A nonlinear load measure method, system, device and medium

Abstract

The application belongs to the technical field of data processing, and specifically discloses a nonlinear load measurement method, system, device and medium, wherein the method comprises the following steps: collecting physiological signal data from a multi-source sensor chain through an edge computing node, extracting key features in a time sequence by using a signal decomposition algorithm, and generating an initial load feature vector; segmenting individual motion behavior data according to the initial load feature vector, extracting motion mode features by using a sliding window technology, judging the similarity between the converted feature vector and the initial load feature vector, and determining a preliminary load match if the similarity exceeds a preset threshold; obtaining real-time environmental parameters from an external environmental data chain, and fusing the environmental parameters and the preliminary load match result by using an adaptive weight algorithm; and the application aims to solve the problems of signal noise interference, insufficient feature extraction and data fusion distortion when processing multi-source heterogeneous data in the national physical fitness monitoring and whole-population health management scenarios.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a nonlinear load measurement method, system, device, and medium. Background Technology

[0002] In scenarios where AI platforms for exercise intervention are applied to national physical fitness monitoring and health management of the entire population, the core technical challenge lies in how to accurately measure and dynamically assess nonlinear loads while ensuring the consistency and reliability of data collection, analysis, and feedback processes. This challenge involves extracting individual exercise indicators from multi-source heterogeneous data and converting them into feature vectors. However, traditional methods may face signal noise interference or insufficient extraction of key features when dealing with complex exercise patterns, making it difficult for feature vectors to accurately reflect actual load characteristics and affecting the accuracy of subsequent analysis results.

[0003] Furthermore, when fusing physiological and behavioral data from different sources and matching them with the load model, differences in data formats and inconsistent evaluation criteria may lead to feature space mismatch, reducing assessment accuracy. Moreover, when acquiring external environmental data (such as temperature and humidity) in real time to correct exercise load calculations, there is a synchronization issue between the data update frequency and the monitoring cycle. If the time window is set improperly or transmission delays cause data to fail to update in a timely manner, it may affect the dynamic adaptability of the load assessment, thereby weakening the effectiveness of comprehensive intervention strategies.

[0004] Ultimately, when executing dynamic load scoring and personalized intervention logic through algorithmic models, the weight allocation mechanism may fail to fully reflect individual differences due to insufficient parameter settings or data fluctuations. Furthermore, if the stored log data is delayed in synchronization or fragmented during retrospective analysis, the analysis path may be incomplete, affecting the scientific nature of decision support.

[0005] This comprehensive issue spans the entire process from data collection and feature extraction to dynamic evaluation, involving complex interactions of signal processing, data fusion, real-time correction, and storage management. It directly affects the effectiveness and promotional value of the sports intervention AI platform in national physical fitness monitoring and its application to the entire population. Summary of the Invention

[0006] This invention provides a nonlinear load measurement method, system, device, and medium, aiming to solve the problems of signal noise interference, insufficient feature extraction, and data fusion distortion when processing multi-source heterogeneous data in the scenarios of national physical fitness monitoring and whole-population health management.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A nonlinear load measurement method includes: collecting physiological signal data from a multi-source sensor chain via edge computing nodes; extracting key features from the time series using a signal decomposition algorithm to generate an initial load feature vector; segmenting individual motion behavior data based on the initial load feature vector; extracting motion pattern features using a sliding window technique; determining the similarity between the transformed feature vector and the initial load feature vector; if the similarity exceeds a preset threshold, a preliminary load match is determined; acquiring real-time environmental parameters from an external environment data chain; fusing the environmental parameters with the preliminary load match results using an adaptive weighting algorithm to generate a comprehensive load vector; and calling an external... The system obtains real-time correction data through the data interface and determines whether the correction data update time is within the monitoring period. If the correction data update time is within the correction time range, the validity of the correction data is determined. Based on the comprehensive load vector and the validity result of the correction data, a weighted fusion calculation is performed to obtain an individual dynamic load score. If the dynamic load score is higher than a preset threshold, an automatic intervention logic is executed through a distributed consensus mechanism. The intervention process record is distributed and stored using a cryptographic hash function to generate an immutable log record. Based on the immutable log record, relevant verification paths are retrieved from the distributed network, and the completeness of the path is determined. If the path is complete, the final output result is determined.

[0009] In one aspect of this disclosure, the step of acquiring physiological signal data from a multi-source sensor chain via edge computing nodes, extracting key features from the time series using a signal decomposition algorithm, and generating an initial load feature vector includes:

[0010] The system retrieves multi-source sensor data stored on the blockchain from edge computing nodes and uses a distributed query protocol to obtain raw signal data.

[0011] The empirical mode decomposition technique in signal decomposition algorithms is used to decompose the original signal data and obtain the sequence of intrinsic mode functions.

[0012] If the intrinsic mode function sequence contains a preset high-frequency noise component, then the high-frequency noise is filtered out by wavelet transform to obtain a denoised mode function sequence;

[0013] The modality function sequence is transformed into a feature vector sequence through a pre-trained feature extraction model, generating an initial load feature vector.

[0014] Based on the initial load feature vector, the principal component analysis algorithm is used to reduce the dimension of the vector to obtain the optimized load feature vector.

[0015] If the similarity between the optimized feature vector and the preset load template exceeds a preset threshold, then the feature vector is determined to meet the load requirements, and a structured load description is generated.

[0016] By using structured load descriptions, clustering algorithms are employed to classify load types and obtain classification results.

[0017] In one aspect of this disclosure, the step of segmenting individual motion behavior data based on the initial load feature vector, extracting motion pattern features using a sliding window technique, determining the similarity between the transformed feature vector and the initial load feature vector, and determining preliminary load matching if the similarity exceeds a preset threshold, includes:

[0018] Initial load characteristic data is obtained, and feature vectors are extracted from it. Keyword extraction and semantic analysis methods are used to generate load feature vectors.

[0019] Individual motion behavior data is segmented and processed, and key fields are extracted using sliding window technology to generate motion pattern data.

[0020] The motion pattern data is converted into motion feature vectors through a pre-trained word vector model.

[0021] Calculate the cosine similarity between the motion feature vector and the load feature vector to obtain a similarity value;

[0022] If the similarity value exceeds a preset threshold, a preliminary load match is determined, and a matching result is generated.

[0023] Based on the matching results, cluster analysis is used to group the individuals to obtain the grouping results;

[0024] Based on the grouping results, the load feature vector of each individual in the group is extracted to generate a sorted list.

