Intelligent anti-skid shoe system based on big data analysis
By adopting technologies such as big data analysis and Bayesian neural networks in the intelligent anti-slip shoe system, the traditional slip risk prediction system has solved the problem of inaccurate identification and unreliable prediction in complex environments, and achieved more efficient and reliable slip risk prediction.
Patent Information
- Application Number
- CN202411876447.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In complex and changeable environments, traditional slip risk prediction systems are difficult to provide efficient, accurate and reliable slip risk prediction due to inaccurate scene identification, insufficient processing capabilities for rare scenes, and lack of uncertainty evaluation and dynamic parameter adjustment.
An intelligent anti-slip shoe system based on big data analysis is adopted, historical and real-time scene data are obtained through the data acquisition module, and a Bayesian neural network is used to train the scene recognition model, probability distribution and uncertainty estimation of different scene types is identified, and a gradient lifting tree is used to train the slip risk prediction model, and the model parameters are dynamically adjusted to adapt to complex environments.
It significantly improves the accuracy of scene recognition and the reliability of slip and fall risk prediction, and can provide more accurate and reliable risk assessment in complex and changing environments, improving users' safety experience.
Smart Images
Figure CN119830126A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of big data analysis, and in particular to an intelligent anti-slip shoe system based on big data analysis. Background Art
[0002] The traditional slip risk prediction system has the following technical problems:
[0003] First, inaccurate scene recognition is a major shortcoming of traditional systems. Traditional systems usually rely on a single scene recognition model. When faced with complex or highly disturbed environments, they are prone to misidentifying wet ground as dry ground, which directly leads to the use of incorrect parameters in the slip risk prediction model, resulting in inaccurate prediction results. Specifically, traditional systems usually rely on a single scene recognition model, such as models based on fixed thresholds or simple classification algorithms. These models are prone to misidentification when faced with environmental interference or complex conditions, misjudging wet ground as dry ground, and thus using incorrect slip risk prediction parameters, resulting in inaccurate prediction results;
[0004] In addition, when training slip risk prediction models, traditional systems often lack effective processing for rare or mixed scene types. For example, there are fewer data samples in complex environments such as indoor and outdoor transition areas and dry and wet mixed ground. Traditional models may not be able to fully learn the characteristics of these scenes, resulting in insufficient prediction capabilities in these rare or mixed scenes, and unable to provide reliable slip risk assessments. At the same time, traditional systems lack an estimation and dynamic adjustment mechanism for the uncertainty of recognition results. When sensor data is fuzzy or environmental conditions are changeable, the system cannot judge the reliability of the current recognition results, and thus cannot provide effective warnings in high-risk environments, which seriously affects the user's safety experience.
[0005] Therefore, in complex and changing environments, traditional slip risk prediction systems are difficult to provide efficient, accurate and reliable slip risk predictions due to inaccurate scene recognition, insufficient processing capabilities for rare scenes, and lack of uncertainty assessment and dynamic parameter adjustment. Summary of the invention
[0006] The purpose of the present invention is to provide an intelligent anti-slip shoe system based on big data analysis to solve the problems raised in the above background technology.
[0007] To achieve the above object, the present invention provides the following technical solution: an intelligent anti-slip shoe system based on big data analysis, comprising a data acquisition module, a scene recognition model training module, a common scene type model training module and a generation and adjustment module, wherein:
[0008] The data acquisition module acquires historical scene data and real-time scene data, where the historical scene data represents sensor data under different scene types;
[0009] The scene recognition model training module uses a Bayesian neural network to train a scene recognition model based on historical scene data; the trained scene recognition model is used to identify the probability distribution of different scene types and the uncertainty estimation of the recognition results;
[0010] The common scenario type model training module divides the historical scenario data into multiple subsets according to the scenario type, and uses the gradient boosting tree to train the slip risk prediction model for each subset, and obtains the slip risk prediction model parameters under different scenario types;
[0011] The generation and adjustment module inputs real-time scene data into the trained scene recognition model, calculates the hybrid model parameters according to the probability distribution of different scene types in the output results and the slip risk prediction model parameters; and selects the scene type with the largest probability distribution according to the uncertainty estimation of the recognition results and the hybrid model parameters, and adjusts the slip risk prediction model parameters corresponding to the scene type.
[0012] As a further improvement of the technical solution, the historical scene data in the data acquisition module includes sensor data, scene type and slip risk.
