Intelligent miner lamp based on brain-computer interface technology

Abstract

The invention discloses a brain-computer interface technology-based intelligent miner lamp, which comprises a miner lamp body, a brain-computer interface acquisition terminal module, a brain-computer interface processing chip module, a voice talkback module, a vibration module and a liquid crystal module, and is characterized in that the miner lamp body comprises a miner lamp box and a miner lamp cap, the brain-computer interface acquisition terminal module is arranged in the miner lamp cap, and the voice talkback module is arranged in the miner lamp cap; and the brain-computer interface processing chip module, the voice talkback module, the vibration module and the liquid crystal module are arranged on the miner lamp box. The intelligent mine lamp management system belongs to the technical field of intelligent mine lamp management systems, not only greatly enhances the functionality of the mine lamp, but also effectively solves the limitation of the traditional mine lamp on health monitoring, communication, positioning and interaction through intelligent upgrading, and creative safety improvement is brought to mine operation management.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent miner's lamp management systems, and specifically refers to an intelligent miner's lamp based on brain-computer interface technology. Background Art

[0002] As a key safety device in mine operations, traditional miner's lamps mainly focus on providing illumination to ensure clear vision for miners during underground operations. However, existing miner's lamp designs usually lack intelligent functions and merely serve as a simple light source device, having limitations in instant communication in emergency situations and monitoring the health status of miners. Especially for fatigue monitoring caused by long-term operations, assessment of consciousness status, and precise tracking of miners' positions in complex mine environments, existing technologies are difficult to provide effective solutions.

[0003] The current situation of existing technologies is as follows:

[0004] 1) Insufficient health monitoring. Traditional miner's lamps cannot real-time monitor the physiological status of miners, such as fatigue level and cognitive load, which is crucial for preventing accidents caused by fatigue.

[0005] 2) Poor emergency communication. Lack of built-in communication modules, miners cannot quickly establish communication with the ground control center or colleagues in case of emergencies.

[0006] 3) Low positioning accuracy. The complex terrain inside the mine limits the accuracy of traditional positioning technologies, making it difficult to quickly locate the position of miners during emergency rescue.

[0007] 4) Single interaction method. Miner's lamps are usually only manually controlled, lacking non-contact interaction methods such as voice control or consciousness control, which is particularly inconvenient when hands are occupied or in emergency situations.

[0008] Therefore, there is an urgent need for an intelligent miner's lamp based on brain-computer interface technology to solve these problems. Summary of the Invention

[0009] In view of the above situation, to overcome the defects of the existing technology, the present invention provides an intelligent miner's lamp based on brain-computer interface technology. By integrating comprehensive applications such as a brain-computer interface acquisition terminal module and a brain-computer interface processing chip module, it can collect and analyze the electroencephalogram data, pulse rate, blood oxygen saturation, acceleration, and positioning data of miners in real time, realizing advantages such as real-time health monitoring, efficient communication, precise positioning, non-contact interaction, and comprehensive display. The present invention not only greatly enhances the functionality of the miner's lamp, but also effectively solves the limitations of traditional miner's lamps in health monitoring, communication, positioning, and interaction through intelligent upgrading, bringing a creative safety improvement to mine operation management.

[0010] The technical solution adopted by the present invention is as follows: An intelligent miner's lamp based on brain-computer interface technology provided by the present invention includes a miner's lamp body, a brain-computer interface acquisition terminal module, a brain-computer interface processing chip module, a voice intercom module, a vibration module, and a liquid crystal module. The miner's lamp body includes a miner's lamp box and a miner's lamp cap. The brain-computer interface acquisition terminal module is arranged in the miner's lamp cap, and the brain-computer interface processing chip module, the voice intercom module, the vibration module, and the liquid crystal module are arranged on the miner's lamp box;

[0011] The brain-computer interface acquisition terminal module is used to collect comprehensive information of electroencephalogram data, pulse rate data, blood oxygen data, acceleration data, and positioning data, and transmit the information to the brain-computer interface processing chip module;

[0012] The brain-computer interface processing chip module is used to receive the information collected by the brain-computer interface acquisition terminal module, and perform real-time calculations in the cloud or the brain-computer interface processing chip module, and calculate the comprehensive information of electroencephalogram data, pulse rate data, blood oxygen data, acceleration data, and positioning data in real time, so as to realize the real-time monitoring functions of consciousness state, fatigue state, cognitive load, work intensity, and positioning;

[0013] The voice intercom module receives the data of the brain-computer interface processing chip module and is used for voice intercom;

[0014] The vibration module receives the data of the brain-computer interface processing chip module and is used for vibration reminder;

[0015] The liquid crystal module receives the data of the brain-computer interface processing chip module and is used for liquid crystal display.

[0016] Furthermore, the brain-computer interface acquisition terminal module includes a non-invasive EEG acquisition module, an optical pulse sensor, a blood oxygen meter, an accelerometer, and a positioning module.

[0017] The non-invasive EEG acquisition module adopts a non-invasive EEG acquisition method, through a single-channel or more advanced multi-channel EEG acquisition module (including the TGAM module or its subsequent evolved version), and these modules have miniaturization and portability, and are suitable for long-term wearing by miners. Using forehead patch electrodes and ear clip electrodes, dry electrode technology is used to reduce skin irritation and improve wearing comfort and signal quality.