[0025] In one aspect of this disclosure, the step of obtaining real-time environmental parameters from an external environmental data chain and fusing the environmental parameters with the preliminary load matching results using an adaptive weighting algorithm to generate a comprehensive load vector includes:

[0026] Real-time environmental parameters are obtained from the external environment data chain, and a structured environmental parameter dataset is obtained by using preset parameter extraction rules.

[0027] Using a structured environmental parameter dataset, the weight values ​​of each parameter are calculated, and a weighted average method is used to generate a weighted set of environmental parameters.

[0028] Individual load data are obtained from the initial load matching results. The vector embedding method is used to fuse the weighted set of environmental parameters with the individual load data to generate an initial comprehensive load vector.

[0029] If the dimension of the initial composite load vector exceeds the preset threshold, principal component analysis is used to reduce the dimension and obtain the optimized composite load vector.

[0030] Based on the optimized comprehensive load vector, the similarity between vectors is calculated, and the cosine similarity algorithm is used to determine the degree of load matching between individuals;

[0031] By assessing the degree of load matching between individuals and combining it with preset classification rules, the load level of each individual is determined, and the final load assessment result is generated.

[0032] Based on the final load assessment results, a data association analysis method is used to associate the assessment results with on-chain environmental parameter records to generate a verification dataset of the comprehensive load vector.

[0033] In one aspect of this disclosure, the step of obtaining real-time correction data by calling an external data interface through a distributed communication protocol, determining whether the correction data update time is within the monitoring period, and determining the validity of the correction data if the correction data update time is within the correction time range, includes:

[0034] The timestamps of real-time correction data are obtained from distributed data sources through external data interfaces and stored as first-time data.

[0035] Using time comparison logic, the first-time data is compared with the preset monitoring period. If the first-time data is earlier than the end time of the monitoring period, the correction data update time is determined to be valid, and the validity judgment result is obtained.

[0036] The validity determination result is obtained through the distributed communication protocol interface, and the validity status of the corrected data is generated.

[0037] Based on the validity status, call the external data interface to obtain the complete dataset of real-time correction data and store it as the first correction data;

[0038] The integrity and consistency of the first correction data are verified using a data verification algorithm to obtain the second correction data that has passed the verification.

[0039] A classification algorithm is used to assess the health of the second corrected data and generate a health score.

[0040] The final validity of the real-time corrected data is obtained based on the health score and validity determination results.

[0041] In one aspect of this disclosure, the step of performing a weighted fusion calculation based on the comprehensive load vector and the validity results of the corrected data to obtain an individual dynamic load score includes:

[0042] Obtain individual information and the validity of correction data; determine the integrity and consistency of data through preset data verification rules; and obtain verified correction data quality.

[0043] Based on the verified quality of the corrected data, load assessment indicators were extracted, and factor analysis was used to determine the comprehensive load vector.

[0044] If at least one indicator in the comprehensive load vector exceeds a preset threshold, a preliminary fusion score is obtained by fusing load assessment indicators and correcting data quality using a linear weighting method.

[0045] For the initial fusion score, a weighted fusion calculation is performed using a preset fusion weight allocation to obtain the adjusted fusion score;

[0046] Based on the adjusted fusion score, a logistic regression model is applied to determine the individual's health level and output a dynamic load score;

[0047] If the dynamic load score is lower than the preset threshold, the fusion weight allocation will be adjusted through the score calculation model, and the dynamic load score will be recalculated.

[0048] Obtain the recalculated dynamic load score, and generate an individual comprehensive assessment score by outputting the assessment results.

[0049] In one aspect of this disclosure, the step of retrieving relevant verification paths from the distributed network based on the immutable log records, determining the path integrity, and if complete, determining the final output result includes:

[0050] Log records are obtained from a distributed network, and a distributed node query protocol is used to extract a log dataset containing timestamps and transaction hashes to obtain the original log records.

[0051] For the original log records, parse the log structure, extract the node sequence and hash chain in the verification path, and generate a path dataset;

[0052] If the node sequence in the path dataset is consistent with the preset distributed topology, the Merkle tree verification algorithm is used to calculate the hash value of each node in the path, determine the path integrity, and obtain the integrity verification result.

[0053] Based on the integrity verification results, if the integrity verification passes, the transaction timestamps and associated metadata are extracted from the path dataset to generate the review dataset;

[0054] The SHA-256 algorithm is used to perform hash calculation on the review dataset to generate review hash values ​​and determine the uniqueness of the review results;

[0055] By using network communication protocols, the audit hash value is compared with the preset audit standards to determine whether the audit result meets the standards and output the final audit result.

[0056] In another aspect, this disclosure also relates to a nonlinear load measurement system, comprising:

[0057] The load feature extraction module is used to collect physiological signal data from a multi-source sensor chain through edge computing nodes, extract key features from the time series using a signal decomposition algorithm, and generate an initial load feature vector.

[0058] The load matching module is used to perform time-frequency domain decomposition processing on the historical operating data of nonlinear loads based on the initial load feature vector, calculate the similarity between the decomposed feature vector and the initial load feature vector, and determine the preliminary matching of load characteristics if the similarity exceeds a preset threshold.

[0059] The environment fusion module is used to extract environmental variables and equipment status data from historical operation records on the nonlinear load chain. It uses a multimodal fusion method to associate the environmental variables and equipment status data with the preliminary matching results of load characteristics to generate a comprehensive load feature vector.

[0060] The environment verification module is used to obtain real-time environmental variable updates by calling the external data interface through the smart gateway, and to determine whether the environmental variable update time is within the monitoring period. If it is within the monitoring period, the validity of the environmental variable is determined.

[0061] The scoring calculation module is used to perform weighted fusion calculations based on the comprehensive load characteristic vector and the effectiveness results of environmental variables to obtain a dynamic evaluation score for nonlinear load.

[0062] The verification execution module is used to execute automatic verification logic through the smart gateway if the dynamic evaluation score is higher than the preset threshold. It uses a cryptographic hash algorithm to store the verification process record on the chain and generate an immutable verification log.

[0063] The verification module is used to retrieve relevant verification paths from the distributed network based on the immutable verification log, determine the integrity of the path, and if the path is complete, determine the final output of the evaluation result.

[0064] In another aspect, this disclosure also relates to an electronic device, comprising:

[0065] A memory on which computer programs are stored;

[0066] A processor is configured to execute the computer program in the memory to implement the aforementioned nonlinear load measurement method.

