmental trajectory and the reference fractal trajectory in the fractal phase space is calculated to obtain the developmental deviation value. For example, if the calculated Euclidean distance is 0.2, then the developmental deviation value is 0.2.

[0159] In this embodiment, using a multi-layer spiral spiking neural network to extract the developmental feature vector can more comprehensively reflect the growth and development status of infants, thereby improving the accuracy of developmental assessment; evaluating based on the actual developmental data of individual infants avoids the averaging process of traditional methods and realizes personalized developmental assessment; using a computer for data processing and analysis has a high degree of automation and can significantly improve the efficiency of developmental assessment.

[0160] In an alternative embodiment, the swarm intelligence optimization system constructed according to the developmental deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to achieve dynamic optimization of the breastfeeding volume, including:

[0161] Construct an optimization objective function by weighted summation of the physical development deviation value and the neurodevelopmental deviation value. The weight coefficients of the optimization objective function are allocated based on the importance of the developmental indicators to obtain the objective function value;

[0162] Convert the scale factor of the multi-scale morphological neural network, the parameters of the morphological operator, the network weights and network bias values of the deep hybrid cognitive network into an optimization variable sequence, construct a parameter search space, and initialize the optimization variable sequence based on the objective function value;

[0163] Construct a dual-mode mutation operator, which includes a Gaussian mutation module and a Cauchy mutation module; the Gaussian mutation module performs local parameter search according to the first mutation step size to obtain a local search sequence, and the Cauchy mutation module performs global parameter exploration according to the second mutation step size to obtain a global search sequence, and combines the local search sequence and the global search sequence to form a mutation parameter sequence;

[0164] Input the mutation parameter sequence into the multi-scale morphological neural network and the deep hybrid cognitive network. The multi-scale morphological neural network outputs a feature extraction result, and the deep hybrid cognitive network outputs a development prediction value based on the feature extraction result; substitute the development prediction value into the optimization objective function to calculate the current objective function value, compare the current objective function value with the historical objective function value, and store them in the elite solution set in descending order;

[0165] When the number of iterations reaches the preset value and the difference between adjacent objective function values is less than the first preset threshold, re-execute the dual-mode mutation operator by increasing the mutation step size according to the escape index;

[0166] Select the parameter sequence with the optimal objective function value from the elite solution set as the optimal parameter sequence, drive the multi-scale morphological neural network and the deep hybrid cognitive network, extract the breast milk feature sequence and predict the final development prediction value; calculate the evaluation deviation value between the final development prediction value and the actual development data, and when it is greater than the second preset threshold, return to execute the dual-mode mutation operator, and when it is less than the second preset threshold, output the optimal breast milk feeding amount parameter.

[0167] In a specific implementation manner, first, determine the deviation values of the physical development and neurodevelopment of the infant. For example, the physical development deviation value can be obtained by measuring indicators such as the weight, height, and head circumference of the infant and comparing them with the standard growth curve; the neurodevelopment deviation value can be obtained by evaluating the cognitive, language, motor, etc. abilities of the infant and comparing them with the average level of infants of the same age. Suppose the physical development deviation value of a certain infant is 0.2 and the neurodevelopment deviation value is 0.1.

[0168] Next, according to the importance of the development indicators, perform a weighted sum of the physical development deviation value and the neurodevelopment deviation value to construct an optimization objective function. For example, assume that the importance weight of the physical development indicator is 0.6 and the importance weight of the neurodevelopment indicator is 0.4. Then the optimization objective function value of this infant is 0.2×0.6 + 0.1×0.4 = 0.16. The goal is to make this function value as close to zero as possible.

[0169] Then, the scale factor of the multi-scale morphological neural network, the parameters of the morphological operator, as well as the network weights and network bias values of the deep hybrid cognitive network are converted into an optimization variable sequence using real-number coding to construct a parameter search space. For example, assume that the value range of the scale factor of the multi-scale morphological neural network is [0.5, 1.5], the value range of the morphological operator parameters is [1, 5], and the value ranges of the network weights and bias values of the deep hybrid cognitive network are both [-1, 1]. Then these parameters can be combined into a real-number vector as the optimization variable sequence. Initialize the optimization variable sequence according to the objective function value. For example, a set of parameters can be randomly generated and the corresponding objective function value can be calculated as the initial value.