[0018] In addition to electroencephalogram data, the brain-computer interface acquisition terminal module also integrates advanced sensor technology, including an optical pulse sensor, a blood oxygen meter, an accelerometer, and a positioning module, to synchronously collect pulse rate, blood oxygen saturation, acceleration data, and positioning data, and the brain-computer interface acquisition terminal module has high precision and low power consumption to ensure stable operation in the mine environment.

[0019] Preferably, the brain-computer interface processing chip module adopts a multi-modal fusion neural network algorithm, which specifically includes the following steps:

[0020] (1) Signal preprocessing: Wavelet denoising is used to remove the noise in the EEG data; the specific algorithm formula for wavelet denoising is:

[0021] In the formula, represents the denoised coefficient of the j-th level decomposition at time point n after wavelet denoising; S j [n] represents the original wavelet coefficient, reflecting the characteristics of the signal at different scales after the j-th level decomposition; λ j represents the threshold related to the decomposition level j, which is used to distinguish the boundary between the signal and the noise; n represents the index of the time series, indicating the data at a specific time point; S j represents the approximate coefficient after wavelet decomposition, reflecting the low-frequency part of the signal;

[0022] After the above signal preprocessing steps, the noise can be removed more precisely, while the useful information of the EEG signal is retained, especially the protection of low-intensity signals, which improves the signal-to-noise ratio of the signal. The above wavelet denoising is suitable for real-time signal processing, quickly adapts to the EEG changes of miners in different working states, and improves the response speed and accuracy of real-time monitoring of the system. The design and application of this step reflect the efficiency and precision of EEG signal processing in a complex mine environment and are an indispensable advanced technology in the intelligent miner's lamp system.

[0023] (2) Feature extraction: A convolutional neural network is used to extract features from the EEG data, and at the same time, machine learning algorithms are used for feature engineering of pulse rate data, blood oxygen data, acceleration data, and positioning data;

[0024] (3) Pattern recognition and fusion: A long short-term memory network combined with a support vector machine is used for pattern recognition to achieve the classification of consciousness states and physiological states; the long short-term memory network is used to process time series data,

[0025] while the support vector machine is used for the classified feature vectors, and the fusion strategy adopts a fusion strategy based on the attention mechanism;

[0026] (4) Real-time decision-making and feedback: An optimized decision tree algorithm is used to make decisions quickly, including fatigue state warning and position deviation reminder; the specific optimized decision tree algorithm is: In the formula, P k represents the proportion of the k-th class; Gini(D) represents the impurity of the data set D, that is, the degree of mixing of each class in the data set; k represents the number of classes in the data set;

[0027] In the scenario of intelligent miner's lamps, the optimized decision tree algorithm analyzes the features extracted from multi-dimensional data such as electroencephalogram data and physiological parameters to determine whether the fatigue state or position of the miner deviates from the normal trajectory. The algorithm determines the branch nodes of the tree based on the Gini index, and finds the direction that can make the Gini index drop fastest for division, so as to achieve rapid and accurate classification of the miner's state. When the algorithm identifies that the physiological feature combination of a certain miner points to a high fatigue risk, it will immediately trigger a fatigue state warning, send a reminder through the voice intercom module or vibration module, and at the same time the liquid crystal module displays specific warning information to ensure the safety and operation efficiency of the miner. The advantage of this application lies in its high efficiency and adaptability, which can process complex data streams in real time, make decisions quickly, and effectively improve the safety management of mine operations.

[0028] (5) System feedback and optimization: Introduce an online learning mechanism to adjust the algorithm parameters according to the immediate feedback of the miner, and update the algorithm using the strategy of reinforcement learning; the specific algorithm for updating the strategy of reinforcement learning is: π(α|s), where s represents the state; α represents the action; π(α|s) represents the strategy of taking action α under the given state s.

[0029] For the reinforcement learning strategy update algorithm π(α|s), the specific meaning is that for each state s, there is an action probability distribution to guide how the system responds; when detecting the fatigue state s of the miner, the strategy π can instruct the system to automatically reduce the illumination intensity of the miner's lamp (one of the αs) and send a slight vibration reminder (the other α) to reduce visual stimulation, and at the same time prompt to rest through the voice module.

[0030] Preferably, in step (2), the machine learning algorithm specifically includes the following steps:

[0031] S21: Data preprocessing, standardize or normalize the non-electroencephalogram data to ensure that data from different sources can be compared on a unified scale;

[0032] S22: Feature selection and engineering,

[0033] For pulse rate data, use time-domain analysis to extract features of heart rate variability, including average heart rate and standard deviation;

[0034] For blood oxygen data, pay attention to the stability and change rate of blood oxygen level as an indirect indicator of fatigue;

[0035] For acceleration data, use frequency-domain analysis to extract motion intensity features and identify stationary and active states;

[0036] For positioning data, analyze the pattern of position changes to identify behaviors that deviate from the normal working path;

[0037] S23: Application of machine learning algorithm;

[0038] Deep learning, using a convolutional neural network to perform feature learning on non-EEG data. The convolutional neural network is particularly suitable for processing data with dense time series or spatial features, including processing acceleration data through window sliding of time series to capture dynamic behavior patterns;

[0039] Support vector machine. For classification tasks, the support vector machine is used to handle the division of the feature space, especially when dealing with the classification boundaries of blood oxygen saturation and pulse rate, including distinguishing normal and abnormal states;

[0040] Ensemble learning, using gradient boosting trees, can handle multi-dimensional features, enhance the generalization ability of the model, and is suitable for comprehensive analysis of various physiological parameters.