[0067] In another aspect of this disclosure, the disclosure also relates to a non-transitory computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the aforementioned nonlinear load measurement method.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] First, this invention achieves a qualitative improvement in efficiency and accuracy. The method utilizes natural language processing technology to automatically extract structured feature vectors from bidding requirement texts and performs intelligent vector transformation and similarity matching on enterprise qualification data, replacing the traditional method of manual comparison. This significantly shortens the initial qualification review cycle and, through algorithmic means, avoids subjective oversights, making the matching results more objective and accurate. Second, the system's credibility and transparency are fundamentally enhanced. By constructing a blockchain-based data storage and verification chain, the uploading, verification, and matching processes of enterprise qualification documents, historical performance, and financial data are fully recorded, generating tamper-proof audit logs. This not only ensures the authenticity and traceability of all operational data but also provides a clear verification path for regulatory authorities and participants, effectively preventing information falsification and opaque operations, and creating a fair and transparent bidding environment. Finally, this method achieves automated and intelligent closed-loop business processes. By automatically calling external data and executing preset scoring logic through smart contracts, and triggering automatic approval when conditions are met, qualification review is transformed from a tedious manual approval task into an efficient, reliable, and fully traceable automated process. This significantly reduces the platform's operational costs and risks, providing solid technical support for the large-scale and standardized conduct of bidding activities. Attached Figure Description

[0070] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0071] Figure 1 This is a flowchart of a nonlinear load measurement method according to the present invention. Detailed Implementation

[0072] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments described. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the protection scope of the present invention.

[0073] Please see Figure 1 As shown, this embodiment discloses a nonlinear load measurement method, system, device, and medium. In this embodiment, the nonlinear load measurement method may specifically include:

[0074] S101. Physiological signal data are collected from the multi-source sensor chain through edge computing nodes, and key features in the time series are extracted using a signal decomposition algorithm to generate an initial load feature vector.

[0075] The system acquires multi-source sensor data stored on-chain from edge computing nodes and obtains raw signal data using a distributed query protocol. Empirical Mode Decomposition (EMD) is employed in signal decomposition algorithms to decompose the raw signal data, obtaining an intrinsic mode function (IMF) sequence. If the IMF sequence contains preset high-frequency noise components, wavelet transform is used to filter out the high-frequency noise, resulting in a denoised IMF sequence. A pre-trained feature extraction model is used to transform the IMF sequence into a feature vector sequence, generating an initial load feature vector. Based on the initial load feature vector, principal component analysis (PCA) is used to reduce the vector dimension, obtaining an optimized load feature vector. If the similarity between the optimized feature vector and a preset load template exceeds a preset threshold, the feature vector is determined to meet the load requirements, generating a structured load description. Using the structured load description, a clustering algorithm is employed to classify the load types, obtaining the classification results.

[0076] For example, when obtaining multi-source sensor data from edge computing nodes, physiological signal data stored on the blockchain can be accessed through a distributed query protocol. Assuming the use of IoT edge nodes, the nodes collect data from sensors via the MQTT protocol to obtain raw signals such as electrocardiograms and electromyograms.

[0077] The node uses empirical mode decomposition (EMD) to decompose the original ECG signal, obtaining multiple intrinsic mode functions (IMFs). Assuming that IMF1 contains high-frequency noise, the noise is filtered out using wavelet transform, such as the Daubechies wavelet, to obtain a denoised IMF sequence.

[0078] Subsequently, a pre-trained convolutional neural network (CNN) model is used to transform the IMF sequence into a sequence of feature vectors, generating the initial load feature vector.

[0079] For example, the feature vector might include heart rate variability, muscle fatigue index, etc.; the initial feature vector is 128-dimensional, which is reduced to 32-dimensional through principal component analysis to obtain the optimized load feature vector. The cosine similarity between the optimized feature vector and a preset load template, such as a normal load template, is calculated. If the similarity exceeds 0.85, the feature vector is determined to meet the load requirements, and a structured load description, such as "moderate load," is generated.

[0080] Finally, the K-means clustering algorithm was used to classify the load types, and the classification results were obtained such as "low load", "medium load" and "high load".

[0081] S102. Based on the initial load feature vector, the individual motion behavior data is segmented and processed. The motion pattern features are extracted using the sliding window technique. The similarity between the transformed feature vector and the initial load feature vector is judged. If the similarity exceeds the preset threshold, the load is initially matched.

[0082] Initial load feature data is acquired, and feature vectors are extracted from it. Keyword extraction and semantic analysis methods are used to generate load feature vectors. Individual motion behavior data is segmented, and a sliding window technique is used to extract key fields and generate motion pattern data. A pre-trained word vector model is used to convert the motion pattern data into motion feature vectors. The cosine similarity between the motion feature vectors and the load feature vectors is calculated to obtain a similarity value. If the similarity value exceeds a preset threshold, a preliminary load match is determined, and a matching result is generated. Based on the matching result, cluster analysis is used to group individuals, obtaining grouping results. According to the grouping results, the load feature vector of each group is extracted, generating a sorted list.

[0083] For example, the process of segmenting individual motion behavior data and determining similarity based on the initial load feature vector can be implemented using the following information technology. Assume the initial load feature vector is [0.8, 0.6, 0.9], representing heart rate, cadence, and energy consumption, respectively. Individual motion behavior data comes from an accelerometer. Using a sliding window technique (e.g., a window size of 5 seconds and an overlap rate of 50%), motion pattern features such as average acceleration and peak frequency are extracted. The motion pattern data is then converted into a motion feature vector, for example, [0.7, 0.5, 0.8], using a pre-trained Word2Vec model.

[0084] Calculate the cosine similarity between the motion feature vector and the load feature vector:

[0085] The dot product = 0.8 × 0.7 + 0.6 × 0.5 + 0.9 × 0.8 = 0.56 + 0.3 + 0.72 = 1.58;

[0086] Module length A = = = ≈1.345;

[0087] Length of modulus B = = = ≈1.175;

[0088] Cosine similarity = 1.58 / (1.345×1.175)≈1.58 / 1.580≈1.0.

[0089] If the preset threshold is 0.9, the similarity exceeds the threshold, and a preliminary load match is determined. Individuals are grouped using K-means clustering, such as "Group A: High Load" and "Group B: Medium Load," and a sorted list is generated.

[0090] S103. Obtain real-time environmental parameters from the external environment data link, and use an adaptive weighting algorithm to fuse the environmental parameters with the preliminary load matching results to generate a comprehensive load vector.

