An intelligent solution recommendation system and method based on electromyographic stress and deep learning

Abstract

The present invention discloses an intelligent solution recommendation system and method based on electromyopressure and deep learning, which relates to the field of deep learning technology. The intelligent solution recommendation system synchronously collects pelvic floor electromyography signals and pressure signals through electromyopressure integrated electrodes, extracts evaluation data of various stages such as the pre-resting stage and the fast muscle stage, constructs feature vectors in combination with the user's personal information, and after standardization, uses an encoder to map the features into high-dimensional features, which are input into a decoder through sine-cosine position encoding, and the decoder generates a solution recommendation matrix. In addition, during model training, the model is optimized through a cross-entropy loss function. The present invention improves the efficiency of pelvic floor training through multi-data fusion and the generation of intelligent solutions.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to an intelligent solution recommendation system and method based on electromyographic stress and deep learning. Background Art

[0002] The pelvic floor muscles are a group of muscles located at the base of the pelvis, shaped like a "hanging net." They are primarily responsible for supporting the pelvic organs, controlling excretion, and contributing to core stability. The health of the pelvic floor muscles directly impacts quality of life. If this "net" loses its elasticity and lacks the "hanging force," the organs within it can no longer maintain their normal position. Therefore, pelvic floor muscle exercises are necessary to address pelvic floor dysfunction.

[0003] Existing pelvic floor muscle training programs are mainly formulated by doctors based on the user's pelvic floor muscle conditions. This method of formulating programs based on pelvic floor muscle conditions has the following problems: First, data resources are not fully utilized, such as the user's personal information data and pelvic floor muscle assessment report data; Second, there is no continuous tracking and feedback on the pelvic floor muscle conditions after completing the training according to the proposed program, and the proposed program cannot be evaluated; Third, the level of doctors is uneven, and they make judgments based on their own work experience, which may result in the proposed program not fully meeting the user's own needs, missing the best training time, and thus leading to inefficient pelvic floor muscle training. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent solution recommendation system and method based on electromyographic stress and deep learning to solve the problems raised in the prior art.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: an intelligent solution recommendation system based on electromyographic stress and deep learning, the recommendation system comprising: a data acquisition module, a data processing module, a position encoding module, an encoder module, and a decoder module;

[0006] The data acquisition module is used to collect the user's pelvic floor electromyography and pressure signals, extract evaluation data and collect the user's personal information;

[0007] A data processing module, configured to standardize the evaluation data and personal information, construct feature vectors, and unify the preset component information into a matrix format;

[0008] A position encoding module is used to encode the position of each element in the target matrix and generate a position encoding matrix;

[0009] An encoder module, configured to map the feature vector into a high-dimensional feature vector and expand it into a feature matrix;

[0010] The decoder module is used to fuse the features in the feature matrix with the position code to generate a solution recommendation matrix.

[0011] Furthermore, the data acquisition module includes:

[0012] The pelvic floor electromyography and pressure signals are collected synchronously through the electromyography and pressure integrated electrodes; the evaluation data of each stage are extracted from the collected signals using the mathematical formula of average value and standard deviation;

[0013] The evaluation data for each stage include:

[0014] The static pelvic floor muscle tension assessment data in the pre-resting phase included the pre-resting mean, variability, and comprehensive score;

[0015] The dynamic fast-twitch strength and reaction speed assessment data during the fast-twitch phase include fast-twitch maximum value, fast-twitch contraction time, fast-twitch relaxation time, fast-twitch electromyography score, and fast-twitch pressure score;

[0016] The dynamic slow muscle strength and endurance contraction control stability evaluation data during the slow muscle phase include the slow muscle electromyography average, slow muscle pressure average, slow muscle variability, slow muscle rise time, slow muscle recovery time, slow muscle electromyography score, and slow muscle pressure score;

[0017] The stability evaluation data of muscle endurance contraction control during the endurance phase include the maximum endurance electromyography value, the maximum endurance pressure value, the endurance rise time, the endurance recovery time, the endurance electromyography score, and the endurance pressure score;

[0018] The data of static pelvic floor muscle tension assessment in the post-resting phase include the post-resting mean, variability, and comprehensive score;

[0019] The personal information includes BMI, age at delivery, date of delivery, mode of delivery, number of fetuses, and baby weight.

