Patrol robot based on deep learning algorithm and big data analysis model

Abstract

The invention provides a patrol robot based on a deep learning algorithm and a big data analysis model, and relates to the technical field of medical robots. Comprising a physiological data integration module, a behavior state recognition module, an equipment conflict regulation and control module, a task grading execution module and a control instruction scheduling module. According to the method, the physiological signals of the multiple devices are integrated according to the time pushing sequence, overlapping and synchronous fluctuation sections among the signals are recognized, high-precision time sequence alignment is achieved, linkage features are extracted in combination with patient action tracks and signal fluctuation consistency, the accuracy of abnormal state recognition is improved, the device data processing sequence is arranged according to the sudden change amplitude, and the accuracy of abnormal state recognition is improved. Key signal priority judgment is strengthened, non-emergency operation compresses a calling period according to response time and function classification, high-frequency task execution efficiency is improved, operation instructions are screened and cooperatively distributed according to the emergency degree, the execution process and the field response speed are optimized, and diagnosis and treatment regulation and control capacity and intervention timeliness in a severe scene are enhanced.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, and in particular to a mobile clinic robot based on deep learning algorithms and big data analysis models. Background Technology

[0002] The field of medical robotics encompasses the integrated application of artificial intelligence and information technology in healthcare scenarios, covering multiple areas such as disease prediction and diagnosis, clinical decision support, intelligent image recognition, health management, and telemedicine. In critical care medicine, particularly in intensive care unit (ICU) management, medical robots utilize deep learning, image recognition, speech recognition, IoT sensing, and intelligent decision-making systems as core tools. By analyzing and processing multimodal patient data, they assist doctors in completing high-frequency, high-intensity medical tasks. This technological field emphasizes a data-driven transformation of the medical model, promoting the digitalization and intelligentization of healthcare services. Especially in ICU settings with limited medical resources and high patient density, AI can improve diagnostic and treatment efficiency and data processing capabilities, making it a crucial component of the current smart healthcare system.

[0003] Among them, the mobile ward round robot based on deep learning algorithms and big data analysis models refers to a robot that performs structured analysis and semantic understanding of physiological indicator image data, electronic medical record text information, and real-time data collected by bedside monitoring equipment from ICU patients by constructing functional modules such as image recognition, natural language processing, real-time data parsing, and knowledge reasoning. This ward round robot uses convolutional neural networks to identify abnormal signs in camera data, combines long short-term memory networks to model the patient's historical state trends, completes a comprehensive evaluation of multi-dimensional indicators through multi-source heterogeneous data fusion technology, and generates early warning markers for abnormal situations based on rule-based reasoning algorithms. Simultaneously, the robot integrates a remote control unit and a high-frequency data transmission module, enabling medical staff to remotely access patient data and interact via voice, thereby assisting in remote ward rounds and medical assessments, and supporting the construction of subsequent teaching materials and case classification management.

[0004] In processing multi-source physiological data in intensive care settings, data is often recorded and analyzed separately on a unit-by-unit basis, without establishing temporal overlap and fluctuation correlations between signals. This results in a lack of synergy between data from different sources, making it difficult to identify interactive features in complex pathological states. Behavioral recognition typically relies on single sensor data, ignoring the coupling between physiological signals and movement trajectories, leading to biased judgments and misjudging the patient's true condition. Regarding equipment control logic, the lack of dynamic comparative analysis of the priority and magnitude of changes in data from different channels may cause control interference in scenarios where multiple devices malfunction simultaneously. Task scheduling processes are mostly statically preset, lacking clear functional identification and classification of non-core operations, resulting in wasted execution resources and accumulated waiting time, reducing overall system response efficiency. Operation commands are often executed sequentially according to time, lacking a real-time screening mechanism based on task urgency, making it difficult to adapt to the needs of high-density, high-intensity intensive care scenarios. For example, when a sudden drop in heart rate and bed adjustment trigger commands simultaneously, the existing process cannot prioritize tasks in a timely manner, potentially delaying critical interventions and affecting the patient's survival probability. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a mobile clinic robot based on deep learning algorithms and big data analysis models.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A mobile medical robot based on deep learning algorithms and big data analysis models, the mobile medical robot being equipped with a control system, the system comprising: The physiological data integration module is used to acquire tidal volume, saturation, and heart rate sequences from ventilators, pulse oximeters, and electrocardiogram devices, organize the signals according to the push time sequence, identify and group sequence segments with time overlap, determine the density of data fluctuations within the group, and generate signal synchronous fluctuation segments. The behavior state recognition module is used to call patient posture, hand and chest fluctuation trajectory data based on the synchronous fluctuation segment of the signal, compare the consistency of the direction and time of change of physiological signal and movement trajectory, extract matching segments and classify them into the same group, and generate a linkage action feature set. The equipment conflict control module is used to extract synchronous mutation segments from the infusion pump, infusion monitoring and heart rate signals according to the linkage action feature set, compare the change amplitude of equipment signals, determine the equipment corresponding to the main channel, rearrange the data processing order, and generate a priority control sequence for the patrol node. The task hierarchical execution module is used to retrieve the function response waiting time and function classification based on the priority control sequence of the patrol nodes, identify operations that do not belong to the core monitoring tasks, reorganize them by type, establish the task call order, and generate a patrol execution strategy hierarchy map.

