Traditional Chinese medicine condition dynamic monitoring method and system based on big data time sequence analysis

Abstract

The invention provides a traditional Chinese medicine condition dynamic monitoring method and system based on big data time sequence analysis, and the method comprises the steps: extracting the condition characteristics of a previous sample from a traditional Chinese medicine condition record data set of a patient, a node, and a performance symptom of a prediction sample patient, and carrying out the deep analysis of the data through a condition dynamic development prediction network. Precise and dynamic monitoring of traditional Chinese medicine disease development is achieved, the first confidence coefficient of the patient expression symptom of the node sample at the target time node can be determined, the second confidence coefficient of the patient expression symptom of the prediction sample at the future time node can also be predicted, and therefore the accuracy and perspectiveness of disease prediction are improved, and the disease prediction efficiency is improved. By continuously optimizing the condition dynamic development prediction network, the uncertainty in condition prediction is effectively reduced, more reliable data support is provided for traditional Chinese medicine clinical decision, and the traditional Chinese medicine diagnosis and treatment effect can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for dynamic monitoring of TCM conditions based on big data time series analysis. Background Art

[0002] In the clinical practice of traditional Chinese medicine, accurate monitoring and prediction of the condition is crucial to developing effective treatment plans and improving patient prognosis. However, traditional Chinese medicine diagnostic methods mainly rely on the doctor's experience and the patient's self-report. Although this method can reflect the patient's condition to a certain extent, it is often affected by subjective factors and it is difficult to achieve objective and dynamic monitoring of the condition.

[0003] With the rapid development of big data technology, it has become possible to use time series analysis to dynamically monitor disease conditions. Time series analysis is a statistical technique that can reveal the laws and trends of data changes over time by mining and analyzing historical data. However, most of the time series analysis methods currently used in TCM disease monitoring only focus on data at a single time point, ignoring the continuity and dynamics of disease development, resulting in limited accuracy and timeliness of disease monitoring.

[0004] In addition, when processing TCM disease data, existing disease prediction models often ignore the complex correlations and temporal dependencies between disease characteristics, which affects the reliability of the prediction results. Summary of the invention

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for dynamic monitoring of TCM conditions based on big data time series analysis, the method comprising: Obtaining a past sample condition characteristic data set, a sample patient's symptoms at this node, and a predicted sample patient's symptoms from a patient's TCM condition record data set; the patient's TCM condition record data set includes a plurality of sample patient's symptoms organized based on a time series, the past sample condition characteristic data set includes a plurality of sample patient's symptoms before a target time node, the sample patient's symptoms at this node are sample patient's symptoms corresponding to the target time node, and the predicted sample patient's symptoms are sample patient's symptoms after the target time node; Loading the past sample disease characteristic data set and the symptoms exhibited by the sample patient at this node into the disease dynamic development prediction network, and determining the first confidence level of the symptoms exhibited by the sample patient at this node appearing at the target time node; Loading the past sample disease characteristic data set and the predicted sample patient symptoms into the disease dynamic development prediction network, and determining a second confidence level that the predicted sample patient symptoms will appear after the target time node; Based on the loss between the labeled confidence of the symptoms exhibited by the sample patient of this node and the first confidence, the dynamic disease progression prediction network is optimized, and based on the loss between the labeled confidence of the symptoms exhibited by the predicted sample patient and the second confidence, the dynamic disease progression prediction network is optimized.

[0006] In a possible implementation of the first aspect, the optimizing the disease dynamic development prediction network based on the loss between the annotation confidence of the predicted sample patient's symptoms and the second confidence includes: Obtaining first training supervision data and second training supervision data of the symptoms exhibited by the predicted sample patient, wherein the first training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is 1, and the second training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is x, where x is not less than 0 and x is not greater than 1; Based on the first training supervision data and the second training supervision data, generating a labeling confidence of the predicted sample patient's symptoms; The disease dynamics development prediction network is optimized based on the loss between the labeled confidence of the predicted sample patient's symptoms and the second confidence.

[0007] In a possible implementation of the first aspect, the disease dynamics development prediction network is a taught disease dynamics development prediction network; and obtaining the second training supervision data of the predicted sample patient's symptoms includes: The past sample disease characteristic data set and the predicted sample patient symptoms are loaded into the teaching disease dynamic development prediction network, and a third confidence level that the predicted sample patient symptoms will appear after the target time node is determined; the network parameter amount of the teaching disease dynamic development prediction network is not less than the network parameter amount of the taught disease dynamic development prediction network; The third confidence level is used as second training supervision data for predicting symptoms exhibited by the sample patient.

[0008] In a possible implementation of the first aspect, the taught disease state dynamic development prediction network is the taught disease state dynamic development prediction network in the current round of network parameter optimization phase, and the teaching disease state dynamic development prediction network is the teaching disease state dynamic development prediction network in the current round of network parameter optimization phase; the method further comprises: Based on the taught disease dynamic development prediction network of the previous network parameter optimization stage and the taught disease dynamic development prediction network of the current network parameter optimization stage, the taught disease dynamic development prediction network of the current network parameter optimization stage is generated.

[0009] In a possible implementation of the first aspect, the generating the teaching dynamic disease progression prediction network of the current round of network parameter optimization stage based on the teaching dynamic disease progression prediction network of the previous network parameter optimization stage and the taught dynamic disease progression prediction network of the current round of network parameter optimization stage includes: The neuron weight information of the teaching disease dynamic development prediction network in the previous network parameter optimization stage is fused with the first training degradation factor to generate the first parameter information; The neuron weight information of the trained disease dynamic development prediction network in the current round of network parameter optimization is fused with the second training degradation factor to generate second parameter information; the sum of the first training degradation factor and the second training degradation factor is 1; The first parameter information and the second parameter information are integrated to generate neuron weight information of the disease dynamic development prediction network in the current round of network parameter optimization stage.

[0010] In a possible implementation of the first aspect, generating the annotation confidence of the predicted sample patient symptom based on the first training supervision data and the second training supervision data includes: The second training supervision data is fused with the first influencing factor to generate third training supervision data; The first training supervision data is fused with the second influencing factor to generate fourth training supervision data; the sum of the first influencing factor and the second influencing factor is 1; The third training supervision data and the fourth training supervision data are fused to generate the annotation confidence of the predicted sample patient's symptoms.

[0011] In a possible implementation of the first aspect, the method further includes: Determine a reference influencing factor that matches the second training supervision data from a plurality of candidate first influencing factors; and use the reference influencing factor as the first influencing factor of the second training supervision data.

[0012] In a possible implementation of the first aspect, obtaining the predicted sample patient symptoms includes: The predicted sample patient symptoms are randomly extracted from the sample patient symptoms after the sample patient symptoms at this node in the patient's TCM medical record data set.