[0170] Construct a dual-mode mutation operator, which includes a Gaussian mutation module and a Cauchy mutation module. The Gaussian mutation module performs local parameter search on the optimization variable sequence according to the first mutation step size. For example, assume the first mutation step size is 0.1. Then for each parameter in the optimization variable sequence, a random number that follows a Gaussian distribution with a mean of 0 and a standard deviation of 0.1 is added to its current value to obtain a local search sequence. The Cauchy mutation module performs global parameter exploration on the optimization variable sequence according to the second mutation step size. For example, assume the second mutation step size is 1. Then for each parameter in the optimization variable sequence, a random number that follows a Cauchy distribution with a scale parameter of 1 is added to its current value to obtain a global search sequence. Combine the local search sequence and the global search sequence to form a mutation parameter sequence.

[0171] Input the mutation parameter sequence into the multi-scale morphological neural network and the deep hybrid cognitive network respectively. The multi-scale morphological neural network extracts features from breast milk samples. For example, it extracts information such as the content and proportion of various nutrients in breast milk. The deep hybrid cognitive network outputs the predicted value of infant development based on the extracted features. For example, it predicts indicators such as the weight, height, and cognitive ability of the infant in the future for a period of time.

[0172] Substitute the predicted value of development into the optimization objective function to calculate the current objective function value and compare it with the historical objective function value. Arrange the corresponding mutation parameter sequence in descending order of the objective function value and store it in the elite solution set.

[0173] Set an iteration counter. When the count value of the iteration counter reaches the preset number of iterations and the difference between two adjacent objective function values is less than the first preset threshold, trigger an escape operation. The escape operation increases the first mutation step size and the second mutation step size according to the preset escape exponent, and re-executes the dual-mode mutation operator on the parameter sequences in the elite solution set. For example, assume the escape exponent is 2, then multiply the first mutation step size and the second mutation step size by 2 respectively.

[0174] Select the parameter sequence with the optimal objective function value from the elite solution set as the optimal parameter sequence. Drive the multi-scale morphological neural network based on the optimal parameter sequence to extract the breast milk feature sequence, and input this sequence into the deep hybrid cognitive network to obtain the final development prediction value.

[0175] Calculate the evaluation deviation value between the final development prediction value and the actual development data. When the evaluation deviation value is greater than the second preset threshold, substitute the evaluation deviation value back into the optimization objective function to update the objective function value and return to execute the dual-mode mutation operator. When the evaluation deviation value is less than the second preset threshold, output the optimal breast milk feeding amount parameters based on the optimal parameter sequence. For example, the optimal intake of breast milk, feeding frequency, proportion of nutritional components, etc.

[0176] In Figure 5 shows the performance comparison of this method with the PSO algorithm and the genetic algorithm in two dimensions: feature extraction accuracy and prediction accuracy. From the data distribution, the performance of this method is significantly better than the other two algorithms, and its data points are concentrated in the high-performance area in the upper right corner of the graph.

[0177] Specifically, the feature extraction accuracy of this method is stably maintained between 94.8% and 95.3%, and the fluctuation range is only 0.5 percentage points, indicating that the method has good stability. At the same time, its prediction accuracy reaches a high level of 97.5% - 98.2%, fully proving the effectiveness of the method. In 5 repeated experiments, the best performance point of this method is (95.3%, 98.2%), and the lowest performance point is (94.8%, 97.5%). Even the lowest performance point is significantly better than the best performance of the comparison method.

[0178] In contrast, the performance of the PSO algorithm is relatively low. The feature extraction accuracy is distributed between 87.5% and 88.2%, and the prediction accuracy is between 91.8% and 92.3%. Its best performance point is (88.2%, 92.3%), and the performance gap with this method exceeds 7 percentage points. The performance of the genetic algorithm is the worst. The feature extraction accuracy is only 85.2% - 85.7%, and the prediction accuracy is between 90.1% and 90.5%. The gap between its best performance point (85.7%, 90.5%) and this method is further expanded to about 9 percentage points.

[0179] From the distribution density of the data points, the data points of the three methods are relatively concentrated, indicating that their respective performances are relatively stable, but there are obvious differences in the distribution areas. The data points of this method are distributed in the optimal performance area and are more compactly distributed, further proving the reliability and stability of the method. These data fully illustrate that the method proposed in this paper has significant advantages in two key indicators: feature extraction and prediction accuracy.

