A combined modeling impedance prediction method and system

Abstract

This invention relates to the field of combinatorial modeling technology, specifically to a method and system for impedance prediction using combinatorial modeling. The method includes the following steps: detecting impedance, current, and temperature parameters; comparing impedance measurements with thresholds to generate feature classifications; inputting similar parameter pairs into a Pearson algorithm to construct a correlation matrix; selecting high-scoring feature combinations based on the matrix; optimizing the structure using a K-means model; and assigning values ​​to the parameters with the highest scores after recombination, thus completing the initialization of the prediction sub-model parameters. In this invention, by systematically quantifying the correlation of multi-dimensional feature parameters such as impedance, current, and temperature, the coupling relationships between parameters are revealed. Based on data-driven dynamic recombination of the prediction model structure, key parameter combinations are automatically selected, and the optimal parameter set is used to accurately initialize the prediction sub-model. This eliminates the dependence on fixed equivalent circuit models, enhances model adaptability and prediction reliability, and provides a flexible and solid basis for characterizing the impedance characteristics of complex systems.

Description

Technical Field

[0001] This invention relates to the field of combined modeling technology, and in particular to an impedance prediction method and system for combined modeling. Background Technology

[0002] The field of combined modeling technology involves the organic integration of multiple mathematical or physical models. By establishing a model system within a unified framework, it aims to characterize and analyze the characteristic parameters of complex objects. Core aspects of this technology include the selection and combination of model structures, the determination and adjustment of parameters, the integration of data-driven and mechanistic analysis, and the evaluation of the applicability and reliability of model results under different operating conditions. Overall, this field has formed a systematic modeling method based on multi-model collaboration, widely applied to electrical systems, electrochemical systems, and other engineering objects with complex dynamic characteristics. Traditional combined modeling impedance prediction methods refer to predicting impedance characteristics of electrochemical cells or power components by acquiring frequency response data and fitting equivalent circuit parameters. Common methods include extracting parameters of capacitive and resistive elements using equivalent circuit methods, analyzing frequency response signal characteristics using spectral analysis, and combining numerical fitting techniques to obtain the impedance distribution and variation patterns under different operating conditions.

[0003] Traditional impedance prediction methods based on combinatorial modeling rely heavily on pre-defined equivalent circuit structures. This fixed model framework is difficult to dynamically adjust its representation when faced with nonlinear changes in the internal state of electrical or electrochemical systems. For example, when a battery develops new internal reaction pathways due to aging, the original fixed component combination cannot accurately reflect the real changes in its impedance characteristics, leading to ambiguity in the physical meaning of model parameters. Consequently, the prediction accuracy decreases significantly under a wide range of operating conditions or long-term operation. The fundamental reason is that the model lacks a deep mining and adaptive learning mechanism for the intrinsic correlation of multi-dimensional features in measured data, which limits the universality and reliability of the prediction. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a combined modeling impedance prediction method and system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an impedance prediction method based on combined modeling, comprising the following steps:

[0006] S1: By detecting the impedance-related characteristic parameters of the target group, the impedance measurement value, current parameter and temperature parameter are obtained, and the obtained impedance measurement value is compared with the set threshold range. The comparison result between the impedance measurement value and the set threshold is called to generate the feature classification result.

[0007] S2: By collecting a set of impedance-related feature parameters, calling the feature classification results, comparing pairs of impedance-related feature parameters belonging to the same category, and inputting them into the Pearson correlation coefficient algorithm for correlation calculation, and constructing an impedance feature correlation matrix;

[0008] S3: Based on the impedance feature correlation matrix, select impedance feature combinations with scores exceeding the specified standard, call the selected impedance feature combinations, recombine the structural parameter group according to feature correlation, input it into the K-means clustering model for structural unit optimization, and output the structural reorganization result;

[0009] S4: By calling the structural reorganization result, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and perform parameter assignment operations on this parameter and other parameters in the same group in sequence. Use the assigned parameter set as the parameter initialization input for the prediction sub-model to complete the impedance prediction parameter initialization.

[0010] As a further aspect of the present invention, the feature classification result specifically includes category boundaries, abnormal intervals, and stable intervals; the impedance feature correlation matrix includes the linear strength between parameters, correlation direction, and distribution differences; the structural reorganization result specifically refers to the structural unit division, parameter combination mode, and clustering distribution state; and the impedance prediction parameter initialization includes the core parameter set, weight allocation, and initial threshold.

[0011] As a further aspect of the present invention, the step of obtaining the feature classification result specifically includes:

[0012] S101: Detect impedance-related characteristic parameters of the target group, simultaneously acquire impedance measurement values, current parameters and temperature parameters, and arrange and combine the impedance measurement values, current parameters and temperature parameters as independent dimension data points in a structured manner to establish a multidimensional physiological data record;

[0013] S102: Call the impedance measurement value in the multidimensional physiological data record, obtain the preset impedance reference threshold range, compare the impedance measurement value with the upper limit and lower limit of the impedance reference threshold range one by one, calculate the difference between the impedance measurement value and the closest threshold boundary, and obtain the impedance threshold deviation.

[0014] S103: For the impedance threshold deviation, based on its positive or negative value and amplitude, compare it with the deviation interval corresponding to each category in the classification standard library, perform interval assignment judgment, match a unique category identifier for the current target group, and generate feature classification results.

[0015] As a further aspect of the present invention, the step of obtaining the impedance characteristic correlation matrix specifically includes:

[0016] S201: Call the feature classification result, and for the set of impedance-related feature parameters belonging to the same category identifier, perform a pairwise pairing operation to generate a sequence of feature parameter pairs to be calculated;

[0017] S202: Based on the sequence of feature parameter pairs to be calculated, the two sets of data sequences in each parameter pair are sequentially input into the Pearson correlation coefficient algorithm to calculate the linear strength between the parameters and the direction of the correlation, and generate preliminary correlation data.

[0018] S203: For each parameter pair in the sequence of feature parameters to be calculated, calculate the discreteness and distribution pattern of the data points within it, and obtain the distribution difference measure between parameters.

