A multi-scene fusion perception system and method for a humanoid robot olfactory sensor

Abstract

The application provides a multi-scene fusion perception system and method of a humanoid robot olfactory sensor, and relates to the technical field of intelligent sensors.The system comprises an olfactory perception optimization module for obtaining an olfactory perception optimization network; a first matrix module for monitoring a gas pipeline to obtain a first matrix; a second matrix module for interference compensation according to a real-time environment; a third matrix module for correction according to a sensor state; a leakage detection module for pipeline leakage detection; and a suppression verification module for false alarm suppression verification of the pipeline leakage detection result to obtain a pipeline early warning signal.The application can solve the technical problem of limited gas leakage detection precision in the prior art due to environmental interference and high false alarm rate, and can improve the accuracy of gas leakage detection in a multi-scene environment by real-time adjustment of a multi-node olfactory perception network based on historical data, multi-level correction of monitoring data, and introduction of false alarm suppression verification.

Description

Technical Field

[0001] This application relates to the field of intelligent sensor technology, and in particular to a multi-scene fusion perception system and method for a humanoid robot olfactory sensor. Background Technology

[0002] Artificial intelligence (AI) olfactory sensors are devices that detect, identify, and analyze odors or chemical components by simulating the working principle of the human olfactory system and combining sensor technology with artificial intelligence algorithms. AI olfactory sensors can be used to detect the gas composition in the air surrounding gas pipelines, especially components such as methane and ethane in fuels (such as natural gas or liquefied petroleum gas). When a pipeline leaks, these gases gradually diffuse into the surrounding environment. The olfactory sensor identifies potential leaks or hazards by recognizing the concentration and composition of the gases. However, pipelines are often located underground or in industrial areas, and the surrounding air may contain other interfering gases. During gas pipeline monitoring, intelligent sensors are subject to various external and internal interferences, such as temperature changes, pressure fluctuations, and the mixing of other gases, causing the monitoring data to be affected by background noise or interference, thus affecting accuracy. Furthermore, intelligent sensors may experience performance degradation over long-term use or exhibit inconsistent performance under different environmental conditions, leading to deviations in sensor data and affecting the final monitoring results.

[0003] In summary, existing technologies suffer from limited accuracy in detecting gas leaks due to their susceptibility to interference from various sources in complex and variable pipeline environments, resulting in a high false alarm rate. Summary of the Invention

[0004] The purpose of this application is to provide a multi-scene fusion perception system and method for humanoid robot olfactory sensors, in order to solve the technical problem in the prior art that the detection accuracy of gas leaks is limited due to the high false alarm rate caused by the susceptibility to interference from various parties in complex and variable pipeline environments.

[0005] In view of the above problems, this application provides a multi-scene fusion perception system and method for humanoid robot olfactory sensors.

[0006] In a first aspect, this application provides a multi-scene fusion perception system for a humanoid robot olfactory sensor, wherein the multi-scene fusion perception system for a humanoid robot olfactory sensor includes: an olfactory perception optimization module, used to adaptively adjust a multi-node olfactory perception network of a gas pipeline based on a historical set of pipeline gas in a nearby historical window to obtain an optimized olfactory perception network; a first matrix module, used to monitor the gas pipeline in real time based on the optimized olfactory perception network to obtain a first pipeline gas monitoring matrix; a second matrix module, used to perform background interference compensation on the first pipeline gas monitoring matrix based on a real-time pipeline environment dataset to obtain a second pipeline gas monitoring matrix; a third matrix module, used to correct the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix to obtain a third pipeline gas monitoring matrix; a leak detection module, used to perform pipeline leak detection based on the third pipeline gas monitoring matrix to obtain a pipeline leak detection result; and a suppression verification module, used to perform false alarm suppression verification on the pipeline leak detection result based on the optimized olfactory perception network to obtain a pipeline early warning signal.

[0007] Optionally, the trend prediction unit is used to predict the trend of the gas pipeline based on the historical data of the pipeline gas to obtain the predicted trend of the pipeline gas; the risk identification unit is used to identify the risk of each point of the gas pipeline based on the predicted trend of the pipeline gas to obtain the risk characteristics of the pipeline points; the node optimization unit is used to optimize the nodes of the multi-node olfactory perception network based on the risk characteristics of the pipeline points to obtain the node-optimized olfactory perception network; and the frequency configuration unit is used to configure the sampling frequency of the node-optimized olfactory perception network based on the risk characteristics of the pipeline points to generate the optimized olfactory perception network.

[0008] Optionally, the first parameter extraction unit is used to extract first pipeline gas monitoring parameters based on the first pipeline gas monitoring matrix; the first mapping identification unit is used to map and identify the real-time pipeline environment dataset based on the first pipeline gas monitoring parameters to obtain first pipeline environment data; the first interference identification unit is used to identify interference in the first pipeline gas monitoring parameters based on the first pipeline environment data to determine a first monitoring background interference factor; and the first confidence compensation unit is used to perform confidence compensation on the first pipeline gas monitoring parameters based on the first monitoring background interference factor to generate first updated gas monitoring data and add the first updated gas monitoring data to the second pipeline gas monitoring matrix.

[0009] Optionally, the constraint determination subunit is used to use the first monitoring background interference factor as the first retrieval constraint for interference compensation and the first parameter of pipeline gas monitoring as the second retrieval constraint for interference compensation; the interference compensation subunit is used to perform interference compensation sample retrieval according to the first retrieval constraint and the second retrieval constraint for interference compensation to obtain a first interference compensation sample set; the confidence evaluation subunit is used to evaluate the confidence of the first interference compensation sample set to obtain the confidence coefficient of each compensation sample; the feature screening subunit is used to screen the first interference compensation sample set based on the confidence coefficient of each compensation sample to obtain a confidence interference compensation sample set that meets a predetermined confidence level; and the parameter correction subunit is used to fuse the confidence interference compensation sample set to obtain a first confidence compensation feature, and correct the first parameter of pipeline gas monitoring according to the first confidence compensation feature to obtain the first updated data of gas monitoring.

[0010] Optionally, the anomaly detection unit is used to perform anomaly detection based on the state data of each olfactory sensor and obtain the anomaly detection results of each sensor; the impact analysis unit is used to perform monitoring impact analysis on the gas pipeline based on the anomaly detection results of each sensor and determine the impact characteristics of each sensor anomaly; the adaptive correction unit is used to adaptively correct the second pipeline gas monitoring matrix based on the impact characteristics of each sensor anomaly and generate the third pipeline gas monitoring matrix.

[0011] Optionally, the data extraction unit is used to extract the first node gas monitoring data corresponding to the first node of the pipeline according to the third pipeline gas monitoring matrix; the channel activation unit is used to activate the pipeline leakage risk detection channel, which includes P pipeline leakage risk detection models, where P is a positive integer greater than 1; the risk coefficient determination unit is used to input the first node gas monitoring data into the P pipeline leakage risk detection models to obtain P leakage risk detection coefficients; the proportion calculation unit is used to calculate the proportion of the P leakage risk detection accuracies corresponding to the P pipeline leakage risk detection models to obtain P risk detection weights; and the weighted calculation unit is used to perform weighted calculation on the P leakage risk detection coefficients according to the P risk detection weights to generate the first node leakage risk coefficient and add the first node leakage risk coefficient to the pipeline leakage detection result.

[0012] Optionally, a threshold setting unit is used to set a false alarm suppression verification mechanism, the false alarm suppression verification mechanism including a first leakage risk threshold and a second leakage risk threshold, wherein the first leakage risk threshold is greater than the second leakage risk threshold; a first judgment unit is used to judge whether the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold; and a warning signal generation unit is used to generate the pipeline warning signal based on the leakage risk coefficient of the first node if the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold.

