Hospital logistics robot security inspection path planning system

Abstract

The invention relates to the technical field of hospital robot inspection, and discloses a hospital logistics robot security inspection path planning system. The system comprises a data acquisition module, a region partitioning module, a risk mapping module, a path combination module, a dynamic updating module and an anomaly detection module. The data acquisition module collects robot sensor data and environment data; the area partitioning module identifies a hospital area security risk intention based on the data, and divides a plurality of risk partitions; a risk mapping module obtains risk events corresponding to the risk partitions and constructs a mapping relationship between the risk events; a path combination module extracts key risk partitions and generates a path deviation combination; the dynamic updating module is combined with real-time monitoring data to update the mapping relation; and the anomaly detection module analyzes the behavior data of the robot based on the updated mapping relation to identify the abnormal activity. The system can adapt to the complex environment of a hospital, improves the polling pertinence and timeliness, and guarantees the security and protection safety of the hospital.

Description

Technical Field

[0001] This invention relates to the field of hospital robot inspection technology, specifically a hospital logistics robot security inspection path planning system. Background Technology

[0002] In the operation of modern hospitals, logistics and security patrols are crucial for ensuring normal functioning. With the development of robotics technology, hospital logistics robots are increasingly being used in tasks such as drug transportation, equipment delivery, and routine security patrols. Hospital environments are characterized by complex spatial structures, high personnel flow, and diverse functional areas. Different areas, such as the outpatient hall, inpatient wards, operating rooms, and drug storage rooms, have significantly different security requirements, which places higher demands on the patrol path planning of logistics robots. Most hospital logistics robots plan their inspection routes using pre-set fixed routes, which are difficult to adjust to real-time changes in the hospital environment. For example, when a certain area is obstructed by temporary construction, or when there is a sudden increase in personnel density due to a medical emergency, fixed routes not only affect inspection efficiency but may also fail to detect security risks in that area in a timely manner. Furthermore, current technology has low differentiation of security risks across different areas of the hospital, often treating all areas as having the same risk level for inspection. This results in high-risk areas such as pharmacy warehouses and operating rooms not receiving focused attention, while low-risk areas receive excessive inspection resources, leading to wasted resources. Existing systems lack dynamic update mechanisms for data processing, typically relying solely on initially collected environmental data for path planning. They fail to integrate the latest data acquired by sensors in real time, such as personnel movement data and equipment operating status data. This makes it difficult for the system to promptly identify emerging risk events, such as abnormal equipment overheating in a certain area or prolonged lingering by unfamiliar individuals, thus affecting the timeliness and accuracy of security patrols. Furthermore, in the anomaly detection phase, existing technologies often rely on single data indicators for judgment, lacking comprehensive analysis of multi-dimensional data. This can easily lead to misjudgments or missed detections, failing to effectively guarantee the security of hospitals. Summary of the Invention

[0003] The purpose of this invention is to provide a hospital logistics robot security inspection path planning system to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a hospital logistics robot security inspection path planning system, the system comprising: The data acquisition module is used to collect sensor data and environmental data from the hospital logistics robot. The area partitioning module is used to identify security risk intentions in the hospital area based on the sensor data and environmental data, and to divide the hospital area into multiple risk zones; The risk mapping module is used to obtain the risk events corresponding to the risk partitions and construct the mapping relationship between risk partitions and risk events; The path combination module is used to extract key risk partitions from the risk partitions and risk events, and generate path bias combinations based on the key risk partitions. The dynamic update module is used to update the mapping relationship between risk partitions and risk events based on the path bias combination and real-time monitoring data. The anomaly detection module is used to analyze robot behavior data and identify abnormal activities based on the updated mapping relationship.

[0005] Preferably, the data acquisition module is implemented in the following ways: Receive multi-source sensor data from the hospital logistics robot, including position data, speed data, and environmental perception data; The multi-source sensor data is grouped by type, and the sampling time step of each group of data is synchronized by a time window resampling method. Standardize the units of data from all groups, and detect and mark outliers; Missing values ​​are filled using interpolation methods, and outliers are handled by mean replacement. After verifying the data quality, normalization is used to unify the data of each group to a preset value range.

[0006] Preferably, the implementation of the region partitioning module includes: Based on the sensor data and environmental data, a security risk intent classification model is obtained; The security risk intent classification model is used to classify the data to obtain at least one risk intent category; Identify keywords in the data corresponding to the risk intent category to form a text sample set; Analyze the keywords in the text sample set to obtain the risk word area, parsing area and association area corresponding to the text sample set, which can be used as the risk zoning of the hospital area.

[0007] Preferably, the risk mapping module is implemented in the following ways: The risk event descriptions corresponding to the risk partitions are invoked to generate multiple unlabeled risk identification results. Determine whether multiple unlabeled risk identification results are the target risk identification results; If it is a result of target risk identification, then the target risk identification result shall be regarded as a risk event; Using information from the risk vocabulary area, parsing area, and association area in the risk partition, along with the descriptive information and categories of risk events, a mapping relationship between risk partitions and risk events is constructed.

[0008] Preferably, the path combination module is implemented in the following ways: Risk zones and risk events are clustered according to security type, risk level, and event function; Set the largest cluster center after cluster analysis as the key risk partition; Extract key features from key risk zones and calculate the similarity between each key feature; Establish common sequences related to the similarity between each key feature; Extract existing path segments using common sequences and set the matching degree between each path segment; The matching degree between each path segment is used to generate path bias combinations according to the time distribution probability.