[0019] S204: Based on the preliminary correlation data and the distribution difference measure between the parameters, the linear strength between the parameters, the correlation direction, and the distribution difference are integrated into structured data to construct the impedance feature correlation matrix.

[0020] As a further aspect of the present invention, the step of obtaining the structural reorganization result specifically includes:

[0021] S301: Based on the impedance feature correlation matrix, extract the correlation score corresponding to each group of impedance feature combinations in the matrix, call the preset feature screening standard value, compare the numerical value of each group of correlation scores with the feature screening standard value, select all impedance feature combinations with scores exceeding the feature screening standard value and set them together to generate a high correlation impedance feature set.

[0022] S302: Call the highly correlated impedance feature set, and based on the inherent correlation of each feature, retrieve and match the structural parameters directly corresponding to the features in the original structural parameter set. Recombine and classify the retrieved structural parameters to form a parameter set for the target structural unit and obtain the data cluster of the structural unit to be optimized.

[0023] S303: For the data cluster of structural units to be optimized, initialize K cluster centers, iteratively calculate the Euclidean distance from each structural unit parameter point in the data cluster to each cluster center, and assign each structural unit parameter point to the category of the nearest cluster center. Iteratively update the position of the cluster center point in each category until the cluster center point no longer shifts, and obtain the structural reorganization result.

[0024] As a further aspect of the present invention, the step of obtaining the impedance prediction parameter initialization is specifically as follows:

[0025] S401: Call the structural reorganization results and, based on the impedance characteristic correlation matrix, traverse each reorganized structural parameter group, select the parameter with the highest impedance correlation score, and determine it as the core prediction parameter;

[0026] S402: For the core prediction parameter, calculate its statistical mean in the historical data of the target group, and assign the statistical mean as the initial value to obtain the initial value of the core parameter;

[0027] S403: Based on the initial value of the core parameter, and by calling the correlation scores of other parameters in the same group with the core prediction parameter in the impedance feature correlation matrix, the other parameters in the same group are assigned values ​​proportionally to generate a weighted parameter set.

[0028] S404: Integrate the weighted parameter set, set an initial prediction error threshold, establish the core parameter set, the weight allocation and the initial threshold, and complete the impedance prediction parameter initialization.

[0029] As a further aspect of the present invention, the linear strength between the parameters in the preliminary correlation data is calculated using the following formula:

[0030] ;

[0031] in, The linearity between the parameters is given by n, where n is the total number of data points in the data sequence. For the i-th data point in the first group of data sequences in the sequence to be calculated, For the i-th data point of the second group of data sequences in the sequence to be calculated, The arithmetic mean of all data points in the first set of data sequences is... It is the arithmetic mean of all data points in the second set of data sequences.

[0032] As a further aspect of the present invention, in the step of continuously iteratively updating the position of the cluster center point for each category, the position of the new cluster center point is determined by the following formula:

[0033] ;

[0034] in, These are the updated coordinates of the k-th cluster center. This is the set of all structural unit parameter points that are assigned to the k-th cluster in the current iteration round. For the set The total number of structural element parameter points contained therein For the set The position coordinates of the j-th structural unit parameter point.

[0035] As a further aspect of the present invention, the operation of assigning values ​​to other parameters within the same group proportionally specifically involves:

[0036] First, obtain the initial values ​​of the core parameters and the corresponding correlation scores in the impedance feature correlation matrix;

[0037] Secondly, for any non-core parameter within the structural parameter group, extract its correlation score with the core prediction parameter;

[0038] Then, the initial value of the core parameter is multiplied by the correlation score between the non-core parameter and the core prediction parameter, and the calculation result is used as the baseline value of the non-core parameter.

[0039] Finally, the baseline assignment is adjusted using a parameter adjustment factor to generate the final assignment. This parameter adjustment factor is calculated using the formula... Calculated;

[0040] in, This represents the parameter adjustment factor. This represents a preset scaling weight used to control the adjustment magnitude. This represents the standard deviation of the non-core parameter in the historical data of the target group. The value represents the standard deviation of the core prediction parameter in the historical data of the target group. This adjusted value is the final value of the non-core parameter in the weighted parameter set.

[0041] An impedance prediction system based on combined modeling, the system being used to implement the aforementioned impedance prediction method based on combined modeling, the system comprising:

[0042] The feature parameter classification module is used to obtain the impedance measurement value, current parameter and temperature parameter of the target group by detection, call the obtained impedance measurement value and perform comparison and judgment with the preset threshold range, generate feature classification result, and pass it to the feature correlation calculation module;

[0043] The feature correlation calculation module is used to call the feature classification results, pair impedance-related feature parameters belonging to the same category, and input the paired parameter groups into the Pearson correlation coefficient algorithm for correlation calculation, construct the impedance feature correlation matrix, and pass it to the structural unit optimization module.

[0044] The structural unit optimization module is used to select impedance feature combinations with correlation scores exceeding a specified standard based on the impedance feature correlation matrix, call the structural parameter group corresponding to the selected impedance feature combination, input the structural parameter group into the K-means clustering model for structural optimization, output the structural reorganization result, and pass it to the prediction parameter initialization module.

[0045] The prediction parameter initialization module is used to call the structural reorganization result, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and perform parameter assignment operations on this parameter and other parameters in the same group in sequence. The assigned parameter set is used as the initialization parameter input of the prediction sub-model to complete the impedance prediction parameter initialization.

[0046] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0047] In this invention, by systematically correlating and quantifying the multidimensional characteristic parameters of impedance, current, and temperature within the target group, the intrinsic coupling relationship between the parameters is deeply revealed. Then, based on this data-driven correlation, the structural units of the prediction model are dynamically reorganized, enabling the automatic selection and establishment of key parameter combinations from massive features. The optimal parameter set is then used to accurately initialize the prediction sub-model. This approach eliminates the dependence on fixed equivalent circuit models, significantly enhances the model's adaptability to changes in different operating conditions, and improves the reliability of prediction results. It provides a more flexible and solid basis for the accurate characterization of the impedance characteristics of complex systems. Attached Figure Description

[0048] Figure 1 This is a flowchart of the impedance prediction method based on combined modeling of the present invention.

