Traditional Chinese medicinal material efficacy evaluation method and system based on multi-source element data fusion

Abstract

The invention provides a traditional Chinese medicinal material efficacy evaluation method and system based on multi-source element data fusion. The method comprises the following steps: collecting and preprocessing multi-source element data of traditional Chinese medicinal materials; based on a distributed federated learning model, aggregating the preprocessed multi-source element data of the traditional Chinese medicinal materials to obtain aggregated data; performing data expansion on the aggregated data by adopting a generative adversarial network algorithm based on a quantum coherent enhancement learning strategy to obtain an expanded sample; performing feature extraction and dimension reduction on the expanded sample to obtain dimension reduction features; and performing drug effect classification evaluation on the dimension reduction features to obtain a drug effect evaluation result of the traditional Chinese medicinal materials. The method can accurately evaluate the efficacy of the traditional Chinese medicinal materials while guaranteeing the data safety, and provides a scientific basis for quality control and efficacy prediction of the traditional Chinese medicinal materials.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a traditional Chinese medicine efficacy evaluation method and system based on multi-source factor data fusion. Background Art

[0002] In the quality control and efficacy evaluation of traditional Chinese medicines, the multi-source and complexity of data have always been difficult problems to be solved urgently. The traditional centralized model training method usually relies on a single data source and cannot make full use of the data resources scattered on different nodes, resulting in serious problems of isolated data resources. At the same time, data transmission and sharing under the centralized training mode face challenges of data leakage and privacy protection and cannot meet the requirements of modern data security. In addition, the efficacy evaluation task of traditional Chinese medicines involves a large amount of training data, but the actually collected data is often insufficient, which directly affects the generalization ability and evaluation accuracy of the model.

[0003] The Chinese invention patent with the application number CN202310999925.0 proposes a drug efficacy evaluation method, device, equipment and medium based on a neural network. The method includes: obtaining first body surface physiological data generated by experimental animals that have taken a candidate drug and second body surface physiological data generated by experimental humans that have not taken the candidate drug; respectively inputting the first body surface physiological data and the second body surface physiological data into a preset first neural network and a second neural network, and sequentially outputting a first output feature and a second output feature; inputting the first output feature and the second output feature into a preset loss function based on probability distribution metric to obtain a loss value; based on the loss value, updating the network parameters of the second neural network; wherein, the updated second neural network is used for clinically evaluating the candidate drug. The above technical solution can reduce the clinical evaluation process of the candidate drug.

[0004] The Chinese invention patent with the application number CN202210824838.7 proposes a lead compound screening method based on a graph convolutional neural network, including: obtaining the simplified molecular linear input specification and molecular descriptor features of multiple compounds; constructing a molecular structure diagram model of the compounds, and using the graph convolutional neural network to obtain the structural features of the compounds; merging the structural features and the molecular descriptor features to obtain merged features; sequentially randomly modifying the numerical value of one dimension of the molecular descriptor features in the merged features to obtain corresponding multiple new merged features; respectively inputting the new merged features into a multi-layer perceptron neural network to obtain predicted pharmacodynamic properties; calculating the error between the predicted pharmacodynamic properties and the actual pharmacodynamic properties, and sorting the importance of different dimensions of the molecular descriptor features according to the magnitude of the error value; screening the lead compounds to be screened according to the sorting result. The present invention can quickly screen out compounds with better specific pharmacodynamic properties from a large number of compounds and reduce the drug R & D cost.

[0005] The Chinese invention patent with the application number CN202111085841.3 proposes a drug use prediction system based on a neural network, which mainly includes: a total recovery time prediction module that uses a pre-trained classification model based on basic information to predict the total recovery time of a patient; a remaining time prediction module that uses a pre-trained time series model based on basic information and tracking data to predict the remaining time required for a patient to reach the recovered state; a time setting and judgment module that judges whether the sum of the used drug time Δt and the remaining time T is close to the total recovery time of the patient. When the judgment is no, a pharmacodynamic survival analysis module obtains a pharmacodynamic curve with drug use cessation as the end event through a COX regression model, and then optimizes the pharmacodynamic curve. A time series model fine-tuning module fine-tunes the time series model based on the optimized pharmacodynamic curve to obtain a personalized time series model; a drug use recommendation module uses the personalized time series model to predict a drug use prediction result as a reference for doctors' drug use adjustment.

[0006] The prior art has the following deficiencies:

[0007] 1. In the prior art, in the centralized model training mode, since the data of each node cannot be shared, it leads to the isolation of data resources, unable to make full use of multi-source data, and restricts the generalization ability of the model.

[0008] 2. In terms of data augmentation in the prior art, it often relies on traditional incremental data generation techniques, and the generated data lacks diversity and authenticity, unable to effectively make up for the problem of insufficient training samples.