[0013] In a possible implementation of the first aspect, the method further includes: Based on the optimized disease dynamic development prediction network, any input target disease characteristic data set and target patient symptoms are analyzed to predict the target confidence level of the time node corresponding to the target patient symptoms.

[0014] For example, in a possible implementation of the first aspect, before the steps of obtaining a past sample condition characteristic data set, symptoms exhibited by the sample patient at this node, and predicting symptoms exhibited by the sample patient from the patient's TCM condition record data set, the method further includes: Performing structural analysis on the patient TCM condition record data in the patient TCM condition record data set to determine the different types of data elements contained therein and the relationship structure between the data elements; Based on the prior knowledge in the field of TCM health management and the requirements for data and privacy protection, a semantic classification framework is constructed. The semantic classification framework is used to classify each data element according to its semantic category in TCM theory and define corresponding semantic labels and related attributes. The semantic labels will be used to perform privacy semantic encoding on the data elements. According to the semantic classification framework, each data element in the patient's TCM medical record data is marked, wherein for each data element, the category to which the data element belongs in the semantic classification framework is found, and after the corresponding semantic label is assigned to the data element, privacy weight information is added to each data element on the basis of marking the semantic label, and the privacy weight information is determined based on the importance of the data element in privacy protection, thereby recombining the data elements marked with the semantic label and the privacy weight to form a preliminary data structure after privacy semantic encoding; Extract various health data patterns from the preliminary data structure after privacy semantic encoding, perform feature description on each health data pattern, and establish a health data pattern library, wherein the feature description includes the combination of data elements constituting the health data pattern, the value range or feature value of each data element, and the logical relationship between each data element; For each data subset in the preliminary data structure after privacy semantics encoding, matching is performed with various health data patterns in the health data pattern library, wherein the data subset is a data portion divided according to different diagnostic stages or body systems; For a successfully matched data subset, mark the name or number of the health data pattern matched by the successfully matched data subset, record the degree of similarity when the match is successful, and generate a health data pattern matching result. The degree of similarity is quantitatively evaluated based on the proportion of data element matching and the matching of key data elements. In addition, a data subset that does not match any health data pattern is specially marked to indicate that the data subset is an abnormal data subset or a new unidentified health data pattern; Integrate the preliminary data structure of privacy semantic encoding with the health data pattern matching result, specifically for each data element, associate the semantic label, privacy weight information, matched health data pattern and similarity degree information of the data element, and build a corresponding nested structure based on the association result, wherein the outer structure of the nested structure contains the semantic category information of the data element, and the inner structure contains the data element itself, the privacy weight, and the pattern matching result information; The patient's TCM medical record data is preprocessed for privacy protection based on the nested structure.

[0015] On the other hand, an embodiment of the present invention also provides a dynamic monitoring system for traditional Chinese medicine diseases based on big data time series analysis, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0016] Based on the above aspects, the embodiment of the present application extracts past sample disease characteristics, current node and predicted sample patient symptoms from the patient's TCM disease record data set, and uses the disease dynamic development prediction network to conduct in-depth analysis of these data, thereby achieving accurate dynamic monitoring of the development of TCM disease. It can not only determine the first confidence level of the sample patient's symptoms at the target time node and current node, but also predict the second confidence level of the sample patient's symptoms at future time nodes, thereby improving the accuracy and foresight of disease prediction. By continuously optimizing the disease dynamic development prediction network, the uncertainty in disease prediction is effectively reduced, providing more reliable data support for TCM clinical decision-making, and helping to improve the diagnosis and treatment effects of TCM. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the execution flow of the method for dynamic monitoring of TCM disease conditions based on big data time series analysis provided by an embodiment of the present invention.

[0018] Figure 2 It is a schematic diagram of the hardware architecture of a dynamic monitoring system for TCM conditions based on big data time series analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of a method for dynamic monitoring of TCM conditions based on big data time series analysis provided by an embodiment of the present invention. The method for dynamic monitoring of TCM conditions based on big data time series analysis is introduced in detail below.

[0020] Step S110, obtaining a past sample condition characteristic data set, a sample patient's symptoms at this node, and a predicted sample patient's symptoms from the patient's TCM condition record data set. The patient's TCM condition record data set includes a plurality of sample patient's symptoms organized based on a time series, the past sample condition characteristic data set includes a plurality of sample patient's symptoms before a target time node, the sample patient's symptoms at this node are sample patient's symptoms corresponding to the target time node, and the predicted sample patient's symptoms are sample patient's symptoms after the target time node.

[0021] In this embodiment, the server starts to process the patient condition data of a certain TCM medical institution. The TCM medical condition record data set of the patients in the TCM medical institution contains the TCM diagnosis and treatment records of many patients over the years. These records are sorted according to the time series, and the various symptoms of each patient at different time points are recorded in detail. For example, for patient A, his record starts from the first visit, including pulse (stringy pulse, slippery pulse, etc.), tongue (red tongue with yellow fur, pale tongue with white fur, etc.), physical discomfort symptoms (headache, fatigue, cough, etc.) and other related TCM diagnosis information (such as qi and blood status, viscera function, etc.).

[0022] Now the server needs to determine the target time node for a specific analysis task. Assume that this target time node is the 30th day after patient A's first visit. The server obtains relevant data from the patient's TCM medical record data set. The past sample disease characteristic data set is all the symptom records of patient A before the 30th day. For example, between the 1st day and the 29th day, patient A's pulse has always been stringy, the tongue is red with yellow fur, and he has intermittent headaches and fatigue symptoms. These symptom data constitute the past sample disease characteristic data set.

[0023] The symptoms exhibited by the sample patients at this node are the symptoms corresponding to the target time node of day 30. For example, the record on day 30 shows that patient A's pulse becomes thin, the headache symptoms worsen, and new symptoms such as irritability appear. These are the symptoms exhibited by the sample patients at this node.

[0024] The symptoms of the sample patients are predicted to be those after the 30th day. The server randomly extracts from the subsequent records (according to a possible implementation method), for example, between the 31st and 35th days, the pulse of patient A turns back to a string pulse, the headache symptoms are alleviated but insomnia symptoms appear, and these are used as the symptoms of the predicted sample patients. By accurately obtaining these three types of data from a large-scale patient TCM disease record data set, a basic data source is provided for the subsequent prediction of the dynamic development of the disease.

[0025] Step S120, loading the past sample disease characteristic data set and the symptoms exhibited by the sample patient of the current node into the disease dynamic development prediction network, and determining the first confidence level that the symptoms exhibited by the sample patient of the current node occur at the target time node.

[0026] In this embodiment, the server loads the previously acquired past sample disease characteristic data set (such as patient A's symptom data from day 1 to day 29) and the symptoms of the sample patient at this node (symptom data on day 30) into the pre-built disease dynamic development prediction network. This disease dynamic development prediction network is a neural network model trained and constructed based on a large amount of traditional Chinese medicine disease data. It can learn the complex relationship between different symptoms and the development pattern of the disease over time.