[0091] Real-time environmental parameters are obtained from the external environmental data chain. A structured environmental parameter dataset is obtained using pre-defined parameter extraction rules. The weights of each parameter are calculated using this structured dataset, and a weighted average method is used to generate a weighted set of environmental parameters. Individual load data is obtained from the initial load matching results. A vector embedding method is used to fuse the weighted set of environmental parameters with the individual load data to generate an initial comprehensive load vector. If the dimensionality of the initial comprehensive load vector exceeds a pre-defined threshold, principal component analysis is used for dimensionality reduction to obtain an optimized comprehensive load vector. Based on the optimized comprehensive load vector, the similarity between vectors is calculated, and a cosine similarity algorithm is used to determine the degree of load matching between individuals. Based on the degree of load matching between individuals and pre-defined classification rules, the load level of each individual is determined, generating the final load assessment result. Based on the final load assessment result, a data association analysis method is used to associate the assessment result with on-chain environmental parameter records, generating a validation dataset for the comprehensive load vector.

[0092] For example, real-time environmental parameters, such as temperature, humidity, and altitude, are obtained from an external environmental data chain. Structured data, such as [25, 60, 1000], is obtained through parameter extraction rules. Weight values ​​are calculated using entropy weighting to determine the weights: temperature 0.4, humidity 0.3, and altitude 0.3. A weighted average is then used to generate a weighted environmental parameter set [10, 18, 300]. Individual load data, such as [0.8, 0.6, 0.9], is obtained from the initial load matching results. Vector embedding methods, such as neural networks, are used to fuse environmental parameters and load data to generate an initial comprehensive load vector, for example, 128 dimensions. Dimensionality is reduced to 32 dimensions using PCA to obtain an optimized comprehensive load vector. Cosine similarity between vectors is calculated to determine the degree of load matching between individuals; for example, the similarity between individual A and individual B is 0.92. According to classification rules, a similarity greater than 0.9 indicates a "high match," generating a load assessment result "Individual A: High Load." Environmental parameter records are associated to generate a validation dataset.

[0093] S104. Obtain real-time correction data by calling the external data interface through the distributed communication protocol, and determine whether the correction data update time is within the monitoring period. If the correction data update time is within the correction time range, the validity of the correction data is determined.

[0094] The timestamps of real-time correction data are obtained from a distributed data source via an external data interface and stored as the first-time data. A time comparison logic is used to compare the first-time data with a preset monitoring period. If the first-time data is earlier than the end time of the monitoring period, the correction data update time is determined to be valid, and a validity judgment result is obtained. The validity judgment result is obtained through a distributed communication protocol interface, generating a correction data validity status. Based on the validity status, the complete dataset of real-time correction data is obtained by calling the external data interface and stored as the first correction data. The completeness and consistency of the first correction data are verified using a data validation algorithm to obtain verified second correction data. A classification algorithm is used to perform a health assessment on the second correction data, generating a health score. Based on the health score and the validity judgment result, the final validity of the real-time correction data is obtained.

[0095] For example, a distributed communication protocol is used to call an external data interface to obtain the timestamp of the calibration data, such as "20XX-XX-XX, XX:XX:XX". The monitoring period is from "20XX-XX-XX, XX:XX:XX" to "20XX-XX-XX, XX:XX:XX". If the timestamp is within this range, it is considered valid. Complete calibration data, such as the heart rate calibration value, is obtained. The data integrity is verified using a validation algorithm to obtain the second calibration data. A classification algorithm is used to evaluate the health score of the calibration data, such as 0.9. Combined with the validity determination, the final validity is "passed".

[0096] S105. Based on the comprehensive load vector and the results of the validity of the correction data, perform weighted fusion calculation to obtain the individual dynamic load score.

[0097] The process involves acquiring individual information and validating the calibration data. Using pre-defined data validation rules, data integrity and consistency are assessed to obtain validated calibration data quality. Based on this quality, load assessment indicators are extracted, and factor analysis is used to determine the comprehensive load vector. If at least one indicator in the comprehensive load vector exceeds a pre-defined threshold, a linear weighting method is used to fuse the load assessment indicators and calibration data quality to obtain a preliminary fusion score. For this preliminary fusion score, a pre-defined fusion weight allocation is applied, and a weighted fusion calculation is performed to obtain an adjusted fusion score. Based on the adjusted fusion score, a logistic regression model is applied to determine the individual's health level, and a dynamic load score is output. If the dynamic load score is lower than a pre-defined threshold, the fusion weight allocation is adjusted using a score calculation model, and the dynamic load score is recalculated. The recalculated dynamic load score is obtained, and the assessment results are output to generate an individual comprehensive assessment score.

[0098] For example, individual information includes age and gender, and the corrected data validity is "passed". Data validation rules check data integrity; if all fields are complete, the corrected data quality is "high". Load assessment indicators such as heart rate and energy consumption are extracted, and factor analysis is used to determine the comprehensive load vector [0.8, 0.7]. The indicator threshold is 0.75, and the heart rate indicator exceeds the threshold at 0.8. Linear weighted fusion is performed: load indicator weight 0.6, corrected data quality weight 0.4, and the initial fusion score = 0.8 × 0.6 + 0.9 × 0.4 = 0.48 + 0.36 = 0.84.

[0099] Adjusting the weights, with a load index of 0.7 and a data quality correction of 0.3, the adjusted score = 0.8 × 0.7 + 0.9 × 0.3 = 0.56 + 0.27 = 0.83. The logistic regression model outputs a health level of "Good" and a dynamic load score of 0.83. If the score is lower than 0.8, the weights are adjusted and the calculation is repeated.

[0100] S106. If the dynamic load score is higher than the preset threshold, the automatic intervention logic is executed through the distributed consensus mechanism. The intervention process record is distributed and stored using a cryptographic hash function to generate an immutable log record.

[0101] If the dynamic load score exceeds a preset threshold, an automatic intervention logic is triggered via a distributed consensus mechanism (such as PBFT) to execute intervention measures, such as sending an alert message. A cryptographic hash function (such as SHA-256) is used to hash the intervention process records, generating a hash value. This hash value is then stored in a distributed storage system (such as IPFS), generating an immutable log record.

[0102] For example, a dynamic load score of 0.83 and a threshold of 0.8 trigger an intervention. The intervention record includes the time, action, and user ID, which are hashed and stored in IPFS, returning the storage address.

[0103] S107. Based on the immutable log records, retrieve the relevant verification path from the distributed network, determine the path integrity, and if it is complete, determine the final output result.