[0041] Furthermore, the feature selection of the blood oxygen data includes the mean value M of blood oxygen saturation O2 and the standard deviation SD of blood oxygen saturation O2 ; The specific algorithm formula is as follows:

[0042]

[0043] In the formula, M O2 represents the mean value of blood oxygen saturation, that is, within the time period, the oxygenated hemoglobin in the blood accounts for the total hemoglobin; N represents the total number of observations or samples, which is a statistical parameter used to calculate the mean value and variance; represents the sum of blood oxygen data from the 1st to the Nth time points, which is used to calculate the mean value; O2[i] represents the blood oxygen saturation value at time point i; i represents the index symbol, which is used to mark a specific time point in the time series; SD O2 represents the standard deviation of blood oxygen saturation; represents the sum of the squares of the differences between the blood oxygen values at each time point and the mean value, which is used to measure the volatility of the data.

[0044] Furthermore, in step (5) system feedback and optimization, the policy update algorithm π(α|s) of the reinforcement learning is through iterative learning and optimization, enabling the system to self-adjust according to the miner's immediate feedback. The core lies in using a surrogate objective function, which allows multiple gradient updates without disrupting the policy learning process. The surrogate objective function L clip (θ) The specific algorithm formula is as follows:

[0045]

[0046] In the formula, γ t (θ) represents the ratio of the old and new policies on the action α t ; π θ (α t|s t ) represents the probability of taking action α t in state s t ; represents the probability of the old policy taking action α t in state s t ; represents the advantage estimate, which measures the quality of action α t relative to the average action; α t represents the action taken at time t; ∈ represents a hyperparameter that controls the conservativeness of policy updates; θ represents the parameters of the policy network; represents the expected operation at time t, which is used to evaluate the consistency performance of the policy in different states.

[0047] The surrogate objective function L clip (θ) proposed in this scheme cleverly controls the amplitude of policy updates and ensures the stability of learning. When the value of γ t (θ) exceeds the range of [1 - ∈, 1 + ∈], it is clipped by the clip(·) function to avoid performance collapse caused by sudden policy changes. This mechanism performs well in dealing with continuous action space problems, especially suitable for the fine adjustment of miners' states in intelligent miner's lamp systems. In the actual application of intelligent miner's lamps, the surrogate objective function L clip (θ) is not only used to adjust the strategies of light intensity and vibration reminder, but also can be dynamically adjusted according to the physiological indicators and location information of miners to ensure the most appropriate support and warning in the complex and changeable mine environment.

[0048] Compared with traditional policy gradient methods, the surrogate objective function L clip (θ) effectively avoids the problems of low data sample utilization and poor robustness, and shows better stability and data efficiency in dealing with continuous action space problems. In addition, by dynamically adjusting the clipping probability ratio in the objective function, a gentle transition of policy updates is ensured, significantly improving the robustness of the algorithm.

[0049] As a further preference, the specific steps of the surrogate objective function L clip (θ) are as follows:

[0050] (1) Data collection, using the current policy to interact with the environment and collect a set of experiences (s t , α t , γ t , s t+1 ); α t represents the action taken at time t; s t represents the state at time t; γ t represents the immediate reward at time t;

[0051] (2) Advantage estimation, calculating the advantage function for each time step t Approximately calculated by using the value function V(s t ); V(s t ) represents the estimated value function of state s t at time step t;

[0052] (3) Policy update, in each epoch, using the surrogate objective function L clip (θ) to update the policy parameter θ multiple times until the preset number of updates is reached or the stopping criterion is satisfied;

[0053] (4) Value function update, using the mean squared error loss function to update the value function parameters to improve the accuracy of advantage estimation;

[0054] (5) Repeat, go back to the first step, continue to collect data until the termination condition of the entire training process is satisfied.

[0055] Furthermore, the voice intercom module is based on the CC2500 intercom system to achieve real-time voice communication with miners; when the main control system needs to communicate with miners, the PCM encoded signal stored in the Flash in the system is sent via the CC2500 wireless module. After the receiving end receives the signal, it is processed by the AVR single-chip microcomputer and converted into sound output to ensure clear voice communication even in complex environments.

[0056] Furthermore, the vibration module is used for silent physical reminders; when the system detects specific conditions, including fatigue warnings and position deviations, it will trigger the module to generate vibrations, reminding miners to pay attention to safety through physical touch, without relying on hearing or vision, which is suitable for noisy or sight-restricted mine environments.

[0057] Furthermore, the liquid crystal module displays key information, including miner status, position indication, and warning information. On the intelligent miner's lamp worn by miners, the liquid crystal module provides an intuitive visual interface, enhancing the immediacy and readability of information.