[0013] As a further improvement of the technical solution, the scene recognition model training module performs scene recognition model training specifically including:
[0014] The sensor data in the historical scene data is used as input; the Bayesian neural network contains multiple hidden layers, each of which contains multiple neurons; the scene recognition model outputs two main results, including the scene type probability distribution and the uncertainty estimation of the recognition result, where the scene type probability distribution represents the probability that the current input data belongs to different scene types; the uncertainty estimation of the recognition result represents the degree of uncertainty of the scene recognition model about the current recognition result;
[0015] Normalize the sensor data; initialize the parameters of the Bayesian neural network, including weights and biases; input the processed sensor data into the Bayesian neural network, and finally obtain the scene type probability distribution and uncertainty estimation through calculations of multiple hidden layers;
[0016] Logarithmic loss is selected as the loss function, which not only considers the accuracy of scene type recognition, but also the accuracy of uncertainty estimation. Through the back-propagation algorithm, the parameters of the scene recognition model are updated according to the gradient of the loss function, and the recognition ability and uncertainty estimation of the model are gradually optimized to complete the training.
[0017] As a further improvement of the technical solution, the process of the common scenario type model training module using the gradient boosting tree to train the slip risk prediction model for each subset includes:
[0018] According to different scene types in the historical scene data, the historical scene data is divided into multiple subsets, each subset only contains data of a specific scene type; for example, the historical scene data is divided into four subsets: dry scene data, wet scene data, frozen scene data and mixed scene data;
[0019] Gradient boosting tree is selected as the slip risk prediction model, taking the sensor data in each subset as input;
[0020] The gradient boosting tree is selected to consist of multiple decision trees, and the residual of the previous tree is reduced by gradually adding trees to improve the accuracy of prediction; the sensor data in each subset is normalized; the parameters of the gradient boosting tree are initialized; the normalized sensor data is input into the gradient boosting tree, and the probability of slip risk is finally obtained through the step-by-step calculation of multiple decision trees;
[0021] The mean square error is selected as the loss function. The parameters of the gradient boosting tree are updated according to the gradient of the loss function through the gradient boosting method to gradually optimize the prediction ability of the model. Multiple rounds of training are performed on each subset until the prediction accuracy of the slip risk prediction model reaches the preset standard.
[0022] As a further improvement of the present technical solution, after the training of the slip risk prediction model in the common scenario type model training module is completed, the parameters of the slip risk prediction model generated on each subset are saved, and the parameters include the structure and weight of each decision tree, representing the slip risk prediction model parameters under different scenario types.
[0023] As a further improvement of the technical solution, the generation and adjustment module includes a hybrid model parameter calculation unit, and the process of the hybrid model parameter calculation unit calculating the hybrid model parameters specifically includes:
[0024] Input the real-time sensor data into the scene recognition model to obtain the probability distribution S of different scene types, where S = [s1, s2, ..., sn], n represents the number of all scene types, and sn represents the probability distribution of the nth scene type;
[0025] According to the probability distribution S of different scene types, the mixed model parameter θmix is calculated. The mixed model parameter θmix is the weighted average of the slip risk prediction model parameters under different scene types. The weight is the probability of different scene types in the current scene. The formula is as follows:
[0026] θmix=s1·θ1+s2·θ2+...+sn·θn, where θn is the slip risk prediction model parameter under the nth scenario type.
[0027] As a further improvement of the technical solution, the generation and adjustment module includes an adjustment unit, and the process of the adjustment unit adjusting the parameters of the slip risk prediction model specifically includes:
[0028] Select the scene type smax with the largest probability and its corresponding slip risk prediction model parameter θmax from the probability distribution S of different scene types;
[0029] Obtain the uncertainty estimate u of the recognition result from the scene recognition model;
[0030] Define a weight function α(u) that dynamically adjusts the weights of the hybrid model parameters based on the uncertainty estimate u, where the weight function α(u) is in the following form:
[0031] Where β is an adjustment parameter used to control the degree of influence of uncertainty on the weight;
[0032] According to the weight function α(u) and the mixed model parameter θmix, the slip risk prediction model parameter θmax is adjusted to generate θadjusted. The specific formula is:
[0033] θadjusted=(1-α(u))·θmax+α(u)·θmix.
[0034] Compared with the prior art, the present invention has the following beneficial effects:
[0035] This intelligent anti-slip shoe system based on big data analysis significantly improves the accuracy of scene recognition through multi-source data fusion and Bayesian neural network scene recognition model. The data acquisition module not only collects historical sensor data, but also combines real-time data for comprehensive judgment. The scene recognition model training module uses Bayesian neural network to not only identify the probability distribution of different scene types, but also provides uncertainty estimation of the recognition results. This dual factor ensures that the system can more accurately identify ground conditions in complex and changing environments, reduce the probability of misidentification, and thus improve the reliability of slip risk prediction.