[0049] Figure 2 This is a flowchart of the feature classification result acquisition process of the present invention;

[0050] Figure 3 This is a flowchart illustrating the construction of the impedance characteristic correlation matrix of this invention.

[0051] Figure 4 This is a flowchart of the process for obtaining the structural recombination results of the present invention;

[0052] Figure 5 This is a flowchart of the impedance prediction parameter initialization process for this invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0054] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically stated.

[0055] Please see Figure 1 and Figure 2 This invention provides a technical solution: an impedance prediction method based on combined modeling, comprising the following steps:

[0056] S1: By detecting the impedance-related characteristic parameters of the target group, the impedance measurement value, current parameter and temperature parameter are obtained, and the obtained impedance measurement value is compared with the set threshold range. The comparison result between the impedance measurement value and the set threshold is called to generate the feature classification result.

[0057] The feature classification results specifically include category boundaries, abnormal intervals, and stable intervals;

[0058] The specific steps for obtaining the feature classification results are as follows: S101: Detect the impedance-related feature parameters of the target group, simultaneously acquire impedance measurement values, current parameters and temperature parameters, and arrange and combine the impedance measurement values, current parameters and temperature parameters as independent dimension data points in a structured manner to establish a multidimensional physiological data record;

[0059] S102: Call the impedance measurement value from the multidimensional physiological data record and obtain the preset impedance reference threshold range. Compare the impedance measurement value with the upper and lower limits of the impedance reference threshold range one by one, calculate the difference between the impedance measurement value and the closest threshold boundary, and obtain the impedance threshold deviation.

[0060] S103: For impedance threshold deviation, based on its positive or negative value and amplitude, compare it with the deviation interval corresponding to each category in the classification standard library, perform interval assignment judgment, match a unique category identifier for the current target group, and generate feature classification results.

[0061] The goal of this step is to obtain impedance-related characteristic parameters of the target group and generate feature classification results based on preset thresholds. In a specific implementation scenario, taking the monitoring of physiological changes in a specific muscle group (e.g., the quadriceps) after rehabilitation training as an example, the target group consists of five athletes undergoing rehabilitation training. An IMP-SFB7 multi-frequency bioelectrical impedance analyzer was used to measure each athlete 15 minutes after training in a constant room temperature (25°C) and humidity (50%RH) environment. During the test, four electrodes were placed along the muscle fiber direction in the quadriceps region, and an 800 μA, 50 kHz sinusoidal alternating current was applied. The impedance measurement value, the actual current parameter flowing through the tissue, and the skin surface temperature parameter of each athlete were simultaneously detected and recorded. These three parameters were then structured and combined as independent data points to establish a multidimensional physiological data record. To determine the impedance benchmark threshold range for comparison, an independent benchmark experiment was conducted. The process of establishing the baseline group is as follows: 120 healthy male volunteers aged 20-30 years were recruited. The exclusion criteria included: (1) a history of lower limb muscle injury within the past three months;

[0062] (2) Suffering from a systemic disease that affects fluid balance;

[0063] (3) Currently taking medications that may affect impedance measurement. All participants were prohibited from strenuous exercise and alcohol consumption within 24 hours prior to the test. One hundred eligible volunteers were selected, and their basic physiological parameters, such as height and weight, were recorded to ensure the reproducibility of the experiment. Under measurement conditions identical to the above implementation scenario, the quadriceps femoris of each baseline individual was measured. After collecting all 100 impedance data points, their statistical distribution was calculated. The mean of the baseline data was calculated to be 500.0 ohms (Ω), and the standard deviation was 10.0 ohms (Ω). Based on the principle of normal distribution, a 95% confidence interval was selected as the impedance baseline threshold range under stable physiological conditions. The upper limit of this interval is 500.0 + 1.96 * 10.0 ≈ 519.6 ohms (Ω), and the lower limit is 500.0 - 1.96 * 10.0 ≈ 480.4 ohms (Ω). Therefore, the impedance baseline threshold range was set as [480.4 Ω, 519.6 Ω]. After acquiring multidimensional physiological data records from five athletes, the impedance measurements were retrieved and compared with the pre-defined impedance benchmark threshold range [480.4Ω, 519.6Ω] to calculate the impedance threshold deviation. Specifically, each athlete's impedance measurement was compared with the upper limit (519.6Ω) and lower limit (480.4Ω) of the threshold range, and the difference between this value and the nearest boundary value was taken.

[0064] Table 1. Athlete Physiological Data and Deviation Calculation Table

[0065]

[0066] Table 1 lists the initial measurement data and calculated impedance threshold deviations for five athletes. For example, for athlete 001, the measured impedance value is 478.2Ω, which is closer to the lower limit of 480.4Ω, so its deviation is 478.2 - 480.4 = -2.2Ω. For athlete 004, the measured impedance value is 495.3Ω, which is within the threshold range. The difference between it and the lower limit of 480.4Ω is 14.9Ω, and the difference between it and the upper limit of 519.6Ω is -24.3Ω. Therefore, the smaller absolute value of the difference, 14.9Ω, is chosen as its deviation.

[0067] Next, the data were classified based on the degree of deviation. The classification standard library was established based on the statistical analysis of the deviation data of 100 volunteers in the baseline group, combined with the interpretation results of muscle status by clinical experts. Statistical results showed that 99% of healthy individuals had an absolute deviation value of less than 10.0Ω, so the interval with an absolute deviation value greater than 10.0Ω was defined as the "abnormal interval". At the same time, the data point distribution density was the highest and the corresponding physiological state was the most stable in the interval between the absolute deviation values ​​of [0Ω, 5.0Ω], so this interval was defined as the "category boundary".