[0009] Preferably, the dynamic update module is implemented in the following ways: Extract the temporal distribution probability of each path segment from the path bias combination; The path bias combination is set according to the time period corresponding to the time distribution probability of each path segment, and the target path is set accordingly. The target path of each path segment in the path bias combination is fitted to obtain the fitted target path. Set the probability value of the fitted target path in each time period as the combined occurrence probability; Compare the probability of a combination occurring with the actual occurrence of a risk event and identify the difference. Risk events are classified according to their differences, and the mapping relationship between risk zones and risk events is updated.

[0010] Preferably, the anomaly detection module is implemented in the following ways: Based on the updated mapping relationship between risk partitions and risk events, the robot's behavior records are retrieved; Compare the robot's activity characteristics before and after the update, and analyze the magnitude and frequency of behavioral changes; Calculate the degree of deviation in behavioral patterns; Compare the robot's data access frequency and analyze the changes in access time and frequency; Determine fluctuations in access frequency; Identify robot activities that deviate from the normal pattern and locate the source of the abnormal activity.

[0011] Preferably, the system further includes: a path optimization module, used to optimize the inspection path based on the abnormal activity; Extract robot operating status data; Construct the device state space and motion space; Set a reward function; The optimal path node is selected through a strategy tree generation and iteration process; Construct a complete inspection path optimization strategy based on the optimal path nodes; Dynamically adjust and optimize strategies.

[0012] Preferably, the system further includes: a management module, used to generate a path management library based on the optimized inspection path; Extract security patrol targets; The inspection targets are sorted according to their probability of occurrence to obtain the inspection order; By combining inspection targets and inspection sequences in a structured manner, a path management library is generated.

[0013] Preferably, the system further includes: a security response module, used to adjust the robot's security strategy based on the path management library; Analyze the frequency of risky operations performed by the robot; The scope of impact of the change in computing permissions; Identify robots that frequently engage in abnormal operations and reconfigure their access permissions; Adjust the robot's movement range and speed limits; Optimize the robot's verification method.

[0014] Compared with the prior art, the beneficial effects of the present invention are: The regional zoning module identifies the security risk intent of hospital areas based on the collected multi-dimensional data and divides them into multiple risk zones. This zoning method breaks away from the traditional "one-size-fits-all" approach to regional risks in fixed path planning. It can differentiate the zoning based on the actual risk characteristics of different areas. For example, the drug warehouse and operating room can be divided into high-risk zones, while ordinary corridors can be divided into low-risk zones. This allows inspection resources to be allocated reasonably according to risk levels, avoiding resource waste while ensuring that high-risk areas receive special attention. The risk mapping module establishes a mapping relationship between risk zones and risk events, enabling the system to clearly understand the types of risk events that may occur in different risk zones. For example, high-risk zones may experience drug loss or equipment failure, while low-risk zones may experience crowd congestion or slippery ground. This provides clear risk guidance for subsequent route planning, making the planned routes more targeted and able to prioritize coverage of areas prone to high-risk events. The path combination module extracts key risk zones from risk zones and risk events, and generates path bias combinations based on these key risk zones. This path planning method is not fixed, but rather uses key risk zones as the core to flexibly combine inspection paths. The path bias can be adjusted according to changes in different time periods and risk events. For example, during peak hours, the proportion of inspection paths in densely populated areas such as the outpatient hall can be increased, while at night, the proportion of inspection paths in areas such as the drug warehouse and equipment room can be increased, greatly improving inspection efficiency and flexibility. The dynamic update module is based on path bias combination and combines real-time monitoring data to update the mapping relationship between risk zones and risk events. This enables the system to respond in real time to changes in the hospital environment and promptly capture newly emerging risk factors, such as temporarily added medical equipment storage areas or sudden gatherings of people. It ensures that the risk mapping relationship is always consistent with the actual situation of the hospital, avoids the lag in risk identification caused by environmental changes, and allows the inspection path to be dynamically adjusted with changes in risk, ensuring the timeliness of inspections. The anomaly detection module analyzes robot behavior data based on updated mapping relationships to identify abnormal activities. Because the mapping relationships are updated in real time and cover multi-dimensional risk information, this module can comprehensively judge anomalies by combining multiple data sources. For example, by combining personnel flow data, equipment operation data, and risk zoning characteristics, it can identify anomalies such as unfamiliar personnel staying in high-risk zones for extended periods or equipment operating parameters exceeding normal ranges. This reduces misjudgments or missed judgments caused by judging based on a single data indicator, improves the accuracy of anomaly detection, better protects hospital security, and creates a safe and orderly operating environment for the hospital. The various modules of the entire system form a good collaborative loop. Data collection provides a basis for regional partitioning, regional partitioning supports risk mapping, risk mapping guides path combination, the dynamic update module ensures the real-time data and relationships of each link, and the anomaly detection module can verify the effectiveness of each link in reverse, so that the system can be continuously optimized during operation, continuously improve the quality and efficiency of security inspection, and adapt to the environmental changes and security needs adjustment in the long-term operation of the hospital. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the hospital logistics robot security patrol path planning system described in this invention;

[0016] Figure 2 A flowchart illustrating the working principle of the data acquisition module;

[0017] Figure 3 A flowchart illustrating the working principle of the risk mapping module;

[0018] Figure 4 This is a flowchart illustrating how the dynamic update module works. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 The present invention provides a hospital logistics robot security patrol path planning system, the system comprising: a data acquisition module, a regional partitioning module, a risk mapping module, a path combination module, a dynamic update module, and an anomaly detection module.