[0027] For example, neurons in the network may correspond to different symptom characteristics or symptom combinations, and the connection weights between neurons represent the degree of association between these symptoms. When the data set of past sample disease characteristics and the symptoms of the sample patient at this node are input, the network will perform a series of calculations. Assume that the network analyzes the symptoms of the sample patient at this node, such as the pulse of patient A becoming thin, the headache symptoms worsening, and the irritability on the 30th day, based on the rules learned previously.

[0028] For example, a comprehensive evaluation can be made based on the association between past symptoms (such as previous stringy pulse, red tongue with yellow fur, intermittent headache and fatigue, etc.) and current symptoms, as well as the probability of such symptom transition in a large number of other similar cases. If in a large amount of training data, many patients have similar symptom transitions around the 30th day after similar past symptoms, then the network will give a higher first confidence. Conversely, if such symptom transitions are rare in the training data, then the first confidence will be lower. For example, after calculation, the server obtains the first confidence of 0.7 for the sample patient in this node to show symptoms on the 30th day, which means that the network believes that based on the existing data and learned rules, the probability that patient A will experience these symptoms on the 30th day is 70%.

[0029] Step S130, loading the past sample disease characteristic data set and the predicted sample patient symptoms into the disease dynamic development prediction network, and determining a second confidence level that the predicted sample patient symptoms will appear after the target time node.

[0030] Continue to use the previous sample disease characteristic data set (patient A's symptom data from day 1 to day 29), and load the predicted sample patient symptoms (patient A's pulse turns back to a string pulse, headache symptoms are alleviated, and insomnia symptoms appear between day 31 and day 35) into the disease dynamic development prediction network.

[0031] Similarly, the network will analyze based on its internal structure and learned knowledge. Since the network has learned the association pattern of different symptoms in time series, it will evaluate the possibility of these symptoms appearing after the 30th day based on the previous symptom development of patient A (symptoms from day 1 to day 29) and the symptoms of the current predicted sample patient. For example, the network may find that in previous cases, when patients had similar symptoms before day 30 and developed specific symptoms on day 30 (such as the symptoms on day 30 in step S120), some patients will subsequently experience symptoms such as pulse returning to string pulse, headache relief, and insomnia.

[0032] Based on the proportion of this match and the strength of the association between symptoms, the network calculates the second confidence. Assuming that the calculated second confidence is 0.6, it means that based on the existing data and the learning results of the network, the probability that patient A will show the symptoms of these predicted sample patients after the 30th day is 60%.

[0033] Step S140, optimizing the disease dynamic progression prediction network based on the loss between the labeled confidence of the sample patient's symptoms at this node and the first confidence, and optimizing the disease dynamic progression prediction network based on the loss between the labeled confidence of the sample patient's symptoms at this node and the second confidence.

[0034] First, the annotation confidence of the symptoms of the sample patient in this node is a certain value based on the actual medical diagnosis or more accurate expert evaluation. Assume that for the symptoms that patient A experienced on the 30th day (pulse became thin, headache symptoms worsened, and irritability, etc.), after careful evaluation by experts and combined with a lot of clinical experience, the annotation confidence is determined to be 0.8, which means that the experts believe that the real possibility of the patient experiencing these symptoms in this case is 80%.

[0035] The server calculates the loss between the labeled confidence (0.8) and the first confidence (0.7) of the symptoms of the sample patient in this node. This loss can be calculated using a variety of loss functions, such as the mean square error (MSE). After calculating the loss value, the server uses this loss value to adjust the parameters of the disease dynamics prediction network. For example, if the back propagation algorithm is used, the server calculates the gradient of each neuron weight based on the loss value, and then updates the weight at a certain learning rate, so that the network can give a result closer to the labeled confidence the next time it handles a similar situation.

[0036] For the prediction of the symptoms of the sample patients, the annotation confidence also needs to be determined. According to the method mentioned above, it is assumed that the annotation confidence of the predicted sample patient symptoms (patient A's pulse turns back to string pulse, headache symptoms are alleviated, and insomnia symptoms appear between the 31st and 35th days) is 0.75, which is also based on expert experience or more accurate diagnostic methods.

[0037] The server calculates the loss between the labeled confidence (0.75) and the second confidence (0.6) of the predicted sample patient's symptoms, also calculated using an appropriate loss function. This loss value is then used to optimize the disease dynamics prediction network. For example, the weights of neurons in the network related to symptom association analysis are adjusted to make the network's confidence assessment for predicting future symptom onset more accurate. By continuously optimizing the network based on the loss between the labeled confidence and the confidence calculated by the network, the disease dynamics prediction network can continuously improve its prediction accuracy and better adapt to the complex dynamic changes in TCM disease conditions.

[0038] Throughout the entire process, the ability to predict the dynamic development of TCM diseases is continuously improved through processing of TCM disease record data, calculation in the disease dynamic development prediction network, and network optimization based on confidence loss, thereby providing more accurate and valuable support for TCM clinical diagnosis, health management, etc.

[0039] Based on the above steps, the embodiment of the present application extracts past sample disease characteristics, current node and predicted sample patient symptoms from the patient's TCM disease record data set, and uses the disease dynamic development prediction network to conduct in-depth analysis of these data, thereby achieving accurate dynamic monitoring of the development of TCM disease. It can not only determine the first confidence level of the sample patient's symptoms at the target time node and current node, but also predict the second confidence level of the sample patient's symptoms at future time nodes, thereby improving the accuracy and foresight of disease prediction. By continuously optimizing the disease dynamic development prediction network, the uncertainty in disease prediction is effectively reduced, providing more reliable data support for TCM clinical decision-making, and helping to improve the diagnosis and treatment effects of TCM.

[0040] In a possible implementation, step S140 includes: Step S141, obtaining first training supervision data and second training supervision data of the symptoms exhibited by the predicted sample patient, wherein the first training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is 1, and the second training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is x, where x is not less than 0 and x is not greater than 1.

[0041] Step S142: generating a labeling confidence of the predicted sample patient's symptoms based on the first training supervision data and the second training supervision data.

[0042] Step S143: optimizing the disease dynamics progression prediction network based on the loss between the annotation confidence of the predicted sample patient's symptoms and the second confidence.