[0036] In addition, the slip risk prediction is further optimized through a dynamic adjustment mechanism; the hybrid model parameter calculation unit calculates the probability weighted average of different scenario types based on real-time scenario data to generate hybrid model parameters. The adjustment unit dynamically adjusts the slip risk prediction model parameters based on the uncertainty estimation of scenario recognition, so that the system can flexibly select the most appropriate prediction model parameters and provide more accurate warnings when facing fuzzy sensor data and changing environments; this dynamic adjustment mechanism ensures that the system can provide reliable risk assessments in rare or mixed scenarios, greatly improving the user's safety experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a schematic diagram of the overall module of the present invention;
[0038] Figure 2 It is a schematic diagram of the generation and adjustment module unit of the present invention.
[0039] In the figure: 100, data acquisition module; 200, scene recognition model training module; 300, common scene type model training module; 400, generation and adjustment module; 401, hybrid model parameter calculation unit; 402, adjustment unit. DETAILED DESCRIPTION
[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0041] Next, see Figure 1-Figure 2 The present invention provides a technical solution: an intelligent anti-slip shoe system based on big data analysis, including a data acquisition module 100, a scene recognition model training module 200, a common scene type model training module 300 and a generation and adjustment module 400.
[0042] The data acquisition module 100 acquires real-time scene data using sensors and acquires historical scene data using a public database. The public database contains a large amount of sensor data, which usually comes from actual environments of various scene types and weather conditions; the historical scene data includes but is not limited to the following:
[0043] Sensor data: accelerometer data, gyroscope data, temperature sensor data, humidity sensor data, etc.
[0044] Scene type: Each set of data is marked with the current scene type, such as dry, wet, frozen, mixed, etc.;
[0045] Slip risk: Each set of data also marks the probability of slip risk, that is, the possibility of the user slipping in the current scenario type.
[0046] The scene recognition model training module 200 uses a Bayesian neural network to train a scene recognition model based on historical scene data; the trained scene recognition model is used to identify the probability distribution of different scene types and the uncertainty estimation of the recognition results. The training process specifically includes:
[0047] The sensor data in the historical scene data is used as input, which reflects the activity characteristics of users in different scenes;
[0048] The Bayesian neural network contains multiple hidden layers, each of which contains multiple neurons. By increasing the number of hidden layers and neurons, the expressiveness and accuracy of the scene recognition model are improved.
[0049] The scene recognition model outputs two main results, including the scene type probability distribution and the uncertainty estimate of the recognition result. The scene type probability distribution indicates the probability that the current input data belongs to different scene types. For example, the model outputs that the probability that a set of data belongs to dry, wet, frozen, and mixed scene types is 0.6, 0.3, 0.05, and 0.05, respectively. The uncertainty estimate of the recognition result indicates the degree of uncertainty of the scene recognition model about the current recognition result. The lower the uncertainty estimate, the more confident the scene recognition model is in recognizing the current scene type. The higher the uncertainty estimate, the less confident the scene recognition model is in recognizing the current scene type.
[0050] Normalize the sensor data to ensure that each feature value is within the same numerical range and improve the stability of scene recognition model training;
[0051] Initialize the parameters of the Bayesian neural network, including weights and biases; input the processed sensor data into the Bayesian neural network, and finally obtain the scene type probability distribution and uncertainty estimation through calculations of multiple hidden layers;
[0052] The logarithmic loss is selected as the loss function, which not only considers the accuracy of scene type recognition, but also the accuracy of uncertainty estimation. The parameters of the scene recognition model are updated according to the gradient of the loss function through the back-propagation algorithm, and the recognition ability and uncertainty estimation of the model are gradually optimized to complete the training.
[0053] After training is completed, the scene recognition model can output the probability distribution of the scene type to which the current input data belongs, as well as the uncertainty estimate of the recognition result.