[0058] Adopting technologies such as LCD, it can display numbers, texts, and simple graphics to ensure that miners can clearly view important data even under insufficient underground light. The voice intercom module, vibration module, and liquid crystal module work in coordination through the policy update algorithm π(α|s) of reinforcement learning. Adjust the response strategy according to the immediate feedback of miners. When a fatigue signal is recognized, not only through voice reminders, but also assisted by the vibration module, and at the same time, specific information is displayed on the liquid crystal to ensure the all-round transmission of information.

[0059] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0060] (1) Real-time health monitoring. The present invention can continuously monitor the health status of miners, such as fatigue and cognitive load, timely warn of potential health risks, and improve operation safety.

[0061] (2) Efficient communication. The present invention realizes instant communication and silent reminder even in a noisy environment through the voice intercom module and vibration module, enhancing the instant interaction among miners.

[0062] (3) Precise positioning. The present invention utilizes the built-in positioning data processing to improve the positioning accuracy of miners in complex mine shafts, providing accurate location information for emergency rescue.

[0063] (4) Non-contact interaction. Through the introduction of brain-computer interface technology, the present invention enables miners to control the functions of the miner's lamp through consciousness, greatly improving the operation convenience and safety, especially in scenarios where the hands are not freely usable.

[0064] (5) Comprehensive display. The present invention intuitively displays key information through the liquid crystal module, facilitating miners to understand their own status and communication content at any time, and improving work efficiency and safety.

[0065] (6) In the practical application of the intelligent miner's lamp, the surrogate objective function L clip (θ) is not only used to adjust the strategies of light intensity and vibration reminder, but also can be dynamically adjusted according to the physiological indexes and position information of miners to ensure the most suitable support and warning in the complex and changeable mine environment. Compared with the traditional policy gradient method, through the surrogate objective function L clip (θ), the problems of low data sample utilization rate and poor robustness are effectively avoided, and better stability and data efficiency are shown when dealing with continuous action space problems. In addition, by dynamically adjusting the clipping probability ratio in the objective function, a gentle transition of policy update is ensured, significantly improving the robustness of the algorithm.

[0066] In summary, the intelligent miner's lamp based on brain-computer interface technology provided by the present invention can collect and analyze the electroencephalogram data, pulse rate, blood oxygen saturation, acceleration and positioning data of miners in real time through the integrated application of the brain-computer interface acquisition terminal module, brain-computer interface processing chip module, etc., realizing the advantageous functions of real-time health monitoring, efficient communication, precise positioning, non-contact interaction, comprehensive display, etc.; the present invention not only greatly enhances the functionality of the miner's lamp, but also effectively solves the limitations of traditional miner's lamps in health monitoring, communication, positioning and interaction through intelligent upgrading, bringing a creative safety improvement to mine operation management. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a module schematic diagram of the intelligent miner's lamp based on brain-computer interface technology provided by the present invention.

[0068] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the description. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. Detailed Embodiments

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0070] Embodiment 1, referring to Figure 1 , an intelligent miner's lamp based on brain-computer interface technology provided by the present invention includes a miner's lamp body, a brain-computer interface acquisition terminal module, a brain-computer interface processing chip module, a voice intercom module, a vibration module, and a liquid crystal module. The miner's lamp body includes a miner's lamp box and a miner's lamp cap. The miner's lamp body is a conventional structure, and the specific structure is not shown in the accompanying drawings and will not be described in detail in the detailed embodiments. The brain-computer interface acquisition terminal module is arranged in the miner's lamp cap, and the brain-computer interface processing chip module, the voice intercom module, the vibration module, and the liquid crystal module are arranged on the miner's lamp box;

[0071] The brain-computer interface acquisition terminal module is used to collect comprehensive information of electroencephalogram data, pulse rate data, blood oxygen data, acceleration data, and positioning data, and transmit the information to the brain-computer interface processing chip module;

[0072] The brain-computer interface processing chip module is used to receive the information collected by the brain-computer interface acquisition terminal module and perform real-time calculations in the cloud or the brain-computer interface processing chip module, and calculate the comprehensive information of electroencephalogram data, pulse rate data, blood oxygen data, acceleration data, and positioning data in real time, so as to realize the real-time monitoring functions of consciousness state, fatigue state, cognitive load, work intensity, and positioning;

[0073] The voice intercom module receives the data of the brain-computer interface processing chip module and is used for voice intercom;

[0074] The vibration module receives the data of the brain-computer interface processing chip module and is used for vibration reminder;

[0075] The liquid crystal module receives the data of the brain-computer interface processing chip module and is used for liquid crystal display.

[0076] Embodiment 2, this embodiment is based on the above embodiment, and the brain-computer interface acquisition terminal module includes a non-invasive EEG acquisition module, an optical pulse sensor, a blood oxygen meter, an accelerometer, and a positioning module.

[0077] The non-invasive EEG acquisition module adopts a non-invasive EEG acquisition method. Through a single-channel or more advanced multi-channel EEG acquisition module (including the TGAM module or its subsequent evolved version), these modules are miniaturized and portable, suitable for long-term wearing by miners. Using forehead patch electrodes and ear clip electrodes, dry electrode technology is used to reduce skin irritation and improve wearing comfort and signal quality.