[0007] Optionally, the signal synchronization fluctuation segment includes tidal volume sequence, saturation sequence, and heart rate sequence; the linkage action feature set includes posture trajectory, hand trajectory, and chest rise and fall trajectory; the priority control sequence of the rounds node includes infusion pump signal, infusion monitoring signal, and main channel equipment; and the rounds execution strategy hierarchy map includes function response waiting time, function classification, and task call order.

[0008] Optionally, the physiological data integration module includes: The sequence acquisition submodule is used to acquire the tidal volume sequence, saturation sequence, and heart rate sequence output by the ventilator, pulse oximeter, and electrocardiogram device, classify them according to the push time, call the timestamp field in the classification result, arrange the same type of sequences in chronological order, and generate a multi-source sequence time mapping value set. The overlap identification submodule is used to detect the start and end times of continuous time periods between each group of sequences based on the multi-source sequence time mapping value set, call the start and end time values ​​of adjacent sequence segments, combine and group sequence segments with time intersection relationships, filter the intersection time periods and calculate the duration length value of the intersection interval, and generate a set of intersection sequence time intervals. The fluctuation density judgment submodule is used to determine the frequency of numerical changes per unit time by calling the corresponding interval segments in the tidal volume sequence, saturation sequence, and heart rate sequence based on the cross sequence time interval set, calculating the time density coefficient of the change amplitude, filtering cross segments whose density coefficient exceeds the density judgment threshold, and obtaining the signal synchronous fluctuation segment.

[0009] Optionally, the behavior state recognition module includes: The posture trajectory recognition submodule is used to acquire the patient's posture trajectory, hand trajectory, and chest fluctuation trajectory based on the synchronous fluctuation section of the signal. According to the start and end times of the tidal volume sequence, saturation sequence, and heart rate sequence in the synchronous fluctuation section of the signal, the duration of the trajectory data is filtered. The amplitude of the trajectory change and the frequency of fluctuation within the synchronous section are aggregated. The trajectory information is divided by type and the section is calibrated to generate the average range of trajectory fluctuation. The action consistency comparison submodule is used to call the trajectory change information in the average range of trajectory fluctuations, match the direction of change with the trend of tidal volume, saturation and heart rate signal sequences, compare the degree of fit between the trend of trajectory segment change and the trend of signal sequence change, and filter according to the consistency judgment criteria to obtain the segment where the trajectory and signal direction are consistent. The linkage feature collection submodule is used to call the segments where the trajectory and signal are consistent in direction, identify data segments that meet the consistency of direction and have time alignment features based on the overlap range of the trajectory and signal time periods, classify and integrate them according to the trajectory start and end time range and the overlap ratio, and generate a linkage action feature set.