[0020] Furthermore, the data processing module includes:

[0021] The evaluation data and personal information are used as input feature values, and the input feature values ​​are normalized using the Z-score normalization formula. The Z-score normalization formula is:

[0022] ;

[0023] Among them, x represents the input feature value, x norm represents the input feature matrix after normalization, represents the standard deviation, represents the average value;

[0024] Arranging the eigenvalues ​​corresponding to the evaluation data in order of stages to obtain a first target vector, arranging the eigenvalues ​​corresponding to the personal information in order to obtain a second target vector, and concatenating the first target vector and the second target vector to obtain a eigenvector;

[0025] In the above steps, because the units of evaluation data (such as electromyography values, pressure values, scores, etc.) are completely different from those of personal information (BMI, age of delivery, etc.), the numerical differences are large, and it is necessary to eliminate the dimensional influence, ensure numerical stability, and avoid extreme values ​​that lead to serious model fitting deviations; the above steps also preprocess the data and discard a small amount of missing data in the evaluation report to avoid individual inaccurate evaluation data affecting the overall effect of the model.

[0026] Each preset component is taken as an element, and the elements are arranged according to a preset number of rows and columns. When the number of elements in each row is not equal to the number of columns, zero is added to obtain a target matrix that meets the preset number of rows and columns;

[0027] In the above steps, in order to improve the efficiency of model training, the program information is composed of components. Different components have different training effects on the parts of the pelvic floor muscles, such as type I stimulation 1, type I stimulation 2, muscle contraction and relaxation training 1, muscle contraction and relaxation training 2, etc.

[0028] Furthermore, the position encoding module includes:

[0029] The sine function is used to positionally encode the positions of the even-numbered column elements in the target matrix. The formula is as follows:

[0030] ;

[0031] Among them, PE (pos,2i) Represents the element in the posth row and 2ith column of the position encoding matrix, d model is the hidden layer dimension of the Transformer model;

[0032] The cosine function is used to positionally encode the positions of the odd-numbered column elements in the target matrix. The formula is as follows:

[0033] ;

[0034] Among them, PE (pos,2i+1) Represents the position code of the element in the posth row and 2i+1th column in the position code matrix;

[0035] In the above steps, when generating the solution recommendation matrix, the model needs to understand the order and hierarchical relationship of components in different positions in the solution (such as the semantic difference between the 1st row and the 15th row, and the 3rd column and the 5th column). In conventional sequence processing models, the original attention mechanism cannot perceive the relative position information of elements in the sequence, so the position information of the elements is added here; the present invention adopts a position encoding scheme combining sine and cosine functions. By integrating row and column position information into a single code, the model can distinguish the semantic differences between different rows and columns in the matrix.

[0036] Furthermore, the encoder module includes:

[0037] The feature vector is input into the linear transformation layer for linear transformation as follows:

[0038] ;

[0039] Among them, H enc represents the eigenvector after linear transformation, W e represents the weight matrix of the linear transformation, b e represents the bias vector of the linear transformation;

[0040] The linearly transformed features are processed by a nonlinear activation function to obtain the expanded features, which are recorded as ;in, represents the expanded features, represents the eigenvector after linear transformation, n represents the number of rows of the eigenvector, and m represents the number of columns of the eigenvector; the nonlinear transformation method is as follows:

[0041] ;

[0042] Among them, when the input feature x≥0, the output is f(x)=x; when the input feature x<0, the output is f(x)=0;

[0043] In the above steps, low-dimensional features are converted into high-dimensional features. The purpose of this dimensionality increase is to enable the model to capture the complex relationship between features; the above nonlinear activation function will screen and reorganize the features, highlighting important information and suppressing irrelevant information; when the input is a negative number, the output is "truncated" to 0, suppressing invalid signals; when the input is a positive number, the output is directly equal to the input, retaining valid signals; the features that have undergone nonlinear transformation will be standardized again to ensure that the distribution of features is more stable, which helps to improve the stability and convergence speed of model training.