[0078] In addition to EEG data, the brain-computer interface acquisition terminal module also integrates advanced sensor technologies, including an optical pulse sensor, a blood oxygen meter, an accelerometer, and a positioning module, to synchronously acquire pulse rate, blood oxygen saturation, acceleration data, and positioning data. Moreover, the brain-computer interface acquisition terminal module has high precision and low power consumption to ensure stable operation in the mine environment.

[0079] Embodiment 3. Based on the above embodiment, the brain-computer interface processing chip module adopts a multi-modal fusion neural network algorithm, which specifically includes the following steps:

[0080] (1) Signal preprocessing: Wavelet denoising is used to remove the noise in the EEG data. The specific algorithm formula for wavelet denoising is:

[0081] In the formula, represents the denoised coefficient of the j-th level decomposition at time point n after wavelet denoising processing; S j [n] represents the original wavelet coefficient, reflecting the characteristics of the signal at different scales after the j-th level decomposition; λ j represents the threshold related to the decomposition level j, used to distinguish the boundary between the signal and the noise; n represents the index of the time series, indicating the data at a specific time point; S j represents the approximate coefficient after wavelet decomposition, reflecting the low-frequency part of the signal;

[0082] After the above signal preprocessing steps, the noise can be removed more precisely while retaining the useful information of the EEG signal, especially protecting the low-intensity signal, and improving the signal-to-noise ratio. The above wavelet denoising is suitable for real-time signal processing, quickly adapting to the EEG changes of miners in different working states, and improving the response speed of the system and the accuracy of real-time monitoring. The design and application of this step reflect the efficiency and precision of EEG signal processing in a complex mine environment and are an indispensable advanced technology in the intelligent miner's lamp system.

[0083] (2) Feature extraction: A convolutional neural network is used to extract features from the electroencephalogram data, and at the same time, machine learning algorithms are used to perform feature engineering on the pulse rate data, blood oxygen data, acceleration data, and positioning data;

[0084] (3) Pattern recognition and fusion: Use long short-term memory network combined with support vector machine for pattern recognition to achieve the classification of consciousness state and physiological state; the long short-term memory network is used to process time series data, while the support vector machine is used to classify the processed feature vectors, and the fusion strategy adopts a fusion strategy based on the attention mechanism;

[0085] (4) Real-time decision-making and feedback: Use the optimized decision tree algorithm to make decisions quickly, including fatigue state warning and position deviation reminder; the optimized decision tree algorithm is specifically as follows: In the formula, P k represents the proportion of class k; Gini(D) represents the impurity of dataset D, that is, the degree of mixing of each category in the dataset; k represents the number of categories in the dataset;

[0086] In the scenario of intelligent miner's lamps, the optimized decision tree algorithm analyzes the features extracted from multi-dimensional data such as EEG data and physiological parameters to determine whether the miner's fatigue state or position deviates from the normal trajectory. The algorithm will determine the branch nodes of the tree based on the Gini index, and find the direction that can make the Gini index drop the fastest for division, so as to achieve rapid and accurate classification of the miner's state. When the algorithm identifies that the physiological feature combination of a certain miner points to a high fatigue risk, it will immediately trigger a fatigue state warning, send a reminder through the voice intercom module or vibration module, and at the same time the liquid crystal module displays specific warning information to ensure the safety and operation efficiency of the miner. The advantage of this application lies in its high efficiency and adaptability, which can process complex data streams in real time, make decisions quickly, and effectively improve the safety management of mine operations.

[0087] (5) System feedback and optimization: Introduce an online learning mechanism to adjust the algorithm parameters according to the miner's immediate feedback, and use the strategy of reinforcement learning to update the algorithm; the strategy update algorithm of reinforcement learning is specifically: π(α|s), where s represents the state; α represents the action; π(α|s) represents the strategy of taking action α under the given state s.

[0088] For the strategy update algorithm π(α|s) of reinforcement learning, the specific meaning is that for each state s, there is an action probability distribution to guide how the system responds; when detecting the miner's fatigue state s, the strategy π can instruct the system to automatically reduce the illumination intensity of the miner's lamp (one of the αs) and send a slight vibration reminder (another α) to reduce visual stimulation, and at the same time prompt to rest through the voice module.

[0089] Example 4, this example is based on the above example. In step (2), the machine learning algorithm specifically includes the following steps:

[0090] S21: Data preprocessing, perform standardization or normalization on non-EEG data to ensure that data from different sources can be compared on a unified scale;

[0091] S22: Feature selection and engineering,

[0092] For pulse rate data, using time-domain analysis to extract features of heart rate variability, including mean heart rate and standard deviation;

[0093] For blood oxygen data, paying attention to the stability and change rate of blood oxygen level as an indirect indicator of fatigue;

[0094] For acceleration data, using frequency-domain analysis to extract features of exercise intensity and identify static and active states;

[0095] For positioning data, analyzing the pattern of position changes to identify behaviors deviating from the regular work path;

[0096] S23: Application of machine learning algorithms;

[0097] Deep learning, using a convolutional neural network to perform feature learning on non-EEG data. The convolutional neural network is particularly suitable for processing data with dense time series or spatial features, including processing acceleration data through window sliding of time series to capture dynamic behavior patterns;

[0098] Support vector machine. For classification tasks, the support vector machine is used to handle the partitioning of the feature space, especially when dealing with the classification boundary of blood oxygen saturation and pulse rate, including distinguishing normal and abnormal states;

[0099] Ensemble learning, using gradient boosting trees, which can handle multi-dimensional features, enhance the generalization ability of the model, and is suitable for comprehensive analysis of multiple physiological parameters.