[0009] 3. Feature extraction and dimensionality reduction techniques mostly use fixed paths or simple feature selection algorithms, and fail to dynamically adjust the transmission path of features according to the feature importance of data, resulting in the loss of feature information and low computational efficiency of the model.

[0010] 4. In pharmacodynamic evaluation, the use of classifiers lacks a method for reasonably classifying the data after feature dimensionality reduction, resulting in limited accuracy of the evaluation results. Summary of the Invention

[0011] In order to solve the problems existing in the prior art, the present invention provides a method and system for evaluating the pharmacodynamics of traditional Chinese medicines based on the fusion of multi-source element data, which realizes the effective aggregation of multi-source data on the premise of ensuring data security, and improves the sample quality of model training and the generalization ability of the model.

[0012] To achieve the above object, the present invention provides the following solutions:

[0013] A method for evaluating the pharmacodynamics of traditional Chinese medicines based on the fusion of multi-source element data, the method includes:

[0014] Collect and preprocess multi-source element data of traditional Chinese medicines;

[0015] Based on the distributed federated learning model, aggregate the preprocessed multi-source element data of the Chinese medicinal materials to obtain aggregated data;

[0016] Use the generative adversarial network algorithm based on the quantum coherence enhanced learning strategy to perform data augmentation on the aggregated data to obtain augmented samples;

[0017] Extract features and reduce the dimension of the augmented samples to obtain dimension-reduced features;

[0018] Perform pharmacodynamic classification evaluation on the dimension-reduced features to obtain the pharmacodynamic evaluation results of the Chinese medicinal materials.

[0019] Preferably, the multi-source element data of the Chinese medicinal materials includes sensor monitoring data, laboratory analysis result data, and historical pharmacodynamic data sets;

[0020] The attributes of the multi-source element data of the Chinese medicinal materials include the wet weight of the medicinal materials, the weight after drying, the total alkaloid content, the moisture content, the proportion of nutritional components, the color of the extract, the pH value of the extract, the solubility test results, the analysis of volatile components, and the heavy metal content.

[0021] Preferably, the distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; the process of obtaining the aggregated data includes:

[0022] Allocate the preprocessed multi-source element data of the Chinese medicinal materials to several nodes to train several local models to obtain model parameters;

[0023] Aggregate and update the model parameters based on the federated model to be evaluated to obtain aggregated and updated model parameters;

[0024] Redistribute the aggregated and updated model parameters to each node for iterative training until the federated model to be evaluated meets the preset performance requirements to obtain the final federated model;

[0025] Based on the final federated model, obtain the aggregated data.

[0026] Preferably, the generative adversarial network algorithm structure based on the quantum coherence enhanced learning strategy includes a generator and a discriminator; among them, the generator is used to generate augmented samples; the discriminator is used to evaluate the augmented samples; the method for training the generative adversarial network includes:

[0027] Initialize the weights, biases, and quantum state parameters of the generative adversarial network, decompose the aggregated data to obtain quantum state enhanced features;

[0028] Concatenate the noise vector of the aggregated data with the quantum state enhanced features, and generate augmented samples through the generator;

[0029] Through the discriminator, discriminate the enhanced quantum state features and the augmented samples to obtain a discrimination result;

[0030] Based on the discrimination result and the quantum state parameters, calculate the quantum state information entropy loss;

[0031] Based on the quantum state information loss entropy, adjust the quantum state parameters until a preset condition is met, and complete the training of the generative adversarial network.

[0032] Preferably, use an autoencoder based on importance scores to extract and reduce the dimensions of the augmented samples to obtain reduced-dimensional features; the autoencoder includes a symmetric encoder and decoder; wherein, an automatic routing mechanism based on feature importance scoring is used to automatically determine the transfer paths of each feature in different layers of the encoder.

[0033] The present invention also provides a traditional Chinese medicine efficacy evaluation system based on multi-source factor data fusion for implementing the method, including:

[0034] A data acquisition module for collecting and preprocessing multi-source factor data of traditional Chinese medicine;

[0035] A distributed federated learning module for aggregating the preprocessed multi-source factor data of traditional Chinese medicine based on a distributed federated learning model to obtain aggregated data;

[0036] A data augmentation module for augmenting the aggregated data by using a generative adversarial network algorithm based on a quantum coherence enhanced learning strategy to obtain augmented samples;

[0037] A feature dimension reduction module for extracting and reducing the dimensions of the augmented samples to obtain reduced-dimensional features;

[0038] An efficacy evaluation module for classifying and evaluating the efficacy of the reduced-dimensional features to obtain a traditional Chinese medicine efficacy evaluation result.