[0068] A deviation with an absolute value greater than 5.0Ω and less than or equal to 10.0Ω, and an original impedance measurement within the threshold range of [480.4Ω, 519.6Ω], is defined as a "stable interval". Based on this classification standard, the deviations of five athletes were assigned to intervals: Athlete 001's deviation was -2.2Ω, and its absolute value of 2.2 falls within the range of [0Ω, 5.0Ω], matching the "category boundary" identifier. Athlete 002 has a deviation of -9.1Ω, with an absolute value of 9.1 within the range of (5.0Ω, 10.0Ω), and its original value of 510.5Ω within the range of [480.4Ω, 519.6Ω], thus matching the "stable interval" label. Athlete 003 has a deviation of 6.2Ω, with an absolute value of 6.2 within the range of (5.0Ω, 10.0Ω), but its original value of 525.8Ω exceeds the upper limit of the threshold, therefore matching the "abnormal interval" label. Athlete 004 has a deviation of 14.9Ω, with an absolute value of 14.9... The deviation is 10.0Ω, but the original value of 495.3Ω is within the threshold range. This deviation exceeds the definition of the stable interval, but the original value is within the normal range. It is classified as a non-boundary region of the "stable interval" and matches the "stable interval" label. The deviation of athlete 005 is -10.3Ω, and its absolute value of 10.3 is greater than 10.0Ω, matching the "abnormal interval" label. Through the above judgment, a unique category label is matched for the multidimensional physiological data record of each athlete. The final feature classification result is: Athlete 001 - Category Boundary;

[0069] Athlete 002 - Stable Range;

[0070] Athlete 003 - Abnormal range;

[0071] Athlete 004 - Stable Range;

[0072] Athlete 005 - Abnormal range.

[0073] Please see Figure 1 and Figure 3 S2: By collecting a set of impedance-related feature parameters, calling the feature classification results, comparing pairs of impedance-related feature parameters belonging to the same category, and inputting them into the Pearson correlation coefficient algorithm for correlation calculation, an impedance feature correlation matrix is ​​constructed.

[0074] The impedance characteristic correlation matrix includes the linearity between parameters, the direction of correlation, and the differences in distribution.

[0075] The steps for obtaining the impedance feature correlation matrix are as follows: S201: Call the feature classification results, perform pairwise pairing operations for the impedance-related feature parameter sets belonging to the same category identifier, and generate a sequence of feature parameter pairs to be calculated;

[0076] S202: Based on the sequence of feature parameter pairs to be calculated, the two sets of data sequences in each parameter pair are sequentially input into the Pearson correlation coefficient algorithm to calculate the linear strength and correlation direction between the parameters and generate preliminary correlation data.

[0077] S203: For each parameter pair in the sequence of feature parameter pairs to be calculated, calculate the degree of dispersion and distribution of the data points within it, and obtain the measure of the distribution difference between parameters.

[0078] S204: Based on the preliminary correlation data and the distribution difference measure between parameters, integrate the linear strength, correlation direction and distribution difference between parameters into structured data and construct the impedance characteristic correlation matrix;

[0079] The linear strength between parameters in the preliminary correlation data is calculated using the following formula: ;

[0080] in, Let n be the linear strength between parameters, and n be the total number of data points in the data sequence. For the i-th data point in the first set of data in the sequence, the feature parameters to be calculated are... For the i-th data point in the second set of data in the sequence, where the feature parameters to be calculated are... This is the arithmetic mean of all data points in the first data sequence. This is the arithmetic mean of all data points in the second data sequence.

[0081] The goal of this step is to construct an impedance feature correlation matrix based on the feature classification results generated in S1. Specifically, the classification results of S1 are called, and all data records assigned the "stable interval" category (i.e., the data of athletes 002 and 004) are selected. Additionally, three historical monitoring data records from the same target group, classified as "stable interval" under similar measurement conditions and recorded before the implementation of this invention, are retrieved and merged with the two currently acquired data records to form a set of impedance-related feature parameters containing five sets of data. This set is used for subsequent correlation calculations.

[0082] Table 2. Set of characteristic parameters under the "Stability Interval" category

[0083]

[0084] As shown in Table 2, this set includes three dimensions: impedance measurements (Z), current parameters (I), and temperature parameters (T). Next, a pairwise, non-repeating pairing operation is performed on these three feature parameters to generate a sequence of feature parameter pairs to be calculated, namely (Z,I), (Z,T), and (I,T). Subsequently, the two sets of data sequences in each parameter pair are sequentially input into the Pearson correlation coefficient algorithm to calculate the linear strength and correlation direction between the parameters.

[0085] This section calculates the linearity between the measured impedance (Z) and the temperature parameter (T). Taking this as an example, the calculation process will be explained in detail. Formula This is used here to calculate the linear correlation between two sets of data sequences. Let X be the linear strength between parameters X and Y, and n be the total number of data points in the data sequence. For the i-th data point in the first set of data in the sequence, the feature parameters to be calculated are... For the i-th data point in the second set of data in the sequence, where the feature parameters to be calculated are... This is the arithmetic mean of all data points in the first data sequence. This is the arithmetic mean of all data points in the second set of data. The numerator of the formula calculates the covariance of the two variables, while the denominator is the product of the standard deviations of the two variables. The entire formula, through standardization, quantifies the extent to which the two variables deviate synchronously from their respective means, thus revealing the strength and direction of the linear relationship between them.

[0086] First, prepare the data sequences for the calculation: Z sequence (X): [510.5, 495.3, 505.0, 490.1, 515.2] T sequence (Y): [37.0, 36.9, 36.9, 36.8, 37.1]

[0087] Next, calculate the arithmetic mean of the two sequences:

[0088] Next, the molecular part is calculated. :

[0089] Item 1:

[0090] Item 2:

[0091] Item 3:

[0092] Item 4:

[0093] Item 5:

[0094] Summation:

[0095] Then, calculate the two parts inside the square root in the denominator:

[0096]

[0097] Finally, calculate : This result indicates a highly linear positive correlation between impedance (Z) and temperature (T). In this embodiment, a high correlation is defined as a Pearson correlation coefficient with an absolute value in the range [0.8, 1.0]. Using the exact same calculation procedure, the linear strength of the other two pairs of parameters can be obtained: , These calculation results constitute preliminary correlation data.