[0010] Optionally, the device conflict control module includes: The synchronous mutation identification submodule is used to acquire the infusion pump, infusion monitoring and heart rate signals in the linkage action feature set, divide the signal segments according to the time window, call the mean change, peak difference and slope value for comparison, mark the segment where the mutation amplitude exceeds the heart rate fluctuation benchmark value, and generate a mutation interval identification sequence. The main channel positioning submodule is used to call the mutation interval identifier sequence, extract the change amplitude and direction of the device signal, filter the signal with the largest change amplitude and the highest frequency difference coefficient, and obtain the main channel device identifier value. The control sequence determination submodule is used to retrieve the signal timestamp and response interval within the corresponding time period based on the main channel device identification value, adjust the processing order according to priority, encode and arrange the signal sequence, and obtain the priority control sequence of the patrol node.

[0011] Optionally, the specific calculation formula for adjusting the processing order according to priority is as follows: ; in, This represents the signal priority adjustment value corresponding to the i-th main channel device. This represents the signal timestamp of the i-th main channel device in the j-th record. This represents the weighted value of the response interval corresponding to the i-th main channel device in the j-th record. This represents the weighted average of all response intervals for the i-th main channel device within this time period. This represents the load variation value of the i-th main channel device during this time period. This represents the total number of signal records within that time period.

[0012] Optionally, the task hierarchical execution module includes: The task extraction submodule is used to retrieve the function response waiting time and function category based on the priority control sequence of the patrol node, filter the operation items whose function category does not belong to the core monitoring task, judge whether the waiting time exceeds the patrol response benchmark value, obtain the operation items that have timed out and have no core task association, and generate a non-core operation set. The type aggregation submodule is used to call the non-core operation set, classify it according to the task behavior tags under the functional category, merge operation combinations with related tags, count the number of times operations co-occur, identify frequently co-occurring task clusters, obtain the classification segment corresponding to the task tag, and generate the task type integration segment. The sequence construction submodule is used to integrate segments according to the task type, extract task trigger identifiers and execution time order, sort them by priority identifier values ​​between tasks, identify the task with the lowest priority identifier as the starting node, connect tasks in the order of arrangement to form a chain structure, and generate a hierarchical map of the round-trip execution strategy.

[0013] Optionally, the specific calculation formula for the extraction task trigger identifier and execution time sequence is as follows: ; in, Represents the eigenvalues ​​of order stability. Representing the The trigger identifier value for each task. This represents the average of all task trigger flag values. Representing the The execution timestamp value of each task. This represents the earliest execution timestamp value among all tasks. Representing the The priority identifier integer value for each task. Represents the total number of tasks. Representing the The execution time sequence number of each task. This represents the average of the sequential numbers of all task execution times.

[0014] Optionally, the system further includes: The control command scheduling module is used to retrieve control commands for bed adjustment, voice prompts, and moving mechanisms according to the hierarchical map of the round-trip execution strategy, read the command start time and execution order, compare the urgency of operation commands within the same time period, select priority commands to enter the execution process, and generate a command collaborative distribution control table. The instruction coordination and distribution control table includes bed adjustment instructions, voice prompt instructions, and movement mechanism instructions.

[0015] Optionally, the control command scheduling module includes: The instruction information parsing submodule is used to retrieve control instructions for bed adjustment, voice prompts, and moving mechanisms based on the hierarchical map of the round-trip execution strategy, read the start time and operation sequence of the instructions, extract instructions within the same time period and divide them into time interval groups, and generate operation instruction time interval groups. The priority determination submodule is used to extract the start time, instruction type, function classification, and task level according to the operation instruction time interval group, compare the task level with the core identifier value, select the instruction with the highest level value, and generate a priority instruction identifier sequence. The execution sequence filtering submodule is used to rearrange the start time according to the priority instruction identifier sequence, filter the instruction set whose time does not overlap, combine them according to the sequential numbering, and generate an instruction collaborative distribution control table.