[0043] In this embodiment, the first step is to obtain the first training supervision data and the second training supervision data for predicting the symptoms of the sample patient. In the example of patient A mentioned above, for the predicted sample patient symptoms such as patient A's pulse turning back to string pulse, headache symptoms being alleviated, and insomnia symptoms appearing between the 31st and 35th days, the first training supervision data is used to characterize the positive sample labels of these predicted sample patient symptoms, with a ratio of 1. This means that from an ideal and completely certain perspective, these symptoms are considered to be completely consistent with a certain expected pattern. For example, in a large number of existing case data that have been accurately diagnosed and regarded as standard examples, when the patient previously had a symptom development course similar to that of patient A from the 1st to the 30th day, the appearance of these symptoms from the 31st to the 35th day was clearly identified as a typical development trend of the disease, so the positive sample label ratio was set to 1.

[0044] The second training supervision data is used to characterize the proportion of positive sample labels that predict the symptoms of sample patients, which is x (0 ≤ x ≤ 1). This is a relatively more flexible data that takes into account more practical and complex situations. Continuing with patient A as an example, the server will combine more factors to determine the x value. For example, the server will analyze the uncertainty of the occurrence of these symptoms from the 31st to the 35th day, although they have similar early symptoms, under the influence of individual differences in patients (such as age, physique, living environment, etc.). Assuming that in some young and physically fit patient groups, even if the early symptoms are similar, the proportion of subsequent symptoms is relatively low. After statistical analysis of the data of this part of patients, it is found that the proportion of these symptoms is 0.6, then this 0.6 may be determined as the x value. The determination of this x value is based on a comprehensive consideration of a large number of different types of patient data, not only limited to the symptoms themselves, but also involves the individual characteristics of the patient and various complex associations in the overall data.

[0045] Next, based on the first training supervision data and the second training supervision data, the annotation confidence of the predicted sample patient's symptoms is generated. The server will perform complex calculations based on the two types of training supervision data that have been obtained. For example, suppose that the server uses a weighted fusion method to generate annotation confidence. For the first training supervision data, since it represents the positive sample label ratio of 1, which is a very certain information, the server will give it a relatively high weight. For the second training supervision data, since its x value (assuming it is 0.6) reflects a situation with a certain degree of uncertainty, it will be given a relatively low weight. The specific weight distribution may be determined based on experience or previous data analysis, such as giving the first training supervision data a weight of 0.7 and the second training supervision data a weight of 0.3. Then the annotation confidence of the predicted sample patient's symptoms is calculated as: 1×0.7 + 0.6×0.3 = 0.7 + 0.18 = 0.88. This 0.88 is the annotation confidence obtained by combining the two types of training supervision data, which means that from a comprehensive perspective, patient A will have a comprehensive assessment confidence of the symptoms shown by these predicted sample patients from day 31 to day 35.

[0046] Finally, the dynamic disease progression prediction network is optimized based on the loss between the labeled confidence (0.88) of the predicted sample patient's symptoms and the second confidence (0.6 calculated previously). The server will use a suitable loss function to quantify this loss, such as the mean square error (MSE). The formula for calculating the loss is: MSE = (0.88 - 0.6)² = 0.0784. This loss value reflects the gap between the labeled confidence of the predicted sample patient's symptoms and the second confidence calculated by the network. The server then uses this loss value to optimize the dynamic disease progression prediction network.

[0047] During the optimization process, the server will use technical means such as the back propagation algorithm. The weight of each neuron in the disease dynamic development prediction network will be adjusted according to this loss value. Assume that the disease dynamic development prediction network is a multi-layer neural network, and there are complex connections between neurons in the network, and each connection has a weight. After the loss value is calculated, the server will update the weight according to this loss value. For example, for the neuron connection weights related to the prediction of patient A's symptoms from day 31 to day 35, the server will update the weights according to a certain learning rate (assuming the learning rate is 0.01). If a certain weight is w, the updated weight w' = w - 0.01×(∂MSE / ∂w), where ∂MSE / ∂w is the partial derivative of the loss function MSE with respect to the weight w. By constantly adjusting the weights according to the loss, the disease dynamic development prediction network can continuously learn and improve, so that when processing similar prediction tasks in the future, the confidence can be calculated more accurately, thereby improving the accuracy of the prediction of the dynamic development of TCM diseases. During the entire process, the server accurately processes various data, calculates relevant values, and uses these results to optimize the network to adapt to the complex and changeable actual conditions of TCM diseases.

[0048] In a possible implementation, the disease dynamic development prediction network is a taught disease dynamic development prediction network. Obtaining the second training supervision data of the predicted sample patient's symptoms includes: loading the past sample disease characteristic data set and the predicted sample patient's symptoms into the teaching disease dynamic development prediction network, and determining the third confidence that the predicted sample patient's symptoms will appear after the target time node. The network parameter amount of the teaching disease dynamic development prediction network is not less than the network parameter amount of the taught disease dynamic development prediction network. The third confidence is used as the second training supervision data of the predicted sample patient's symptoms.

[0049] In this embodiment, first, for obtaining the second training supervision data for predicting the symptoms of the sample patient, the server will perform the following operations. Taking the patient A mentioned above as an example, the server has obtained the patient A's past sample disease characteristic data set (such as symptom data from the 1st to the 29th day) and the predicted sample patient's symptoms (the pulse turns back to a string pulse between the 31st and 35th days, the headache symptoms are alleviated, and insomnia symptoms appear, etc.). The server loads these data into the teaching disease dynamic development prediction network. This teaching disease dynamic development prediction network is a relatively complex network with a large number of network parameters, and its network parameter amount is not less than the network parameter amount of the taught disease dynamic development prediction network. Due to the large number of network parameters, more potential relationships and patterns can be mined from the data.

[0050] When the server loads the patient A's past sample disease characteristic data set and the predicted sample patient's symptoms into the teaching disease dynamic development prediction network, the network will calculate according to its internal complex structure and pre-learned knowledge. There are many neurons in this network, each of which corresponds to different disease characteristics or feature combinations, and the connection weights between neurons reflect the degree of correlation between these characteristics. The network will comprehensively consider the relationship between the previous symptom development of patient A (such as pulse, tongue, other symptoms from the 1st to the 29th day, etc.) and the predicted sample patient's symptoms (symptoms from the 31st to the 35th day). For example, the teaching disease dynamic development prediction network may have learned the probability relationship between the subsequent reduction of headache symptoms and insomnia symptoms under certain specific pulse and tongue change patterns. Through a series of complex calculations, the network determines the third confidence level that the predicted sample patient's symptoms will appear after the target time node. Assume that the third confidence level is 0.72 after calculation. This 0.72 means that according to the analysis of the teaching disease dynamic development prediction network, the probability of patient A experiencing these symptoms from the 31st to the 35th day is 72%. The server then uses this third confidence as the second training supervision data for predicting the symptoms exhibited by the sample patient.

[0051] In a possible implementation, the taught disease dynamic development prediction network is the taught disease dynamic development prediction network in the current round of network parameter optimization stage, and the teaching disease dynamic development prediction network is the teaching disease dynamic development prediction network in the current round of network parameter optimization stage. The method further includes: generating the teaching disease dynamic development prediction network in the current round of network parameter optimization stage based on the teaching disease dynamic development prediction network in the previous network parameter optimization stage and the taught disease dynamic development prediction network in the current round of network parameter optimization stage.