[0044] Furthermore, the decoder module includes: self-attention layer, cross-attention layer, feedforward network layer, component prediction layer, and probability matrix reshaping layer;

[0045] The self-attention layer generates a corresponding position index for each position in the position encoding matrix, and each position index is converted into a vector through the embedding layer and recorded as ; Superimpose the position code corresponding to each element in the position code matrix with the position index to generate ; Among them, PE represents the position code corresponding to each element, Indicates the current generated position;

[0046] In the above steps, the generated , making the model aware of position order;

[0047] The cross-attention layer fuses the features in the feature matrix with the position encoding through the cross-attention mechanism. The formula is:

[0048] ;

[0049] Among them, Q represents the input of the current layer of the decoder , K and V both represent the characteristics of the encoder module after expansion , Indicates the degree of attention of the current position to all positions;

[0050] In the above steps, the decoder uses a cross-attention mechanism to allow each target position p to "query" the encoder memory, that is, the encoder generated ; Among them, Q (Query) means "what information is needed at the current position", K (Key) and V (Value) mean "the information stored in the encoder memory";

[0051] The feedforward network layer increases the dimension of the output result of the attention mechanism through nonlinear transformation, and reduces the dimension of the increased output result through linear transformation to obtain a feature matrix after removing redundant information;

[0052] The component prediction layer maps the output of the feedforward network to the component candidate space through nonlinear transformation, and generates a probability distribution through Softmax. The formula is as follows:

[0053] ;

[0054] in, represents the probability of selecting k components at the jth position of the i-th user, and N represents the number of components in the candidate space;

[0055] The probability matrix reshaping layer divides the fused feature matrix into n rows and m columns, transforming it into an n×m matrix. In the component list, the component with the largest predicted probability is selected to fill the feature matrix to obtain the solution recommendation matrix.

[0056] Furthermore, in order to better implement the above system, an intelligent solution recommendation method based on electromyographic pressure and deep learning is also provided. The recommendation method includes:

[0057] Step S1: The user's pelvic floor electromyographic and pressure signals are collected through the electromyographic and pressure integrated electrodes, and the evaluation data of each stage is extracted through the mean, standard deviation and variability, and the user's personal information is collected;

[0058] Step S2: Clean the collected data, standardize the cleaned assessment data and personal information, construct a feature vector, and organize the preset component information into a matrix format to obtain a target matrix;

[0059] Step S3: performing sine coding on the even-numbered column elements and cosine coding on the odd-numbered column elements in the target matrix to obtain a position coding matrix;

[0060] Step S4: performing linear transformation and nonlinear transformation on the feature vector in sequence to obtain an expanded feature vector;

[0061] Step S5: The expanded feature vector and the position coding matrix are fused through a decoder to obtain a solution recommendation matrix.

[0062] Furthermore, step S1 includes:

[0063] The formula for calculating the average value is:

[0064] ;

[0065] in, represents the average value, x i represents the value of the i-th sample collected, and N represents the number of samples collected in each stage;

[0066] The formula for calculating the standard deviation is:

[0067] ;

[0068] in, represents the standard deviation;

[0069] The formula for calculating variability is:

[0070] ;

[0071] Here, CV stands for variability.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] 1. The electromyography and pressure integrated electrodes synchronously collect multi-dimensional signals to improve the comprehensiveness and accuracy of pelvic floor assessment.

[0074] 2. Combine the user's personal information with deep learning algorithms to generate personalized pelvic floor muscle training plans to improve training efficiency and effectiveness.

[0075] 3. The position encoding and attention mechanism are used to ensure the rationality of the structure and order of the recommendation scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 This is a schematic diagram of the system structure of an intelligent solution recommendation system and method based on electromyographic stress and deep learning of the present invention;

[0077] Figure 2 This is a flow chart of a method for an intelligent solution recommendation system and method based on electromyographic stress and deep learning according to the present invention;

[0078] Figure 3 This is a pelvic floor electromyography report evaluation diagram of an intelligent recommendation system and method based on electromyography pressure and deep learning in the present invention. DETAILED DESCRIPTION