[0104] Log records are retrieved from a distributed network. A distributed node query protocol is used to extract a log dataset containing timestamps and transaction hashes, yielding the raw log records. For these raw log records, the log structure is parsed, and the node sequence and hash chain in the verification path are extracted to generate a path dataset. If the node sequence in the path dataset matches a preset distributed topology, a Merkle tree verification algorithm is used to calculate the hash value of each node in the path, determining the path's integrity and obtaining an integrity verification result. Based on the integrity verification result, if the integrity verification passes, transaction timestamps and associated metadata are extracted from the path dataset to generate an audit dataset. The SHA-256 algorithm is used to hash the audit dataset, generating audit hash values ​​to determine the uniqueness of the audit results. Through a network communication protocol, the audit hash values ​​are compared with preset audit standards to determine if the audit results meet the standards, and the final audit result is output.

[0105] For example, log records are retrieved from a distributed network, and the node sequence [A,B,C] and hash chain [H1,H2,H3] are parsed. A Merkle tree is used to verify that the root hash matches and the path is complete. Timestamps and metadata are extracted to generate an audit dataset. SHA-256 hashes are used to generate audit hash values, which are compared with audit standards. If they match, "audit passed" is output.

[0106] In its implementation, the system of this invention includes multiple functional modules: a load feature extraction module, a load matching module, an environment fusion module, an environment verification module, a scoring calculation module, a verification execution module, and a verification module. These modules are connected through a distributed network architecture and achieve data interaction and functional collaboration through communication protocols. The specific implementation steps of each module are based on multi-level feature extraction, dynamic correction mechanisms, and distributed storage management techniques, thereby ensuring the accuracy and reliability of nonlinear load measurement.

[0107] The load feature extraction module is the core starting point of the entire system. Its main task is to collect physiological signal data from a multi-source sensor chain and generate an initial load feature vector. Specifically, this module first establishes a connection with the multi-source sensors through edge computing nodes. The sensor chain includes devices such as heart rate monitors, blood pressure monitors, and accelerometers, used to collect individual physiological signal data in real time. This data is transmitted to the edge computing nodes via a distributed query protocol, forming the raw signal dataset. Subsequently, empirical mode decomposition (EMD) is used to decompose the raw signal data to obtain an intrinsic mode function (IMF) sequence. If the IMF sequence contains high-frequency noise components, wavelet transform is used to filter them out, resulting in a denoised IMF sequence. Next, a pre-trained feature extraction model is used to transform the IMF sequence into a feature vector sequence, and principal component analysis (PCA) is used to reduce the dimensionality of the feature vectors, ultimately generating an optimized load feature vector. Each step in this process must strictly adhere to standardized data transmission and processing procedures to ensure data integrity and consistency.

[0108] The task of the load matching module is to segment individual motion behavior data based on the initial load feature vector, extract motion pattern features using the sliding window technique, and then determine the similarity between the transformed feature vector and the initial load feature vector. Specifically, the module first acquires the initial load feature data generated by the load feature extraction module and extracts feature vectors from it. Then, keyword extraction and semantic analysis methods are used to generate load feature vectors. For individual motion behavior data, the module segments the data and applies the sliding window technique to extract key fields, generating motion pattern data. This data is converted into motion feature vectors using a pre-trained word vector model and then cosine similarity is calculated with the load feature vector. If the similarity value exceeds a preset threshold, a preliminary load match is determined, and a matching result is generated. Based on this, the module further uses clustering analysis to group individuals and extracts the load feature vector for each group based on the grouping results, generating a sorted list. Both the segmentation processing and the sliding window technique in the above process rely on high-precision timestamp records to ensure the accuracy of data segmentation.

[0109] The environmental fusion module's function is to acquire real-time environmental parameters from an external environmental data chain and fuse them with the preliminary load matching results using an adaptive weighting algorithm to generate a comprehensive load vector. Specifically, the module first acquires real-time environmental parameters, such as temperature, humidity, and air pressure, from the external environmental data chain and generates a structured environmental parameter dataset using preset parameter extraction rules. Then, it calculates the weight values ​​of each parameter using a weighted average method, generating a weighted set of environmental parameters. Based on this, the module obtains individual load data from the preliminary load matching results and fuses the weighted set of environmental parameters with the individual load data using a vector embedding method to generate an initial comprehensive load vector. If the dimensionality of the initial comprehensive load vector exceeds a preset threshold, it is dimensionality-reduced using principal component analysis to obtain an optimized comprehensive load vector. Finally, the module calculates the similarity between vectors using a cosine similarity algorithm and, combined with preset classification rules, determines the individual's load level to generate the final load assessment result. Parameter extraction and weight allocation in this process are dynamically adjusted based on historical data analysis and a real-time feedback mechanism.

[0110] The main task of the environment verification module is to obtain real-time correction data by calling an external data interface through a distributed communication protocol and determine whether the update time of the correction data is within the monitoring period. Specifically, the module first obtains the timestamp of the real-time correction data from a distributed data source through the external data interface and stores it as the first-time data. Then, it uses time comparison logic to compare the first-time data with a preset monitoring period. If the first-time data is earlier than the end time of the monitoring period, the update time of the correction data is determined to be valid, and a validity judgment result is generated. Based on this judgment result, the module obtains the complete dataset of real-time correction data through the distributed communication protocol interface and verifies its completeness and consistency through a data verification algorithm, generating verified second-time correction data. Finally, the module uses a classification algorithm to perform a health assessment on the second-time correction data, generates a health score, and outputs the final validity of the real-time correction data based on the validity judgment result. Both timestamp management and data verification in this process rely on a high-precision time synchronization mechanism and a distributed consensus algorithm.

[0111] The scoring calculation module's task is to perform a weighted fusion calculation based on the comprehensive load vector and the validity results of the corrected data to obtain an individual's dynamic load score. In practice, the module first acquires individual information and the validity of the corrected data, and uses preset data validation rules to determine data integrity and consistency, generating validated corrected data quality. Subsequently, the module extracts load assessment indicators and uses factor analysis to determine the comprehensive load vector. If at least one indicator in the comprehensive load vector exceeds a preset threshold, the load assessment indicators and corrected data quality are fused using a linear weighting method to generate a preliminary fusion score. Based on this, the module performs a weighted fusion calculation using preset fusion weights to generate an adjusted fusion score. Finally, the module uses a logistic regression model to determine the individual's health level and outputs the dynamic load score. If the dynamic load score is lower than a preset threshold, the fusion weight allocation is adjusted using the score calculation model, and the dynamic load score is recalculated until the requirements are met. Both the weight allocation and score calculation in this process require dynamic optimization based on historical data and real-time feedback.