[0013] Optionally, the second judgment subunit is used to determine whether the leakage risk coefficient of the first node is greater than or equal to the second leakage risk threshold if the leakage risk coefficient of the first node is less than the first leakage risk threshold, and obtain the first node risk judgment result; the update monitoring subunit is used to generate a first node feature monitoring instruction based on the first node risk judgment result, and control the olfactory perception optimization network to continuously monitor the first node of the pipeline based on the first node feature monitoring instruction, and obtain the first node update monitoring data; the risk coefficient calculation subunit is used to input the first node update monitoring data into the pipeline leakage risk detection channel, obtain the first node leakage risk update coefficient, and calculate the average of the first node leakage risk coefficient and the first node leakage risk update coefficient to generate a first reliable leakage risk coefficient; the signal generation subunit is used to generate the pipeline early warning signal if the first reliable leakage risk coefficient is greater than or equal to the first leakage risk threshold.

[0014] Optionally, a real-time monitoring unit is used to monitor the gas pipeline in real time according to the olfactory perception optimization network to obtain a pipeline gas monitoring set; a data filtering unit is used to filter the pipeline gas monitoring set to obtain a pipeline gas dataset; and a matrix processing unit is used to perform matrix processing on the pipeline gas dataset to generate the first pipeline gas monitoring matrix.

[0015] Secondly, this application also provides a multi-scene fusion perception method for a humanoid robot olfactory sensor. This method includes: adaptively adjusting a multi-node olfactory perception network for a gas pipeline based on a historical set of pipeline gases from adjacent historical windows to obtain an optimized olfactory perception network; monitoring the gas pipeline in real-time using the optimized olfactory perception network to obtain a first pipeline gas monitoring matrix; compensating for background interference on the first pipeline gas monitoring matrix based on a real-time pipeline environment dataset to obtain a second pipeline gas monitoring matrix; correcting the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix to obtain a third pipeline gas monitoring matrix; detecting pipeline leaks using the third pipeline gas monitoring matrix to obtain a pipeline leak detection result; and verifying the false alarm suppression of the pipeline leak detection result using the optimized olfactory perception network to obtain a pipeline early warning signal.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] The system employs a multi-node olfactory perception optimization module to adaptively adjust the multi-node olfactory perception network of the gas pipeline based on the historical gas data of adjacent historical windows, thereby obtaining an optimized olfactory perception network. A first matrix module monitors the gas pipeline in real time using the optimized olfactory perception network to obtain a first pipeline gas monitoring matrix. A second matrix module compensates for background interference in the first pipeline gas monitoring matrix based on the real-time pipeline environment dataset, resulting in a second pipeline gas monitoring matrix. A third matrix module corrects the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix, resulting in a third pipeline gas monitoring matrix. A leak detection module performs pipeline leak detection based on the third pipeline gas monitoring matrix, obtaining a pipeline leak detection result. A suppression and verification module performs false alarm suppression verification on the pipeline leak detection result based on the optimized olfactory perception network, obtaining a pipeline early warning signal. In other words, by adaptively adjusting the multi-node olfactory sensing network of the gas pipeline based on the historical gas data of the adjacent historical window, the olfactory sensing optimization network is used to monitor the gas pipeline in real time. The monitoring data is corrected and compensated by combining the real-time environment and sensor status. A multi-modal fusion leak detection algorithm is executed. By setting a false alarm suppression and verification mechanism with multi-level risk thresholds, multi-level pipeline early warning signals are generated, realizing accurate prediction and intelligent response of gas leaks. This improves the accuracy and precision of gas leak detection in multiple scenarios and has the advantages of adaptive adjustment, strong anti-interference, and accurate intelligent identification.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the structure of a multi-scene fusion perception system for a humanoid robot olfactory sensor according to this application;

[0021] Figure 2 This is a flowchart illustrating a multi-scene fusion perception method for a humanoid robot olfactory sensor according to this application.

[0022] Explanation of reference numerals in the attached diagram: olfactory perception optimization module 11, first matrix module 12, second matrix module 13, third matrix module 14, leakage detection module 15, suppression verification module 16. Detailed Implementation

[0023] This application provides a multi-scene fusion perception system and method for humanoid robot olfactory sensors, solving the technical problem of limited detection accuracy of gas leaks in existing technologies due to high false alarm rates caused by susceptibility to interference from various parties in complex and variable pipeline environments. By adaptively adjusting the multi-node olfactory perception network of the gas pipeline based on the historical gas data of adjacent historical windows, the system utilizes the optimized olfactory perception network to monitor the gas pipeline in real time. It then corrects and compensates the monitoring data by combining real-time environmental and sensor status data, executes a multi-modal fusion leak detection algorithm, and generates multi-level pipeline early warning signals through a false alarm suppression and verification mechanism with multi-level risk thresholds. This achieves accurate prediction and intelligent response to gas leaks, improving the accuracy and precision of gas leak detection in various environments, and possessing advantages such as adaptive adjustment, strong anti-interference, and accurate intelligent identification.

[0024] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0025] Example 1, please refer to the appendix. Figure 1 This application provides a multi-scene fusion perception system for a humanoid robot olfactory sensor. The system is used to implement a multi-scene fusion perception method for a humanoid robot olfactory sensor. The multi-scene fusion perception system for a humanoid robot olfactory sensor includes:

[0026] The olfactory perception optimization module 11 is used to adaptively adjust the multi-node olfactory perception network of the gas pipeline based on the pipeline gas history set of the adjacent historical window to obtain the olfactory perception optimization network.

[0027] Furthermore, the olfactory perception optimization module 11 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0028] The system includes a trend prediction unit for predicting the gas pipeline trend based on the historical gas data, and a risk identification unit for identifying the risk at each point of the gas pipeline based on the predicted gas trend, and obtaining the risk characteristics of each pipeline point. A node optimization unit is used to optimize the nodes of the multi-node olfactory sensing network based on the risk characteristics of each pipeline point, and obtain a node-optimized olfactory sensing network. A frequency configuration unit is used to configure the sampling frequency of the node-optimized olfactory sensing network based on the risk characteristics of each pipeline point, and generate the optimized olfactory sensing network.

[0029] Specifically, the process involves acquiring historical data on gas parameters in the gas pipeline, including historical data on changes in gas parameters (such as concentration, flow rate, and pressure) over time. Based on this historical data, time series analysis is used to predict future trends in the pipeline. For example, an LSTM model can be used to predict the concentration trend for the next month based on gas concentration data from the past year. The historical data is preprocessed, including standardization (such as Z-score normalization) and correction or interpolation for outliers (such as breakpoints or abrupt changes). A sliding time window method is used to construct training samples; for example, using the previous 30 days as input, the concentration trend for the next 7 days is predicted, and the prediction is compared with the original concentration data in the historical data set to determine the accuracy of the prediction. The dataset is divided into training, validation, and test sets in a 70:15:15 ratio. A multi-layer LSTM neural network is constructed, including determining the number of layers, the number of neurons in each layer, and the input and output sizes. The preprocessed data is used to train the LSTM model. During training, the model learns to recognize patterns and trends in the data. The training process typically involves multiple iterations, in which the model adjusts its weights based on the error. A validation dataset is used to evaluate the model's performance, ensuring it doesn't overfit (i.e., perform well on training data but poorly on unseen data). The model is then optimized using a loss function (such as mean squared error). Training stops when the model's prediction accuracy reaches a preset threshold (e.g., above 95%). The model at this stage is then used to predict gas trends in gas pipelines, obtaining predicted gas trends.