[0039] Preferably, the distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; the distributed federated learning module includes:

[0040] A model parameter acquisition unit for distributing the preprocessed multi-source factor data of traditional Chinese medicine to several nodes to train several local models to obtain model parameters;

[0041] A parameter aggregation unit for aggregating and updating the model parameters based on the federated model to be evaluated to obtain aggregated and updated model parameters;

[0042] An iterative training unit for re - distributing the aggregated and updated model parameters to each node for iterative training until the to - be - evaluated federated model meets the preset performance requirements, obtaining the final federated model;

[0043] An aggregated data acquisition unit for obtaining the aggregated data based on the final federated model.

[0044] Preferably, the generative adversarial network algorithm structure based on the quantum coherence enhanced learning strategy includes a generator and a discriminator; wherein, the generator is used to generate augmented samples; the discriminator is used to evaluate the augmented samples; the data augmentation module includes:

[0045] An initialization unit for initializing the weights, biases, and quantum state parameters of the generative adversarial network, decomposing the aggregated data to obtain quantum state enhanced features;

[0046] An augmented sample generation unit for concatenating the noise vector of the aggregated data with the quantum state enhanced features and generating augmented samples through the generator;

[0047] An augmented sample evaluation unit for discriminating the quantum state enhanced features and the augmented samples through the discriminator to obtain a discrimination result;

[0048] A loss calculation unit for calculating the quantum state information entropy loss based on the discrimination result and the quantum state parameters;

[0049] A parameter adjustment unit for adjusting the quantum state parameters based on the quantum state information loss entropy until the preset conditions are met, completing the training of the generative adversarial network.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] The present invention adopts a distributed federated learning architecture, and solves the problem of data resource islands existing in traditional centralized model training through the collaborative training of distributed node models. Each node completes data processing and model training locally, and only uploads the model parameters to the central server for parameter aggregation and update, avoiding the risk of data leakage. In order to expand the training data, the present invention uses a generative adversarial network algorithm based on the quantum coherence enhanced learning strategy to generate realistic sample data, improving the problem of insufficient training samples. In addition, during the feature dimensionality reduction process, feature selection and dimensionality reduction are performed through an auto - encoder algorithm based on importance scores, improving the accuracy of feature representation and the computational efficiency of the model. Finally, in the pharmacodynamic evaluation stage, the Softmax classification function is used to classify the dimensionality - reduced features, realizing the accurate classification and evaluation of the pharmacodynamics of traditional Chinese medicines.

[0052] Through the above technical means, the present invention realizes the effective aggregation of multi-source data while ensuring data security, improves the sample quality of model training and the generalization ability of the model. The combination of the distributed federated learning architecture and the quantum-enhanced generative adversarial network algorithm enhances the diversity and authenticity of the data and strengthens the adaptability of the model when facing new data. The feature dimensionality reduction method of the autoencoder effectively reduces redundant features, reduces the complexity of the model, and thus improves the efficiency and accuracy of the efficacy evaluation. The use of the classifier ensures the scientific evaluation of the efficacy of traditional Chinese medicinal materials and helps to effectively control and predict the quality of traditional Chinese medicinal materials in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0054] Figure 1 It is a flowchart of the method for evaluating the efficacy of traditional Chinese medicinal materials based on multi-source element data fusion according to an embodiment of the present invention;

[0055] Figure 2 It is the distributed federated learning training process according to an embodiment of the present invention;

[0056] Figure 3 It is the training process of the feature dimensionality reduction model according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. 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.

[0058] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0059] Embodiment 1

[0060] As Figure 1 shown, the method for evaluating the efficacy of traditional Chinese medicinal materials based on multi-source element data fusion includes:

[0061] S1: Collect and preprocess the multi-source element data of traditional Chinese medicinal materials; further embodiments are that the multi-source element data of traditional Chinese medicinal materials include sensor monitoring data, laboratory analysis result data, and historical efficacy data sets;

[0062] The attributes of the multi-source element data of traditional Chinese medicinal materials include R a1 is the wet weight of the medicinal material, R a2 is the weight after drying, R a3 is the total alkaloid content, R a4 is the moisture content, R a5 is the proportion of nutrient components, R a6 is the color of the extract, R a7 is the pH value of the extract, R a8 is the result of the solubility test, R a9 is the analysis of volatile components, R a10 is the heavy metal content.

[0063] Since the efficacy evaluation task requires a large amount of effective training data for feature extraction and analysis learning, the present invention realizes the distributed model training and data mining of the multi-node joint model by adopting the model training architecture of distributed federated learning, so as to break the problem of data resource islands in the traditional centralized training mode and aggregate and obtain more effective data. At the same time, the distributed federated learning architecture can prevent the leakage of local data of each node. The data is used locally and the model is trained locally, effectively ensuring data security.

[0064] S2: Based on the distributed federated learning model, aggregate the preprocessed multi-source element data of traditional Chinese medicinal materials to obtain aggregated data; in the federated learning framework, each node does not need to share the data of its local training model, but trains the local model and sends the updated model to the centralized learning unit for summary. The model training architecture of distributed federated learning is as Figure 2 shown.