[0098] To obtain a measure of the distributional difference between parameters, for each parameter pair, the dispersion (e.g., standard deviation or coefficient of variation) and distribution shape (e.g., skewness and kurtosis) of its internal data points are calculated. Taking the (Z,T) parameter pair as an example, the standard deviation of the Z sequence is... The standard deviation of the T sequence is The coefficients of variation (CVs) of the two sequences were calculated. The CV_Z coefficient of variation for the Z sequence was approximately 0.0185, and the CV_T coefficient of variation for the T sequence was approximately 0.102 / 36.94. The absolute value of the difference between the two CVs, |CV_Z - CV_T| = 0.0157, was used as a measure of the difference in the dispersion of the distributions between the quantified parameters.

[0099] Finally, the calculated linear strength (e.g., 0.933), correlation direction (positive correlation), and distribution difference measures (e.g., the difference in the coefficients of variation of Z and T) between parameters are integrated into structured data to construct an impedance characteristic correlation matrix. This matrix is ​​arranged with parameters as rows and columns, and each element in the matrix records the complete correlation information of the corresponding parameter pair. For example, the element at position (Z, T) in the matrix is ​​structured data containing {linear strength: 0.933, direction: positive, distribution difference: 0.016}.

[0100] Please see Figure 1 and Figure 4 S3: Based on the impedance feature correlation matrix, select impedance feature combinations with scores exceeding the specified standard, call the selected impedance feature combinations, recombine the structural parameter group according to feature correlation, input it into the K-means clustering model for structural unit optimization, and output the structural reorganization result;

[0101] The results of structural reorganization specifically refer to the division of structural units, parameter combination patterns, and cluster distribution.

[0102] The specific steps for obtaining the structural reorganization results are as follows: S301: Based on the impedance feature correlation matrix, extract the correlation score corresponding to each group of impedance feature combinations in the matrix, call the preset feature screening standard value, compare the numerical value of each group of correlation score with the feature screening standard value, select all impedance feature combinations whose scores exceed the feature screening standard value and set them together to generate a high correlation impedance feature set.

[0103] S302: Call the highly correlated impedance feature set, and based on the inherent correlation of each feature, retrieve and match the structural parameters directly corresponding to the features in the original structural parameter set. Recombine and classify the retrieved structural parameters to form a parameter set for the target structural unit and obtain the data cluster of the structural unit to be optimized.

[0104] S303: For the data cluster of structural units to be optimized, initialize K cluster centers, iteratively calculate the Euclidean distance from each structural unit parameter point in the data cluster to each cluster center point, and assign each structural unit parameter point to the category of the nearest cluster center point. Iterate and update the position of the cluster center point of each category until the cluster center point no longer shifts, and obtain the structural reorganization result.

[0105] In the step of iteratively updating the position of the cluster center point for each category, the position of the new cluster center point is determined by the following formula: ;

[0106] in, These are the updated coordinates of the k-th cluster center. This is the set of all structural unit parameter points that are assigned to the k-th cluster in the current iteration round. For set The total number of structural element parameter points contained therein For set The position coordinates of the j-th structural unit parameter point.

[0107] The goal of this step is to optimize the structural parameter set based on the impedance characteristic correlation matrix constructed by S2 and output the structural reorganization results. First, a feature selection criterion needs to be set to select strongly correlated impedance characteristic combinations from the matrix. This criterion is set based on: ranking and statistically analyzing the correlation coefficients of each parameter pair in the benchmark experiment, determining that the minimum correlation coefficient for the strongly correlated combinations ranked in the top 5% is 0.89. To select the most core feature combinations, the criterion is rounded down and set to 0.9.

[0108] The impedance feature correlation matrix generated by S2 is used to extract the correlation score (i.e., the absolute value of the Pearson correlation coefficient) corresponding to each impedance feature combination in the matrix. These scores are compared with the feature selection standard value of 0.9. The correlation score of the -(Z,I) combination is |0.852|=0.852, which is less than 0.9. The correlation score of the -(Z,T) combination is |0.933|=0.933, which is greater than 0.9. The correlation score of the -(I,T) combination is |0.795|=0.795, which is less than 0.9. By comparison, all impedance feature combinations with scores greater than 0.9 are selected. In this example, only the (Z,T) combination is selected. This combination is then grouped together to generate a high-correlation impedance feature set.

[0109] Next, the highly correlated impedance feature set {(Z,T)} is invoked, and based on the inherent correlation of the features, the original structural parameter sets directly corresponding to these features are retrieved and matched. In the application scenario of this invention, the original structural parameter sets are preset to assess muscle fatigue state, and this set includes {parameter set A: [intracellular fluid content, extracellular fluid content], parameter set B: [muscle fiber temperature, metabolite concentration], parameter set C: [nerve electrical signal intensity]}. The impedance measurement value (Z) mainly reflects the ratio and conductivity of intracellular and extracellular fluid, and therefore directly corresponds to "intracellular fluid content" and "extracellular fluid content" in parameter set A. The temperature parameter (T) directly corresponds to "muscle fiber temperature" in parameter set B. Therefore, the three retrieved structural parameters "intracellular fluid content", "extracellular fluid content", and "muscle fiber temperature" are recombined and classified to form a new parameter set oriented towards the target structural unit (here referring to a comprehensive unit reflecting hydration and thermal state), and the data cluster of the structural unit to be optimized is obtained.

[0110] For this data cluster of structural units to be optimized, a K-means clustering model was used for structural unit optimization, with the number of clusters set to K=2. Structural parameters corresponding to the highly correlated feature set {Z,T} were extracted from the historical monitoring data of the target group, forming a data cluster containing 20 samples. The specific values ​​for this data cluster are shown in Table 3. Initially, two data points were randomly selected as the initial cluster centers. The initial cluster centers were set as follows: C1_initial=(40.1,25.5,36.9) C2_initial=(38.2,27.8,37.1)

[0111] Table 3. Data clusters of structural units to be optimized (partial)

[0112]

[0113] The table above lists a portion of the sample data for the structural unit data clusters to be optimized.

[0114] Then, the iterative calculation process begins. The Euclidean distance from each structural unit parameter point within the data cluster (e.g., sample point P1=(41.0,25.0,37.0)) to the two cluster centers is calculated iteratively. distance Since 1.034 < 3.961, sample point P1 is assigned to the first cluster. The same calculation and assignment operation is performed on all 20 sample points in the data cluster.