[0180] In this embodiment, according to the specific development situation of each infant, a personalized breastfeeding plan is provided to avoid the "one-size-fits-all" feeding method and more pertinently promote the healthy growth of the infant; it can be dynamically adjusted according to the real-time development data of the infant, continuously optimize the feeding plan, and ensure that the feeding plan always conforms to the actual needs of the infant; by optimizing the breastfeeding parameters through intelligent algorithms, it can help parents breastfeed more scientifically and efficiently, saving time and energy.

[0181] Figure 6 It is a schematic structural diagram of the intelligent calculation and dynamic adjustment system for breastfeeding volume in the embodiment of the present invention, as Figure 6 shown, the system includes:

[0182] The first unit is used to collect the real-time physiological parameters of the infant to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to construct an infant digital model, perform feature dimensionality reduction and clustering through a self-organizing competitive mapping network to obtain a physiological feature set; calculate the parameter entropy value of the physiological feature set to construct an energy metabolism feature vector, input it into a dynamic fuzzy inference system, combine with a biological rhythm prediction model to calculate the basic energy consumption value, and calculate the benchmark breastfeeding volume through an adaptive fuzzy neural network;

[0183] The second unit is used to collect the physiological state data and breast milk component data of the mother, input them into a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain time-series features through a chaotic memory mapping; input the feature vectors and time-series features into an immune evolutionary self-organizing network to calculate the breast milk nutrition density coefficient, correct the benchmark breastfeeding volume according to the breast milk nutrition density coefficient, and calculate the actual breastfeeding volume through an adaptive resonance theory network;

[0184] The third unit is used to collect the growth and development data of the infant after ingesting the actual breastfeeding volume, input it into a multi-layer spiral pulse neural network to extract development feature vectors; use a fractal dynamics model to map with standard growth and development parameters to calculate the development deviation value; the biotic swarm intelligence optimization system constructed according to the development deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on a mutation algorithm to realize the dynamic optimization of the breastfeeding volume.

[0185] In the third aspect of the embodiment of the present invention,

[0186] An electronic device is provided, including:

[0187] A processor;

[0188] A memory for storing instructions executable by the processor;

[0189] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0190] In a fourth aspect of the embodiments of the present invention,

[0191] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the foregoing method is implemented.

[0192] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present invention.

[0193] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent calculation and dynamic adjustment of breast milk feeding amount, characterized in that: include: The real-time physiological parameters of infants are collected to form a dynamic data stream, which is then input into a multi-scale morphological neural network to construct a digital model of the infant. The physiological feature set is obtained by performing feature dimension reduction and clustering through a self-organizing competitive mapping network. Calculate the parameter entropy value of the physiological characteristic set to construct the energy metabolism feature vector, input it into the dynamic fuzzy inference system, calculate the basic energy consumption value in combination with the biological rhythm prediction model, and calculate the breast milk baseline feeding amount through the adaptive fuzzy neural network; The mother's physiological status data and breast milk composition data are collected and input into a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism to extract feature vectors, and the temporal features are obtained through chaotic memory mapping. Inputting the characteristic vector and the time series feature into the immune evolution self-organizing network to calculate the breast milk nutritional density coefficient, correcting the breast milk baseline feeding amount according to the breast milk nutritional density coefficient, and calculating the actual breast milk feeding amount through the adaptive resonance theory network; The growth and development data of infants after they ingest the actual amount of breast milk were collected and input into a multi-layer spiral pulse neural network to extract the developmental feature vector. The fractal dynamics model was mapped with standard growth and development parameters to calculate the developmental deviation value. The bioswarm intelligent optimization system constructed according to the developmental deviation value adjusted the multi-scale morphological neural network and deep hybrid cognitive network parameters based on the mutation algorithm to achieve dynamic optimization of the breastfeeding amount.