[0021] The data acquisition module collects sensor and environmental data from the hospital's logistics robot, including location, speed, and environmental perception information. The area zoning module identifies security risk intentions within the hospital area based on this data, dividing the area into multiple risk zones, such as high-risk, medium-risk, and low-risk zones. The risk mapping module acquires risk events corresponding to each risk zone, such as equipment malfunctions or personnel gatherings, and constructs a mapping relationship between zones and events. The path combination module extracts key risk zones from the mapping relationship and generates path-biased combinations, i.e., priority inspection paths for high-risk areas. The dynamic update module combines real-time monitoring data, such as sensor live streams, to update the mapping relationship to reflect the current risk status. The anomaly detection module analyzes robot behavior data based on the updated mapping relationship, identifying abnormal activities deviating from normal patterns, such as abnormal pauses or sudden speed changes. Through modular collaboration, the system achieves adaptive path planning, improving the efficiency and responsiveness of security inspections.

[0022] Example 1: See Figure 2 The data acquisition module of the hospital logistics robot's security patrol path planning system is responsible for receiving multi-source data streams from various sensors on the robot. This data includes centimeter-level position coordinates obtained via GPS receivers and indoor UWB positioning beacons, real-time speed readings generated by encoders embedded in the drive motors, and raw data captured by environmental perception devices such as visible light cameras, infrared thermal imagers, and LiDAR. These heterogeneous data are transmitted using different sampling frequencies and communication protocols; position data is typically updated at 10Hz, speed data is acquired at 50Hz, and high-resolution image data may be generated at 30fps.

[0023] To coordinate these multi-rate data streams, the system employs a sliding time window-based resampling mechanism. A fixed time window of one second is set to align the timing of all sensor data streams to a unified clock reference. Within each time window, high-frequency data (such as velocity readings) is downsampled, and the arithmetic mean within the window is calculated to generate a composite velocity value representing that time period. For low-frequency data (such as complete image frames), interpolation is used to calculate the characteristic values ​​of the current time window based on the data states at preceding and following time points. This timing synchronization method allows previously asynchronous data streams to be correlated and analyzed at the same timestamp.

[0024] After timing synchronization is complete, the system performs standardization processing on various types of data, unifying all length units to the international standard metric system, speed units to meters per second, and temperature readings to the Celsius scale. For image information in the environmental sensing data, it is converted into a standardized range of pixel intensity values. During this process, the system detects the numerical distribution characteristics of each data stream and uses statistical anomaly detection algorithms to identify outliers that deviate from the normal range. These outliers may manifest as sudden jumps in location coordinates, drastic fluctuations in speed readings, or abnormal spikes in temperature values. All detected outliers are marked with a special flag, but are not immediately removed from the data stream.

[0025] To address the potential issue of missing values ​​in the data stream, the system employs multiple interpolation strategies for imputation. For continuous data such as location and velocity information, a linear interpolation algorithm is used to calculate the estimated location of the missing data based on the preceding and following valid data points. For discrete missing values ​​in environmental perception data, the nearest neighbor interpolation method is used to select the most temporally closest valid data for imputation. Marked outliers are handled using a sliding window mean replacement strategy, employing the average value of normal data within a certain time window before and after the outlier as the replacement value.

[0026] After all data has been cleaned and filled, the system executes a data quality verification process. This verification includes data integrity checks to ensure all data types have valid values ​​within each time window; temporal consistency verification to confirm that timestamps of all data streams remain synchronized; and numerical range verification to ensure all data falls within the physically possible range of values. The verified data is then normalized, mapping all data types to a numerical range of [0,1]. Location coordinates are normalized using a min-max normalization method, scaled according to the actual map size; velocity data is linearly normalized based on the robot's maximum design speed; and environmental perception data is scaled according to the maximum range of each sensor. After these processing steps, all sensor data is converted into a standardized, dimensionless numerical form, providing a consistent input format for subsequent analysis and processing.

[0027] This data processing approach adapts to the complexity of hospital environments. For example, in high-dynamic areas such as the emergency department, sensor data often contains more noise and missing data. The system ensures data quality by increasing the interpolation window size and adjusting the anomaly detection threshold. In relatively stable inpatient areas, standard parameters can be used for processing. The entire data acquisition process adopts a pipeline architecture, with each processing step executed in parallel, ensuring that the system can process massive amounts of sensor data generated by multiple robots in real time. All intermediate states and parameter adjustments during processing are recorded in detail, forming a complete data traceability chain, facilitating subsequent analysis modules to track the data transformation process. The final output standardized data stream not only contains numerical information but also carries data quality markers and processing history metadata, providing a rich information foundation for other modules of the system.