[0052] In a possible implementation, the generating of the teaching disease dynamic development prediction network of the current round of network parameter optimization phase based on the teaching disease dynamic development prediction network of the previous network parameter optimization phase and the taught disease dynamic development prediction network of the current round of network parameter optimization phase includes: The neuron weight information of the teaching disease dynamic development prediction network in the previous network parameter optimization stage is fused with the first training degradation factor to generate the first parameter information.

[0053] The neuron weight information of the trained disease dynamic development prediction network in the current round of network parameter optimization is fused with the second training degradation factor to generate second parameter information. The sum of the first training degradation factor and the second training degradation factor is 1.

[0054] Fuse the first parameter information and the second parameter information to generate the neuron weight information of the teaching disease dynamic development prediction network in this round of network parameter optimization stage.

[0055] Next, in a possible implementation, the taught disease dynamic development prediction network is the taught disease dynamic development prediction network in this round of network parameter optimization stage, the teaching disease dynamic development prediction network is the teaching disease dynamic development prediction network in this round of network parameter optimization stage, and there is an operation of generating the teaching disease dynamic development prediction network in this round of network parameter optimization stage based on the teaching disease dynamic development prediction network in the previous network parameter optimization stage and the taught disease dynamic development prediction network in this round of network parameter optimization stage.

[0056] When the server performs this operation, it first fuses the neuron weight information of the teaching disease dynamic development prediction network in the previous network parameter optimization stage with the first training degradation factor to generate the first parameter information. Assume that the weight of a certain neuron in the teaching disease dynamic development prediction network in the previous network parameter optimization stage is w1, and the first training degradation factor is a (0 < a < 1). Then the weight w1' of this neuron after fusion is w1 * a. Such an operation is performed on all neurons in the network to generate the first parameter information. In this process, the role of the first training degradation factor is to adjust the neuron weight information of the previous teaching disease dynamic development prediction network to a certain extent to meet the new network optimization requirements.

[0057] At the same time, the server fuses the neuron weight information of the taught disease dynamic development prediction network in this round of network parameter optimization stage with the second training degradation factor to generate the second parameter information. Assume that the weight of a certain neuron in the taught disease dynamic development prediction network in this round of network parameter optimization stage is w2, and the second training degradation factor is b (0 < b < 1 and a + b = 1). Then the weight w2' of this neuron after fusion is w2 * b. Similarly, such an operation is performed on all neurons in the network to obtain the second parameter information.

[0058] Finally, the server fuses the first parameter information and the second parameter information to generate the neuron weight information of the teaching disease dynamic development prediction network in this round of network parameter optimization stage. For each neuron in the network, the weight w1' in the first parameter information and the weight w2' in the second parameter information are fused. For example, the fused neuron weight w3 = w1' + w2'. By performing such operations on all neurons in the network, the generation of neuron weight information of the teaching disease dynamic development prediction network in this round of network parameter optimization stage is completed. This fusion method can combine the advantages of the teaching disease dynamic development prediction network in the previous network parameter optimization stage and the taught disease dynamic development prediction network in this round of network parameter optimization stage, so that the newly generated teaching disease dynamic development prediction network in this round of network parameter optimization stage can better adapt to the needs of disease dynamic development prediction and improve the accuracy and reliability of prediction. Throughout the process, the server accurately processes various data, network weight information and different factors to optimize and improve the disease dynamic development prediction network, so as to better cope with the complex and changeable actual situation of traditional Chinese medicine disease.

[0059] When handling these tasks, the server needs to process a large amount of patient condition data and perform various calculations and operations accurately. From obtaining the second training supervision data from the teaching disease dynamic development prediction network to generating a new teaching disease dynamic development prediction network based on networks at different stages, each step involves an in-depth understanding and precise operation of the data and network structure. For example, in the process of processing patient A's disease data, the server must ensure the accuracy of the data at each stage, as well as the correctness of the parameter conversion and fusion between different networks. This requires the server to have strong computing power and data processing capabilities to cope with the complexity and diversity of TCM disease data. At the same time, this complex network construction and optimization process is also to improve the accuracy of disease prediction and provide more valuable support for TCM clinical diagnosis, health management, etc. By continuously optimizing the teaching disease dynamic development prediction network, the server can make better use of the existing disease data and dig out more valuable disease development laws, thereby providing more accurate services for patients' disease prediction and health management.

[0060] In a possible implementation, step S142 includes: Step S1421: fuse the second training supervision data with the first influencing factor to generate third training supervision data.

[0061] Step S1422: The first training supervision data is fused with the second influencing factor to generate fourth training supervision data. The sum of the first influencing factor and the second influencing factor is 1.

[0062] Step S1423: The third training supervision data and the fourth training supervision data are merged to generate the annotation confidence of the predicted sample patient's symptoms.

[0063] In a possible implementation, the method further includes: determining a reference influence factor matching the second training supervision data from a plurality of candidate first influence factors, and using the reference influence factor as the first influence factor of the second training supervision data.

[0064] In this embodiment, taking the case of patient A as an example, the first training supervision data (the ratio of positive sample labels used to characterize the symptoms of the predicted sample patients is 1) and the second training supervision data (assuming that the third confidence level of patient A's symptoms from day 31 to day 35 obtained in teaching the disease dynamic development prediction network is 0.72, and this value is used as the second training supervision data) have been obtained in previous operations.

[0065] First, the server determines a reference influencing factor that matches the second training supervision data from multiple candidate first influencing factors, and uses it as the first influencing factor of the second training supervision data. The server maintains a set of candidate first influencing factors, which are determined based on a large amount of previous data processing and analysis experience. For each candidate first influencing factor, there are some data features or condition ranges associated with it. For example, when the value of the second training supervision data is in a certain interval, it may correspond to a specific candidate first influencing factor. Assume that when the value of the second training supervision data is in the interval of 0.7-0.8, the corresponding candidate first influencing factor is 0.3. Since the second training supervision data obtained previously is 0.72, this 0.3 is determined as the reference influencing factor that matches the second training supervision data, that is, the first influencing factor of the second training supervision data.

[0066] Next, the server merges the second training supervision data with the first influencing factor to generate the third training supervision data. For the example of patient A, the second training supervision data is 0.72, and the first influencing factor is 0.3, then the third training supervision data = 0.72 * 0.3 = 0.216. This calculation process is to adjust the second training supervision data according to the proportion of the first influencing factor to obtain a new intermediate data, namely the third training supervision data.