[0079] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0080] Example 1: Figure 1 As shown, the present invention provides a technical solution, an intelligent solution recommendation system based on electromyographic stress and deep learning, the system comprising: a data acquisition module, a data processing module, a position encoding module, an encoder module, and a decoder module;

[0081] The data acquisition module is used to collect the user's pelvic floor electromyography and pressure signals, extract evaluation data and collect the user's personal information;

[0082] A data processing module, configured to standardize the evaluation data and personal information, construct feature vectors, and unify the preset component information into a matrix format;

[0083] A position encoding module is used to encode the position of each element in the target matrix and generate a position encoding matrix;

[0084] An encoder module, configured to map the feature vector into a high-dimensional feature vector and expand it into a feature matrix;

[0085] The decoder module is used to fuse the features in the feature matrix with the position encoding to generate a solution recommendation matrix;

[0086] The data acquisition module includes:

[0087] The pelvic floor electromyography and pressure signals are collected synchronously through the electromyography and pressure integrated electrodes; the evaluation data of each stage are extracted from the collected signals using the mathematical formula of average value and standard deviation;

[0088] In an embodiment of the present invention, pelvic floor assessment is performed using an integrated myoelectric and pressure pelvic floor electrode, with a sampling rate of 128 times per second for a total of 6 minutes and 17 seconds, collecting a total of 128×377=48256 myoelectric signals and 48256 pressure signals simultaneously.

[0089] See Figure 3 , the evaluation data for each stage include:

[0090] The static pelvic floor muscle tension assessment data in the pre-resting phase included the pre-resting mean, variability, and comprehensive score;

[0091] The dynamic fast-twitch strength and reaction speed assessment data during the fast-twitch phase include fast-twitch maximum value, fast-twitch contraction time, fast-twitch relaxation time, fast-twitch electromyography score, and fast-twitch pressure score;

[0092] The dynamic slow muscle strength and endurance contraction control stability evaluation data during the slow muscle phase include the slow muscle electromyography average, slow muscle pressure average, slow muscle variability, slow muscle rise time, slow muscle recovery time, slow muscle electromyography score, and slow muscle pressure score;

[0093] The stability evaluation data of muscle endurance contraction control during the endurance phase include the maximum endurance electromyography value, the maximum endurance pressure value, the endurance rise time, the endurance recovery time, the endurance electromyography score, and the endurance pressure score;

[0094] The data of static pelvic floor muscle tension assessment in the post-resting phase include the post-resting mean, variability, and comprehensive score;

[0095] Personal information included BMI, age at delivery, date of delivery, mode of delivery, number of fetuses gestated, and infant weight;

[0096] In an embodiment of the present invention, a user's personal information is collected through a form for comprehensive assessment of pelvic floor muscle condition, including BMI, age at delivery, date of delivery, mode of delivery, number of fetuses, and baby weight, a total of 6 items;

[0097] The data processing module includes:

[0098] The evaluation data and personal information are used as input feature values, and the input feature values ​​are standardized using the Z-score standardization formula. The Z-score standardization formula is:

[0099] ;

[0100] Among them, x represents the input feature value, x normrepresents the input feature matrix after normalization, represents the standard deviation, represents the average value;

[0101] Arrange the eigenvalues ​​corresponding to the evaluation data in order of stages to obtain a first target vector, arrange the eigenvalues ​​corresponding to the personal information in order to obtain a second target vector, and concatenate the first target vector and the second target vector to obtain a eigenvector;

[0102] Each preset component is taken as an element, and the elements are arranged according to a preset number of rows and columns. When the number of elements in each row is not equal to the number of columns, zero is added to obtain a target matrix that meets the preset number of rows and columns;

[0103] In the embodiment of the present invention, to ensure the effectiveness of the program, a total of 300 components were developed. Each program has 15 sections, and each section consists of 1 to 5 components, that is, a 15×5 matrix of information. To ensure the optimal treatment effect, the training data collected by the present invention are all excellent training programs ranked in the top 20% in terms of training effect. The training effect evaluation method is: the top 20% with the highest improvement in the comprehensive score of the pelvic floor assessment report before and after the completion of the training program. To ensure the consistency of model input content, when there are less than 5 components, they are padded with 0 to ensure that the output is a 15×5 matrix.