[0028] Example 2: See Figure 3 The regional partitioning module of the hospital logistics robot security patrol path planning system identifies and divides the hospital environment into zones by analyzing standardized multi-source sensor data. This module first loads a pre-trained security risk intent classification model, which is based on a deep neural network architecture and trained using historical security event data and multimodal sensor data. The model receives standardized data streams from the data acquisition module, including position coordinates, velocity vectors, thermal imaging temperature data, and visual feature vectors. The model jointly analyzes these multimodal inputs and outputs the categories of potential risk intents in the hospital environment, such as typical risk scenarios like equipment malfunction, excessive personnel gathering, abnormal ambient temperature, or suspicious object presence. Each identified risk intent is accompanied by a confidence score to measure the reliability of the identification results.

[0029] After obtaining the risk intent classification results, the module performs textual processing on the raw sensor data, converting numerical sensor readings and image features into natural language descriptions. Location coordinates are mapped to specific department names within the hospital; speed patterns are described as behavioral characteristics such as "steady patrol" or "abrupt stop and quick walk"; temperature data is converted into state descriptions such as "normal temperature" or "localized overheating"; and visual data is used to generate text descriptions through image annotation algorithms. These text descriptions, combined with the risk intent categories, form a set of text samples containing rich semantic information. The system employs natural language processing techniques to perform in-depth analysis of these text samples, extracting key semantic elements and correlation patterns.

[0030] The keywords in the text sample set were processed through a multi-level analysis process, including word frequency statistics and importance assessment, to identify key terms that frequently appeared under specific risk intent, such as "high temperature alarm," "crowded areas," and "abnormal equipment noise," which have clear risk indication significance. These words were categorized into different functional areas based on their semantic features and context: the risk vocabulary area stores core words with direct risk indication significance, the parsing area contains auxiliary words for risk explanation and context description, and the association area stores words related to the relationships between different risk elements. The words in each area are labeled and indexed in multiple dimensions according to risk level, urgency, and scope of impact.

[0031] Based on these semantic analysis results, the module divides the hospital's physical space into corresponding risk zones. High-risk terms in the risk term zone are mapped to key areas within the hospital, such as operating rooms, ICUs, and pharmacies—areas requiring high-level security monitoring. Medium-risk terms in the resolution zone correspond to general wards and corridors—areas requiring routine monitoring. Low-risk terms in the association zone correspond to relatively safe areas such as offices and rest areas. Each risk zone has clearly defined boundaries and risk characteristics, including the expected personnel density range, equipment operating status indicators, and environmental parameter thresholds. This zone information is stored and managed in the form of a spatial grid, with each grid cell associated with corresponding risk attribute metadata.

[0032] The risk mapping module further constructs the correspondence between risk zones and specific risk events. The module accesses the hospital's security knowledge base to obtain standardized descriptions of various risk events, including detailed information such as event type, severity, and handling procedures. The system matches the real-time identified risk intent with event templates in the knowledge base to generate preliminary risk identification results. These results are initially unlabeled and require a verification process to be confirmed as valid risk events. The verification process employs a multi-evidence fusion strategy, comprehensively evaluating the collaborative evidence and support from different sensors. Only when multiple independent sensor data point to the same risk judgment, and the confidence level exceeds a set threshold, will the preliminary identification result be confirmed as the target risk identification result.

[0033] Once a risk is identified as a target risk, it is formally marked as a risk event instance and mapped to the corresponding risk zone. The mapping is constructed using a multi-dimensional association model, which not only records the simple correspondence between risk zones and risk events but also includes rich contextual information. The risk vocabulary area provides specific numerical values ​​and status descriptions of core risk indicators; the parsing area records the contextual conditions and influencing factors of the risk occurrence; and the association area stores the association patterns and propagation paths between different risk events. These mappings are stored in a graph structure, where nodes represent risk zones or risk events, edges represent the associations between them, edge weights indicate the association strength, and edge attributes record the specific characteristics and historical evidence of the association.

[0034] The entire mapping process is dynamically evolving. The system continuously monitors new sensor data inputs and automatically adjusts existing mappings when a change in risk patterns is detected. For example, if the population density in a certain area consistently exceeds historical patterns, the system will correspondingly increase the risk level of that area and adjust its mapping weight with related risk events. Simultaneously, the system maintains the version history of the mapping relationships, recording the reasons and basis for each adjustment, forming a complete audit trail. This dynamic mapping mechanism enables the system to adapt to changes in the hospital environment, maintaining the accuracy of risk identification and zoning management. The final risk zoning and risk event mapping relationship provides a decision-making basis for subsequent path planning. Each risk zoning is associated with a set of possible risk events and their probabilities, and each risk event corresponds to a spatial area and contextual conditions. This rich association information enables the system to perform accurate risk prediction and targeted inspection planning, ensuring that the logistics robot's security inspection work can effectively cover various risk points in the hospital environment, achieving early detection and timely handling of potential safety hazards.

[0035] Example 3: See Figure 4 The path combination module of the hospital logistics robot security patrol path planning system performs in-depth analysis of risk zones and risk events with established mapping relationships, aiming to generate patrol path solutions with risk-specific targets. This module first performs multi-dimensional cluster analysis on risk zones and related events, considering dimensions including security type (e.g., physical security, network security, environmental security), risk level (high, medium, and low classification), and event function (monitoring, early warning, and response). The clustering algorithm employs an improved adaptive density clustering method, which can automatically determine the optimal number of clusters and dynamically adjust the cluster radius based on data distribution characteristics. Each cluster center represents a set of zones with similar risk characteristics. The cluster with the largest number of zones and the highest density of risk events is identified as the key risk cluster, and its core area is set as the key risk zone.