2. The method according to claim 1, characterized in that The real-time physiological parameters of infants are collected to form a dynamic data stream, which is then input into a multi-scale morphological neural network to construct an infant digital model. The physiological feature set is obtained by performing feature dimension reduction and clustering through a self-organizing competitive mapping network, including: Collecting physiological parameters of the infant, including body temperature parameters, heart rate parameters, respiratory rate parameters, blood oxygen saturation parameters and weight parameters; classifying the physiological parameters according to the change characteristics of the physiological parameters to determine fast-changing parameters and slow-changing parameters; determining the sampling frequency of the fast-changing parameters according to the ratio of the standard deviation of the fast-changing parameters to the mean, and determining the sampling frequency of the slow-changing parameters according to the ratio of the change amount of the slow-changing parameters to the time interval; sampling the physiological parameters at the sampling frequency to obtain raw data, and performing noise reduction processing on the raw data using wavelet threshold filtering to obtain a dynamic data stream; Input the dynamic data stream into a multi-scale morphological neural network, determine a scale factor according to the ratio of the characteristic period of the dynamic data stream to the basic sampling interval, and use the scale factor to perform multi-scale decomposition on the dynamic data stream to obtain multi-scale data; calculate the variance of the multi-scale data in a local window, determine the size of the structural element of the morphological operator according to the variance, construct a dilation operator and an erosion operator using the size of the structural element, process the multi-scale data through the dilation operator and the erosion operator, and construct a digital model of the baby; The initial number of nodes of the self-organizing competitive mapping network is determined according to the number of samples and the input dimension of the infant digital model, and the self-organizing competitive mapping network adopts an adaptive growth structure; the infant digital model is input into the self-organizing competitive mapping network, the Euclidean distance and the Manhattan distance between samples are calculated, an improved distance metric is constructed according to the Euclidean distance and the Manhattan distance, and the local response of the infant digital model is normalized using the improved distance metric to obtain normalized features; a timing constraint function is constructed, the timing constraint value of the normalized feature is calculated according to the timing constraint function, and the timing constraint value is fused with the normalized feature to obtain a physiological feature set.

3. The method according to claim 1, characterized in that Calculate the parameter entropy value of the physiological feature set to construct the energy metabolism feature vector, input it into the dynamic fuzzy inference system, calculate the basic energy consumption value in combination with the biorhythm prediction model, and calculate the breast milk baseline feeding amount through the adaptive fuzzy neural network, including: The feature sequence in the physiological feature set is segmented, and the similarity threshold is determined according to the standard deviation of the feature sequence; the number of matching patterns in adjacent dimensions is counted to construct a conditional probability matrix; the single feature entropy value is obtained by local sensitive hash mapping; the mutual information between different features in the physiological feature set is calculated to obtain an interactive entropy matrix; the single feature entropy value is fused with the interactive entropy matrix to obtain an energy metabolism feature vector; A fuzzy rule base of a dynamic fuzzy inference system is constructed, wherein the rule weights of the fuzzy rule base include historical data weights, current state weights, and future trend weights; an energy metabolism feature vector is input into the dynamic fuzzy inference system, and a biorhythm prediction model is constructed to predict the energy metabolism pattern in each time window to obtain a time series energy feature; a second-order filter is used to process the nonlinear time-varying features of the time series energy feature, and the basic energy consumption value is calculated in combination with the output results of the dynamic fuzzy inference system; A five-layer adaptive fuzzy neural network including an input layer, a fuzzy layer, a rule layer, a normalization layer and an output layer is constructed; the input layer receives the basic energy consumption value, and dynamically adjusts the weight coefficient according to the input time and the importance of the basic energy consumption value to obtain a weighted feature; the fuzzy layer uses a bell-shaped membership function to fuzzify the weighted feature to obtain a fuzzy feature, and optimizes the parameters of the bell-shaped membership function by a gradient descent method; the rule layer receives the fuzzy feature and performs rule reasoning, and triggers rule splitting when the description accuracy of the existing rule for the fuzzy feature is lower than the rule threshold, and triggers rule merging when the number of rules exceeds the redundancy threshold to obtain a rule reasoning result; the normalization layer uses a soft maximization function to normalize the rule reasoning result to obtain a normalized feature; the output layer uses a hybrid learning algorithm to process the normalized feature, uses the least squares method to estimate the consequent parameter in the forward propagation process, and uses the gradient descent method to optimize the antecedent parameter in the back propagation process, dynamically adjusts the learning step size according to the network output error, and finally outputs the breast milk benchmark feeding amount.