[0065] A further implementation manner is that the distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; in the model training architecture of distributed federated learning, in each round of iteration, each node respectively performs local model training and uploads the trained model parameters to the central server (global model). The central server completes parameter aggregation and update, and sends the updated parameters to each node to start a new round of iteration until the training converges.

[0066] The process of obtaining the aggregated data includes:

[0067] Allocate the preprocessed multi-source element data of traditional Chinese medicinal materials to several nodes to train several local models to obtain model parameters; in this embodiment, the federated average algorithm is used as the model aggregation algorithm, and the federated average algorithm realizes the collaborative training of local models through multiple global iterations. Specifically, for each global iteration, let the number of nodes be N leb , the total number of samples it owns is D leb , and the number of samples of the kth node is The objective function of federated learning is defined as:

[0068]

[0069] In the formula, f leb () is the objective function of federated learning, ω leb is the model parameter, is the loss prediction of the j-th sample data for the model parameter ω leb , L tra () is the training loss, is the sample feature of the j-th sample data, and the label of the j-th sample data. Preferably, the training loss L tra () is calculated using the cross-entropy loss function.

[0070] Furthermore, for the k-th node, the objective function of this node is defined as:

[0071]

[0072] In the formula, is the number of samples of the k-th node, is the objective function of the k-th node, is the data distribution of the k-th node.

[0073] Based on the federated model (global model) to be evaluated, aggregate and update the model parameters to obtain the aggregated and updated model parameters;

[0074] Redistribute the aggregated and updated model parameters to each node for iterative training until the federated model to be evaluated meets the preset performance requirements to obtain the final federated model; In this embodiment, taking the parameter update method of the k-th node at the t-th iteration as an example, let the parameter gradient of the k-th node be Then the way to update the model parameters at the t-th iteration is expressed as:

[0075]

[0076] In the formula, is the model parameter at the t-th iteration, is the model parameter at the (t + 1)-th iteration, R nod is the learning rate of the current parameter update, is the number of samples of the k-th node.

[0077] Furthermore, the parameter update method of the global model of the central server is expressed as:

[0078]

[0079] In the formula, are the parameters of the global model of the central server at the (t + 1)-th iteration.

[0080] Furthermore, repeating the iteration operation indicates that the training of the global model of the central server and the models of each node is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 5000 times.

[0081] Based on the final federated model, aggregated data is obtained. It should be noted that in the distributed federated learning architecture proposed by the present invention, the data of each node is collected from traditional Chinese medicine with multi-source factor data fusion. The data collection sources are extensive, including but not limited to direct sensor monitoring, laboratory chemical analysis results, and historical efficacy data sets. The data is stored in vector form and uses the JSON format to support efficient data reading and processing. This embodiment is only to illustrate a data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds.

[0082] Furthermore, the collected data is labeled. The labeling method of the present invention is manual labeling. In one embodiment, the labeling categories include "very effective", "effective", "average", and "ineffective", a total of 4 categories.

[0083] It can be understood that in the task of the present invention, the acquisition, labeling, and preprocessing of training data are time-consuming and laborious, and insufficient training samples are likely to lead to poor generalization ability of the model and affect the accuracy of the model at the same time. The present invention uses a generative adversarial network algorithm based on a quantum coherence enhanced learning strategy for sample generation, and then realizes data augmentation. The generative adversarial network algorithm based on the quantum coherence enhanced learning strategy includes two parts: a generator G c and a discriminator D c . The generator G c is responsible for generating realistic data samples, and the discriminator D c evaluates the authenticity of the samples. In particular, the input of the generator G c not only includes the noise vector z, but also includes the feature enhancement signal obtained through quantum state correlation matrix decomposition to guide the generator to more accurately reproduce the data distribution.

[0084] S3: Use the generative adversarial network algorithm based on the quantum coherence enhanced learning strategy to perform data augmentation on the aggregated data to obtain augmented samples; a further implementation manner is that the generative adversarial network algorithm structure based on the quantum coherence enhanced learning strategy includes a generator and a discriminator; wherein, the generator is used to generate augmented samples; the discriminator is used to evaluate the augmented samples; the method for training the generative adversarial network includes:

[0085] S31: Initialize the weights, biases, and quantum state parameters of the generative adversarial network, decompose the aggregated data, and obtain quantum state enhanced features; in this embodiment, the quantum state parameters and network weights are initialized, and the initialization method is expressed as:

[0086]

[0087] where, is the initial weight of the generative adversarial network, is the initial bias of the generative adversarial network, is the initial quantum state parameter; represents a normal distribution with a mean of 0 and a covariance of the identity matrix; represents a uniform distribution in the interval [0, 2π].