[0054] The common scene type model training module 300 divides the historical scene data into multiple subsets according to the scene type, and uses the gradient boosting tree to train the slip risk prediction model for each subset, and obtains the slip risk prediction model parameters under different scene types, specifically including:
[0055] According to different scene types in the historical scene data, the historical scene data is divided into multiple subsets, each subset only contains data of a specific scene type; for example, the historical scene data is divided into four subsets: dry scene data, wet scene data, frozen scene data and mixed scene data;
[0056] Gradient boosting tree is selected as the slip risk prediction model, and the sensor data in each subset is used as input, which reflects the activity characteristics of users in a specific scenario type;
[0057] The gradient boosting tree is selected to consist of multiple decision trees. The residual of the previous tree is reduced by gradually adding trees, thereby improving the accuracy of the prediction; the sensor data in each subset is normalized to ensure that each feature value is in the same numerical range to improve the stability of model training; the parameters of the gradient boosting tree are initialized, including the number of trees, the depth of the tree, the learning rate, etc.; the normalized sensor data is input into the gradient boosting tree, and the probability of slip risk is finally obtained through the step-by-step calculation of multiple decision trees;
[0058] The mean square error is selected as the loss function. The parameters of the gradient boosting tree are updated according to the gradient of the loss function through the gradient boosting method to gradually optimize the prediction ability of the model. Multiple rounds of training are performed on each subset until the prediction accuracy of the slip risk prediction model reaches the preset standard.
[0059] After the training is completed, the parameters of the slip risk prediction model generated on each subset are saved, including the structure and weight of each decision tree, which will be used for subsequent hybrid model generation and slip risk prediction.
[0060] The hybrid model parameter calculation unit 401 in the generation and adjustment module 400 inputs the real-time scene data into the trained scene recognition model, and calculates the hybrid model parameters according to the probability distribution of different scene types and the slip risk prediction model parameters in the output results, specifically including:
[0061] The real-time sensor data is input into the scene recognition model to obtain the probability distribution S of different scene types, where S = [s1, s2, ..., sn], n represents the number of all scene types, and sn represents the probability distribution of the nth scene type; for example, the scene types include dry, wet, frozen and mixed, and the probability distribution of different scene types S = [s1, s2, s3, s4], which respectively represent the probabilities of belonging to the dry, wet, frozen and mixed scene types;
[0062] According to the probability distribution S of different scene types, the mixed model parameter θmix is calculated. The mixed model parameter θmix is the weighted average of the slip risk prediction model parameters under different scene types. The weight is the probability of different scene types in the current scene. The formula is as follows:
[0063] θmix=s1·θ1+s2·θ2+...+sn·θn, where θn is the slip risk prediction model parameter under the nth scenario type.
[0064] The adjustment unit 402 in the generation and adjustment module 400 selects the largest scene type in the probability distribution according to the uncertainty estimation of the recognition result and the hybrid model parameters, and adjusts the slip risk prediction model parameters corresponding to the scene type, specifically including:
[0065] Select the scene type smax with the largest probability and its corresponding slip risk prediction model parameter θmax from the probability distribution S of different scene types;
[0066] Obtain the uncertainty estimate u of the recognition result from the scene recognition model. The lower the uncertainty estimate, the more confident the model is in recognizing the current scene type; the higher the uncertainty estimate, the less confident the model is in recognizing the current scene type.
[0067] Define a weight function α(u) that dynamically adjusts the weights of the hybrid model parameters based on the uncertainty estimate u, where the weight function α(u) is in the following form:
[0068] Where β is an adjustment parameter used to control the degree of influence of uncertainty on the weight;
[0069] According to the weight function α(u) and the mixed model parameter θmix, the slip risk prediction model parameter θmax is adjusted to generate θadjusted. The specific formula is:
[0070] θadjusted=(1-α(u))·θmax+α(u)·θmix.
[0071] The adjusted slip risk prediction model parameter θadjusted is used to predict the slip risk, and the real-time sensor data is input into the adjusted slip risk prediction model to obtain the probability of slip risk.
[0072] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. Intelligent anti-slip shoe system based on big data analysis, characterized in that: The system comprises a data acquisition module (100), a scene recognition model training module (200), a common scene type model training module (300) and a generation and adjustment module (400), wherein: The data acquisition module (100) acquires historical scene data and real-time scene data, wherein the historical scene data represents sensor data under different scene types; The scene recognition model training module (200) uses a Bayesian neural network to train a scene recognition model based on historical scene data; the trained scene recognition model is used to identify the probability distribution of different scene types and estimate the uncertainty of the recognition result; The common scene type model training module (300) divides the historical scene data into multiple subsets according to the scene type, and uses a gradient boosting tree to train a slip risk prediction model for each subset, and obtains slip risk prediction model parameters under different scene types; The generation and adjustment module (400) inputs real-time scene data into a trained scene recognition model, calculates mixed model parameters according to the probability distribution of different scene types in the output results and the slip risk prediction model parameters; and selects the scene type with the largest probability distribution according to the uncertainty estimation of the recognition results and the mixed model parameters, and adjusts the slip risk prediction model parameters corresponding to the scene type.