[0104] Among them, the position encoding module includes:

[0105] The sine function is used to positionally encode the positions of the even-numbered column elements in the target matrix. The formula is as follows:

[0106] ;

[0107] Among them, PE (pos,2i) Represents the element in the posth row and 2ith column of the position encoding matrix, d model The hidden layer dimension of the Transformer model is designed to be 128 dimensions in the present invention;

[0108] The cosine function is used to positionally encode the positions of the odd-numbered column elements in the target matrix. The formula is as follows:

[0109] ;

[0110] Among them, PE (pos,2i+1) Represents the position code of the element in the posth row and 2i+1th column in the position code matrix;

[0111] The encoder module includes:

[0112] The feature vector is input into the linear transformation layer for linear transformation as follows:

[0113] ;

[0114] Among them, H enc represents the eigenvector after linear transformation, W e represents the weight matrix of the linear transformation, b e represents the bias vector of the linear transformation;

[0115] The linearly transformed features are processed by a nonlinear activation function to obtain the expanded features, which are recorded as ;in, represents the expanded features, represents the eigenvector after linear transformation, n represents the number of rows of the eigenvector, and m represents the number of columns of the eigenvector; the nonlinear transformation method is as follows:

[0116] ;

[0117] Among them, when the input feature x≥0, the output is f(x)=x; when the input feature x<0, the output is f(x)=0;

[0118] The decoder module includes: self-attention layer, cross-attention layer, feedforward network layer, component prediction layer, and probability matrix reshaping layer;

[0119] The self-attention layer generates a corresponding position index for each position in the position encoding matrix, and converts each position index into a vector through the embedding layer. ; Superimpose the position code corresponding to each element in the position code matrix with the position index to generate ; Among them, PE represents the position code corresponding to each element, Indicates the current generated position;

[0120] The cross-attention layer fuses the features in the feature matrix with the position encoding through the cross-attention mechanism. The formula is:

[0121] ;

[0122] Among them, Q represents the input of the current layer of the decoder , K and V both represent the characteristics of the encoder module after expansion , Indicates the degree of attention of the current position to all positions;

[0123] The feedforward network layer increases the dimension of the output result of the attention mechanism through nonlinear transformation, and reduces the dimension of the increased output result through linear transformation to obtain a feature matrix after removing redundant information;

[0124] The component prediction layer maps the output of the feedforward network to the component candidate space through nonlinear transformation, and generates a probability distribution through Softmax. The formula is as follows:

[0125] ;

[0126] in, represents the probability of selecting component k for the jth position of the i-th user;

[0127] In an embodiment of the present invention, a cross entropy loss function is added to the model during the training phase, and its formula is:

[0128] ;

[0129] in, Represents the loss function value, which measures the difference between the model prediction and the true value, by minimizing Optimize model parameters; N represents the batch size, that is, the number of samples contained in a training batch; 75 represents the sequence length of each sample, corresponding to the 75 positions after the 15×5 matrix is ​​expanded; Represents a double sum operation, where the outer layer traverses each sample n in the batch, and the inner layer traverses the 75 positions p of each sample; represents the true component label of the p-th position of the n-th sample (the component id with a value of (1~300) or 0 for padding); X (n) Represents the input features of the nth sample (i.e., 30-dimensional user feature vector); Indicates that the model is input X (n) When , the nth sample pth position is predicted to be the true label probability; Express the natural logarithm of the above probabilities, converting the probabilities to a logarithmic scale for ease of calculation and optimization; It means averaging the log-probability of all samples and positions and adding a negative sign, which turns the loss function into a "negative log-likelihood". The probability of the model predicting the true label is maximized by minimizing the loss.

[0130] The above cross entropy loss function calculates the loss by calculating the average negative value of the logarithmic probability of the model predicting the true label, driving the model to adjust parameters to increase the probability of predicting the true label, thereby optimizing the model performance;

[0131] The probability matrix reshaping layer divides the fused feature matrix into n rows and m columns, transforming it into an n×m matrix. In the component list, the component with the largest predicted probability is selected to fill the feature matrix, thus obtaining the solution recommendation matrix.