[0088] Use the quantum state correlation matrix to decompose the input data to obtain an enhanced feature representation X c , providing deep feature information of the original data for the generator. Specifically, for the input data X c Apply the quantum state correlation matrix decomposition to obtain the enhanced feature representation The method is expressed as:

[0089]

[0090] where, X c is the original data matrix; U c , Σ c , are respectively the left singular vector matrix, the singular value diagonal matrix, and the transposed right singular vector matrix obtained by the singular value decomposition of the data, is the singular value diagonal matrix obtained by the singular value decomposition of the data based on quantum characteristics; exp(iθ c ) represents that each element on the diagonal matrix is applied to the iθ c th power of e, and i * is the imaginary unit.

[0091] In one embodiment, the calculation method of the singular value diagonal matrix obtained by the singular value decomposition of the data based on quantum characteristics is expressed as:

[0092]

[0093] where, ⊙ represents the Hadamard product, that is, element-wise multiplication; γ c and δ c are respectively the first adjustment parameter and the second adjustment parameter, used to control the non-linear expansion of the singular value and the quantum phase adjustment; exp() represents the exponential function with the natural number e as the base.

[0094] S32: Concatenate the noise vector of the aggregated data with the quantum state enhanced features, and generate augmented samples through the generator; in this embodiment, the generator G c receives the noise vector z and the quantum state enhanced features X c , and generates new data samples (augmented samples) which is expressed as:

[0095]

[0096] In the formula, G c () is the generator function; represents the concatenation of the noise vector and the enhanced feature vector; tanh() is the hyperbolic tangent activation function, which is used to output the generated data in the range of [-1, 1]; W c is the weight of the generative adversarial network, and b c is the bias of the generative adversarial network.

[0097] Furthermore, during the training process, an adaptive weight update mechanism is adopted to update the weights of the generative adversarial network, which is expressed as:

[0098] W c ←α c W c,prev +(1 - α c )ΔW c

[0099] In the formula, ← is the parameter update operation, α c is the adaptive learning rate, W c,prev is the weight of the previous iteration, and ΔW c is the weight update amount calculated based on the quantum state gradient descent. Preferably, α c is set to decay according to the simulated annealing strategy.

[0100] Furthermore, the calculation method of the weight update amount calculated based on the quantum state gradient descent is expressed as:

[0101]

[0102] where μ c is the momentum parameter, M c is the momentum value of the previous iteration, which is used to introduce historical gradient information and enhance the training stability; η cds is the learning rate of the weight; L c is the quantum state information entropy loss. Preferably, η cds is set to 0.1, and μ c is set to 0.3.

[0103] S33: Through the discriminator, discriminate the quantum state enhanced features and the augmented samples to obtain the discrimination result; the discriminator Dc Discriminate between real samples and generated samples, and output the probability that the sample is true, expressed as:

[0104]

[0105] In the formula, D c () is the discriminator function; Sig() is the Sigmoid activation function, which is used to convert the output into a probability value, and the range is [0,1].

[0106] S34: Based on the discrimination result and the quantum state parameters, calculate the quantum state information entropy loss, and the calculation method is:

[0107] L c = -[y c log(P c )+(1 - y c )log(1 - P c )]+λ c ∥θ c ∥ 2

[0108] In the formula, y c is the real sample label, P c is the probability output by the discriminator; λ c is the regularization coefficient, which is used to control the influence of the quantum state parameters. Preferably, λ c is set to 0.3.

[0109] In one embodiment, the quantum term λ of the loss function c ∥θ c ∥ 2 considers the quantum decoherence and entanglement characteristics, and the calculation method is expressed as:

[0110] ∥θ c ∥ 2 =(θ c -θ ref ) T K c (θ c -θ ref )

[0111] In the formula, θ ref is the reference quantum state, which is used to simulate the quantum coherence in the ideal state; K c is the quantum state adjustment matrix, which is used to adjust the interaction and influence strength between each quantum state parameter.

[0112] S35: Adjust the quantum state parameters based on the quantum state information loss entropy until the preset conditions are met, and complete the training of the generative adversarial network. Specifically, according to the information entropy difference between the generated samples and the real samples, feedback is used to adjust the quantum state parameters to reduce the difference between the generated samples and the real samples. The adjustment method is expressed as:

[0113]

[0114] where η c is the learning rate of the quantum state parameters, and are the gradients of the loss function L c with respect to θ c and W c respectively; β c is the learning step size, which controls the rate of quantum state adjustment; D diff is the discriminator D c outputs the difference measure between the generated data and the real data; σ c is the normalization factor, which is used to adjust the influence range of the difference measure. Preferably, σ c is set to the standard deviation of the current batch of data, β c is set to 0.1, and η c is set to 0.01.