[0030] Based on predicted pipeline gas trends, potential risks at various points within the gas pipeline are identified. The pipeline is divided into N key points according to its geographical structure. For each point, a risk value is calculated based on predicted concentration changes, historical leakage frequency, and gas flow direction. Different weights are assigned according to the influence of predicted concentration changes, historical leakage frequency, and gas flow direction on gas leakage. These three results are standardized and then weighted to obtain the risk value for each point, thus determining the risk characteristics of each pipeline point—that is, the risk level of each pipeline point (or monitoring point) based on historical data and predicted trends. For example, after standardization, the predicted concentration change at point A is 0.52, the historical leakage frequency is 0.3, and the airflow direction influence coefficient is 0.9; at point B, the predicted concentration change is 0.11, the historical leakage frequency is 0.2, and the airflow direction influence coefficient is 0.3; at point C, the predicted concentration change is 0.35, the historical leakage frequency is 0.1, and the airflow direction influence coefficient is 0.6; and at point D, the predicted concentration change is 0.9, the historical leakage frequency is 0.4, and the airflow direction influence coefficient is 0.4. Assuming weights of 0.5 for concentration, 0.3 for frequency, and 0.2 for direction, the calculated risk values ​​are 0.53 for point A, 0.175 for point B, 0.325 for point C, and 0.65 for point D. Therefore, point D has the highest risk, requiring more sensors and a higher sampling frequency.

[0031] Olfactory sensors are installed on the heads, shoulders, or arms of multiple humanoid robots to capture odors and gas concentrations in the surrounding environment, forming a multi-node olfactory sensing network. The robots are interconnected through the network to achieve data sharing and real-time information exchange. Based on the risk characteristics of pipeline locations, the multi-node olfactory sensing network is optimized to generate a node-optimized olfactory sensing network. In other words, the operating parameters of the olfactory sensors are dynamically adjusted according to the risk characteristics of pipeline locations to ensure that each node's olfactory sensing network can provide the most accurate and timely data under any environmental conditions. Specifically, for nodes located at high-risk locations, the sensitivity of the olfactory sensors on each humanoid robot is adjusted to make them more sensitive to subtle changes in gas. For example, near pipeline interfaces and valves, where the risk of gas leaks is higher, the sensitivity of these nodes needs to be increased to detect even minor gas leaks promptly. The monitoring range of different nodes is adjusted according to their location to ensure that each node covers its specific gas distribution area. In the multi-node olfactory sensing network, node optimization involves not only adjusting the operating parameters of the sensors themselves but also optimizing data transmission between nodes and improving data processing efficiency.

[0032] Based on the risk level of each location, the sampling frequency of the sensors is configured. For example, the sampling frequency is increased at high-risk locations to ensure real-time monitoring, while lower-risk locations can have a lower sampling frequency to reduce resource consumption. For instance, based on historical data of gas pipeline pressure and flow over the past year, it was determined that section M of the pipeline has a high risk of leakage due to its long service life. Therefore, the sampling frequency of the sensors on this section of the pipeline was increased from once every ten minutes to once per minute. Through trend prediction and risk identification, node optimization and sampling frequency configuration of the humanoid robot's multi-node olfactory perception network were achieved, improving the targeting and real-time performance of gas pipeline monitoring.

[0033] The olfactory perception optimization network in this application can be applied to various key locations in the home environment, such as the kitchen, living room, and bathroom, to ensure effective monitoring of different gases (food odors, toxic gases, explosive gases, etc.). Based on the types of odors detected in the home, auxiliary functions such as odor identification, cooking guidance, and ventilation are implemented. Odor types include not only odors generated from daily household activities (such as cooking fumes, soy sauce and vinegar, tea aromas, cleaning agent smells, pet excrement smells, etc.), but also toxic and harmful gases and potentially explosive gases that are difficult or completely imperceptible to humans, such as methane, ammonia, and volatile organic compounds. In conjunction with a humanoid robot, when harmful gases or abnormal odors are detected, the robot will automatically activate an early warning mechanism, generate an alarm signal, and execute preset emergency operations (such as activating the ventilation system).

[0034] Combining AI algorithms (such as machine learning and deep learning), this method classifies and identifies gases in the environment, acquiring and labeling environmental datasets to indicate which gases are toxic, which are kitchen odors, etc. Machine learning (such as supervised learning) is used to classify and identify gases in the environment. Based on the environmental dataset and labeled gas sets, a model is trained to identify the characteristics of different gases and determine their concentrations. Specifically, the environmental dataset and labeled gas sets are normalized and divided into training and validation sets. The model is trained using the training set, continuously optimizing the loss function to ensure it can output the correct gas type and concentration given inputs. The training process is iterative until the loss function converges or reaches a set accuracy threshold. The validation set data is input into the trained model to evaluate the consistency between its predictions and the true labels, such as an accuracy rate of 95%. The trained model is then used as a gas classification and identification model to identify gas types and concentrations. Based on continuously collected new data, the model is adjusted to better adapt to different environmental changes. For example, if the odor characteristics in the home environment change (e.g., new food or changes in ventilation), the model will automatically update to adapt to the new odor environment.

[0035] During real-time operation, the olfactory perception optimization network continuously acquires new data and analyzes gas concentrations using a trained gas classification and recognition model. It identifies potential hazards such as leaks and fires, and automatically assesses leak risk based on changes in gas type and concentration. The AI ​​algorithm automatically adjusts its sensitivity to leak risk for different environments. For example, in industrial environments, toxic gases with higher leak risks require more sensitive detection, while in residential environments with lower gas concentrations, monitoring parameters are adjusted according to actual needs.

[0036] When a leak or harmful gas is detected, an alarm is issued, and the type of alarm and response may vary depending on the environmental scenario. For example, in a home, an audible alarm is activated in conjunction with a push notification from a mobile app, while the robot begins to perform preset emergency operations (such as activating the ventilation system); in an industrial environment, an emergency shutdown system is activated, and ventilation equipment is activated to prevent disasters, adapting to different home and industrial environments.

[0037] The olfactory perception optimization network is not only suitable for high-precision monitoring of industrial gas pipelines, but can also be extended to humanoid robot systems. By simulating human olfactory functions, it can broadly identify common odors in the home environment (such as cooking fumes, soy sauce, vinegar, tea, and cleaning agent residue), and can also detect gaseous components that are imperceptible to humans or have extremely low levels of danger, such as carbon monoxide, methane, and volatile organic compounds. Based on the olfactory perception optimization network, humanoid robots can perform tasks such as home security protection, intelligent ventilation control, and odor classification.

[0038] The humanoid robot has a built-in warning threshold and uses an optimized network to detect indoor gases through olfactory perception. Once the risk level determined by the indoor gas exceeds the set threshold, the humanoid robot will automatically activate a preset emergency response mechanism, including reminding the user via voice broadcast, turning on the exhaust fan and opening windows in conjunction with the home IoT system, and sending real-time warning information to the user's mobile app or community security platform when necessary. In scenarios of serious leaks or a continuous rise in gas concentration, it can also activate air purification devices and shut off gas valves as emergency measures.

[0039] As the home environment changes (such as formaldehyde release from new renovations or lingering odors from new food), the odor recognition model is continuously corrected using newly sampled data to form an adaptive odor perception system. Newly collected samples are added to a dynamic training set, and the gas recognition model parameters are automatically updated through online incremental learning. Through the collaboration of an olfactory perception optimization network and AI algorithms, the humanoid robot can not only perform intelligent functions such as protecting home air quality, identifying odor sources, providing cooking guidance and ventilation strategy suggestions, but also construct a dynamic odor knowledge graph for the home. The graph records the odor distribution characteristics and corresponding device status at different time periods and spatial locations, assisting the robot in autonomously identifying potential problems without external prompts. For example, it can detect food spoilage in the refrigerator based on subtle odor changes, determine if carpets are damp or moldy, and identify pet excrement areas.

[0040] The first matrix module 12 is used to monitor the gas pipeline in real time according to the olfactory perception optimization network to obtain the first matrix for pipeline gas monitoring.