[0115] After completing one round of assignment, the cluster center positions for each category are recalculated. Assume that after the first round of assignment, 12 points were assigned to category 1 and 8 points to category 2. At this point, the formula is applied... Update the cluster centers. Among them, These are the coordinates of the updated k-th cluster center. It is the set of all structural unit parameter points that are assigned to the k-th cluster in the current iteration round. It is a set The total number of structural element parameter points contained therein It is a set The position coordinates of the j-th structural unit parameter point. This formula finds the "centroid" of a category in multidimensional space by calculating the arithmetic mean of all data points within that category. This centroid, as the new cluster center, can better represent the overall position of the data points within that category.

[0116] Taking updating C1 as an example, assume the sum of the coordinates of the 12 points assigned to category 1 is (486.0, 300.0, 443.4). The coordinates of the new cluster center C1_new are: C1_new = (486.0 / 12, 300.0 / 12, 443.4 / 12) = (40.5, 25.0, 36.95). Similarly, calculate the new cluster center C2_new. Then, using C1_new and C2_new as the new center points, repeat the iterative process of distance calculation, sample point assignment, and center point update until, after two consecutive iterations, the positions of all cluster center points no longer change, or the change is less than a very small preset value (e.g., 1e-4). At this point, the clustering process converges, and the resulting structural unit is divided into two stable clusters. The centroid coordinates of each cluster are calculated. The centroid of cluster 1 has a value (e.g., 40.5) higher than the average level in the "intracellular fluid content" dimension, while its value (e.g., 36.95) in the "muscle fiber temperature" dimension is within the normal range. Therefore, it is identified as a "high hydration-normal temperature" state.

[0117] Similarly, the centroid of cluster 2 has a value below average in the "intracellular fluid content" dimension (e.g., 38.0) and a value slightly above the normal range in the "muscle fiber temperature" dimension (e.g., 37.2), thus it is identified as a "low hydration - slightly high temperature" state. These parameter combination patterns and the clustering distribution of these clusters in the three-dimensional parameter space together constitute the structural reorganization result.

[0118] Please see Figure 1 and Figure 5 S4: By calling the structural reorganization results, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and sequentially assign values ​​to this parameter and other parameters in the same group. Use the assigned parameter set as the initialization input for the prediction sub-model parameters to complete the impedance prediction parameter initialization. Impedance prediction parameter initialization includes the core parameter set, weight allocation, and initial threshold.

[0119] The specific steps for obtaining the impedance prediction parameter initialization are as follows: S401: Call the structural reorganization results, and based on the impedance characteristic correlation matrix, traverse each reorganized structural parameter group, select the parameter with the highest impedance correlation score, and determine it as the core prediction parameter;

[0120] S402: For the core prediction parameters, calculate their statistical mean in the historical data of the target group, and assign this statistical mean as the initial value to obtain the initial value of the core parameters;

[0121] S403: Based on the initial assignment of core parameters, and by calling the correlation scores of other parameters in the same group with the core prediction parameters in the impedance feature correlation matrix, assign values ​​to other parameters in the same group proportionally to generate a weighted parameter set.

[0122] S404: Integrate the weighted parameter set, set the initial prediction error threshold, establish the core parameter set, weight allocation and initial threshold, and complete the impedance prediction parameter initialization;

[0123] The operation of assigning values ​​to other parameters in the same group proportionally is as follows: First, obtain the correlation score between the initial value of the core parameter and the corresponding correlation matrix of impedance characteristics.

[0124] Secondly, for any non-core parameter in the structural parameter group, extract its correlation score with the core prediction parameter;

[0125] Then, the initial value of the core parameter is multiplied by the correlation score between the non-core parameter and the core prediction parameter, and the calculation result is used as the baseline value of the non-core parameter.

[0126] Finally, the baseline assignment is adjusted using a parameter adjustment factor to generate the final assignment. The parameter adjustment factor is calculated using the formula... Calculated;

[0127] in, Representative parameter adjustment factor, This represents a preset scaling weight used to control the adjustment magnitude. This represents the standard deviation of non-core parameters in the historical data of the target population. This represents the standard deviation of the core prediction parameter in the historical data of the target population. The adjusted value is the final value of the non-core parameter in the weighted parameter set.

[0128] The goal of this step is to call the structural reorganization results from S3 to initialize the impedance prediction parameters. First, the structural unit partitioning, parameter combination patterns, and cluster distribution states output from S3 are called. Based on the impedance feature correlation matrix constructed in S2, each reorganized structural parameter group (i.e., the two clusters ultimately formed in S3) is traversed. Within each group, the parameter with the highest correlation score to impedance (Z) is selected. According to the calculation results of S2, the correlation score (Z,T) is the highest (0.933). Therefore, in the cluster containing the parameter "muscle fiber temperature," "muscle fiber temperature" is determined as the core prediction parameter.

[0129] Next, for the core prediction parameter "muscle fiber temperature," its statistical mean in the historical data of the target population is calculated. Muscle fiber temperature data from 100 healthy adults in the baseline group of S1 are retrieved, and their arithmetic mean is calculated to be 36.9°C. This statistical mean of 36.9°C is used as the initial value for the core prediction parameter, resulting in the initial value of the core parameter.

[0130] Then, based on the initial values ​​of the core parameters, and by calling the correlation scores between other parameters in the same group and the core predicted parameter in the impedance feature correlation matrix, the other parameters in the same group are assigned values ​​proportionally. Assume that the cluster containing the core predicted parameter "muscle fiber temperature" also includes two non-core parameters: "intracellular fluid content" and "extracellular fluid content." The correlation scores between these two non-core parameters and the core parameter "muscle fiber temperature" need to be calculated. Using a method similar to that in S2, the correlation score between "muscle fiber temperature" and "intracellular fluid content" is calculated to be 0.75, and the correlation score between "muscle fiber temperature" and "extracellular fluid content" is -0.68.