[0115] In one embodiment, a quantum feedback unit is used to dynamically adjust the quantum state parameter θ c based on the difference between the generated data and the real data. Based on the difference between the generated data and the real data, the calculation method of the difference measure between the generated data and the real data output by the discriminator D c is expressed as:

[0116]

[0117] where is the discrimination result of the generated data, and D c (X c ) is the discrimination result of the real data.

[0118] Repeat the above steps iteratively until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0119] S4: Extract and reduce the dimensions of the augmented samples to obtain the reduced-dimensional features; as Figure 3 shown.

[0120] In a further embodiment, an autoencoder based on importance scores is used to extract features and reduce the dimension of the augmented samples, obtaining dimension-reduced features; the autoencoder includes a symmetric encoder and decoder; wherein, a dynamic routing mechanism based on feature importance scoring automatically determines the transmission path of each feature in different layers of the encoder, enhancing the model's ability to reconstruct the original data while reducing information loss.

[0121] S41: Initialize the weights and biases of the autoencoder. In one embodiment, the initialization method is expressed as:

[0122]

[0123] In the formula, is the initial weight of the i-th layer of the autoencoder; is the initial bias of the i-th layer of the autoencoder; n r,i is the number of neurons in the i-th layer of the autoencoder; n r,i-1 is the number of neurons in the (i - 1)-th layer of the autoencoder;

[0124] randn(n r,i , n r,i-1 ) generates a standard normal distribution random number with a shape of n r,i × n r,i-1 ; zeros(n r,i ) generates a zero vector with a length of n r,i .

[0125] S42: The input data is transmitted through each layer of the encoder. Each layer makes a dynamic routing decision based on the output of the previous layer and the feature importance score of the current layer, selectively transmitting the most informative features, expressed as:

[0126] z r,i = W r,i · x r,i-1 + b r,i

[0127] a r,i = Re(z r,i )

[0128] x r,i = route(a r,i , imp(a r,i ))

[0129] In the formula, z r,i is the output of the linear transformation of the i-th layer of the autoencoder; a r,i is the output feature of the activation function of the autoencoder; W r,i is the weight of the i-th layer of the autoencoder; b r,iis the bias of the i-th layer of the autoencoder; Re() is the ReLU activation function; x r,i-1 is the input feature of the i-th layer of the autoencoder; route() represents the dynamic routing function based on feature importance; imp(a r,i ) calculates the importance scoring function of the output features of the activation function of the autoencoder.

[0130] In one embodiment, the dynamic routing function integrates the activation outputs of each feature by weighted average, where the weights are determined by the importance scores of the features, and the calculation method is expressed as:

[0131]

[0132] Furthermore, the calculation method of the importance score of the output features of the activation function of the autoencoder is expressed as:

[0133]

[0134] In the formula, a r,i,j is the activation output of the j-th neuron in the i-th layer, a r,i,k is the activation output of the k-th neuron in the i-th layer; w r,imp,j is the weight corresponding to the feature importance, obtained through training, representing the importance of the j-th feature; w r,imp,k is the weight corresponding to the feature importance, obtained through training, representing the importance of the k-th feature; n r,i is the total number of neurons in the i-th layer.

[0135] S43: In the decoder stage, use the features from different encoding layers to perform data reconstruction to calculate the reconstruction error. Then, during the training process, the calculation method of the loss function of the autoencoder is expressed as:

[0136]

[0137] In the formula, L r is the loss function of the autoencoder; m r is the number of samples in the current batch input; is the j-th sample input to the autoencoder; is the j-th sample after reconstruction.

[0138] S44: Calculate the gradient of the loss function with respect to each parameter of the autoencoder through the backpropagation algorithm, and update the weights and biases of the autoencoder. The update method is expressed as:

[0139]

[0140] In the formula, ΔW r,i is the update increment of the weight of the i-th layer of the autoencoder; Δbr,i The update increment of the bias for the i-th layer of the autoencoder; η r The learning rate of the autoencoder; The weight of the i-th layer of the updated autoencoder; The bias of the i-th layer of the updated autoencoder; S r,i The sparsity measure of the neurons in the i-th layer of the autoencoder; λ r,s The sparsity pruning threshold; The indicator function, which takes 1 when the condition in the parentheses holds and 0 otherwise. Preferably, λ r,s Is set to 0.8, indicating that when the sparsity measure is higher than 80%, the weights will be pruned.

[0141] In one embodiment, the computational efficiency and generalization ability of the model are improved by dynamically pruning redundant weights, that is, the weights corresponding to the neurons with small activation values in most cases will be dynamically pruned, thereby simplifying the model structure. Specifically, the importance of neurons is judged by statistically analyzing the sparsity of activation values. The calculation method of the sparsity measure of the neurons in the i-th layer of the autoencoder is expressed as:

[0142]

[0143] In the formula, S r,i Is the sparsity measure of the neurons in the i-th layer; a r,i,j Is the activation value of the j-th sample in the i-th layer; τ r Is the sparsity threshold. Preferably, τ r Is set to 0.1.