[0041] Furthermore, the first matrix module 12 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0042] A real-time monitoring unit is used to monitor the gas pipeline in real time according to the olfactory perception optimization network to obtain a pipeline gas monitoring set; a data filtering unit is used to filter the pipeline gas monitoring set to obtain a pipeline gas dataset; and a matrix processing unit is used to perform matrix processing on the pipeline gas dataset to generate the first pipeline gas monitoring matrix.

[0043] Specifically, the olfactory perception optimization network is a sensor set formed by configuring various types of olfactory sensors (such as MQ-5 sensors, electrochemical sensors, etc.) at multiple locations on multiple robots through adaptive node optimization and sampling frequency configuration in the previous stage. This sensor set is used to monitor the target gas and its interfering gases in the air surrounding the pipeline in real time. Based on the determined olfactory perception optimization network, the gas pipeline is monitored in real time at the configured sampling frequency (e.g., 1Hz to 10Hz), resulting in a pipeline gas monitoring set. The pipeline gas monitoring set includes gas data monitored by multiple olfactory sensors at each pipeline location.

[0044] Data filtering is applied to the pipeline gas monitoring dataset to remove random noise, spikes, and other interference, resulting in a smoother and more reliable signal. For example, a low-pass filter can remove high-frequency noise while retaining useful low-frequency signals. The filtered pipeline gas dataset is then matrixed to generate the first pipeline gas monitoring matrix. The filtered multi-channel data is mapped into a three-dimensional array according to sensor number, gas type, and timestamp, forming the first pipeline gas monitoring matrix, typically denoted as M[i,j,t], where i represents the i-th sensor, j represents the gas type (e.g., methane, carbon monoxide), and t represents the time point.

[0045] The second matrix module 13 is used to perform background interference compensation on the first pipeline gas monitoring matrix based on the pipeline real-time environment dataset to obtain the second pipeline gas monitoring matrix.

[0046] Furthermore, the second matrix module 13 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0047] A first parameter extraction unit is used to extract a first parameter for pipeline gas monitoring based on the first pipeline gas monitoring matrix; a first mapping and identification unit is used to map and identify the real-time pipeline environment dataset based on the first parameter for pipeline gas monitoring to obtain first pipeline environment data; a first interference identification unit is used to identify interference in the first parameter for pipeline gas monitoring based on the first pipeline environment data to determine a first monitoring background interference factor; a first confidence compensation unit is used to perform confidence compensation on the first parameter for pipeline gas monitoring based on the first monitoring background interference factor to generate first updated gas monitoring data, and add the first updated gas monitoring data to the second pipeline gas monitoring matrix.

[0048] The system includes a constraint determination subunit, which uses the first monitoring background interference factor as the first retrieval constraint for interference compensation and the first parameter of pipeline gas monitoring as the second retrieval constraint for interference compensation; an interference compensation subunit, which performs interference compensation sample retrieval based on the first and second retrieval constraints for interference compensation to obtain a first interference compensation sample set; a confidence evaluation subunit, which evaluates the confidence level of the first interference compensation sample set to obtain a confidence coefficient for each compensation sample; a feature filtering subunit, which filters the first interference compensation sample set based on the confidence coefficients of each compensation sample to obtain a confidence interference compensation sample set that meets a predetermined confidence level; and a parameter correction subunit, which fuses the confidence interference compensation sample set to obtain a first confidence compensation feature and corrects the first parameter of pipeline gas monitoring based on the first confidence compensation feature to obtain the first updated data of gas monitoring.

[0049] Specifically, a first parameter for pipeline gas monitoring is arbitrarily extracted from the first pipeline gas monitoring matrix, selecting a specific gas parameter. Based on the extracted first parameter, relevant environmental data is searched in the real-time pipeline environmental dataset to identify the environmental data related to this gas parameter, i.e., the first pipeline environmental data, including temperature, humidity, air pressure, wind speed, etc., to determine which environmental factors may affect the gas sensor readings. The Pearson correlation coefficient is used to calculate the linear correlation between each environmental variable and the monitoring parameter, assessing the correlation between environmental variables and monitoring parameters, and screening out candidate interference factors. Variables with significant influence are selected, and their interference direction and coefficient are recorded, constituting the first monitoring background interference factors. For example, suppose the methane concentration extracted from the first pipeline gas monitoring matrix is ​​100 ppm. Through mapping identification, it is found that the current pipeline environment temperature is 25℃ and the humidity is 60%. Interference identification analysis shows that temperature fluctuations have a significant impact on methane concentration monitoring. Based on a compensation model established using historical data, when the temperature rises by 1℃, the methane concentration monitoring value will be 2 ppm higher. Therefore, if the current temperature is 3°C higher than the reference temperature, a confidence compensation is needed for the methane concentration, subtracting 6 ppm to obtain a corrected methane concentration of 94 ppm. This updated data is then added to the second matrix of pipeline gas monitoring. In field tests, the confidence compensation method reduced the error in methane concentration monitoring from an average of 5% to 2%.

[0050] The primary monitoring background interference factor is the identified main interference factor, which is the interference factor that has the greatest impact on the primary parameter of pipeline gas monitoring. Based on this interference factor, confidence compensation is performed on the primary parameter of pipeline gas monitoring, the original gas value is corrected according to the degree of interference, and a confidence level (reliability) weight is given.

[0051] The first background interference factor is used as the first retrieval constraint for interference compensation, and the first parameter of pipeline gas monitoring is used as the second retrieval constraint for interference compensation. Samples satisfying both constraints are retrieved from the historical database to form the first interference compensation sample set. For example, gas concentration data measured under similar temperature conditions are searched. The first interference compensation sample set is a set of samples that satisfy both the first and second retrieval constraints for interference compensation, reflecting the sensor response under similar interference conditions and specific gas parameter values.

[0052] For each sample in the first interference compensation sample set, a confidence level is evaluated, and a confidence coefficient is calculated for each sample. The confidence coefficient is determined by comparing the deviation between the sample data and the actual measurement data. Samples with confidence coefficients greater than or equal to a predetermined confidence threshold are selected to form a confidence interference compensation sample set. For example, only samples with confidence coefficients greater than 0.8 are retained.

[0053] The samples in the confidence interference compensation sample set all possess high reliability and can effectively reflect the gas response characteristics under the current monitoring conditions. A weighted average algorithm is applied to weight and fuse the confidence interference compensation sample set according to confidence level, yielding the first confidence compensation feature. Using this first confidence compensation feature as a reference, the currently acquired pipeline gas monitoring first parameter is corrected, generating the first updated gas monitoring data. Assuming the offset caused by interference is a linear residual, residual correction is performed directly; alternatively, the offset between the compensation feature and the current value is recursively corrected. For example, if it is found that for every 1°C increase in temperature, the gas concentration increases by an average of 2%, the gas concentration measurement value can be adjusted accordingly based on the current temperature. Through the retrieval, confidence evaluation, and screening of interference compensation samples, the pipeline gas monitoring first parameter is corrected, reducing the impact of background interference. Confidence evaluation and screening ensure the quality of the samples used for interference compensation, improving the accuracy and reliability of the corrected data.

[0054] The third matrix module 14 is used to correct the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix, so as to obtain the third pipeline gas monitoring matrix.

[0055] Furthermore, the third matrix module 14 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0056] An anomaly detection unit is used to perform anomaly detection based on the state data of each olfactory sensor and obtain the anomaly detection results of each sensor; an impact analysis unit is used to perform monitoring impact analysis on the gas pipeline based on the anomaly detection results of each sensor and determine the impact characteristics of each sensor anomaly; an adaptive correction unit is used to adaptively correct the second pipeline gas monitoring matrix based on the impact characteristics of each sensor anomaly and generate the third pipeline gas monitoring matrix.