[0100] Example 5. This example is based on the above example. The feature selection of blood oxygen data includes the mean value M of blood oxygen saturation O2 and the standard deviation SD of blood oxygen saturation O2 ; The specific algorithm formula is as follows:

[0101]

[0102] In the formula, M O2 represents the mean value of blood oxygen saturation, that is, within a time period, the proportion of oxyhemoglobin in total hemoglobin in the blood; N represents the total number of observations or samples, which is a statistical parameter used to calculate the mean value and variance; represents the sum of blood oxygen data from the 1st to the Nth time point, which is used to calculate the mean value; O2[i] represents the blood oxygen saturation value at time point i; i represents an index symbol used to mark a specific time point in the time series; SD O2 represents the standard deviation of blood oxygen saturation; represents the sum of the squares of the differences between the blood oxygen values at each time point and the mean value, which is used to measure the volatility of the data.

[0103] Example 6. This example is based on the above example. In step (5) of system feedback and optimization, the policy update algorithm π(α|s) of reinforcement learning is iteratively learned and optimized so that the system can self-adjust according to the immediate feedback of miners. The core lies in using a surrogate objective function, which allows multiple gradient updates without disrupting the policy learning process. The surrogate objective function L clip (θ) has the following specific algorithm formula:

[0104]

[0105] In the formula, γ t (θ) represents the ratio of the old and new policies in action α t ; π θ (α t |s t ) represents the probability of taking action α t in state s t ; represents the probability of the old policy taking action α t in state s t ; represents the advantage estimate, which measures the superiority or inferiority of action α t relative to the average action; α t represents the action taken at time t; ∈ represents a hyperparameter that controls the conservatism of policy updates; θ represents the parameters of the policy network; represents the expected operation at time t, which is used to evaluate the consistency performance of the policy in different states.

[0106] The surrogate objective function L clip (θ) cleverly controls the amplitude of policy updates and ensures the stability of learning. When the value of γ t (θ) exceeds the range of [1 - ∈, 1 + ∈], it is clipped by the clip(·) function to avoid performance collapse caused by sudden policy changes. This mechanism performs well in dealing with continuous action space problems, especially suitable for the fine adjustment of miners' states in the intelligent miner's lamp system. In the actual application of the intelligent miner's lamp, the surrogate objective function L clip (θ) is not only used to adjust the policies of light intensity and vibration reminder, but also can dynamically adjust according to the physiological indicators and location information of miners to ensure the most appropriate support and warning in the complex and changeable mine environment.

[0107] Compared with the traditional policy gradient method, through the surrogate objective function L clip(θ) effectively avoids the problems of low data sample utilization and poor robustness, and shows better stability and data efficiency when dealing with continuous action space problems. In addition, by dynamically adjusting the clipping probability ratio in the objective function, a gentle transition of policy update is ensured, significantly improving the robustness of the algorithm.

[0108] Example 7. This example is based on the above example, and the surrogate objective function L clip (θ) The specific steps are as follows:

[0109] (1) Data collection. Use the current policy to interact with the environment and collect a set of experiences (s t , α t , γ t , s t+1 ); α t represents the action taken at time t; s t represents the state at time t; γ t represents the immediate reward at time t;

[0110] (2) Advantage estimation. Calculate the advantage function at each time step t by using the value function V(s t ) for approximate calculation; V(s t ) represents the value function estimate of state s t t;

[0111] (3) Policy update. In each epoch, use the surrogate objective function L clip (θ) to update the policy parameters θ multiple times until the preset number of updates is reached or the stopping criterion is met;

[0112] (4) Value function update. Use the mean squared error loss function to update the value function parameters to improve the accuracy of advantage estimation;

[0113] (5) Repeat. Go back to the first step and continue to collect data until the termination condition of the entire training process is met.

[0114] Example 8. This example is based on the above example. The voice intercom module is based on the CC2500 intercom system to achieve real-time voice communication with miners; when the main control system needs to communicate with miners, the PCM encoded signal stored in the Flash in the system is sent via the CC2500 wireless module. After the receiving end receives the signal, it is processed by the AVR single-chip microcomputer and converted into sound output to ensure clear voice communication in complex environments.

[0115] Embodiment Nine. This embodiment is based on the above embodiments, and the vibration module is used for silent physical reminders. When the system detects specific conditions, including fatigue warnings and position deviations, it will trigger the module to generate vibrations, reminding miners to pay attention to safety through physical touch, without relying on hearing or vision, which is suitable for noisy or sight-restricted mine environments.