[0067] At the same time, the server fuses the first training supervision data with the second impact factor to generate the fourth training supervision data. Since the sum of the first impact factor and the second impact factor is 1, the first impact factor has been determined to be 0.3, so the second impact factor is 1 - 0.3 = 0.7. The first training supervision data is 1 (the ratio of positive sample labels is 1), then the fourth training supervision data = 1 * 0.7 = 0.7. This process also adjusts the first training supervision data according to the corresponding ratio to obtain the fourth training supervision data.

[0068] Finally, the server merges the third training supervision data and the fourth training supervision data to generate the annotation confidence of the predicted sample patient's symptoms. For patient A, the third training supervision data is 0.216 and the fourth training supervision data is 0.7, so the annotation confidence = 0.216 + 0.7 = 0.916. This annotation confidence represents the comprehensive evaluation confidence of patient A's predicted sample patient symptoms from day 31 to day 35 after comprehensively considering the first training supervision data and the second training supervision data and their respective influencing factors.

[0069] Throughout the process, the server accurately determines the first influencing factor that matches the second training supervision data, and adjusts and fuses the first training supervision data and the second training supervision data according to rigorous calculation rules, thereby obtaining the annotation confidence of the predicted sample patient's symptoms. This annotation confidence will be used to optimize the dynamic development prediction network of the disease in the future to improve the accuracy of the network's prediction of the disease development. When processing these operations, the server needs to accurately manage and calculate a large amount of data and various parameters, which depends on its powerful computing power and data processing capabilities. For example, when determining the matching reference influencing factor from the candidate first influencing factor, it is necessary to quickly and accurately find the appropriate influencing factor among many candidates based on the value of the second training supervision data. When performing data fusion calculations, the accuracy of the calculations must also be ensured to ensure that the final generated annotation confidence can accurately reflect the comprehensive evaluation of the disease prediction. This precise data processing and calculation process is an important guarantee for the effective operation of the entire disease dynamic development prediction system, and helps to improve the application value of TCM disease prediction in clinical and health management.

[0070] In a possible implementation, obtaining the predicted sample patient symptoms includes: The predicted sample patient symptoms are randomly extracted from the sample patient symptoms after the sample patient symptoms at this node in the patient's TCM medical record data set.

[0071] In a possible implementation, the method further includes: Based on the optimized disease dynamic development prediction network, any input target disease characteristic data set and target patient symptoms are analyzed to predict the target confidence level of the time node corresponding to the target patient symptoms.

[0072] In this embodiment, when obtaining the symptoms of the predicted sample patient, the sample patient's symptoms after the sample patient's symptoms at this node are randomly extracted from the patient's TCM disease record data set. Taking the patient A mentioned above as an example, the TCM disease record data set of patient A contains a lot of symptom information organized in time series since the first visit. After determining the symptoms of the sample patient at this node (assuming that the pulse of patient A on the 30th day becomes a thin pulse, the headache symptoms worsen, and the symptoms of irritability), the server will check all the sample patient's symptoms after the sample patient at this node shows symptoms, that is, the symptom records on the 31st day and later. These records contain various possible symptom changes, such as further changes in pulse, relief or aggravation of headache symptoms, whether new symptoms (such as insomnia, cough, etc.) appear, and changes in other TCM-related symptoms such as tongue image, qi and blood, etc. The server randomly extracts from these subsequent symptom records to obtain the symptoms of the predicted sample patient. For example, the server may randomly extract the symptom records of patient A between the 31st and 35th days, in which the pulse turns back to a string pulse, the headache symptoms are alleviated, and insomnia symptoms appear. These extracted symptoms become the symptoms of the predicted sample patient. This random extraction method can, to a certain extent, represent the various possibilities of the patient's disease development and provide different sample data for the subsequent prediction of the dynamic development of the disease.

[0073] After completing the above operations and optimizing the disease dynamic development prediction network based on the previous series of steps, the server also needs to perform an operation based on the optimized disease dynamic development prediction network to analyze any input target disease characteristic data set and target patient symptoms, and predict the target confidence level of the target patient's symptoms at the time node. Assume that there is a new patient B, whose target disease characteristic data set contains the initial symptom information of patient B, such as the pulse (such as deep pulse), tongue (such as pale tongue with white fur), physical discomfort symptoms (such as fatigue, chills, etc.) and other information related to TCM diagnosis (such as qi and blood deficiency, etc.). The target patient's symptoms may be symptoms that may appear at a specific time point (assuming it is the 15th day after the first visit), such as possible cough aggravation, pulse changes, etc.

[0074] The server inputs the target disease characteristic data set of patient B and the symptoms of the target patient into the optimized disease dynamic development prediction network. This optimized network has been trained and optimized with many previous patient data (such as the data processing process of patient A), and the internal neuron weights and other parameters have been adjusted to a relatively accurate state. The neurons in the network correspond to different disease characteristics or feature combinations, and the connection weights between neurons indicate the degree of correlation between these features. When the target disease characteristic data set of patient B and the symptoms of the target patient are input, the network will perform complex calculations based on its internal structure and learned rules.

[0075] First, the network will identify the initial condition pattern of patient B based on the initial symptom information in the target condition feature data set (such as deep pulse, pale tongue with white fur, fatigue, chills, etc.). Then, combined with the symptoms of the target patient (symptoms that may appear on the 15th day), the network will look for similar patterns in the large number of disease development patterns it has learned. For example, the network may find that in the previously processed patient data, a portion of patients who initially had similar pulse, tongue, and physical discomfort symptoms had similar symptoms such as worsening cough and pulse changes around the 15th day.

[0076] Based on these matching situations and factors such as the strength of association between different symptoms learned previously, the network calculates the target confidence that the target patient's symptoms will appear on the 15th day. Assuming that the target confidence is calculated to be 0.65, it means that based on the existing data and the network's learning results, the probability that patient B will experience these target patient symptoms on the 15th day after the initial visit is 65%. In this way, the server uses the optimized disease dynamic development prediction network to predict the condition of new patients, providing a valuable reference for TCM clinical diagnosis, health management, etc. Throughout the process, the server accurately obtains the symptoms of the predicted sample patients and uses the optimized network to predict the condition of new patients, reflecting the effectiveness and practicality of the entire system in TCM disease analysis and prediction.

[0077] For example, in a possible implementation, before step S110, the method further includes: Step S101, performing structural analysis on the patient's TCM condition record data in the patient's TCM condition record data set to determine the different types of data elements contained therein and the relationship structure between the data elements.

[0078] Step S102, constructing a semantic classification framework based on prior knowledge in the field of TCM health management and data and privacy protection requirements. The semantic classification framework is used to classify each data element according to its semantic category in TCM theory, and define corresponding semantic labels and related attributes. The semantic labels will be used to perform privacy semantic encoding on the data elements.