[0057] Specifically, the system acquires the status data of each olfactory sensor corresponding to the first matrix of pipeline gas monitoring, i.e., the operating data of the olfactory sensors, including parameters such as sensor voltage, temperature, humidity, and response time. Based on the sensor's factory specifications and historical averages, reasonable thresholds are set to detect anomalies in the status data of each olfactory sensor, determining whether the sensor's output exceeds a predetermined range, and whether there is a sensor malfunction or environmental anomaly. Each sensor's operating parameter is checked to see if it exceeds the set reasonable threshold; if so, an anomaly is considered to exist. Anomaly detection is performed on each sensor to obtain its abnormal state, recording which sensors are faulty and which data are normal.

[0058] Based on the abnormal detection results of sensors, the impact of these anomalies on the overall monitoring results is analyzed. For example, if a sensor malfunctions, it may lead to incorrect gas concentration data at a certain location; if multiple sensors malfunction, it may affect the leak detection of the entire pipeline. By analyzing the impact of abnormal sensor states on the entire gas pipeline monitoring system, it is determined which sensor anomalies will have a greater impact on the monitoring results and which will have a smaller impact, obtaining the characteristics of the impact of each sensor anomaly, such as sensor drift, noise interference, and deterministic bias. Sensor drift is the degree to which the sensor deviates from its normal response; noise interference is the fluctuation range of the sensor signal; deterministic bias is the systematic error between the sensor output and the actual gas concentration.

[0059] Based on the characteristics of the anomalies, the pipeline gas monitoring matrix is ​​corrected, including replacing data from abnormal sensors and using weighted averaging. For example, values ​​from adjacent normal sensors can be used to interpolate and fill in the values ​​from abnormal sensors, or weighted averaging can be used to correct the data from abnormal sensors. For instance, if a sensor measures a concentration that is significantly higher (e.g., 10 ppm), while neighboring sensors measure a concentration of 5 ppm, a weighted correction method can be used to correct it to 6 ppm.

[0060] By establishing a data correction model and combining it with the characteristics of sensor anomalies, the monitoring data is automatically adjusted, and elements in the monitoring matrix containing abnormal data are dynamically adjusted. For example, based on the state data of adjacent sensors, the sensor data with anomalies is weighted and corrected. After adaptive correction, all data in the monitoring matrix is ​​updated, ultimately resulting in a corrected third matrix for pipeline gas monitoring, which includes the corrected sensor data and reflects a more accurate pipeline gas concentration.

[0061] By employing anomaly detection, monitoring impact analysis, and adaptive correction, the impact of sensor anomalies on gas pipeline monitoring results is reduced, ensuring timely detection of sensor faults or drift. Adaptive correction improves the accuracy and reliability of monitoring data, contributing to more stable pipeline monitoring, reducing false alarms and missed alarms caused by sensor anomalies, and enhancing pipeline operation safety and monitoring efficiency.

[0062] Leak detection module 15 is used to perform pipeline leak detection based on the pipeline gas monitoring third matrix and obtain pipeline leak detection results.

[0063] Furthermore, the leakage detection module 15 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0064] The system includes a data extraction unit for extracting gas monitoring data of the first node corresponding to the first node of the pipeline based on the third pipeline gas monitoring matrix; a channel activation unit for activating the pipeline leakage risk detection channel, which includes P pipeline leakage risk detection models, where P is a positive integer greater than 1; a risk coefficient determination unit for inputting the gas monitoring data of the first node into the P pipeline leakage risk detection models to obtain P leakage risk detection coefficients; a proportion calculation unit for calculating the proportion of the P leakage risk detection accuracies corresponding to the P pipeline leakage risk detection models to obtain P risk detection weights; and a weighted calculation unit for performing weighted calculations on the P leakage risk detection coefficients based on the P risk detection weights to generate a first node leakage risk coefficient and adding the first node leakage risk coefficient to the pipeline leakage detection result.

[0065] Specifically, the pipeline gas monitoring third matrix is ​​a monitoring data matrix that has undergone background interference compensation and adaptive correction. It contains processed gas concentration data from each sensor, reflecting the gas distribution in the pipeline and the sensor status. Gas monitoring data corresponding to the first node of the pipeline is extracted from the pipeline gas monitoring third matrix, including the gas concentration at that node (e.g., ppm concentration of gases such as CO and CH4). The pipeline leakage risk detection channel is a model used to assess pipeline leakage risk, comprising P pipeline leakage risk detection models, each using a different method to assess leakage risk. P is a positive integer greater than 1; for example, P = 3 indicates the use of 3 different risk detection models. Typically, the P pipeline leakage risk detection models are constructed using different algorithms to assess leakage risk from different perspectives.

[0066] For example, a regression model and a deep learning model are used. A historical dataset of pipeline gas data from the past year is obtained, including data on gas concentrations (e.g., methane concentration), temperature, pressure, etc., at multiple nodes, as well as whether a leak event occurred (labeled data). Data preprocessing techniques such as Z-score normalization are used to process the data to make it uniform in scale. Support Vector Regression (SVR) is chosen as the algorithm to predict leak risk based on gas concentration and other environmental variables. A regression model f(x) is constructed, where x are the input features (gas concentration, temperature, pressure, etc.), and f(x) is the predicted leak risk coefficient. The model is trained by minimizing the loss function of SVR using 80% of the dataset as the training set and 20% as the validation set. The model is then tested using the validation set, and errors such as mean squared error (MSE) and coefficient of determination (R²) are calculated. 2 Assume the model's R... 2A value of 0.85 indicates that the model can predict leakage risk relatively well. The trained regression model is deployed to a real-time monitoring system, and leakage risk is calculated based on real-time gas concentration and other data. By using features such as gas concentration, temperature, and pressure, the regression model can quantitatively predict pipeline leakage risk, and its output is a leakage risk coefficient, which can help determine the probability or magnitude of a leak.

[0067] For the deep learning model, the historical gas data from the pipeline is standardized (e.g., using min-max standardization to scale the data to the [0,1] range), and the leak label (whether there is a leak) is treated as a binary classification problem (0 or 1). A multilayer perceptron (MLP) is used for modeling, constructing a three-layer neural network: the input layer contains features such as gas concentration, temperature, and pressure; the hidden layer consists of two hidden layers, each containing 64 neurons, using the ReLU activation function; the output layer outputs a binary classification probability using the Sigmoid activation function. Training is performed using the cross-entropy loss function and the Adam optimizer. The ratio of training to validation sets is 80:20. During training, the model learns the impact of each feature on leak risk and continuously adjusts the weights. Model performance is evaluated using metrics such as accuracy and F1 score. For example, assuming the trained model has an accuracy of 92% and an F1 score of 0.9, it indicates that the model has high accuracy in leak detection classification. The trained neural network model is deployed to a real-time monitoring system to predict leak risk and issue alarms in real time using pipeline sensor data. Deep learning models, such as multilayer perceptrons (MLP), are used to perform binary classification of leakage risks and output the probability (0 or 1) of whether a leakage has occurred, which is used for accurate leakage prediction and classification tasks.

[0068] The gas monitoring data from the first node is input into P pipeline leakage risk detection models, resulting in P different leakage risk detection coefficients. Each model derives a risk coefficient based on gas concentration, historical data, and other factors. For example, the risk coefficients output by the three models might be: Model 1: 0.8; Model 2: 0.6; Model 3: 0.9.

[0069] The accuracy of each leakage risk detection model is evaluated using historical or experimental data (e.g., by comparing model predictions with actual leakage situations), resulting in P risk detection weights. For example, the accuracy weight of Model 1 is 0.4, Model 2 is 0.3, and Model 3 is 0.3.