[0131] The assignment operation is as follows: Multiply the initial core parameter value (36.9) by these two correlation scores respectively to obtain the baseline values. The baseline value for "Intracellular Fluid Content" = 36.9 * 0.75 = 27.675; the baseline value for "Extracellular Fluid Content" = 36.9 * (-0.68) = -25.092.

[0132] Subsequently, an adjustment factor is used. The baseline value is adjusted. The formula for calculating this adjustment factor is as follows: .in, It is an adjustment factor for non-core parameters. It is a preset scaling weight used to control the adjustment range. It is the standard deviation of non-core parameters in the historical data of the target group. This is the standard deviation of the core prediction parameter in the historical data of the target population. This adjusted assignment is the final assignment of this non-core parameter in the weighted parameter set.

[0133] To determine the weights A dataset containing 1000 historical samples was constructed and divided into a training set and a validation set in a 7:3 ratio. On the training set, [the following methods were used] respectively. The parameters are initialized using the formula {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0}, and a baseline linear regression model is trained. The root mean square error (RMSE) of each model is calculated on the validation set. The results show that when... When Ω=0.4, RMSE reaches its minimum value of 3.25Ω, therefore, the value is selected. The standard deviation is 0.4. This was calculated from historical data. The standard deviation of the non-core parameter "intracellular fluid content". (Unit quantity). Standard deviation of the non-core parameter "extracellular fluid content". (Unit quantity).

[0134] Calculate the adjustment factor for "intracellular fluid content" : = = = = = Its final assignment = baseline assignment

[0135] Calculate the adjustment factor for "extracellular fluid content" : Its final assignment = baseline assignment These adjusted assignments constitute the set of weighted parameters.

[0136] Finally, the weighted parameter set is integrated, and the initial prediction error threshold is set to 5%. This threshold is based on the reference [specific reference title, such as "Application and Challenges of Bioelectrical Impedance in Human Physiological Monitoring"], which indicates that the average relative error of similar bioelectrical impedance prediction models is between 4% and 6%. To balance the sensitivity and specificity of the prediction, this embodiment selects the median value of 5% as the initial prediction error threshold. The core parameter set is established by combining the core prediction parameter (muscle fiber temperature), non-core parameters (intracellular fluid content, extracellular fluid content) and their corresponding initial values ​​(36.9, 154.98, -105.3864).

[0137] The correlation score and adjustment factor used in the calculation process will be used as the basis for weight allocation;

[0138] A 5% error rate was used as the initial threshold. Through the above steps, the impedance prediction parameters, including the core parameter set, weight allocation, and initial threshold, were initialized.

[0139] An impedance prediction system based on combined modeling is provided for performing the aforementioned impedance prediction method based on combined modeling. The system includes:

[0140] The feature parameter classification module is used to obtain the impedance measurement value, current parameter and temperature parameter of the target group by detection, call the obtained impedance measurement value and perform comparison and judgment with the preset threshold range, generate feature classification results, and pass them to the feature correlation calculation module;

[0141] The feature correlation calculation module is used to call the feature classification results, pair impedance-related feature parameters belonging to the same category, and input the paired parameter groups into the Pearson correlation coefficient algorithm for correlation calculation, construct the impedance feature correlation matrix, and pass it to the structural unit optimization module.

[0142] The structural unit optimization module is used to select impedance feature combinations with correlation scores exceeding a specified standard based on the impedance feature correlation matrix, call the structural parameter group corresponding to the selected impedance feature combination, input the structural parameter group into the K-means clustering model for structural optimization, output the structural reorganization result, and pass it to the prediction parameter initialization module.

[0143] The prediction parameter initialization module is used to call the structural reorganization results, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and perform parameter assignment operations on this parameter and other parameters in the same group in sequence. The assigned parameter set is used as the initialization parameter input of the prediction sub-model to complete the impedance prediction parameter initialization.

[0144] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. An impedance prediction method based on combined modeling, characterized in that, Includes the following steps: S1: By detecting the impedance-related characteristic parameters of the target group, the impedance measurement value, current parameter and temperature parameter are obtained, and the obtained impedance measurement value is compared with the set threshold range. The comparison result between the impedance measurement value and the set threshold is called to generate the feature classification result. The specific steps for obtaining the feature classification results are as follows: S101: Detect impedance-related characteristic parameters of the target group, simultaneously acquire impedance measurement values, current parameters and temperature parameters, and arrange and combine the impedance measurement values, current parameters and temperature parameters as independent dimension data points in a structured manner to establish a multidimensional physiological data record; S102: Call the impedance measurement value in the multidimensional physiological data record, obtain the preset impedance reference threshold range, compare the impedance measurement value with the upper limit and lower limit of the impedance reference threshold range one by one, calculate the difference between the impedance measurement value and the closest threshold boundary, and obtain the impedance threshold deviation. S103: For the impedance threshold deviation, based on its positive or negative value and amplitude, compare it with the deviation interval corresponding to each category in the classification standard library, perform interval assignment judgment, match a unique category identifier for the current target group, and generate feature classification results; S2: By collecting a set of impedance-related feature parameters, calling the feature classification results, comparing pairs of impedance-related feature parameters belonging to the same category, and inputting them into the Pearson correlation coefficient algorithm for correlation calculation, and constructing an impedance feature correlation matrix; S3: Based on the impedance feature correlation matrix, select impedance feature combinations with scores exceeding the specified standard, call the selected impedance feature combinations, recombine the structural parameter group according to feature correlation, input it into the K-means clustering model for structural unit optimization, and output the structural reorganization result; S4: By calling the structural reorganization result, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and perform parameter assignment operations on the parameter and other parameters in the same group in sequence. Use the assigned parameter set as the parameter initialization input of the prediction sub-model to complete the impedance prediction parameter initialization. The feature classification results specifically include category boundaries, abnormal intervals, and stable intervals. The impedance feature correlation matrix includes the linear strength between parameters, correlation direction, and distribution differences. The structural reorganization results specifically refer to the structural unit division, parameter combination mode, and clustering distribution state. The impedance prediction parameter initialization includes the core parameter set, weight allocation, and initial threshold.