[0144] S45: Repeat the above steps iteratively until the preset stop iteration condition is satisfied, which means the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0145] S5: Perform a pharmacodynamic classification evaluation on the dimensionality-reduced features to obtain the Chinese medicinal material pharmacodynamic evaluation result.

[0146] Furthermore, after the feature dimensionality reduction model training is completed, use the trained feature dimensionality reduction model to perform feature dimensionality reduction on the data, and classify the dimensionality-reduced features through a preset Softmax classification function to obtain the pharmacodynamic evaluation result. In one embodiment, the evaluation results include "very effective", "effective", "general", and "ineffective", a total of 4 categories.

[0147] Through the above steps, the present invention can effectively utilize the trained model to quickly and accurately evaluate the efficacy of new traditional Chinese medicine samples, not only improving the evaluation efficiency, but also ensuring the generalization ability of the model when facing new data, and can be widely applied to the fields of quality control and efficacy prediction of traditional Chinese medicine.

[0148] Example Two

[0149] The present invention also provides a traditional Chinese medicine efficacy evaluation system based on multi-source factor data fusion for implementing the method, including:

[0150] A data acquisition module, configured to acquire and preprocess multi-source factor data of traditional Chinese medicine;

[0151] A distributed federated learning module, configured to aggregate the preprocessed multi-source factor data of traditional Chinese medicine based on a distributed federated learning model to obtain aggregated data;

[0152] A data augmentation module, configured to perform data augmentation on the aggregated data by using a generative adversarial network algorithm based on a quantum coherence enhanced learning strategy to obtain augmented samples;

[0153] A feature dimensionality reduction module, configured to perform feature extraction and dimensionality reduction on the augmented samples to obtain dimensionality-reduced features;

[0154] An efficacy evaluation module, configured to perform efficacy classification evaluation on the dimensionality-reduced features to obtain a traditional Chinese medicine efficacy evaluation result.

[0155] A further implementation manner lies in that the distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; the distributed federated learning module includes:

[0156] A model parameter acquisition unit, configured to distribute the preprocessed multi-source factor data of traditional Chinese medicine to several nodes to train several local models to obtain model parameters;

[0157] A parameter aggregation unit, configured to aggregate and update the model parameters based on the federated model to be evaluated to obtain aggregated and updated model parameters;

[0158] An iterative training unit, configured to redistribute the aggregated and updated model parameters to each node for iterative training until the federated model to be evaluated meets the preset performance requirements to obtain the final federated model;

[0159] An aggregated data acquisition unit, configured to obtain aggregated data based on the final federated model.

[0160] A further implementation manner lies in that the generative adversarial network algorithm structure based on a quantum coherence enhanced learning strategy includes a generator and a discriminator; wherein, the generator is configured to generate augmented samples; the discriminator is configured to evaluate the augmented samples; the data augmentation module includes:

[0161] An initialization unit, configured to initialize the weights, biases of the generative adversarial network, and quantum state parameters, decompose and aggregate data, and obtain quantum state enhanced features;

[0162] An augmented sample generation unit, configured to splice the noise vector of the aggregated data with the quantum state enhanced features, and generate augmented samples through a generator;

[0163] An augmented sample evaluation unit, configured to discriminate the quantum state enhanced features and the augmented samples through a discriminator, and obtain a discrimination result;

[0164] A loss calculation unit, configured to calculate the quantum state information entropy loss based on the discrimination result and the quantum state parameters;

[0165] A parameter adjustment unit, configured to adjust the quantum state parameters based on the quantum state information loss entropy until a preset condition is met, and complete the training of the generative adversarial network.

[0166] In this embodiment, it further includes:

[0167] A model update and management module:

[0168] This module is used to manage and update the models in the system. After each round of training, it decides whether to update the model according to the evaluation results and the set iteration conditions. In addition, this module is also responsible for the version management of the models and the monitoring of the model performance.

[0169] A user interface module:

[0170] This module provides a friendly interaction interface for users. Users can view the training progress of each node, the model evaluation results, the data acquisition status, etc. through the interface. At the same time, this interface also supports users to adjust the system parameters, such as modifying the number of training rounds, setting the learning rate, managing the data sources, etc.

[0171] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for evaluating the efficacy of Chinese medicinal materials based on multi-source factor data fusion, characterized in that: The method comprises: Collect and pre-process multi-source data of Chinese medicinal materials; Based on a distributed federated learning model, aggregating the preprocessed multi-source element data of the Chinese medicinal materials to obtain aggregated data; Using a generative adversarial network algorithm based on quantum coherence reinforcement learning strategy to perform data expansion on the aggregated data to obtain expanded samples; Performing feature extraction and dimensionality reduction on the expanded samples to obtain dimensionality reduction features; The reduced dimension features are subjected to efficacy classification evaluation to obtain efficacy evaluation results of Chinese medicinal materials.