[0070] The P leakage risk detection coefficients are weighted and calculated based on P risk detection weights to obtain the leakage risk coefficient for the first node. This first node leakage risk coefficient is added to the leakage detection results of the entire pipeline system as part of the pipeline leakage monitoring results. For example, the weighted calculation yields a first node leakage risk coefficient of 0.77. This calculation is performed on the gas monitoring data of any node in the third matrix of pipeline gas monitoring, ultimately obtaining the leakage risk coefficient for each node. This constitutes the pipeline leakage detection results, summarizing the leakage risk data of all monitoring nodes, including the leakage risk coefficients of each node, which helps to assess the leakage risk of the entire pipeline system. By using a multi-model fusion approach, the accuracy and reliability of pipeline leakage risk detection are improved. Each model assesses leakage risk from different perspectives; the weighted calculation integrates the judgments of these models, reducing the bias that may exist in a single model, helping to more accurately identify leakage risks, issue timely warnings, and thus improve the safety and maintenance efficiency of pipeline operation.

[0071] The suppression verification module 16 is used to perform false alarm suppression verification on the pipeline leak detection results based on the olfactory perception optimization network to obtain a pipeline early warning signal.

[0072] Furthermore, the suppression verification module 16 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0073] A threshold setting unit is used to set a false alarm suppression verification mechanism, which includes a first leakage risk threshold and a second leakage risk threshold, wherein the first leakage risk threshold is greater than the second leakage risk threshold; a first judgment unit is used to judge whether the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold; and a warning signal generation unit is used to generate the pipeline warning signal based on the leakage risk coefficient of the first node if the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold.

[0074] Specifically, the false alarm suppression and verification mechanism avoids false alarms under normal circumstances by setting thresholds. Sometimes, olfactory sensors react to environmental changes or short-term fluctuations, issuing false alarms. By setting appropriate thresholds, it can be ensured that an alarm is only triggered when the leakage risk increases significantly. The false alarm suppression and verification mechanism includes a first leakage risk threshold and a second leakage risk threshold, used to set the risk assessment criteria for pipeline leaks. The first leakage risk threshold is usually high, indicating that when the leakage risk coefficient exceeds this threshold, the leakage risk is considered very high, worthy of issuing an alarm; the second leakage risk threshold is lower, used to judge the trend of risk changes, and further confirms whether there are false alarms by comparing it with the first threshold.

[0075] When the leakage risk coefficient at the first node is greater than or equal to the first leakage risk threshold, the criteria for triggering an alarm have been met, indicating a high probability of leakage, and an alarm should be issued. Based on the leakage risk coefficient at the first node, a pipeline early warning signal is generated, such as issuing an alarm sound, displaying a warning message, or automatically initiating an emergency response procedure. By setting a first and second leakage risk threshold, false alarms for short-term fluctuations or non-leakage events are effectively avoided, improving the accuracy and reliability of early warnings. An alarm is only issued immediately when the leakage risk coefficient reaches a high threshold, thus preventing frequent false alarms.

[0076] Furthermore, the suppression verification module 16 in the multi-scene fusion perception system of the humanoid robot olfactory sensor is also used for:

[0077] The second judgment subunit is used to determine whether the leakage risk coefficient of the first node is greater than or equal to the second leakage risk threshold if the leakage risk coefficient of the first node is less than the first leakage risk threshold, and obtain the first node risk judgment result; the update monitoring subunit is used to generate a first node feature monitoring instruction based on the first node risk judgment result, and control the olfactory perception optimization network to continuously monitor the first node of the pipeline based on the first node feature monitoring instruction, and obtain the first node update monitoring data; the risk coefficient calculation subunit is used to input the first node update monitoring data into the pipeline leakage risk detection channel, obtain the first node leakage risk update coefficient, and calculate the average of the first node leakage risk coefficient and the first node leakage risk update coefficient to generate a first reliable leakage risk coefficient; the signal generation subunit is used to generate the pipeline early warning signal if the first reliable leakage risk coefficient is greater than or equal to the first leakage risk threshold.

[0078] Specifically, when the leakage risk coefficient of the first node is less than the first leakage risk threshold, it indicates that there is a risk, but it has not yet reached the level requiring immediate warning. At this time, it is necessary to further determine whether it is greater than or equal to the second leakage risk threshold to obtain the risk assessment result of the first node. Based on the risk assessment result of the first node, a feature monitoring instruction for the first node is generated: if the risk assessment result of the first node is that the leakage risk coefficient of the first node is less than the second leakage risk threshold, then monitoring continues; if the risk assessment result of the first node is that the leakage risk coefficient of the first node is greater than or equal to the second leakage risk threshold, and the leakage risk coefficient of the first node is less than the first leakage risk threshold, then monitoring is enhanced.

[0079] The system executes the first-node feature monitoring command, controlling the olfactory perception optimization network to monitor the first node of the pipeline at the corresponding level, obtaining updated monitoring data for the first node. This updated monitoring data is then input into the pipeline leakage risk detection channel to calculate the first-node leakage risk update coefficient. Specifically, the data is input into P pipeline leakage risk detection models within the channel, yielding corresponding leakage risk detection coefficients for each model. These coefficients are then weighted to obtain the final first-node leakage risk update coefficient. Finally, the mean of the first-node leakage risk coefficient and the first-node leakage risk update coefficient is calculated to generate the first reliable leakage risk coefficient, achieving a fusion of trend correction and current observation results.

[0080] If the first credible leakage risk coefficient is greater than or equal to the first leakage risk threshold, it indicates that the updated risk coefficient has reached the standard for triggering an alarm, suggesting a high probability of leakage. An alarm should be issued to initiate emergency procedures such as manual re-inspection and valve closure. Otherwise, enhanced monitoring should continue or the frequency should be appropriately reduced according to the strategy.

[0081] By setting two thresholds to differentiate between different levels of risk and respond accordingly, the likelihood of false alarms is reduced while ensuring that no potential leaks are missed. For risks below the highest threshold but still worthy of attention, monitoring continues or is intensified, using updated monitoring data and calculating an average risk coefficient to more accurately assess leak risk and adjust based on new information. By requiring the credible leak risk coefficient to exceed a first threshold before generating an alert, alarms are only triggered when there is a highly credible actual leak, thus improving the accuracy of warnings.

[0082] In summary, the multi-scene fusion perception system for a humanoid robot olfactory sensor provided in this application has the following technical effects:

[0083] The system employs a multi-node olfactory perception optimization module to adaptively adjust the multi-node olfactory perception network of the gas pipeline based on the historical gas data of adjacent historical windows, thereby obtaining an optimized olfactory perception network. A first matrix module monitors the gas pipeline in real time using the optimized olfactory perception network to obtain a first pipeline gas monitoring matrix. A second matrix module compensates for background interference in the first pipeline gas monitoring matrix based on the real-time pipeline environment dataset, resulting in a second pipeline gas monitoring matrix. A third matrix module corrects the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix, resulting in a third pipeline gas monitoring matrix. A leak detection module performs pipeline leak detection based on the third pipeline gas monitoring matrix, obtaining a pipeline leak detection result. A suppression and verification module performs false alarm suppression verification on the pipeline leak detection result based on the optimized olfactory perception network, obtaining a pipeline early warning signal. In other words, by adaptively adjusting the multi-node olfactory sensing network of the gas pipeline based on the historical gas data of the adjacent historical window, the olfactory sensing optimization network is used to monitor the gas pipeline in real time. The monitoring data is corrected and compensated by combining the real-time environment and sensor status. A multi-modal fusion leak detection algorithm is executed. By setting a false alarm suppression and verification mechanism with multi-level risk thresholds, multi-level pipeline early warning signals are generated, realizing accurate prediction and intelligent response of gas leaks. This improves the accuracy and precision of gas leak detection in multiple scenarios and has the advantages of adaptive adjustment, strong anti-interference, and accurate intelligent identification.