2. The intelligent anti-slip shoe system based on big data analysis according to claim 1 is characterized in that: The historical scene data in the data acquisition module (100) includes sensor data, scene type and slip risk.
3. The intelligent anti-slip shoe system based on big data analysis according to claim 2 is characterized in that: The scene recognition model training module (200) performs scene recognition model training specifically including: The sensor data in the historical scene data is used as input; the Bayesian neural network contains multiple hidden layers, each of which contains multiple neurons; the scene recognition model outputs two main results, including the scene type probability distribution and the uncertainty estimation of the recognition result, where the scene type probability distribution represents the probability that the current input data belongs to different scene types; the uncertainty estimation of the recognition result represents the degree of uncertainty of the scene recognition model about the current recognition result; Normalize the sensor data; initialize the parameters of the Bayesian neural network, including weights and biases; input the processed sensor data into the Bayesian neural network, and finally obtain the scene type probability distribution and uncertainty estimation through calculations of multiple hidden layers; Logarithmic loss is selected as the loss function, which not only considers the accuracy of scene type recognition, but also the accuracy of uncertainty estimation. Through the back-propagation algorithm, the parameters of the scene recognition model are updated according to the gradient of the loss function, and the recognition ability and uncertainty estimation of the model are gradually optimized to complete the training.
4. The intelligent anti-slip shoe system based on big data analysis according to claim 3 is characterized in that: The process of the common scenario type model training module (300) using the gradient boosting tree to train the slip risk prediction model for each subset includes: According to different scene types in the historical scene data, the historical scene data is divided into multiple subsets, each subset only contains data of a specific scene type; for example, the historical scene data is divided into four subsets: dry scene data, wet scene data, frozen scene data and mixed scene data; Gradient boosting tree is selected as the slip risk prediction model, taking the sensor data in each subset as input; The gradient boosting tree is selected to consist of multiple decision trees, and the residual of the previous tree is reduced by gradually adding trees to improve the accuracy of prediction; the sensor data in each subset is normalized; the parameters of the gradient boosting tree are initialized; the normalized sensor data is input into the gradient boosting tree, and the probability of slip risk is finally obtained through the step-by-step calculation of multiple decision trees; The mean square error is selected as the loss function. The parameters of the gradient boosting tree are updated according to the gradient of the loss function through the gradient boosting method to gradually optimize the prediction ability of the model. Multiple rounds of training are performed on each subset until the prediction accuracy of the slip risk prediction model reaches the preset standard.
5. The intelligent anti-slip shoe system based on big data analysis according to claim 4 is characterized in that: After the slip risk prediction model training in the common scenario type model training module (300) is completed, the parameters of the slip risk prediction model generated on each subset are saved, and the parameters include the structure and weight of each decision tree, representing the slip risk prediction model parameters under different scenario types.
6. The intelligent anti-slip shoe system based on big data analysis according to claim 5 is characterized in that: The generation and adjustment module (400) comprises a hybrid model parameter calculation unit (401), and the process of the hybrid model parameter calculation unit (401) calculating the hybrid model parameters specifically comprises: Input the real-time sensor data into the scene recognition model to obtain the probability distribution S of different scene types, where S = [s1, s2, ..., sn], n represents the number of all scene types, and sn represents the probability distribution of the nth scene type; According to the probability distribution S of different scene types, the mixed model parameter θmix is calculated. The mixed model parameter θmix is the weighted average of the slip risk prediction model parameters under different scene types. The weight is the probability of different scene types in the current scene. The formula is as follows: θmix=s1·θ1+s2·θ2+...+sn·θn, where θn is the slip risk prediction model parameter under the nth scenario type.
7. The intelligent anti-slip shoe system based on big data analysis according to claim 6 is characterized in that: The generation and adjustment module (400) comprises an adjustment unit (402), and the process of adjusting the parameters of the slip risk prediction model by the adjustment unit (402) specifically comprises: Select the scene type smax with the largest probability and its corresponding slip risk prediction model parameter θmax from the probability distribution S of different scene types; Obtain the uncertainty estimate u of the recognition result from the scene recognition model; Define a weight function α(u) that dynamically adjusts the weights of the hybrid model parameters based on the uncertainty estimate u, where the weight function α(u) is in the following form: Where β is an adjustment parameter used to control the degree of influence of uncertainty on the weight; According to the weight function α(u) and the mixed model parameter θmix, the slip risk prediction model parameter θmax is adjusted to generate θadjusted. The specific formula is: θadjusted=(1-α(u))·θmax+α(u)·θmix.