[0016] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects: In this invention, physiological signals from multiple devices are integrated in chronological order to identify overlapping and synchronous fluctuation segments between signals, achieving high-precision time alignment. Linkage features are extracted by combining patient movement trajectories with signal fluctuation consistency, improving the accuracy of abnormal state identification. The data processing order of devices is rearranged by comparing the magnitude of sudden changes, strengthening the priority determination of key signals. Non-urgent operations are categorized by response time and function to compress the call cycle, improving the execution efficiency of high-frequency tasks. Operation instructions are filtered and distributed collaboratively according to their urgency, optimizing the execution process and on-site response speed, and enhancing the diagnostic and treatment control capabilities and intervention timeliness in critical care scenarios. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] Please see Figure 1 A mobile clinic robot based on deep learning algorithms and big data analysis models, the mobile clinic robot is equipped with a control system, the system comprising: The physiological data integration module is used to acquire tidal volume, saturation, and heart rate sequences from ventilators, pulse oximeters, and electrocardiogram devices, organize the signals according to the push time sequence, identify and group sequence segments with time overlap, determine the density of data fluctuations within the group, and generate signal synchronous fluctuation segments. The behavior state recognition module is used to call patient posture, hand and chest fluctuation trajectory data based on the synchronous fluctuation segment of the signal, compare the consistency of the direction and time of change of physiological signals and movement trajectory, extract matching segments and classify them into the same group, and generate a set of linkage action features. The equipment conflict control module is used to extract synchronous mutation segments from the infusion pump, infusion monitoring and heart rate signals based on the linkage action feature set, compare the change amplitude of equipment signals, determine the corresponding equipment of the main channel, rearrange the data processing order, and generate the priority control sequence of the patrol node. The task hierarchical execution module is used to retrieve the function response waiting time and function classification based on the priority control sequence of the round of visits node, identify operations that do not belong to the core monitoring tasks, reorganize them by type, establish the task call order, and generate a hierarchical map of the round of visits execution strategy. The control command scheduling module is used to retrieve control commands for bed adjustment, voice prompts, and moving mechanisms based on the hierarchical map of the round-trip execution strategy, read the start time and execution order of the commands, compare the urgency of operation commands within the same time period, select priority commands to enter the execution process, and generate a command collaborative distribution control table. The signal synchronization fluctuation range includes tidal volume sequence, saturation sequence, and heart rate sequence; the linkage action feature set includes posture trajectory, hand trajectory, and chest rise and fall trajectory; the priority control sequence for rounds nodes includes infusion pump signal, infusion monitoring signal, and main channel equipment; the rounds execution strategy hierarchy map includes function response waiting time, function classification, and task call order; and the instruction coordination distribution control table includes bed adjustment instructions, voice prompt instructions, and movement mechanism instructions.

[0021] Please see Figure 2 The physiological data integration module includes: The sequence acquisition submodule acquires tidal volume, saturation, and heart rate sequences output by ventilators, pulse oximeters, and electrocardiogram devices, categorizes them according to the push time, calls the timestamp field in the categorization results, arranges the same type of sequences in chronological order, and generates a multi-source sequence time mapping value set. The sequence acquisition submodule uses ventilators, pulse oximeters, and electrocardiogram (ECG) devices as data sources. In actual operation, it acquires signal sequences such as tidal volume, pulse oximetry, and heart rate, respectively, and sets a uniform sampling frequency, such as once per second. All acquired data is accompanied by the original device timestamps. After uniform formatting, different time representation formats are converted into a unified standard time representation format, such as converting to integer timestamps in seconds. Then, sequences from the same type of device are classified and organized according to the timestamps. Tidal volume values ​​are sorted chronologically to form a tidal volume sequence. Simultaneously, pulse oximetry and heart rate values ​​are... The same operation ensures the time consistency within the data sequence. After processing, the values ​​corresponding to the three devices at the same moment are extracted and grouped according to time to construct a unified time mapping set. In actual operation scenarios, if the tidal volume collection records include four sets of data: 450, 470, 430, and 460, which are four points in four consecutive seconds, and the corresponding saturation and heart rate values ​​are also provided within this time period, then the three types of data are matched accordingly. Each second corresponds to a combination of three data elements, thus forming a continuous time mapping value set in seconds. This set is the basic data structure for subsequent sequence synchronization analysis.