[0112] The function of the verification execution module is to execute automatic intervention logic through a distributed consensus mechanism when the dynamic load score exceeds a preset threshold, and to use a cryptographic hash function to distribute the intervention process records, generating immutable log records. In specific implementation, the module first obtains the data to be intervened through the triggering logic of the distributed consensus mechanism, executes the automatic intervention operation, and generates the intervention result. Subsequently, the module uses a cryptographic hash function to encrypt the intervention result and intervention record, generating encrypted data. This data is written to on-chain storage through a distributed network, generating storage addresses. The module retrieves the on-chain data through the storage addresses and verifies its integrity. If the verification status is complete, a log record is generated through the distributed consensus mechanism. Finally, the module distributes the log data through a decentralized network and obtains consensus feedback from network nodes to generate the final on-chain record status. Both the encryption algorithm and consensus mechanism in this process must be designed based on highly secure distributed storage technology.

[0113] The verification module's task is to retrieve relevant verification paths from a distributed network based on immutable log records and determine path integrity. In practice, the module first obtains log records from the distributed network and uses a distributed node query protocol to extract a log dataset containing timestamps and transaction hashes, generating raw log records. Then, the module parses the log structure and extracts the node sequences and hash chains from the verification paths, generating a path dataset. If the node sequences in the path dataset match a preset distributed topology, a Merkle tree verification algorithm is used to calculate the hash value of each node in the path and determine path integrity. If integrity verification passes, the module extracts transaction timestamps and associated metadata from the path dataset to generate an audit dataset. Finally, the module uses the SHA-256 algorithm to hash the audit dataset, generating audit hash values, and compares them with preset audit standards via a network communication protocol, outputting the final audit result. Both path verification and hash calculation in this process must be designed based on highly reliable distributed network technology.

[0114] The collaboration between these modules is achieved through distributed communication protocols and distributed storage technology, ensuring efficient data transmission and storage. Furthermore, the design of each module must consider the hardware configuration and network environment of the actual application scenario to ensure system stability and scalability.

[0115] To enable those skilled in the art to fully understand and implement this invention, the following supplementary explanation of the implementation principle of this invention is provided in conjunction with a specific application scenario.

[0116] In the actual deployment of a certain sports intervention AI platform, the system collects individual physiological signal data in real time through a multi-source sensor chain and dynamically evaluates nonlinear loads in combination with external environmental parameters. The entire process starts from the load feature extraction module and gradually completes data processing, feature matching, environmental fusion, calibration and verification, score calculation, and intervention execution, ultimately generating tamper-proof log records to ensure traceability of results.

[0117] First, during the operation of the load feature extraction module, edge computing nodes establish connections with multi-source sensors such as heart rate monitors, blood pressure monitors, and accelerometers to acquire individual physiological signal data in real time. This data is transmitted to the edge computing nodes via a distributed query protocol, forming the raw signal dataset. Subsequently, empirical mode decomposition (EMD) is used to decompose the raw signal data, obtaining an intrinsic mode function (IMF) sequence. If the IMF sequence contains high-frequency noise components, wavelet transform is used to filter them out to improve data quality. Next, a pre-trained feature extraction model is used to transform the denoised IMF sequence into a feature vector sequence, and principal component analysis (PCA) is used to reduce the dimensionality of the feature vectors, ultimately generating an optimized load feature vector. This process ensures that the key features extracted from the complex signal accurately reflect the individual's load characteristics.

[0118] Next, the load matching module segments the individual motion behavior data based on the initial load feature vector and extracts motion pattern features using a sliding window technique. Specifically, the module first obtains initial load feature data from the load feature extraction module and generates load feature vectors using keyword extraction and semantic analysis methods. For individual motion behavior data, the module segments the data and extracts key fields using a sliding window technique to generate motion pattern data. This data is converted into motion feature vectors using a pre-trained word vector model and then cosine similarity is calculated with the load feature vector. If the similarity value exceeds a preset threshold, a preliminary load match is determined, and a matching result is generated. Based on this, the module further uses clustering analysis to group individuals and extracts the load feature vector for each group based on the grouping results, generating a ranked list. This process ensures the accuracy of data segmentation through high-precision timestamp recording, thereby improving the reliability of load matching.

[0119] Subsequently, the environment fusion module acquires real-time environmental parameters from the external environmental data chain and fuses them with the preliminary load matching results using an adaptive weighting algorithm. Specifically, the module first acquires real-time environmental parameters such as temperature, humidity, and air pressure from the external environmental data chain and generates a structured environmental parameter dataset using preset parameter extraction rules. Then, it calculates the weight values ​​of each parameter using a weighted average method and generates a weighted environmental parameter set. Based on this, the module obtains individual load data from the preliminary load matching results and fuses the weighted environmental parameter set with the individual load data using a vector embedding method to generate an initial comprehensive load vector. If the dimension of the initial comprehensive load vector exceeds a preset threshold, it is dimensionality-reduced using principal component analysis to obtain an optimized comprehensive load vector. Finally, the module calculates the similarity between vectors using a cosine similarity algorithm and, combined with preset classification rules, determines the individual load level to generate the final load assessment result. This process, by dynamically adjusting parameter weights, ensures that the impact of environmental factors on load assessment is reasonably reflected.

[0120] In the operation of the calibration and verification module, the module uses a distributed communication protocol to call an external data interface to obtain real-time calibration data and determines whether the update time of the calibration data is within the monitoring period. Specifically, the module first obtains the timestamp of the real-time calibration data from the distributed data source through the external data interface and stores it as the first-time data. Then, it uses time comparison logic to compare the first-time data with a preset monitoring period. If the first-time data is earlier than the end time of the monitoring period, the calibration data update time is determined to be valid, and a validity judgment result is generated. Based on this judgment result, the module obtains the complete dataset of real-time calibration data through the distributed communication protocol interface and verifies its completeness and consistency through a data verification algorithm, generating verified second-time calibration data. Finally, the module uses a classification algorithm to perform a health assessment on the second-time calibration data, generates a health score, and outputs the final validity of the real-time calibration data based on the validity judgment result. This process ensures the timeliness and reliability of the calibration data through a high-precision time synchronization mechanism and a distributed consensus algorithm.

[0121] The scoring calculation module's task is to perform a weighted fusion calculation based on the comprehensive load vector and the validity results of the corrected data to obtain an individual dynamic load score. In practice, the module first acquires individual information and the validity of the corrected data, and uses preset data validation rules to determine data integrity and consistency, generating validated corrected data quality. Subsequently, the module extracts load assessment indicators and uses factor analysis to determine the comprehensive load vector. If at least one indicator in the comprehensive load vector exceeds a preset threshold, the load assessment indicators and corrected data quality are fused using a linear weighting method to generate a preliminary fusion score. Based on this, the module performs a weighted fusion calculation using preset fusion weights to generate an adjusted fusion score. Finally, the module uses a logistic regression model to determine the individual's health level and outputs a dynamic load score. If the dynamic load score is lower than a preset threshold, the fusion weight allocation is adjusted using the score calculation model, and the dynamic load score is recalculated until the requirements are met. This process, through dynamic optimization of weight allocation, ensures that the scoring results comprehensively reflect individual differences.