[0084] Example 2: Based on the same inventive concept as the multi-scene fusion perception system for a humanoid robot olfactory sensor in Example 1, this application also provides a multi-scene fusion perception method for a humanoid robot olfactory sensor. Please refer to the appendix. Figure 2 The multi-scene fusion perception method for a humanoid robot olfactory sensor includes:

[0085] S100: Adaptively adjust the multi-node olfactory sensing network of the gas pipeline based on the historical gas data of the adjacent historical window to obtain an optimized olfactory sensing network; S200: Monitor the gas pipeline in real time based on the optimized olfactory sensing network to obtain a first pipeline gas monitoring matrix; S300: Compensate for background interference on the first pipeline gas monitoring matrix based on the real-time pipeline environment dataset to obtain a second pipeline gas monitoring matrix; S400: Correct the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix to obtain a third pipeline gas monitoring matrix; S500: Detect pipeline leaks based on the third pipeline gas monitoring matrix to obtain pipeline leak detection results; S600: Perform false alarm suppression verification on the pipeline leak detection results based on the optimized olfactory sensing network to obtain a pipeline early warning signal.

[0086] Furthermore, the adaptive adjustment of the multi-node olfactory perception network of the gas pipeline based on the pipeline gas history set of the nearest historical window to obtain the olfactory perception optimized network includes:

[0087] Based on the historical data of pipeline gas, trend prediction is performed on the gas pipeline to obtain the pipeline gas prediction trend; based on the pipeline gas prediction trend, risk identification is performed at each point of the gas pipeline to obtain pipeline point risk characteristics; based on the pipeline point risk characteristics, the nodes of the multi-node olfactory perception network are optimized to obtain a node-optimized olfactory perception network; based on the pipeline point risk characteristics, the sampling frequency of the node-optimized olfactory perception network is configured to generate the olfactory perception optimization network.

[0088] Furthermore, the step of performing background interference compensation on the first pipeline gas monitoring matrix based on the real-time pipeline environment dataset to obtain the second pipeline gas monitoring matrix includes:

[0089] Based on the pipeline gas monitoring first matrix, extract the pipeline gas monitoring first parameter; map and identify the pipeline real-time environment dataset based on the pipeline gas monitoring first parameter to obtain first pipeline environment data; identify interference in the pipeline gas monitoring first parameter based on the first pipeline environment data to determine a first monitoring background interference factor; perform confidence compensation on the pipeline gas monitoring first parameter based on the first monitoring background interference factor to generate gas monitoring first update data, and add the gas monitoring first update data to the pipeline gas monitoring second matrix.

[0090] Further, the step of performing confidence compensation on the first parameter of pipeline gas monitoring based on the first monitoring background interference factor to generate the first updated gas monitoring data includes:

[0091] Using the first monitoring background interference factor as the first retrieval constraint for interference compensation and the first parameter of pipeline gas monitoring as the second retrieval constraint for interference compensation, interference compensation samples are retrieved according to the first and second retrieval constraints to obtain a first interference compensation sample set. The confidence level of the first interference compensation sample set is evaluated to obtain the confidence coefficient of each compensation sample. Based on the confidence coefficients of each compensation sample, the first interference compensation sample set is filtered to obtain a confidence interference compensation sample set that meets a predetermined confidence level. The confidence interference compensation sample set is fused to obtain a first confidence compensation feature, and the first parameter of pipeline gas monitoring is corrected according to the first confidence compensation feature to obtain the first updated data of gas monitoring.

[0092] Further, the step of correcting the pipeline gas monitoring second matrix based on the state data of each olfactory sensor corresponding to the pipeline gas monitoring first matrix to obtain the pipeline gas monitoring third matrix includes:

[0093] Anomaly detection is performed based on the status data of each olfactory sensor to obtain the anomaly detection results of each sensor; the monitoring impact analysis of the gas pipeline is performed based on the anomaly detection results of each sensor to determine the impact characteristics of each sensor anomaly; the second pipeline gas monitoring matrix is ​​adaptively corrected based on the impact characteristics of each sensor anomaly to generate the third pipeline gas monitoring matrix.

[0094] Furthermore, the step of performing pipeline leak detection based on the third pipeline gas monitoring matrix to obtain pipeline leak detection results includes:

[0095] Based on the third pipeline gas monitoring matrix, extract the first node gas monitoring data corresponding to the first node of the pipeline; activate the pipeline leakage risk detection channel, which includes P pipeline leakage risk detection models, where P is a positive integer greater than 1; input the first node gas monitoring data into the P pipeline leakage risk detection models to obtain P leakage risk detection coefficients; calculate the proportion of the P leakage risk detection accuracy corresponding to the P pipeline leakage risk detection models to obtain P risk detection weights; perform a weighted calculation on the P leakage risk detection coefficients based on the P risk detection weights to generate the first node leakage risk coefficient, and add the first node leakage risk coefficient to the pipeline leakage detection result.

[0096] Furthermore, the step of performing false alarm suppression verification on the pipeline leak detection results based on the olfactory perception optimization network to obtain a pipeline early warning signal includes:

[0097] A false alarm suppression verification mechanism is set up, which includes a first threshold for leakage risk and a second threshold for leakage risk, wherein the first threshold for leakage risk is greater than the second threshold for leakage risk; it is determined whether the leakage risk coefficient of the first node is greater than or equal to the first threshold for leakage risk; if the leakage risk coefficient of the first node is greater than or equal to the first threshold for leakage risk, the pipeline early warning signal is generated based on the leakage risk coefficient of the first node.

[0098] Furthermore, determining whether the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold includes:

[0099] If the leakage risk coefficient of the first node is less than the first leakage risk threshold, determine whether the leakage risk coefficient of the first node is greater than or equal to the second leakage risk threshold to obtain the risk judgment result of the first node; based on the risk judgment result of the first node, generate a first node feature monitoring instruction, and control the olfactory perception optimization network to continuously monitor the first node of the pipeline based on the first node feature monitoring instruction to obtain updated monitoring data of the first node; input the updated monitoring data of the first node into the pipeline leakage risk detection channel to obtain the first node leakage risk update coefficient, and calculate the average of the first node leakage risk coefficient and the first node leakage risk update coefficient to generate a first reliable leakage risk coefficient; if the first reliable leakage risk coefficient is greater than or equal to the first leakage risk threshold, generate the pipeline early warning signal.

[0100] Furthermore, the step of obtaining a first pipeline gas monitoring matrix by real-time monitoring of the gas pipeline based on the olfactory perception optimization network includes:

[0101] The gas pipeline is monitored in real time using the olfactory perception optimization network to obtain a pipeline gas monitoring set; the pipeline gas monitoring set is filtered to obtain a pipeline gas dataset; and the pipeline gas dataset is matrixed to generate the first pipeline gas monitoring matrix.

[0102] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1 The multi-scene fusion perception system and specific examples of a humanoid robot olfactory sensor in Embodiment 1 are also applicable to the multi-scene fusion perception method of a humanoid robot olfactory sensor in this embodiment. Through the foregoing detailed description of the multi-scene fusion perception system of a humanoid robot olfactory sensor, those skilled in the art can clearly understand the multi-scene fusion perception method of a humanoid robot olfactory sensor in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here. As for the method disclosed in the embodiments, since it corresponds to the system disclosed in the embodiments, the description is relatively simple; relevant details can be found in the system section description.