[0079] Step S103: mark each data element in the patient's TCM medical record data according to the semantic classification framework, wherein for each data element, find the category to which the data element belongs in the semantic classification framework, assign the corresponding semantic label to the data element, and then add privacy weight information to each data element based on the marked semantic label. The privacy weight information is determined based on the importance of the data element in privacy protection, thereby recombining the data elements marked with semantic labels and privacy weights to form a preliminary data structure after privacy semantic encoding.

[0080] Step S104, extract various health data patterns from the preliminary data structure after privacy semantic encoding, perform feature description on each health data pattern, and establish a health data pattern library. The feature description includes the combination of data elements constituting the health data pattern, the value range or characteristic value of each data element, and the logical relationship between each data element.

[0081] Step S105, for each data subset in the preliminary data structure after privacy semantic encoding, each data subset is matched with various health data patterns in the health data pattern library, and the data subset is a data part divided according to different diagnostic stages or body systems.

[0082] Step S106, for the successfully matched data subset, mark the name or number of the health data pattern matched by the successfully matched data subset, record the degree of similarity when the match is successful, and generate a health data pattern matching result. The degree of similarity is quantitatively evaluated based on the proportion of data element matching and the matching of key data elements. In addition, the data subset that does not match any health data pattern is specially marked to indicate that the data subset is an abnormal data subset or a new unidentified health data pattern.

[0083] Step S107, integrate the preliminary data structure of the privacy semantic encoding with the health data pattern matching result. Specifically, for each data element, associate the semantic label, privacy weight information, matched health data pattern and similarity degree information of the data element, and build a corresponding nested structure based on the association result. The outer structure of the nested structure contains the semantic category information of the data element, and the inner structure contains the data element itself, the privacy weight, and the pattern matching result information.

[0084] Step S108: performing privacy protection preprocessing on the patient's TCM medical record data based on the nested structure.

[0085] In this embodiment, first, the server performs structural analysis on the patient's TCM condition record data in the patient's TCM condition record data set. Taking the patient data of a large TCM medical institution as an example, these patient TCM condition record data contain rich information. The server will carefully comb through these data to determine the different types of data elements contained therein and the relationship structure between each data element. For example, the patient's basic information (name, age, gender, etc.), TCM diagnosis information (pulse, tongue, qi and blood status, etc.), symptom description (headache, fatigue, cough, etc.) and treatment process (medication status, acupuncture points, etc.) are all different types of data elements. There are complex relationship structures between these data elements. For example, age may be related to the probability of onset of certain diseases, pulse and tongue may jointly reflect the functional state of the patient's internal organs, and medication status will be adjusted according to the diagnosis results and symptoms.

[0086] Next, based on the prior knowledge in the field of TCM health management and the requirements for data and privacy protection, the server constructs a semantic classification framework. This framework classifies each data element according to the semantic categories in TCM theory. For example, for pulse data elements, they are classified according to the meanings represented by different pulses in TCM theory. For example, string pulses may be classified as pulse categories related to liver function, and their semantic labels are defined as "liver pulse related-string pulses", and related attributes such as characteristic descriptions of pulses (straight and long, such as pressing strings, etc.) are set. For headaches in symptom descriptions, they may be classified according to the location and nature of the pain, such as "head yangming meridian headache-forehead pain" as a semantic label, and related attributes include the frequency of pain attacks, accompanying symptoms, etc. These semantic labels will be used to encode privacy semantics for data elements.

[0087] Then, according to the constructed semantic classification framework, the server marks each data element in the patient's TCM medical record data. For each data element, the server will find the category to which the data element belongs in the semantic classification framework and assign it a corresponding semantic label. After that, on the basis of marking the semantic label, privacy weight information is added to each data element. This privacy weight information is determined based on the importance of the data element in privacy protection. For example, basic information such as the patient's name and ID number has a higher privacy weight, because once this information is leaked, it may directly lead to the patient's privacy exposure, so it may be assigned a higher value, such as 0.9; while some general symptom descriptions, such as occasional mild coughs, have relatively low privacy weights and may be assigned 0.3. In this way, the data elements marked with semantic labels and privacy weights are recombined to form a preliminary data structure after privacy semantic encoding.

[0088] From the preliminary data structure after privacy semantic encoding, the server begins to extract various health data patterns. Taking a certain type of common disease (such as a cold) as an example, there may be a health data pattern whose characteristic description includes a combination of data elements such as a floating pulse, a thin and white tongue, and symptoms such as fever, headache, and cough. The value range of the floating pulse is 90-100 pulses per minute, and the characteristic value of the thin and white tongue is a thin and white tongue coating. The logical relationship between the data elements is that the floating pulse, the thin and white tongue, and the symptoms such as fever, headache, and cough have a causal relationship or a concomitant relationship. The server performs such a detailed characteristic description for each identified health data pattern, thereby establishing a health data pattern library.

[0089] For each data subset in the preliminary data structure after privacy semantic encoding (these data subsets are data parts divided according to different diagnostic stages or body systems, for example, symptoms, diagnosis and treatment data related to the respiratory system are divided into one data subset, and data related to the digestive system are divided into another data subset, etc.), the server matches them with various health data patterns in the health data pattern library. For the data subset related to the respiratory system, it contains data elements such as pulse, tongue, cough, and respiratory rate. The server compares these data elements with various patterns in the health data pattern library. If the pulse of one of the data subsets is floating, the tongue is thin and white, the cough is frequent, and the respiratory rate is slightly accelerated, it matches the previously established cold health data pattern.

[0090] For the successfully matched data subset, the server marks the name or number of the health data pattern matched by the successfully matched data subset (such as marked as "Cold-Pattern 1"), and records the degree of similarity when the match is successful. This degree of similarity is quantitatively evaluated based on the proportion of data element matching and the matching of key data elements. For example, if the three key data elements of pulse, tongue, and cough in the data subset are completely matched with the cold health data pattern, and the change in respiratory rate is slightly different from the standard value in the pattern, the similarity degree may be calculated to be 0.8. For data subsets that do not match any health data pattern, the server makes a special mark to indicate that the data subset is an abnormal data subset or a new unidentified health data pattern. For example, a data subset contains a special pulse and a combination of symptoms that have never been recorded. The server marks it as a new unidentified health data pattern for further research.

[0091] Finally, the server integrates the preliminary data structure of privacy semantic encoding with the health data pattern matching results. Specifically, for each data element, the semantic label, privacy weight information, matched health data pattern and similarity degree information of the data element are associated, and the corresponding nested structure is constructed based on the association results. For example, for a certain data element (such as pulse floating number), the outer structure contains its semantic category information (pulse - pulse related to exogenous infection), and the inner structure contains the data element itself (floating number), privacy weight (0.3), and pattern matching result information (matched to cold - pattern 1, similarity degree is 0.8). Through such a nested structure, the server can better perform privacy protection preprocessing on the patient's TCM condition record data. For example, in the subsequent data processing process, if data sharing or analysis is involved, the server can decide how to process the data based on information such as privacy weight and pattern matching results, which can not only meet the needs of data analysis, but also effectively protect the patient's privacy. At the same time, this structure also helps to more accurately identify and analyze condition data, laying a solid foundation for subsequent operations such as obtaining relevant condition characteristic data from the patient's TCM condition record data set. During the entire process, the server uses a rigorous data processing process to achieve effective management, privacy protection, and data mining preparation of TCM medical record data.