[0122] The intervention execution module's function is to execute automatic intervention logic through a distributed consensus mechanism when the dynamic load score exceeds a preset threshold, and to use a cryptographic hash function to distribute the intervention process records. In practice, the module first obtains the data to be intervened through the triggering logic of the distributed consensus mechanism, executes the automatic intervention operation, and generates the intervention result. Subsequently, the module uses a cryptographic hash function to encrypt the intervention result and the intervention record, generating encrypted data. This data is written to on-chain storage through a distributed network, generating a storage address. The module retrieves the on-chain data through the storage address and verifies its integrity. If the verification status is complete, a log record is generated through the distributed consensus mechanism. Finally, the module distributes the log data through a decentralized network and obtains consensus feedback from network nodes to generate the final on-chain record status. This process, through highly secure distributed storage technology, ensures the immutability of the intervention records.

[0123] The log verification module's task is to retrieve relevant verification paths from a distributed network based on immutable log records and determine the path's integrity. In practice, the module first obtains log records from the distributed network and uses a distributed node query protocol to extract a log dataset containing timestamps and transaction hashes, generating raw log records. Then, the module parses the log structure and extracts the node sequences and hash chains from the verification paths, generating a path dataset. If the node sequences in the path dataset match a preset distributed topology, a Merkle tree verification algorithm is used to calculate the hash value of each node in the path and determine path integrity. If integrity verification passes, the module extracts transaction timestamps and associated metadata from the path dataset to generate an audit dataset. Finally, the module uses the SHA-256 algorithm to hash the audit dataset, generating audit hash values, and compares them with preset audit standards via a network communication protocol, outputting the final audit result. This process, through highly reliable distributed network technology, ensures the integrity and traceability of log records.

[0124] The collaboration between the above modules is achieved through distributed communication protocols and distributed storage technology, ensuring efficient data transmission and storage. Simultaneously, the design of each module must consider the hardware configuration and network environment in the actual application scenario to ensure system stability and scalability. Through the detailed implementation of the above steps, this invention achieves accurate measurement and dynamic evaluation of nonlinear loads, significantly improving the effectiveness and promotional value of the exercise intervention AI platform in national physical fitness monitoring and population-wide health management.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A nonlinear load measurement method, characterized in that, include: Physiological signal data are collected from a multi-source sensor chain through edge computing nodes, and key features in the time series are extracted using a signal decomposition algorithm to generate an initial load feature vector. Based on the initial load feature vector, the individual motion behavior data is segmented and processed. The motion pattern features are extracted using the sliding window technique. The similarity between the transformed feature vector and the initial load feature vector is judged. If the similarity exceeds a preset threshold, the load is initially matched. Real-time environmental parameters are obtained from the external environment data chain, and an adaptive weighting algorithm is used to fuse the environmental parameters with the preliminary load matching results to generate a comprehensive load vector. The process of obtaining real-time environmental parameters from an external environmental data chain and fusing the environmental parameters with the preliminary load matching results using an adaptive weighting algorithm to generate a comprehensive load vector includes: Real-time environmental parameters are obtained from the external environment data chain, and a structured environmental parameter dataset is obtained by using preset parameter extraction rules. Using a structured environmental parameter dataset, the weight values ​​of each parameter are calculated, and a weighted average method is used to generate a weighted set of environmental parameters. Individual load data are obtained from the initial load matching results. The vector embedding method is used to fuse the weighted set of environmental parameters with the individual load data to generate an initial comprehensive load vector. If the dimension of the initial composite load vector exceeds the preset threshold, principal component analysis is used to reduce the dimension and obtain the optimized composite load vector. Based on the optimized comprehensive load vector, the similarity between vectors is calculated, and the cosine similarity algorithm is used to determine the degree of load matching between individuals; By assessing the degree of load matching between individuals and combining it with preset classification rules, the load level of each individual is determined, and the final load assessment result is generated. Based on the final load assessment results, a data association analysis method is used to associate the assessment results with on-chain environmental parameter records to generate a verification dataset of the comprehensive load vector. Real-time correction data is obtained by calling an external data interface through a distributed communication protocol. It is then determined whether the correction data update time is within the monitoring period. If the correction data update time is within the correction time range, the validity of the correction data is determined. Based on the comprehensive load vector and the validity results of the corrected data, a weighted fusion calculation is performed to obtain the individual dynamic load score; If the dynamic load score is higher than the preset threshold, an automatic intervention logic is executed through a distributed consensus mechanism. The intervention process records are distributed and stored using a cryptographic hash function to generate an immutable log record. Based on the immutable log records, relevant verification paths are retrieved from the distributed network, and the integrity of the paths is determined. If the paths are complete, the final output result is determined.

2. The nonlinear load measurement method according to claim 1, characterized in that: The process involves collecting physiological signal data from a multi-source sensor chain via edge computing nodes, extracting key features from the time series using a signal decomposition algorithm, and generating an initial load feature vector, including: The system retrieves multi-source sensor data stored on the blockchain from edge computing nodes and uses a distributed query protocol to obtain raw signal data. The empirical mode decomposition technique in signal decomposition algorithms is used to decompose the original signal data and obtain the sequence of intrinsic mode functions. If the intrinsic mode function sequence contains a preset high-frequency noise component, then the high-frequency noise is filtered out by wavelet transform to obtain a denoised mode function sequence; The modality function sequence is transformed into a feature vector sequence through a pre-trained feature extraction model, generating an initial load feature vector. Based on the initial load feature vector, the principal component analysis algorithm is used to reduce the dimension of the vector to obtain the optimized load feature vector. The optimized load feature vector is used to update the initial load feature vector. If the similarity between the optimized feature vector and the preset load template exceeds a preset threshold, then the feature vector is determined to meet the load requirements, and a structured load description is generated. By using structured load descriptions, clustering algorithms are employed to classify load types and obtain classification results.