[0103] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0104] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A multi-scene fusion perception system for a humanoid robot's olfactory sensor, characterized in that, include: The olfactory perception optimization module is used to adaptively adjust the multi-node olfactory perception network of the gas pipeline based on the pipeline gas history set of the adjacent historical window to obtain the olfactory perception optimization network. The first matrix module is used to monitor the gas pipeline in real time according to the olfactory perception optimization network to obtain the first matrix for pipeline gas monitoring. The second matrix module is used to perform background interference compensation on the first pipeline gas monitoring matrix based on the real-time pipeline environment dataset to obtain the second pipeline gas monitoring matrix. The third matrix module is used to correct the second pipeline gas monitoring matrix based on the state data of each olfactory sensor corresponding to the first pipeline gas monitoring matrix, and obtain the third pipeline gas monitoring matrix. The leak detection module is used to perform pipeline leak detection based on the pipeline gas monitoring third matrix and obtain pipeline leak detection results. The suppression verification module is used to perform false alarm suppression verification on the pipeline leak detection results based on the olfactory perception optimization network, and obtain a pipeline early warning signal. The olfactory perception optimization module includes: The trend prediction unit is used to predict the trend of the gas pipeline based on the historical data of the pipeline gas, and obtain the predicted trend of the pipeline gas. The risk identification unit is used to identify the risk at each point of the gas pipeline based on the predicted gas pipeline trend, and to obtain the risk characteristics of the pipeline points. The node optimization unit is used to optimize the nodes of the multi-node olfactory perception network according to the risk characteristics of the pipeline location, so as to obtain a node-optimized olfactory perception network. A frequency configuration unit is used to configure the sampling frequency of the node-optimized olfactory perception network according to the risk characteristics of the pipeline location, and generate the olfactory perception optimization network. The third matrix module includes: An anomaly detection unit is used to perform anomaly detection based on the state data of each olfactory sensor and obtain the anomaly detection results of each sensor. The impact analysis unit is used to perform monitoring impact analysis on the gas pipeline based on the abnormal detection results of each sensor, and to determine the impact characteristics of each sensor abnormality. An adaptive correction unit is used to adaptively correct the second pipeline gas monitoring matrix based on the influence characteristics of each sensor anomaly, and generate the third pipeline gas monitoring matrix. The leakage detection module includes: The data extraction unit is used to extract the first node gas monitoring data corresponding to the first node of the pipeline based on the pipeline gas monitoring third matrix. The channel activation unit is used to activate the pipeline leakage risk detection channel, which includes P pipeline leakage risk detection models, where P is a positive integer greater than 1. The risk coefficient determination unit is used to input the gas monitoring data of the first node into the P pipeline leakage risk detection models to obtain P leakage risk detection coefficients. The proportion calculation unit is used to calculate the proportion of the P leakage risk detection accuracies corresponding to the P pipeline leakage risk detection models, and obtain the P risk detection weights. The weighted calculation unit is used to perform weighted calculation on the P leakage risk detection coefficients according to the P risk detection weights, generate a first node leakage risk coefficient, and add the first node leakage risk coefficient to the pipeline leakage detection result.

2. The multi-scene fusion perception system for a humanoid robot olfactory sensor as described in claim 1, characterized in that, The second matrix module includes: The first parameter extraction unit is used to extract the first parameter of pipeline gas monitoring based on the pipeline gas monitoring first matrix; The first mapping and identification unit is used to map and identify the real-time environment dataset of the pipeline according to the first parameter of pipeline gas monitoring to obtain the first pipeline environment data. The first interference identification unit is used to identify interference in the first parameter of pipeline gas monitoring based on the first pipeline environment data, and determine the first monitoring background interference factor. The first confidence compensation unit is used to perform confidence compensation on the first parameter of pipeline gas monitoring according to the first monitoring background interference factor, generate first updated data of gas monitoring, and add the first updated data of gas monitoring to the second matrix of pipeline gas monitoring.

3. The multi-scene fusion perception system for a humanoid robot olfactory sensor as described in claim 2, characterized in that, The first confidence compensation unit includes: The constraint determination subunit is used to take the first monitoring background interference factor as the first retrieval constraint for interference compensation and the first parameter of pipeline gas monitoring as the second retrieval constraint for interference compensation. The interference compensation subunit is used to perform interference compensation sample retrieval based on the first interference compensation retrieval constraint and the second interference compensation retrieval constraint to obtain a first interference compensation sample set. The confidence evaluation subunit is used to evaluate the confidence level of the first interference compensation sample set and obtain the confidence coefficient of each compensation sample. The feature filtering subunit is used to filter the first interference compensation sample set based on the confidence coefficient of each compensation sample to obtain a confidence interference compensation sample set that meets a predetermined confidence level. The parameter correction subunit is used to fuse the confidence interference compensation sample set to obtain a first confidence compensation feature, and to correct the first parameter of the pipeline gas monitoring according to the first confidence compensation feature to obtain the first updated data of the gas monitoring.

4. The multi-scene fusion perception system for a humanoid robot olfactory sensor as described in claim 1, characterized in that, The suppression verification module includes: A threshold setting unit is used to set a false alarm suppression verification mechanism, wherein the false alarm suppression verification mechanism includes a first threshold for leakage risk and a second threshold for leakage risk, wherein the first threshold for leakage risk is greater than the second threshold for leakage risk. The first judgment unit is used to determine whether the leakage risk coefficient of the first node is greater than or equal to the first leakage risk threshold. The early warning signal generation unit is used to generate the pipeline early warning signal based on the leakage risk coefficient of the first node if the leakage risk coefficient of the first node is greater than or equal to the first threshold of leakage risk.

5. A multi-scene fusion perception system for a humanoid robot olfactory sensor as described in claim 4, characterized in that, The first determination unit includes: The second judgment subunit is used to determine whether the leakage risk coefficient of the first node is greater than or equal to the second leakage risk threshold if the leakage risk coefficient of the first node is less than the first leakage risk threshold, and to obtain the first node risk judgment result. The update monitoring subunit is used to generate a first node feature monitoring instruction based on the first node risk judgment result, and control the olfactory perception optimization network to continuously monitor the first node of the pipeline based on the first node feature monitoring instruction to obtain the first node update monitoring data. The risk coefficient calculation subunit is used to input the first node update monitoring data into the pipeline leakage risk detection channel, obtain the first node leakage risk update coefficient, and calculate the average of the first node leakage risk coefficient and the first node leakage risk update coefficient to generate a first reliable leakage risk coefficient. The signal generation subunit is used to generate the pipeline warning signal if the first credible leakage risk coefficient is greater than or equal to the first leakage risk threshold.

6. The multi-scene fusion perception system for a humanoid robot olfactory sensor as described in claim 1, characterized in that, The first matrix module includes: A real-time monitoring unit is used to monitor the gas pipeline in real time according to the olfactory perception optimization network to obtain a pipeline gas monitoring set. A data filtering unit is used to filter the pipeline gas monitoring set to obtain a pipeline gas dataset. The matrix processing unit is used to perform matrix processing on the pipeline gas dataset to generate the first pipeline gas monitoring matrix.

7. A multi-scene fusion perception method for an olfactory sensor of a humanoid robot, characterized in that, The multi-scene fusion perception system of a humanoid robot olfactory sensor according to any one of claims 1 to 6 is used for execution, and the multi-scene fusion perception method of the humanoid robot olfactory sensor includes: The multi-node olfactory perception network of the gas pipeline is adaptively adjusted based on the pipeline gas history set of the adjacent historical window to obtain the olfactory perception optimization network. The gas pipeline is monitored in real time using the olfactory perception optimization network to obtain the first matrix of pipeline gas monitoring. Based on the real-time pipeline environment dataset, the first pipeline gas monitoring matrix is ​​subjected to background interference compensation to obtain the second pipeline gas monitoring matrix. The pipeline gas monitoring second matrix is ​​corrected based on the status data of each olfactory sensor corresponding to the pipeline gas monitoring first matrix to obtain the pipeline gas monitoring third matrix; Pipeline leak detection is performed based on the third pipeline gas monitoring matrix to obtain pipeline leak detection results; The pipeline leak detection results are verified by the olfactory perception optimization network to suppress false alarms and obtain a pipeline early warning signal.

Images