[0116] Embodiment Ten. This embodiment is based on the above embodiments, and the liquid crystal module displays key information, including miner status, position indication, and warning messages. On the intelligent miner's lamp worn by miners, the liquid crystal module provides an intuitive visual interface, enhancing the immediacy and readability of information.

[0117] Adopting technologies such as LCD can display numbers, text, and simple graphics, ensuring that miners can clearly view important data even under insufficient underground light conditions. The voice intercom module, vibration module, and liquid crystal module work in coordination through the policy update algorithm π(α|s) of reinforcement learning. Adjust the response strategy according to the immediate feedback of miners. When a fatigue signal is recognized, not only give a voice reminder but also assist with a reminder through the vibration module, and at the same time display specific information on the liquid crystal to ensure the all-round transmission of information.

[0118] The present invention realizes advantageous functions such as real-time health monitoring, efficient communication, precise positioning, non-contact interaction, and comprehensive display. The present invention not only greatly enhances the functionality of the miner's lamp but also effectively solves the limitations of traditional miner's lamps in health monitoring, communication, positioning, and interaction through intelligent upgrading, bringing a creative safety improvement to mine operation management.

[0119] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0120] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0121] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, creatively design structural manners and embodiments similar to the technical solution, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent miner's lamp based on brain-computer interface technology, characterized in that: It includes a miner's lamp body, a brain-computer interface acquisition terminal module, and a brain-computer interface processing chip module. The miner's lamp body includes a miner's lamp box and a miner's lamp cap. The brain-computer interface acquisition terminal module is arranged inside the miner's lamp cap, and the brain-computer interface processing chip module is arranged on the miner's lamp box; The brain-computer interface acquisition terminal module is used to collect electroencephalogram data; The brain-computer interface processing chip module is used to receive the information collected by the brain-computer interface acquisition terminal module and perform real-time calculations in the cloud or on the brain-computer interface processing chip module to realize the brain state monitoring function related to the working safety state.

2. The intelligent miner's lamp based on brain-computer interface technology according to claim 1, characterized in that: It further includes a voice intercom module, a vibration module, and a liquid crystal module. The voice intercom module, the vibration module, and the liquid crystal module are arranged on the miner's lamp box, The voice intercom module receives the data of the brain-computer interface processing chip module and is used for voice intercom; The vibration module receives the data of the brain-computer interface processing chip module and is used for vibration reminder; The liquid crystal module receives the data of the brain-computer interface processing chip module and is used for liquid crystal display; The brain-computer interface acquisition terminal module is further used to collect comprehensive information of pulse rate data, blood oxygen data, acceleration data, and positioning data, and transmit the information to the brain-computer interface processing chip module; The real-time calculation includes real-time calculation of comprehensive information of electroencephalogram data, pulse rate data, blood oxygen data, acceleration data, and positioning data; The brain state monitoring function is a real-time monitoring function for real-time monitoring of brain state consciousness state, fatigue state, cognitive load, working intensity, and positioning.

3. The intelligent miner's lamp based on brain-computer interface technology according to claim 2, wherein: The brain-computer interface acquisition terminal module includes a non-invasive EEG acquisition module, an optical pulse sensor, a blood oxygen meter, an accelerometer, and a positioning module.

4. The intelligent miner's lamp based on brain-computer interface technology according to claim 2, characterized in that: The brain-computer interface processing chip module adopts a multi-modal fusion neural network algorithm, which specifically includes the following steps: (1) Signal preprocessing: Wavelet denoising is used to remove the noise in the EEG data; the specific algorithm formula for wavelet denoising is: In the formula, represents the value of the denoising coefficient at the n-th time point after the j-th level of decomposition after wavelet denoising; S j [n] represents the original wavelet coefficient, reflecting the characteristics of the signal at different scales after the j-th level of decomposition; λ j represents the threshold related to the decomposition level j, which is used to distinguish the boundary between the signal and the noise; n represents the index of the time series, Shows the data at a specific point in time; S j Represents the approximation coefficient after wavelet decomposition, It reflects the low-frequency part of the signal; (2) Feature extraction: Use a convolutional neural network to extract features from electroencephalogram data, and at the same time use machine learning algorithms to perform feature engineering on pulse rate data, blood oxygen data, acceleration data, and positioning data; (3) Pattern recognition and fusion: Use a long short-term memory network combined with a support vector machine for pattern recognition to realize the classification of consciousness state and physiological state; the long short-term memory network is used to process time series data, and the support vector machine is used to classify the processed feature vectors. The fusion strategy adopts a fusion strategy based on the attention mechanism; (4) Real-time decision-making and feedback: Utilize the optimized decision tree algorithm to make decisions quickly, including fatigue status warnings and position deviation reminders; the specific optimized decision tree algorithm is as follows: In the formula, P k represents the proportion of class k; Gini(D) represents the impurity of dataset D, that is, the degree of mixing of each category in the dataset; k represents the number of categories in the dataset. (5) System feedback and optimization: Introduce an online learning mechanism, adjust the algorithm parameters according to the immediate feedback of the miner, and update the algorithm using the strategy of reinforcement learning; the strategy update algorithm of the reinforcement learning is specifically: π(α|s), where s represents the state; α represents the action; π(α|s) represents the strategy of taking action α under the given state s.