[0036] After identifying key risk zones, the module extracts multidimensional feature vectors for these zones, including spatial location features, temporal features, risk intensity features, and environmental association features. These feature vectors are transformed into comparable numerical representations through feature engineering, and the comprehensive similarity between each feature is calculated. The similarity calculation considers not only static numerical matching of features but also the comparison of dynamic change patterns, employing a sequence similarity measurement method based on dynamic time warping, which can effectively handle feature sequences of different time lengths. Based on feature similarity analysis, the module constructs a common sequence model describing the evolution of risks. This model captures the order of occurrence and dependencies of different risk features in the temporal dimension, forming a temporal pattern template for risk development. Using these common sequences, the system extracts path segments from historical inspection path data that match the current risk pattern. Each path segment contains spatial trajectory information, temporal distribution characteristics, and risk coverage features. The matching degree between segments is calculated using a multidimensional similarity function, comprehensively considering factors such as spatial overlap, temporal synchronization, and risk response level.

[0037] The matching degree analysis results of path segments are converted into a time distribution probability model, which describes the applicability and effectiveness probability of each path segment at different time periods. The system employs time series analysis technology based on Hidden Markov Models to capture the dynamic patterns of path segment selection probabilities changing over time. The final generated path bias combination is a set of time-adaptive path solutions, each labeled with its applicable time range, expected risk coverage effect, and execution priority score. The path solutions in the combination are ranked according to risk response effectiveness, ensuring that high-priority risk areas receive focused coverage.

[0038] The dynamic update module is responsible for continuously optimizing and adjusting the path bias combinations. This module extracts the temporal distribution probability features of each path segment from the path combinations and establishes a correlation model between path selection and time factors. Based on these temporal probability distributions, the system sets corresponding target path schemes for different time periods. These target paths not only consider the current risk situation but also include predictions of future risk trends. Target paths from multiple time periods are integrated and optimized using a path fitting algorithm to generate smooth and continuous global inspection paths. The fitting process employs an adaptive spline curve method to ensure the feasibility and execution efficiency of the paths while maximizing coverage of key risk areas. The fitted target path corresponds to a combination occurrence probability value at each time point, reflecting the applicability of the path scheme under the current spatiotemporal conditions. The system calculates the dynamic fitness of the path scheme using the following formula: ; in: This represents the overall fitness of path scheme P at time t. This represents the weight coefficient of the i-th risk factor. It is the predicted coverage probability of path P for risk i at time t. It is the actual probability of risk i occurring at time t. This represents the total number of risk factors considered. This fitness function effectively quantifies the degree of match between the proposed path and the actual risk situation.

[0039] The system continuously compares the predicted performance of path solutions with the actual occurrence of risks, calculating the difference between the two. The difference calculation employs a multi-indicator approach, including response delays in the time dimension, coverage gaps in the spatial dimension, and insufficient response in the intensity dimension. These difference values ​​are used to reassess and classify risk events, adjusting the risk level and handling priority of events based on the actual response results. Based on these adjustments, the module dynamically updates the mapping relationship between risk zones and risk events, including modifying mapping weights, adjusting correlation strength, and updating response strategies.

[0040] The entire update process employs an incremental learning mechanism, continuously incorporating new monitoring data into the existing model while retaining representative samples of historical data. This ensures the model can adapt to environmental changes without forgetting important historical patterns. The updated mapping relationship is immediately fed back to the path combination module, forming a closed-loop optimization process. This dynamic adjustment mechanism enables the system to adapt to changes in the hospital environment, such as adjustments to departmental functions, changes in patient flow patterns, or the emergence of new risk patterns, always maintaining a high degree of match between the inspection path plan and the actual risk situation. Through this continuous optimization approach, the system can provide intelligent security inspection path guidance for hospital logistics robots, making inspection work more focused on actual risks and improving the accuracy and timeliness of security response. The dynamic adaptability of the path plan ensures the optimal allocation of limited inspection resources, achieving a balance between maximizing risk coverage and minimizing inspection costs.

[0041] Example 4: The anomaly detection module of the hospital logistics robot security patrol path planning system performs multi-dimensional monitoring and analysis of the robot's operational behavior based on dynamically updated risk mapping relationships. This module first retrieves historical behavior records of the robot from the central data warehouse. These records are stored in time-series format and include detailed information such as the robot's movement trajectory coordinates, speed change curves, sensor activation status logs, and task execution timestamps. The system establishes a behavioral baseline model for the robot. This model extracts the normal behavioral pattern characteristics of each robot in different time periods and areas by analyzing long-term historical data, including typical speed distribution ranges, common movement path sets, and standard environmental interaction frequencies.

[0042] The system compares the changes in robot activity characteristics before and after the update, focusing on the evolution trend of behavioral patterns. For movement trajectory characteristics, it analyzes changes in path deviation and calculates the spatial difference coefficient between the actual path and the planned path; for speed characteristics, it monitors changes in acceleration patterns and identifies abnormal rapid acceleration or deceleration; for task execution characteristics, it compares changes in task time distribution and detects abnormal fluctuations in execution efficiency. These changes are analyzed using a sliding time window mechanism, calculating the magnitude and frequency of behavioral changes in the short term (last hour), medium term (last 24 hours), and long term (last 7 days) to establish a multi-dimensional behavioral evolution map.