2. The method according to claim 1, characterized in that The multi-source element data of Chinese medicinal materials include sensor monitoring data, laboratory analysis result data and historical efficacy data set; The attributes of the multi-source element data of traditional Chinese medicine include wet weight of medicinal materials, weight after drying, total alkaloid content, moisture content, nutrient ratio, extract color, extract pH value, solubility test results, volatile component analysis and heavy metal content.

3. The method according to claim 1, characterized in that The distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; the process of obtaining the aggregated data includes: Distribute the preprocessed multi-source element data of Chinese medicinal materials to several nodes to train several local models to obtain model parameters; Aggregating and updating the model parameters based on the federated model to be evaluated to obtain aggregated and updated model parameters; The aggregated and updated model parameters are re-sent to each node for iterative training until the federated model to be evaluated meets the preset performance requirements, thereby obtaining the final federated model; Based on the final federated model, the aggregated data is obtained.

4. The method according to claim 1, characterized in that: The algorithm structure of the generative adversarial network based on the quantum coherent reinforcement learning strategy includes a generator and a discriminator; wherein the generator is used to generate an augmented sample; the discriminator is used to evaluate the augmented sample; and the method for training the generative adversarial network includes: Initializing the weights, biases, and quantum state parameters of a generative adversarial network, decomposing the aggregated data, and obtaining quantum state enhancement features; splicing the noise vector of the aggregated data with the quantum state enhancement feature, and generating an augmented sample through the generator; The quantum state enhancement feature and the expanded sample are discriminated by the discriminator to obtain a discrimination result; Based on the discrimination result and the quantum state parameter, calculating the quantum state information entropy loss; The quantum state parameters are adjusted based on the quantum state information loss entropy until preset conditions are met, thereby completing the training of the generative adversarial network.

5. The method according to claim 1, characterized in that An importance score-based autoencoder is used to perform feature extraction and dimensionality reduction on the expanded samples to obtain reduced-dimensional features; the autoencoder includes a symmetrical encoder and decoder; wherein a transmission path of each feature in different layers of the encoder is automatically determined by a dynamic routing mechanism based on feature importance scores.

6. A Chinese herbal medicine efficacy evaluation system based on multi-source factor data fusion, used to implement the method according to any one of claims 1 to 5, characterized in that: include: Data collection module, used to collect and pre-process multi-source element data of Chinese medicinal materials; A distributed federated learning module, used for aggregating the pre-processed multi-source element data of the Chinese medicinal materials based on a distributed federated learning model to obtain aggregated data; A data expansion module, used to perform data expansion on the aggregated data using a generative adversarial network algorithm based on a quantum coherent reinforcement learning strategy to obtain an expanded sample; A feature dimension reduction module is used to extract features and reduce the dimension of the expanded samples to obtain reduced dimension features; The drug efficacy evaluation module is used to classify and evaluate the drug efficacy of the dimensionality reduction features to obtain drug efficacy evaluation results of traditional Chinese medicines.

7. The system according to claim 6, characterized in that The distributed federated learning model includes a local model, a federated model to be evaluated, and a final federated model; the distributed federated learning module includes: A model parameter acquisition unit, used for distributing the pre-processed multi-source element data of traditional Chinese medicine to several nodes to train several local models and obtain model parameters; A parameter aggregation unit, used to aggregate and update the model parameters based on the federated model to be evaluated to obtain the aggregated and updated model parameters; An iterative training unit, used to re-send the aggregated updated model parameters to each node for iterative training until the federated model to be evaluated meets the preset performance requirements, thereby obtaining a final federated model; The aggregate data acquisition unit is used to obtain the aggregate data based on the final federated model.

8. The system according to claim 6, characterized in that The algorithm structure of the generative adversarial network based on quantum coherence reinforcement learning strategy includes a generator and a discriminator; wherein the generator is used to generate an expanded sample; the discriminator is used to evaluate the expanded sample; the data expansion module includes: An initialization unit, used to initialize the weights, biases and quantum state parameters of the generative adversarial network, decompose the aggregated data, and obtain quantum state enhancement features; An extended sample generating unit, used for splicing the noise vector of the aggregated data with the quantum state enhancement feature, and generating an extended sample through the generator; An extended sample evaluation unit, used to discriminate the quantum state enhancement feature and the extended sample through the discriminator to obtain a discrimination result; A loss calculation unit, used to calculate the quantum state information entropy loss based on the discrimination result and the quantum state parameter; A parameter adjustment unit is used to adjust the quantum state parameters based on the quantum state information loss entropy until a preset condition is met to complete the training of the generative adversarial network.

Images