[0022] The overlap identification submodule is based on the multi-source sequence time mapping value set. It detects the start and end times of continuous time periods between each group of sequences, calls the start and end time values ​​of adjacent sequence segments, combines and groups sequence segments with time intersection relationships, filters the intersection time periods and calculates the duration of the intersection interval, and generates a set of intersection sequence time intervals. The overlap identification submodule, within the established multi-source time mapping set, sequentially detects continuous intervals in the time dimension of each physiological data sequence, identifies the start and end times of each data segment, and records these start and end time periods. It then cross-compares the time periods of all data sequences to determine if there are overlapping intervals in the output data from multiple devices within the same time period. During cross-judgment, the sequence with the largest start time in each comparison group is selected as the starting point of the potential overlapping segment, while the sequence with the smallest end time in each group is selected as the ending point. If the starting point is smaller than the ending point, it indicates a time overlap, and this time segment is identified as a single overlapping segment. The crossover segment is calculated by subtracting the start time from the end time to obtain the duration in seconds. If three devices record sequences that last for several seconds, and one segment of tidal volume data appears between the 23rd and 27th seconds, one segment of blood oxygen saturation data appears between the 25th and 29th seconds, and one segment of heart rate data appears between the 24th and 28th seconds, then all three sequences have data recorded between the 25th and 27th seconds, thus forming an overlapping segment. This module extracts the overlapping segment and records its start and end times and duration, completing the generation of the crossover interval set for further analysis in subsequent processing steps.

[0023] The fluctuation density judgment submodule, based on the time interval set of the cross sequence, calls the numerical variation amplitude of the corresponding interval segment in the tidal volume sequence, saturation sequence, and heart rate sequence to judge the frequency of numerical change per unit time, calculates the time density coefficient of the variation amplitude, filters the cross segments whose density coefficient exceeds the density judgment threshold, and obtains the synchronous fluctuation segment of the signal. The fluctuation density determination submodule performs fluctuation analysis on various physiological signal data within the crossover time period identified by the previous module. During the analysis, it calculates the difference between every two adjacent values ​​of tidal volume, saturation, and heart rate within that time period to identify their amplitude of change. It also counts the number of times these values ​​change within the period. Finally, it sums all the amplitude values ​​and divides them by the number of seconds covered by the time period to obtain the average fluctuation value per second. This value is the density coefficient of the signal within that time period. For example, if the tidal volume values ​​at four consecutive time points in a certain crossover period are 450, 480, 440, and 46... The density coefficient is 30, with each change being 30, 40, and 20, averaging 30 after three cumulative changes. If the saturation is recorded as 97, 96, 95, and 97 in the same time period, with each change being 1, 1, and 2, averaging 1.33, and the heart rate is recorded as 75, 77, 78, and 76, with changes being 2, 1, and 2, averaging 1.67, under the premise of setting the density judgment threshold to 1.5, the density coefficients of tidal volume and heart rate both exceed the threshold, while the saturation does not reach the threshold. Therefore, tidal volume and heart rate in this cross segment are identified as having significant synchronous fluctuation characteristics, and this synchronous fluctuation segment is finally output as the synchronous analysis result.

[0024] Please see Figure 2 The behavior state recognition module includes: The posture trajectory recognition submodule acquires the patient's posture trajectory, hand trajectory, and chest fluctuation trajectory based on the synchronous fluctuation section of the signal. According to the start and end times of the tidal volume sequence, saturation sequence, and heart rate sequence in the synchronous fluctuation section of the signal, the duration of the trajectory data is filtered. The amplitude of the trajectory change and the frequency of fluctuation within the synchronous section are aggregated and processed. The trajectory information is divided by type and the section is calibrated to generate the mean range of trajectory fluctuation. The posture trajectory recognition submodule needs to acquire multiple key trajectory information based on the synchronous fluctuation segments of the signal. First, it needs to capture posture trajectory, hand trajectory, and chest rise and fall trajectory through synchronous recording. For example, multi-point infrared cameras and accelerometers can be deployed around the bed to monitor the displacement of the shoulder, wrist, and chest in real time. During the patient's natural breathing, tidal volume, blood oxygen saturation, and heart rate data are collected simultaneously. The acquired signal sequence is used to extract the synchronous segments with significant fluctuations as a reference time range. For example, if tidal volume fluctuations are detected between 3.5 and 9.8 seconds, all trajectory segments within this time period are selected for analysis. The trajectories of chest, hand, and torso displacement are categorized separately, and variation features are extracted based on the direction and amplitude of movement. For example, the most significant forward and backward movement of the chest within this time period... The amplitude was 4.2 cm, with six rhythmic fluctuations. The maximum left-right movement of the hand trajectory was 6.1 cm, with five regular movements. Combining this information, the average amplitude of the trajectory changes was calculated to characterize the trend of body part changes within that time period. Different trajectory types were classified, such as chest as respiratory fluctuations, hands as active movements, and body tilting as posture adjustments. Next, the corresponding time intervals were marked. For example, the hand trajectory fluctuations were mainly concentrated between 4.2 seconds and 6.8 seconds, so this segment could be characterized as an active movement trajectory. By summarizing the range of trajectory fluctuations and time distribution, the average fluctuation range of the chest trajectory was finally determined to be 3.9 cm, corresponding to a time range of 3.5 to 9.8 seconds, and the average fluctuation of the hand trajectory was 5.8 cm, corresponding to a time range of 4.2 to 6.8 seconds.