[0043] The degree of deviation in robot behavior patterns is quantitatively assessed using composite indicators. The system constructs a behavior feature vector space, mapping various activity parameters of the robot into a high-dimensional feature space, and calculates the comprehensive distance between the current behavior state and the historical baseline state. This distance metric considers not only numerical differences but also morphological and temporal pattern differences in the behavior sequence. Simultaneously, the system analyzes the persistence and trend of behavior changes, distinguishing between temporary fluctuations and systematic deviations, avoiding misjudging occasional normal changes as anomalies.

[0044] The monitoring of data access patterns focuses on analyzing the communication behavior between robots and the system server. The system records information such as the time point of each robot's data request, access frequency, data transmission volume, and access target, establishing a communication behavior profile. By analyzing the distribution patterns of access time points, it identifies access behavior during abnormal time periods, such as data requests outside of working hours; by statistically analyzing the changing patterns of access frequency, it detects abnormally frequent or sparse communication activities; and by analyzing the correlation of access targets, it discovers abnormal data access patterns, such as sudden access to data resources in unrelated areas. Fluctuations in communication frequency are assessed using a multi-level monitoring mechanism. The system sets dynamic threshold ranges and establishes upper and lower boundaries of normal fluctuation ranges based on the statistical characteristics of historical communication patterns. For short-term fluctuations, real-time monitoring is used to detect abnormal peaks or troughs in instantaneous communication volume; for medium- and long-term trend changes, time series analysis is used to identify systematic shifts in communication patterns. Fluctuation assessment also considers environmental context factors, such as differences in normal communication patterns under different scenarios, such as hospital working hours and special events.

[0045] The identification and localization of abnormal activities employs a multi-evidence fusion decision-making mechanism. The system comprehensively analyzes behavioral changes, assesses communication patterns, and incorporates environmental context information to calculate anomaly confidence scores. When multiple independent indicators simultaneously display abnormal characteristics, the system generates an anomaly alarm and performs root cause analysis. The localization process involves retrospectively analyzing the time series of abnormal behavior, correlating it with corresponding environmental events and system state changes, and identifying the triggering factors and propagation paths of abnormal activities. The system also establishes an anomaly type classification system, distinguishing between anomalies caused by equipment failure, anomalies caused by environmental interference, and potential safety threats. The path optimization module dynamically adjusts the inspection path based on anomaly detection results. The module extracts real-time operating status data of the robot, including parameters such as battery level, sensor operating status, computational load, and network connection quality. These status parameters, combined with anomaly detection results, construct a complete description of the equipment's state space, with state dimensions including the robot's physical state, functional state, and environmental interaction state.

[0046] The action space is constructed to include all possible action choices for the robot, including options for adjusting movement direction, speed control, path selection, and task execution. Each action option is associated with execution conditions and expected effect evaluation. The reward function is set using a multi-objective optimization principle, comprehensively considering multiple optimization objectives such as risk coverage, energy efficiency, anomaly response speed, and task completion quality. Each objective is assigned a corresponding weight coefficient, which is dynamically adjusted according to the current environmental risk and system state. For example, during high-risk periods, the weight of risk coverage is increased accordingly; when battery power is low, the weight of energy efficiency is appropriately increased. The reward value is calculated based on the degree of matching between the actual effect after action execution and the expected goal. The policy tree is generated using a depth-first search algorithm, starting from the current state, simulating the state transition paths that different action choices may lead to, and constructing a multi-step look-ahead decision tree. Each tree node represents a system state, edges represent action choices, and leaf nodes contain the cumulative reward value estimate for that path. Through an iterative optimization process, the system selects the path node sequence with the largest reward value as the optimal decision scheme.

[0047] Based on the optimal path nodes, the module constructs a complete inspection path optimization strategy. This strategy includes not only spatial path planning but also comprehensive optimization aspects such as time scheduling schemes, speed control strategies, and task execution sequences. The strategy implementation adopts a hierarchical control architecture, with higher-level strategies determining the overall direction and objectives, and lower-level strategies handling specific execution details. A dynamic adjustment mechanism for the optimization strategy can respond in real-time to environmental changes and anomalies. The system continuously monitors the strategy execution effect, and when a significant deviation is detected between the actual effect and the expected target, the strategy re-optimization process is automatically triggered. Adjustment methods include multi-level adjustment mechanisms such as parameter fine-tuning, local strategy correction, and global strategy reconstruction, ensuring that the system always maintains an optimal operating state.

[0048] Table 1: Evaluation Index of Robot Behavior Anomalies.

[0049]

[0050] Referring to Table 1, through this comprehensive anomaly detection and path optimization mechanism, the system can ensure that the hospital logistics robot maintains efficient and reliable operation in complex security inspection environments, promptly detects and handles anomalies, and continuously improves inspection efficiency and risk response capabilities through dynamic path optimization. The entire system forms a complete closed loop of detection-evaluation-optimization-execution, realizing intelligent and adaptive security inspection work.