[0132] Example 2: Figure 2As shown, in order to better implement the above system, a solution intelligent recommendation method based on electromyographic pressure and deep learning is also provided, and the recommendation method includes:

[0133] Step S1: The user's pelvic floor electromyographic and pressure signals are collected through the electromyographic and pressure integrated electrodes, and the evaluation data of each stage is extracted through the mean, standard deviation and variability, and the user's personal information is collected;

[0134] Step S2: Clean the collected data, standardize the cleaned assessment data and personal information, construct a feature vector, and organize the preset component information into a matrix format to obtain a target matrix;

[0135] Step S3: performing sine coding on the even-numbered column elements and cosine coding on the odd-numbered column elements in the target matrix to obtain a position coding matrix;

[0136] Step S4: performing linear transformation and nonlinear transformation on the feature vector in sequence to obtain an expanded feature vector;

[0137] Step S5: The expanded feature vector and the position coding matrix are fused through the decoder to obtain a solution recommendation matrix;

[0138] Wherein, step S1 includes:

[0139] The formula for calculating the average value is:

[0140] ;

[0141] in, represents the average value, x i represents the value of the i-th sample collected, and N represents the number of samples collected in each stage;

[0142] The formula for calculating the standard deviation is:

[0143] ;

[0144] in, represents the standard deviation;

[0145] The formula for calculating variability is:

[0146] ;

[0147] Here, CV stands for variability.

[0148] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An intelligent solution recommendation system based on electromyographic stress and deep learning, characterized by: The recommendation system includes: a data acquisition module, a data processing module, a position encoding module, an encoder module, and a decoder module; The data acquisition module is used to collect the user's pelvic floor electromyography signals and pressure signals, extract evaluation data and collect the user's personal information; The data processing module is used to standardize the evaluation data and personal information, construct feature vectors and unify the preset component information into a matrix format; The position encoding module is used to encode the position of each element in the target matrix to generate a position encoding matrix; The encoder module is used to map the feature vector into a high-dimensional feature vector and expand it into a feature matrix; The decoder module is used to fuse the features in the feature matrix with the position code to generate a solution recommendation matrix; The position encoding module includes: The sine function is used to positionally encode the positions of the even-numbered column elements in the target matrix. The formula is as follows: ； Among them, PE (pos,2i) Represents the element in the posth row and 2ith column of the position encoding matrix, d model is the hidden layer dimension of the Transformer model; The cosine function is used to positionally encode the positions of the odd-numbered column elements in the target matrix. The formula is as follows: ； Among them, PE (pos,2i+1) Represents the position code of the element in the posth row and 2i+1th column in the position code matrix; The encoder module comprises: The feature vector is input into the linear transformation layer for linear transformation as follows: ； Among them, H enc represents the eigenvector after linear transformation, W e represents the weight matrix of the linear transformation, b e represents the bias vector of the linear transformation; The linearly transformed features are processed by a nonlinear activation function to obtain the expanded features, which are recorded as ;in, represents the expanded features, represents the eigenvector after linear transformation, n represents the number of rows of the eigenvector, and m represents the number of columns of the eigenvector; the nonlinear transformation method is as follows: ； Among them, when the input feature x≥0, the output is f(x)=x; when the input feature x<0, the output is f(x)=0; The decoder module includes: self-attention layer, cross-attention layer, feedforward network layer, component prediction layer, and probability matrix reshaping layer; The self-attention layer generates a corresponding position index for each position in the position encoding matrix, and each position index is converted into a vector by the embedding layer. ; Superimpose the position code corresponding to each element in the position code matrix with the position index to generate ; Among them, PE represents the position code corresponding to each element, Indicates the current generated position; The cross-attention layer fuses the features in the feature matrix with the position encoding through the cross-attention mechanism. The formula is: ； Among them, Q represents the input of the current layer of the decoder , K and V both represent the characteristics of the encoder module after expansion , Indicates the degree of attention of the current position to all positions; The feedforward network layer increases the dimension of the output result of the attention mechanism through nonlinear transformation, and reduces the dimension of the output result after the increase in dimension through linear transformation to obtain a feature matrix after removing redundant information; The component prediction layer maps the output of the feedforward network to the component candidate space through nonlinear transformation, and generates a probability distribution through Softmax. The formula is as follows: ； in, represents the probability of selecting k components at the jth position of the i-th user, and N represents the number of components in the candidate space; The probability matrix reshaping layer divides the fused feature matrix into n rows and m columns, transforming it into an n×m matrix, and selects the component with the largest predicted probability from the component list to fill the feature matrix to obtain a solution recommendation matrix.