2. The impedance prediction method based on combined modeling according to claim 1, characterized in that, The specific steps for obtaining the impedance characteristic correlation matrix are as follows: S201: Call the feature classification result, and for the set of impedance-related feature parameters belonging to the same category identifier, perform a pairwise pairing operation to generate a sequence of feature parameter pairs to be calculated; S202: Based on the sequence of feature parameter pairs to be calculated, the two sets of data sequences in each parameter pair are sequentially input into the Pearson correlation coefficient algorithm to calculate the linear strength between the parameters and the direction of the correlation, and generate preliminary correlation data. S203: For each parameter pair in the sequence of feature parameter pairs to be calculated, calculate the degree of dispersion and distribution pattern of the data points within it, and obtain the distribution difference measure between parameters. S204: Based on the preliminary correlation data and the distribution difference measure between the parameters, the linear strength between the parameters, the correlation direction, and the distribution difference are integrated into structured data to construct the impedance feature correlation matrix.

3. The impedance prediction method based on combined modeling according to claim 2, characterized in that, The specific steps for obtaining the structural reorganization results are as follows: S301: Based on the impedance feature correlation matrix, extract the correlation score corresponding to each group of impedance feature combinations in the matrix, call the preset feature screening standard value, compare the numerical value of each group of correlation scores with the feature screening standard value, select all impedance feature combinations with scores exceeding the feature screening standard value and set them together to generate a high correlation impedance feature set. S302: Call the highly correlated impedance feature set, and based on the inherent correlation of each feature, retrieve and match the structural parameters directly corresponding to the features in the original structural parameter set. Recombine and classify the retrieved structural parameters to form a parameter set for the target structural unit and obtain the data cluster of the structural unit to be optimized. S303: For the data cluster of structural units to be optimized, initialize K cluster centers, iteratively calculate the Euclidean distance from each structural unit parameter point in the data cluster to each cluster center, and assign each structural unit parameter point to the category of the nearest cluster center. Iteratively update the position of the cluster center point in each category until the cluster center point no longer shifts, and obtain the structural reorganization result.

4. The impedance prediction method based on combined modeling according to claim 3, characterized in that, The specific steps for obtaining the impedance prediction parameter initialization are as follows: S401: Call the structural reorganization results and, based on the impedance characteristic correlation matrix, traverse each reorganized structural parameter group, select the parameter with the highest impedance correlation score, and determine it as the core prediction parameter; S402: For the core prediction parameter, calculate its statistical mean in the historical data of the target group, and assign the statistical mean as the initial value to obtain the initial value of the core parameter; S403: Based on the initial value of the core parameter, and by calling the correlation scores of other parameters in the same group with the core prediction parameter in the impedance feature correlation matrix, the other parameters in the same group are assigned values ​​proportionally to generate a weighted parameter set. S404: Integrate the weighted parameter set, set an initial prediction error threshold, establish the core parameter set, the weight allocation and the initial threshold, and complete the impedance prediction parameter initialization.

5. The impedance prediction method based on combined modeling according to claim 2, characterized in that, The linearity between the parameters in the preliminary correlation data is calculated using the following formula: ； in, The linearity between the parameters is given by n, where n is the total number of data points in the data sequence. For the i-th data point in the first group of data sequences in the sequence to be calculated, For the i-th data point of the second group of data sequences in the sequence to be calculated, The arithmetic mean of all data points in the first set of data sequences is... It is the arithmetic mean of all data points in the second set of data sequences.

6. The impedance prediction method based on combined modeling according to claim 3, characterized in that, In the step of iteratively updating the position of the cluster center point for each category, the position of the new cluster center point is determined by the following formula: ； in, These are the updated coordinates of the k-th cluster center. This is the set of all structural unit parameter points that are assigned to the k-th cluster in the current iteration round. For the set The total number of structural element parameter points contained therein For the set The position coordinates of the j-th structural unit parameter point.

7. The impedance prediction method based on combined modeling according to claim 4, characterized in that, The operation of assigning values ​​to other parameters within the same group proportionally is specifically as follows: First, obtain the initial values ​​of the core parameters and the corresponding correlation scores in the impedance feature correlation matrix; Secondly, for any non-core parameter in the structural parameter group, extract its correlation score with the core prediction parameter; Then, the initial value of the core parameter is multiplied by the correlation score between the non-core parameter and the core prediction parameter, and the calculation result is used as the baseline value of the non-core parameter. Finally, the baseline assignment is adjusted using a parameter adjustment factor to generate the final assignment. This parameter adjustment factor is calculated using the formula... Calculated; in, This represents the parameter adjustment factor. This represents a preset scaling weight used to control the adjustment magnitude. This represents the standard deviation of the non-core parameter in the historical data of the target group. The value represents the standard deviation of the core prediction parameter in the historical data of the target group. This adjusted value is the final value of the non-core parameter in the weighted parameter set.

8. An impedance prediction system based on combined modeling, characterized in that, The system is used to implement the impedance prediction method of combined modeling as described in any one of claims 1-7, and the system comprises: The feature parameter classification module is used to obtain the impedance measurement value, current parameter and temperature parameter of the target group by detection, call the obtained impedance measurement value and perform comparison and judgment with the preset threshold range, generate feature classification result, and pass it to the feature correlation calculation module; The feature correlation calculation module is used to call the feature classification results, pair impedance-related feature parameters belonging to the same category, and input the paired parameter groups into the Pearson correlation coefficient algorithm for correlation calculation, construct the impedance feature correlation matrix, and pass it to the structural unit optimization module. The structural unit optimization module is used to select impedance feature combinations with correlation scores exceeding a specified standard based on the impedance feature correlation matrix, call the structural parameter group corresponding to the selected impedance feature combination, input the structural parameter group into the K-means clustering model for structural optimization, output the structural reorganization result, and pass it to the prediction parameter initialization module. The prediction parameter initialization module is used to call the structural reorganization result, select the parameter with the highest impedance correlation score in the reorganized structural parameter group, and perform parameter assignment operations on this parameter and other parameters in the same group in sequence. The assigned parameter set is used as the initialization parameter input of the prediction sub-model to complete the impedance prediction parameter initialization.

Images