[0051] Example 5: The management module of the hospital logistics robot security inspection path planning system is responsible for transforming the optimized inspection path into an executable path management scheme. This module first receives dynamically adjusted optimal path schemes from the path optimization module. These schemes include complete spatial trajectory information, time scheduling plans, and task execution sequences. The system extracts the security inspection targets involved in the path schemes. These targets include not only physical location points but also specific security inspection items to be performed, such as equipment status detection, environmental parameter collection, and anomaly observation—a diverse range of task elements. Each inspection target is accompanied by detailed execution requirements and quality standards, forming a structured task description system.

[0052] The system performs occurrence probability analysis on the extracted inspection targets, calculating the importance and urgency scores of each target in the current time period based on historical execution data and real-time environmental conditions. The occurrence probability calculation comprehensively considers multiple dimensions, including the frequency of historical risk events in the target area, risk indicators reflected in current sensor monitoring data, the influence weight of environmental context information, and the complexity of task dependencies. These probability values ​​are generated through a multi-source information fusion algorithm and continuously updated with new data input, ensuring the timeliness and accuracy of the probability assessment. Based on the occurrence probability assessment results, the system dynamically optimizes the ranking of inspection targets. The ranking algorithm considers not only the occurrence probability of individual targets but also the spatial correlation, temporal continuity, and resource constraints between targets. Spatial correlation analysis ensures that geographically close targets are arranged in the same inspection batch, reducing robot travel distance and time consumption; temporal continuity optimization ensures that time-sensitive targets are prioritized; and resource constraint considerations ensure that the ranking scheme meets the actual limitations of robot endurance, sensor performance, and other practical constraints. The final generated inspection sequence is an optimal sequence that comprehensively considers the balance of multiple factors, maximizing risk coverage efficiency while ensuring the feasibility of task execution.

[0053] The system generates a complete path management database by combining inspection targets and sequences in a structured format. This database employs a hierarchical storage architecture: the top layer stores a summary of global path solutions, the middle layer stores specific task execution sequences and time-sharing scheduling plans, and the bottom layer stores detailed execution parameters and quality standards. All data is encoded and indexed in a unified format, supporting fast querying and dynamic updates. The path management database also includes version control functionality, recording the history of each modification and the reasons for changes, facilitating traceability and analysis of the decision-making process. Each entry in the database contains rich metadata information, such as creation time, last update time, validity period, and applicable conditions.

[0054] The safety response module, based on the path management library, fine-tunes the robot's security strategy. The module analyzes risk operation frequency data recorded by the robot during inspection tasks, including detailed information such as device access records, number of area entries, and special operation executions. Risk operation analysis employs a multi-level classification method, categorizing and statistically analyzing operations according to operation type, risk level, and scope of impact to create a profile of the robot's operational behavior. Through long-term data accumulation, the system identifies the typical operating patterns and behavioral characteristics of each robot.

[0055] The impact assessment of permission changes employs a network analysis approach, modeling factors such as robots, access targets, and operation types as nodes and edges in a complex network. When considering adjusting the access permissions of a particular robot, the system simulates the potential cascading effects of this change throughout the network, including indirect impacts on other robots, potential impacts on system functionality, and the overall impact on the security system. The impact calculation considers not only direct effects but also second- and third-order indirect effects, ensuring the comprehensiveness and security of permission adjustment decisions.

[0056] The system identifies robots exhibiting frequent abnormal operating patterns. These anomalies may manifest as an unusual number of device accesses beyond the normal range, operational activities during abnormal time periods, or operational sequences that do not meet task requirements. For these robots, the system initiates a permission reconfiguration process, finely adjusting their access permissions based on their actual task needs and historical behavior records. Permission configuration follows the principle of least privilege, granting only the minimum permissions required to perform necessary tasks, thus reducing potential security risks.

[0057] The robot's movement range is optimized based on risk zoning mapping and actual task requirements. The system dynamically sets the permitted spatial boundaries based on the robot's role and current risk assessment results. These boundaries include not only physical space limitations but also temporal constraints, such as allowing access to certain areas only during specific time periods. The movement range adjustment employs a progressive optimization strategy, gradually adjusting the activity range based on the robot's behavioral compliance and task performance, ensuring both security and uninterrupted task execution. Speed ​​limit adjustments comprehensively consider environmental factors and task requirements. In high-risk or densely populated areas, the system appropriately reduces the robot's maximum permissible speed to improve movement safety; in low-risk areas or emergency situations, the speed limit can be appropriately increased to ensure task efficiency. Speed ​​limit settings also consider the robot's mechanical characteristics, sensor performance, and current environmental conditions to ensure the rationality and feasibility of the limits. Optimization of the verification method focuses on improving the security and convenience of identity authentication. The system employs a multi-factor authentication mechanism, combining robot hardware feature recognition, digital certificate verification, and dynamic passwords. For high-risk operations, stricter verification processes are required; for routine operations, simplified verification methods are used to improve efficiency. The verification mechanism also includes anomaly detection functionality, automatically increasing verification strength or triggering security alerts when abnormal login attempts or suspicious operation patterns are detected. The entire security response process forms a closed-loop control mechanism, continuously monitoring and evaluating the effects of permission adjustments and policy optimizations, and making further adjustments based on actual results. The system maintains a complete record of security policy changes, including the reasons, basis, and implementation effects of each adjustment, providing data support for security audits and policy optimization. Through this dynamic security response mechanism, the hospital logistics robot system is ensured to maintain a safe and reliable operating state in complex environments, effectively completing security patrol tasks while minimizing security risks.