2. The intelligent solution recommendation system based on electromyographic stress and deep learning according to claim 1, characterized in that: The data acquisition module includes: The pelvic floor electromyography and pressure signals are collected synchronously through the electromyography and pressure integrated electrodes; the evaluation data of each stage are extracted from the collected signals using the mathematical formula of average value and standard deviation; The evaluation data for each stage include: The static pelvic floor muscle tension assessment data in the pre-resting phase included the pre-resting mean, variability, and comprehensive score; The dynamic fast-twitch strength and reaction speed assessment data during the fast-twitch phase include fast-twitch maximum value, fast-twitch contraction time, fast-twitch relaxation time, fast-twitch electromyography score, and fast-twitch pressure score; The dynamic slow muscle strength and endurance contraction control stability evaluation data during the slow muscle phase include the slow muscle electromyography average, slow muscle pressure average, slow muscle variability, slow muscle rise time, slow muscle recovery time, slow muscle electromyography score, and slow muscle pressure score; The stability evaluation data of muscle endurance contraction control during the endurance phase include the maximum endurance electromyography value, the maximum endurance pressure value, the endurance rise time, the endurance recovery time, the endurance electromyography score, and the endurance pressure score; The data of static pelvic floor muscle tension assessment in the post-resting phase include the post-resting mean, variability, and comprehensive score; The personal information includes BMI, age at delivery, date of delivery, mode of delivery, number of fetuses, and baby weight.

3. The intelligent solution recommendation system based on electromyographic stress and deep learning according to claim 1, characterized in that: The data processing module includes: The evaluation data and personal information are used as input feature values, and the input feature values ​​are normalized using the Z-score normalization formula. The Z-score normalization formula is: ； Among them, x represents the input feature value, x norm represents the input feature matrix after normalization, represents the standard deviation, represents the average value; Arranging the eigenvalues ​​corresponding to the evaluation data in order of stages to obtain a first target vector, arranging the eigenvalues ​​corresponding to the personal information in order to obtain a second target vector, and concatenating the first target vector and the second target vector to obtain a eigenvector; Each preset component is taken as an element, and the elements are arranged according to the preset number of rows and columns. When the number of elements in each row is not equal to the number of columns, 0 is added to obtain a target matrix that meets the preset number of rows and columns.

4. A method for intelligent solution recommendation based on electromyographic stress and deep learning, for implementing the intelligent solution recommendation system based on electromyographic stress and deep learning according to any one of claims 1 to 3, characterized in that: The recommended methods include: Step S1: The user's pelvic floor electromyographic and pressure signals are collected through the electromyographic and pressure integrated electrodes, and the evaluation data of each stage is extracted through the mean, standard deviation and variability, and the user's personal information is collected; Step S2: Clean the collected data, standardize the cleaned assessment data and personal information, construct a feature vector, and organize the preset component information into a matrix format to obtain a target matrix; Step S3: performing sine coding on the even-numbered column elements and cosine coding on the odd-numbered column elements in the target matrix to obtain a position coding matrix; Step S4: performing linear transformation and nonlinear transformation on the feature vector in sequence to obtain an expanded feature vector; Step S5: The expanded feature vector and the position coding matrix are fused through a decoder to obtain a solution recommendation matrix.

5. The method for intelligent solution recommendation based on electromyographic pressure and deep learning according to claim 4, characterized in that: Step S1 includes: The calculation formula of the average value is: ； in, represents the mean value, x i represents the value of the i-th sample collected, and N represents the number of samples collected in each stage; The standard deviation is calculated as follows: ； in, represents the standard deviation; The calculation formula of the variability is: ； Here, CV stands for variability.

Images