4. The method according to claim 1, characterized in that: The mother's physiological state data and breast milk composition data are collected and input into a deep hybrid cognitive network with a dynamic morphological topology adaptive mechanism to extract feature vectors. The temporal features obtained through chaotic memory mapping include: Collect the mother's physiological status data and breast milk composition data and input them into the deep hybrid cognitive network; A dynamic morphological topology adaptive mechanism is constructed. The dynamic morphological topology adaptive mechanism is based on a deep hybrid cognitive network. A morphological gradient operator is used to perform edge enhancement to obtain key point data. A data window is determined by the key point data. An adaptive threshold is calculated based on a weighted sum of variance and mean in the data window to screen out feature points. A dynamic connection matrix is ​​constructed based on the feature points to determine the topological relationship between features. An annealing strategy is used to adjust the connection strength of the dynamic connection matrix. The temperature parameter of the annealing strategy decreases with the training rounds. The size of the structural element of the morphological operator is determined based on the local distribution of the feature points. The feature points are processed to obtain an initial feature vector. The initial feature vector is transformed through chaotic mapping to obtain nonlinear features; a multi-layer memory unit is constructed according to the preset total number of memory units, and each layer of memory units calculates the correlation between the current moment and the historical state through the attention unit to obtain the attention weight; the forget gate unit controls the retention degree of historical information based on the time interval and state similarity to obtain the memory feature; the memory feature is forward propagated to extract the causal relationship feature and the long-term dependency feature is extracted by backward propagation, and the chaotic dynamics of the forward propagation and the backward propagation are both calculated using piecewise linear approximation; residual connections are constructed between two adjacent layers of the multi-layer memory unit to transmit gradient information; the causal relationship feature and the long-term dependency feature are fused to obtain the time series feature.

5. The method according to claim 1, characterized in that: Inputting the characteristic vector and the time series feature into the immune evolution self-organizing network to calculate the breast milk nutritional density coefficient, correcting the breast milk baseline feeding amount according to the breast milk nutritional density coefficient, and calculating the actual breast milk feeding amount through the adaptive resonance theory network includes: The feature vector and the time series feature are combined to form the feature to be optimized, which is used as the antigen in the immune evolution self-organizing network to initialize and construct an antibody group, where each antibody corresponds to a set of nutrient density coefficients, which are represented by real number coding; An affinity function is constructed based on the nutritional component index and dynamic change trend in the feature to be optimized, the degree of matching between the antibody and the antigen is calculated to obtain an affinity value, and the antibody group is cloned and amplified according to the affinity value to obtain a cloned antibody group, and the number of clones of each antibody is proportional to the corresponding affinity value; a Gaussian mutation operation is performed on the cloned antibody group to obtain a mutant antibody group, and the mutation probability is inversely proportional to the corresponding affinity value; and an evolved antibody group is screened according to a preset affinity threshold value; The optimal antibody with the highest affinity is selected to determine the optimal nutrient density coefficient, which is then multiplied by the baseline breast milk feeding amount to obtain the initial corrected feeding amount; An adaptive resonance theory network including a comparison layer and a recognition layer is constructed, wherein the comparison layer receives the preliminary correction of the feeding amount, the features to be optimized, and the feedback signal of the recognition layer; the matching degree is calculated based on an adaptive threshold, and when the matching degree exceeds a preset warning parameter, it is classified into an existing category and the feature template is updated, otherwise a new category is created; a lateral inhibition mechanism is constructed between the neurons in the recognition layer for competitive learning to ensure a single winning neuron, and the weight vector of the winning neuron is updated, and the learning rate of the weight vector decreases with each training round; the actual breastfeeding amount is calculated based on the weight vector of the winning neuron and the feature template.

6. The method according to claim 1, characterized in that The growth and development data of infants after they have taken in the actual amount of breast milk are collected and input into a multi-layer spiral pulse neural network to extract the developmental feature vector; Using the fractal dynamics model and standard growth and development parameter mapping, the development deviation values ​​are calculated including: Collect growth and development data of infants after they have consumed the actual amount of breast milk, including physical development data and neurodevelopmental data; A multi-layer spiral pulse neural network is constructed to convert growth and development data into a pulse sequence, where the pulse frequency is proportional to the value; an ion channel model is constructed through hidden layer neurons, and the pulse sequence is received through synaptic weights and integrated into a postsynaptic potential. When the postsynaptic potential exceeds a preset potential threshold, an action potential is triggered to form an action potential sequence; The spiral synaptic transmission mechanism is constructed based on the action potential sequence, which includes a short-term regulation module and a long-term regulation module; the short-term regulation module calculates the short-term intensity regulation factor based on the amplitude, and the long-term regulation module calculates the long-term intensity regulation factor based on the frequency and duration, and the combination forms the total regulation coefficient; Based on the total regulation coefficient, a recursive loop is constructed to realize the reciprocal transmission of information between layers, the synaptic weights are dynamically adjusted to form a weight matrix, and the discharge pattern of neurons in the output layer is combined with the weight matrix to construct a developmental feature vector. A fractal dynamics model is constructed, and the mapping trajectory of the developmental characteristic vector in the fractal phase space is calculated through an iterative function to obtain the actual developmental trajectory; a reference fractal trajectory with the same fractal dimension distribution and iterative mapping rules is constructed based on standard growth and development parameters; and the Euclidean distance between the actual developmental trajectory and the reference fractal trajectory in the fractal phase space is calculated to obtain a developmental deviation value.