[0058] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A hospital logistics robot security patrol path planning system, characterized in that, include: The data acquisition module is used to collect sensor data and environmental data from the hospital logistics robot. The area partitioning module is used to identify security risk intentions in the hospital area based on the sensor data and environmental data, and to divide the hospital area into multiple risk zones; The risk mapping module is used to obtain the risk events corresponding to the risk partitions and construct the mapping relationship between risk partitions and risk events; The path combination module is used to extract key risk partitions from the risk partitions and risk events, and generate path bias combinations based on the key risk partitions. The dynamic update module is used to update the mapping relationship between risk partitions and risk events based on the path bias combination and real-time monitoring data. The anomaly detection module is used to analyze robot behavior data and identify abnormal activities based on the updated mapping relationship.

2. The hospital logistics robot security patrol path planning system according to claim 1, characterized in that, The data acquisition module is implemented in the following ways: Receive multi-source sensor data from the hospital logistics robot, including position data, speed data, and environmental perception data; The multi-source sensor data is grouped by type, and the sampling time step of each group of data is synchronized by a time window resampling method. Standardize the units of data from all groups, and detect and mark outliers; Missing values ​​are filled using interpolation methods, and outliers are handled by mean replacement. After verifying the data quality, normalization is used to unify the data of each group to a preset value range.

3. The hospital logistics robot security inspection path planning system according to claim 2, characterized in that, The implementation methods of the region partitioning module include: Based on the sensor data and environmental data, a security risk intent classification model is obtained; The security risk intent classification model is used to classify the data to obtain at least one risk intent category; Identify keywords in the data corresponding to the risk intent category to form a text sample set; Analyze the keywords in the text sample set to obtain the risk word area, parsing area and association area corresponding to the text sample set, which can be used as the risk zoning of the hospital area.

4. The hospital logistics robot security inspection path planning system according to claim 3, characterized in that, The risk mapping module is implemented in the following ways: The risk event descriptions corresponding to the risk partitions are invoked to generate multiple unlabeled risk identification results. Determine whether multiple unlabeled risk identification results are the target risk identification results; If it is a result of target risk identification, then the target risk identification result shall be regarded as a risk event; Using information from the risk vocabulary area, parsing area, and association area in the risk partition, along with the descriptive information and categories of risk events, a mapping relationship between risk partitions and risk events is constructed.

5. The hospital logistics robot security patrol path planning system according to claim 4, characterized in that, The path combination module is implemented in the following ways: Risk zones and risk events are clustered according to security type, risk level, and event function; Set the largest cluster center after cluster analysis as the key risk partition; Extract key features from key risk zones and calculate the similarity between each key feature; Establish common sequences related to the similarity between each key feature; Extract existing path segments using common sequences and set the matching degree between each path segment; The matching degree between each path segment is used to generate path bias combinations according to the time distribution probability.

6. The hospital logistics robot security inspection path planning system according to claim 5, characterized in that, The implementation methods of the dynamic update module include: Extract the temporal distribution probability of each path segment from the path bias combination; The path bias combination is set according to the time period corresponding to the time distribution probability of each path segment, and the target path is set accordingly. The target path of each path segment in the path bias combination is fitted to obtain the fitted target path. Set the probability value of the fitted target path in each time period as the combined occurrence probability; Compare the probability of a combination occurring with the actual occurrence of a risk event and identify the difference. Risk events are classified according to their differences, and the mapping relationship between risk zones and risk events is updated.

7. The hospital logistics robot security inspection path planning system according to claim 6, characterized in that, The anomaly detection module is implemented in the following ways: Based on the updated mapping relationship between risk partitions and risk events, the robot's behavior records are retrieved; Compare the robot's activity characteristics before and after the update, and analyze the magnitude and frequency of behavioral changes; Calculate the degree of deviation in behavioral patterns; Compare the robot's data access frequency and analyze the changes in access time and frequency; Determine fluctuations in access frequency; Identify robot activities that deviate from the normal pattern and locate the source of the abnormal activity.

8. The hospital logistics robot security inspection path planning system according to claim 7, characterized in that, It also includes: a path optimization module, used to optimize the inspection path based on the abnormal activity; Extract robot operating status data; Construct the device state space and motion space; Set a reward function; The optimal path node is selected through a strategy tree generation and iteration process; Construct a complete inspection path optimization strategy based on the optimal path nodes; Dynamically adjust and optimize strategies.

9. The hospital logistics robot security inspection path planning system according to claim 8, characterized in that, Also includes: The management module is used to generate a path management library based on the optimized inspection path; Extract security patrol targets; The inspection targets are sorted according to their probability of occurrence to obtain the inspection order; By combining inspection targets and inspection sequences in a structured manner, a path management library is generated.

10. The hospital logistics robot security inspection path planning system according to claim 9, characterized in that, Also includes: A safety response module is used to adjust the robot's security strategy based on the path management library; Analyze the frequency of risky operations performed by the robot; The scope of impact of the change in computing permissions; Identify robots that frequently engage in abnormal operations and reconfigure their access permissions; Adjust the robot's movement range and speed limits; Optimize the robot's verification method.

Images