3. The nonlinear load measurement method according to claim 1, characterized in that: The process involves segmenting individual motion behavior data based on the initial load feature vector, extracting motion pattern features using a sliding window technique, and determining the similarity between the transformed feature vector and the initial load feature vector. If the similarity exceeds a preset threshold, a preliminary load match is determined. This includes: Initial load characteristic data is obtained, and feature vectors are extracted from it. Keyword extraction and semantic analysis methods are used to generate load feature vectors. Individual motion behavior data is segmented and processed, and key fields are extracted using sliding window technology to generate motion pattern data. The motion pattern data is converted into motion feature vectors through a pre-trained word vector model. Calculate the cosine similarity between the motion feature vector and the load feature vector to obtain a similarity value; If the similarity value exceeds a preset threshold, a preliminary load match is determined, and a matching result is generated. Based on the matching results, cluster analysis is used to group the individuals to obtain the grouping results; Based on the grouping results, the load feature vector of each individual in the group is extracted to generate a sorted list.

4. The nonlinear load measurement method according to claim 1, characterized in that: The process of obtaining real-time correction data by calling an external data interface through a distributed communication protocol, determining whether the correction data update time is within the monitoring period, and confirming the validity of the correction data if the update time is within the correction time range includes: The timestamps of real-time correction data are obtained from distributed data sources through external data interfaces and stored as first-time data. Using time comparison logic, the first-time data is compared with the preset monitoring period. If the first-time data is earlier than the end time of the monitoring period, the correction data update time is determined to be valid, and the validity judgment result is obtained. The validity determination result is obtained through the distributed communication protocol interface, and the validity status of the corrected data is generated. Based on the validity status, call the external data interface to obtain the complete dataset of real-time correction data and store it as the first correction data; The integrity and consistency of the first correction data are verified using a data verification algorithm to obtain the second correction data that has passed the verification. A classification algorithm is used to assess the health of the second corrected data and generate a health score. The final validity of the real-time corrected data is obtained based on the health score and validity determination results.

5. The nonlinear load measurement method according to claim 1, characterized in that: The step of performing a weighted fusion calculation based on the comprehensive load vector and the validity results of the corrected data to obtain an individual dynamic load score includes: Obtain individual information and the validity of correction data; determine the integrity and consistency of data through preset data verification rules; and obtain verified correction data quality. Based on the verified quality of the corrected data, load assessment indicators were extracted, and factor analysis was used to determine the comprehensive load vector. If at least one indicator in the comprehensive load vector exceeds a preset threshold, a preliminary fusion score is obtained by fusing load assessment indicators and correcting data quality using a linear weighting method. For the initial fusion score, a weighted fusion calculation is performed using a preset fusion weight allocation to obtain the adjusted fusion score; Based on the adjusted fusion score, a logistic regression model is applied to determine the individual's health level and output a dynamic load score; If the dynamic load score is lower than the preset threshold, the fusion weight allocation will be adjusted through the score calculation model, and the dynamic load score will be recalculated. Obtain the recalculated dynamic load score, and generate an individual comprehensive assessment score by outputting the assessment results.

6. The nonlinear load measurement method according to claim 1, characterized in that: The step of retrieving relevant verification paths from the distributed network based on the immutable log records, determining the path integrity, and if complete, determining the final output result includes: Log records are obtained from a distributed network, and a distributed node query protocol is used to extract a log dataset containing timestamps and transaction hashes to obtain the original log records. For the original log records, parse the log structure, extract the node sequence and hash chain in the verification path, and generate a path dataset; If the node sequence in the path dataset is consistent with the preset distributed topology, the Merkle tree verification algorithm is used to calculate the hash value of each node in the path, determine the path integrity, and obtain the integrity verification result. Based on the integrity verification results, if the integrity verification passes, the transaction timestamps and associated metadata are extracted from the path dataset to generate the review dataset; The SHA-256 algorithm is used to perform hash calculation on the review dataset to generate review hash values ​​and determine the uniqueness of the review results; By using network communication protocols, the audit hash value is compared with the preset audit standards to determine whether the audit result meets the standards and output the final audit result.

7. A nonlinear load measurement system, characterized in that, include: The load feature extraction module is used to collect physiological signal data from a multi-source sensor chain through edge computing nodes, extract key features from the time series using a signal decomposition algorithm, and generate an initial load feature vector. The load matching module is used to segment individual motion behavior data according to the initial load feature vector, extract motion pattern features using sliding window technology, and determine the similarity between the transformed feature vector and the initial load feature vector. If the similarity exceeds a preset threshold, the load is initially matched. The environment fusion module is used to obtain real-time environmental parameters from the external environment data chain, and use an adaptive weighting algorithm to fuse the environmental parameters with the preliminary load matching results to generate a comprehensive load vector. The process of obtaining real-time environmental parameters from an external environmental data chain and fusing the environmental parameters with the preliminary load matching results using an adaptive weighting algorithm to generate a comprehensive load vector includes: Real-time environmental parameters are obtained from the external environment data chain, and a structured environmental parameter dataset is obtained by using preset parameter extraction rules. Using a structured environmental parameter dataset, the weight values ​​of each parameter are calculated, and a weighted average method is used to generate a weighted set of environmental parameters. Individual load data are obtained from the initial load matching results. The vector embedding method is used to fuse the weighted set of environmental parameters with the individual load data to generate an initial comprehensive load vector. If the dimension of the initial composite load vector exceeds the preset threshold, principal component analysis is used to reduce the dimension and obtain the optimized composite load vector. Based on the optimized comprehensive load vector, the similarity between vectors is calculated, and the cosine similarity algorithm is used to determine the degree of load matching between individuals; By assessing the degree of load matching between individuals and combining it with preset classification rules, the load level of each individual is determined, and the final load assessment result is generated. Based on the final load assessment results, a data association analysis method is used to associate the assessment results with on-chain environmental parameter records to generate a verification dataset of the comprehensive load vector. The calibration verification module is used to call an external data interface through a distributed communication protocol to obtain real-time calibration data, determine whether the calibration data update time is within the monitoring period, and determine the validity of the calibration data if the calibration data update time is within the calibration time range. The scoring calculation module is used to perform weighted fusion calculation based on the comprehensive load vector and the validity results of the correction data to obtain an individual dynamic load score; The intervention execution module is used to execute automatic intervention logic through a distributed consensus mechanism if the dynamic load score is higher than a preset threshold. It uses a cryptographic hash function to distribute the intervention process records and generate immutable log records. The log verification module is used to retrieve relevant verification paths from the distributed network based on the immutable log records, determine the integrity of the paths, and determine the final output result if the paths are complete.

8. An electronic device, characterized in that, include: A memory on which computer programs are stored; A processor for executing the computer program in the memory to implement a nonlinear load measurement method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a nonlinear load measurement method as described in any one of claims 1 to 6.

Images