7. The method according to claim 1, characterized in that The bioswarm intelligent optimization system constructed according to the developmental deviation value adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to achieve dynamic optimization of breastfeeding amount, including: The physical development deviation value and the neural development deviation value are weighted and summed to construct an optimization objective function, wherein the weight coefficient of the optimization objective function is allocated based on the importance of the development index to obtain the objective function value; Convert the scale factor of the multi-scale morphological neural network, the morphological operator parameters, and the network weights and network bias values ​​of the deep hybrid cognitive network into an optimization variable sequence, construct a parameter search space, and initialize the optimization variable sequence based on the objective function value; A dual-mode mutation operator is constructed, including a Gaussian mutation module and a Cauchy mutation module; the Gaussian mutation module performs local parameter search according to the first variable step length to obtain a local search sequence, and the Cauchy mutation module performs global parameter exploration according to the second variable step length to obtain a global search sequence, and the local search sequence and the global search sequence are combined to form a mutation parameter sequence; Inputting the mutation parameter sequence into the multi-scale morphological neural network and the deep hybrid cognitive network, the multi-scale morphological neural network outputs a feature extraction result, and the deep hybrid cognitive network outputs a development prediction value based on the feature extraction result; substituting the development prediction value into the optimization objective function to calculate the current objective function value, comparing the current objective function value with the historical objective function value, and storing them in an elite solution set in descending order; When the number of iterations reaches a preset value and the difference between adjacent objective function values ​​is less than a first preset threshold, the dual-mode mutation operator is re-executed with a variable step length according to the escape index; A parameter sequence with the best objective function value is selected from the elite solution set as the optimal parameter sequence, and the multi-scale morphological neural network and the deep hybrid cognitive network are driven to extract the breast milk feature sequence and predict the final development prediction value; an evaluation deviation value between the final development prediction value and the actual development data is calculated, and when it is greater than a second preset threshold, a dual-mode mutation operator is returned to execute, and when it is less than the second preset threshold, the optimal breast milk feeding amount parameter is output.

8. A system for intelligent calculation and dynamic adjustment of breast milk feeding amount, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect real-time physiological parameters of infants to form a dynamic data stream, input the data stream into a multi-scale morphological neural network to build an infant digital model, and obtain a physiological feature set by performing feature dimension reduction and clustering through a self-organizing competitive mapping network; Calculate the parameter entropy value of the physiological characteristic set to construct the energy metabolism feature vector, input it into the dynamic fuzzy inference system, calculate the basic energy consumption value in combination with the biological rhythm prediction model, and calculate the breast milk baseline feeding amount through the adaptive fuzzy neural network; The second unit is used to collect the mother's physiological state data and breast milk composition data, input the deep hybrid cognitive network with dynamic morphological topology adaptive mechanism to extract feature vectors, and obtain the time series characteristics through chaotic memory mapping; Inputting the characteristic vector and the time series feature into the immune evolution self-organizing network to calculate the breast milk nutritional density coefficient, correcting the breast milk baseline feeding amount according to the breast milk nutritional density coefficient, and calculating the actual breast milk feeding amount through the adaptive resonance theory network; The third unit is used to collect growth and development data of infants after they have consumed the actual amount of breast milk, and input it into a multi-layer spiral pulse neural network to extract developmental feature vectors; use the fractal dynamics model to map with standard growth and development parameters to calculate the developmental deviation value; and construct a bioswarm intelligent optimization system based on the developmental deviation value, which adjusts the parameters of the multi-scale morphological neural network and the deep hybrid cognitive network based on the mutation algorithm to achieve dynamic optimization of the amount of breast milk fed.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images