5. The intelligent miner's lamp based on brain-computer interface technology according to claim 4, wherein: In step (2), the machine learning algorithm specifically includes the following steps: S21: Data preprocessing, perform standardization or normalization processing on non-electroencephalogram data to ensure that data from different sources can be compared on a unified scale; S22: Feature selection and engineering, Pulse rate data, using time-domain analysis, to extract the features of heart rate variability, including mean heart rate and standard deviation; Blood oxygen data, focusing on the stability and change rate of blood oxygen level, as an indirect indicator of fatigue; Acceleration data, through frequency-domain analysis, to extract the features of exercise intensity and identify stationary and active states; Location data, analyzing the pattern of location changes to identify behaviors deviating from the regular work path; S23: Application of machine learning algorithms; Deep learning, using a convolutional neural network to perform feature learning on non-EEG data. The convolutional neural network is particularly suitable for processing data with dense time series or spatial features, including processing acceleration data through window sliding of the time series to capture dynamic behavior patterns; Support vector machine, for classification tasks, the support vector machine is used to handle the division of the feature space, especially when dealing with the classification boundary of blood oxygen saturation and pulse rate, including distinguishing normal and abnormal states; Ensemble learning, adopting gradient boosting trees, which can handle multi-dimensional features and enhance the generalization ability of the model, and is suitable for comprehensive analysis of various physiological parameters.

6. The intelligent miner's lamp based on brain-computer interface technology according to claim 5, characterized in that: The feature selection of the blood oxygen data includes the mean value M of the blood oxygen saturation O2 and the standard deviation SD of the blood oxygen saturation O2 ; The specific algorithm formula is as follows: where M O2 represents the average value of the blood oxygen saturation, that is, within the time period, the oxyhemoglobin in the blood accounts for the total hemoglobin; N represents the total number of observations or samples, which is a statistical parameter used to calculate the average value and variance; represents the sum of blood oxygen data from the 1st to the Nth time points, which is used to calculate the average value; O2[i] represents the blood oxygen saturation value at time point i; i represents an index symbol used to mark a specific time point in the time series; SD O2 represents the standard deviation of blood oxygen saturation; represents the sum of the squares of the differences between the blood oxygen values at each time point and the average value, which is used to measure the volatility of the data.

7. The intelligent miner's lamp based on brain-computer interface technology according to claim 4, characterized in that: In step (5) system feedback and optimization, the policy update algorithm π(α|s) of the reinforcement learning enables the system to self-adjust according to the miners' immediate feedback through iterative learning and optimization. The core lies in using a surrogate objective function, which allows multiple gradient updates without disrupting the policy learning process. The surrogate objective function L clip (θ) has the following specific algorithm formula: wherein, γ t (θ) represents the ratio of the new and old policies for action α t π θ (α t |s t ) represents the probability of taking action α t under state s t ; represents the probability of the old policy taking action α t under state s t ; represents the advantage estimate, measuring the goodness or badness of action α t relative to the average action; α t represents the action taken at time t; ∈ represents a hyperparameter that controls the conservatism of policy update; θ represents the parameters of the policy network; represents the expected action at time t, which is used to evaluate the consistent performance of the policy in different states.

8. The intelligent miner's lamp based on brain-computer interface technology according to claim 7, characterized in that: The proxy objective function L clip (θ) is as follows: (1) Data collection, using the current strategy Interact with the environment to collect a set of experiences (s t , α t , γ t , s t+1 ); α t represents the action taken at time t; s t represents the state at time t; γ t represents the immediate reward at time t; (2) Advantage estimation, calculating the advantage function for each time step t by using the value function V(s t ) to approximately calculate; V(s t ) represents the estimated value function of state s t t; (3) Policy update. In each epoch, the policy parameter θ is updated multiple times using the surrogate objective function L clip (θ) until the preset number of updates is reached or the stopping criterion is met; (4) Value function update, using the mean squared error loss function to update the parameters of the value function to improve the accuracy of advantage estimation; (5) Repeat, go back to the first step and continue to collect data until the termination conditions of the entire training process are met.

9. An intelligent miner's lamp based on brain-computer interface technology according to claim 2, characterized in that: The voice intercom module is based on the CC2500 intercom system to achieve real-time voice communication with miners; when the main control system needs to communicate with miners, the PCM-encoded signal stored in the Flash in the system is sent via the CC2500 wireless module. After the receiving end receives the signal, it is processed by the AVR single-chip microcomputer and converted into sound output to ensure clear voice communication in complex environments.

10. The intelligent miner's lamp based on brain-computer interface technology according to claim 2, characterized in that: The vibration module is used for silent physical reminders; when the system detects specific conditions, including fatigue warnings and location deviations, it will trigger the module to generate vibrations, reminding miners to pay attention to safety through physical touch, without relying on hearing or vision, which is suitable for noisy or sight-limited mine environments; the liquid crystal module displays key information, including miner status, location indication, and warning messages. On the intelligent miner's lamp worn by miners, the liquid crystal module provides an intuitive visual interface, enhancing the immediacy and readability of information.

Images