[0092] Figure 2 The hardware structure of the system 100 for dynamically monitoring a TCM condition based on big data time series analysis provided by the embodiment of the present invention for implementing the above-mentioned method for dynamically monitoring a TCM condition based on big data time series analysis is shown. Figure 2 As shown, the TCM disease dynamic monitoring system 100 based on big data time series analysis may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0093] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the TCM disease dynamic monitoring system based on big data time series analysis 100 uses to execute or use to complete the exemplary method described in the present invention.

[0094] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the method for dynamic monitoring of traditional Chinese medicine diseases based on big data time series analysis in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0095] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned traditional Chinese medicine disease dynamic monitoring system 100 based on big data time series analysis. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0096] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned method for dynamic monitoring of traditional Chinese medicine conditions based on big data time series analysis is implemented.

[0097] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A method for dynamic monitoring of TCM disease based on big data time series analysis, characterized in that: The method comprises: Obtaining a past sample condition characteristic data set, a sample patient's symptoms at this node, and a predicted sample patient's symptoms from a patient's TCM condition record data set; the patient's TCM condition record data set includes a plurality of sample patient's symptoms organized based on a time series, the past sample condition characteristic data set includes a plurality of sample patient's symptoms before a target time node, the sample patient's symptoms at this node are sample patient's symptoms corresponding to the target time node, and the predicted sample patient's symptoms are sample patient's symptoms after the target time node; Loading the past sample disease characteristic data set and the symptoms exhibited by the sample patient at this node into the disease dynamic development prediction network, and determining the first confidence level of the symptoms exhibited by the sample patient at this node appearing at the target time node; Loading the past sample disease characteristic data set and the predicted sample patient symptoms into the disease dynamic development prediction network, and determining a second confidence level that the predicted sample patient symptoms will appear after the target time node; Based on the loss between the labeled confidence of the symptoms exhibited by the sample patient of this node and the first confidence, the dynamic disease progression prediction network is optimized, and based on the loss between the labeled confidence of the symptoms exhibited by the predicted sample patient and the second confidence, the dynamic disease progression prediction network is optimized.

2. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 1 is characterized in that: The optimizing the disease dynamic development prediction network based on the loss between the labeled confidence of the predicted sample patient's symptoms and the second confidence includes: Obtaining first training supervision data and second training supervision data of the symptoms exhibited by the predicted sample patient, wherein the first training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is 1, and the second training supervision data is used to characterize that the ratio of positive sample labels of the symptoms exhibited by the predicted sample patient is x, where x is not less than 0 and x is not greater than 1; Based on the first training supervision data and the second training supervision data, generating a labeling confidence of the predicted sample patient's symptoms; The disease dynamics development prediction network is optimized based on the loss between the labeled confidence of the predicted sample patient's symptoms and the second confidence.

3. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 2 is characterized in that: The disease dynamics development prediction network is a taught disease dynamics development prediction network; obtaining second training supervision data of symptoms exhibited by the predicted sample patient includes: The past sample disease characteristic data set and the predicted sample patient symptoms are loaded into the teaching disease dynamic development prediction network, and a third confidence level that the predicted sample patient symptoms will appear after the target time node is determined; the network parameter amount of the teaching disease dynamic development prediction network is not less than the network parameter amount of the taught disease dynamic development prediction network; The third confidence level is used as second training supervision data for predicting symptoms exhibited by the sample patient.

4. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 3 is characterized in that: The taught disease dynamic development prediction network is the taught disease dynamic development prediction network in the current round of network parameter optimization stage, and the teaching disease dynamic development prediction network is the teaching disease dynamic development prediction network in the current round of network parameter optimization stage; the method further includes: Based on the taught disease dynamic development prediction network of the previous network parameter optimization stage and the taught disease dynamic development prediction network of the current network parameter optimization stage, the taught disease dynamic development prediction network of the current network parameter optimization stage is generated.

5. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 4 is characterized in that: The method of generating the teaching disease dynamic development prediction network of the current round of network parameter optimization stage based on the teaching disease dynamic development prediction network of the previous network parameter optimization stage and the taught disease dynamic development prediction network of the current round of network parameter optimization stage comprises: The neuron weight information of the teaching disease dynamic development prediction network in the previous network parameter optimization stage is fused with the first training degradation factor to generate the first parameter information; The neuron weight information of the trained disease dynamic development prediction network in the current round of network parameter optimization is fused with the second training degradation factor to generate second parameter information; the sum of the first training degradation factor and the second training degradation factor is 1; The first parameter information and the second parameter information are integrated to generate neuron weight information of the disease dynamic development prediction network in the current round of network parameter optimization stage.

6. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 2 is characterized in that: The step of generating the annotation confidence of the predicted sample patient's symptoms based on the first training supervision data and the second training supervision data includes: The second training supervision data is fused with the first influencing factor to generate third training supervision data; The first training supervision data is fused with the second influencing factor to generate fourth training supervision data; the sum of the first influencing factor and the second influencing factor is 1; The third training supervision data and the fourth training supervision data are fused to generate the annotation confidence of the predicted sample patient's symptoms.

7. The method for dynamic monitoring of TCM disease based on big data time series analysis according to claim 6 is characterized in that: The method further comprises: Determine a reference influencing factor that matches the second training supervision data from a plurality of candidate first influencing factors; and use the reference influencing factor as the first influencing factor of the second training supervision data.

8. The method for dynamic monitoring of TCM disease based on big data time series analysis according to any one of claims 1 to 7, characterized in that: Obtaining the symptoms of the predicted sample patient includes: The predicted sample patient symptoms are randomly extracted from the sample patient symptoms after the sample patient symptoms at this node in the patient's TCM medical record data set.

9. The method for dynamic monitoring of TCM disease based on big data time series analysis according to any one of claims 1 to 7, characterized in that: The method further comprises: Based on the optimized disease dynamic development prediction network, any input target disease characteristic data set and target patient symptoms are analyzed to predict the target confidence level of the time node corresponding to the target patient symptoms.

10. A dynamic monitoring system for TCM illness based on big data time series analysis, characterized in that: The system for dynamic monitoring of TCM diseases based on big data time series analysis includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the method for dynamic monitoring of TCM diseases based on big data time series analysis as described in any one of claims 1 to 9.

Images