[0025] The action consistency comparison submodule calls the trajectory change information in the average interval of trajectory fluctuation and performs directional matching with the changing trends of tidal volume, saturation and heart rate signal sequences. Based on the fit between the trend of trajectory segment changes and the trend of signal sequence changes, it compares the degree of conformity of directional changes and filters according to the consistency judgment criteria to obtain the segments where the trajectory and signal directions are consistent. The motion consistency comparison submodule needs to extract the direction of motion trajectory change within the mean range of trajectory fluctuations and compare its trend with the direction of change of synchronous signals such as tidal volume, blood oxygen saturation, and heart rate. First, the trajectory change direction needs to be serialized; for example, the forward and backward movement trajectory along the chest Z-axis can be organized into a sequence, and the fluctuation of tidal volume over time can be constructed into a sequence of the same dimension. These sequences are then compared in terms of trend direction to determine whether their change directions are consistent within the same time period. When comparing trends, certain matching criteria can be set, such as a correlation greater than 85% for trend consistency, using a sliding window. The method involves gradually comparing the changes between each trajectory sequence and the signal sequence to see if they are synchronous. If it is found that the chest moves outward during a certain period of time, and the tidal volume also shows an upward trend during that period, then the trajectory and the signal direction are considered to be consistent. For example, in a certain analysis window, the chest trajectory shows a continuous outward movement, with data changes of -2, -1, zero, +1, and +2, which is relatively consistent with the tidal volume rising from 200 ml to 300 ml and then falling to 260 ml. By calculating the degree of consistency of the changes and finding that it is higher than a set threshold, it is determined that the trajectory segment is consistent with the tidal volume direction. These segments that meet the conditions are retained as a basis for further classification.

[0026] The linkage feature collection submodule calls the segments where the trajectory and signal direction are consistent. Based on the overlap range of the trajectory and signal time periods, it identifies data segments that meet the direction consistency and have time alignment features. It then classifies and integrates these segments according to the trajectory start and end time range and the overlap ratio to generate a linkage action feature set. The linkage feature collection submodule needs to further analyze the overlap of the segments with the same trajectory and signal direction identified in the previous stage on the time axis, comparing the trajectory and signal according to their time ranges. For example, if a trajectory segment starts at 4 seconds and ends at 9 seconds, while a signal change segment starts at 5.5 seconds and ends at 8 seconds, they intersect on the time axis, with the intersection lasting from 5.5 seconds to 8 seconds, corresponding to an overlap duration of 2.5 seconds. The total trajectory duration is 5 seconds, so the overlap ratio is calculated to be 50%, based on the set classification criteria, such as... If the overlap ratio reaches 40% or more, it can be identified as a high overlap segment. Trajectories and signal pairs with high overlap characteristics are grouped into the same feature set. After classification and organization, the trajectory start and end time range, amplitude change value, and signal change trend are integrated to form a unified data record. For example, the left hand trajectory shows a rightward swinging trend. Within the time period of 4.2 seconds to 6.3 seconds, the horizontal displacement amplitude is 5.8 cm. At the same time, the heart rate increases from 80 beats per minute to 92 beats per minute, and the blood oxygen saturation decreases from 96% to 94%. The above data is merged into the linkage feature set and used for subsequent processing.

[0027] Please see Figure 2 The equipment conflict control module includes: The synchronous mutation identification submodule acquires the infusion pump, infusion monitoring and heart rate signals from the linkage action feature set, divides the signal segments according to the time window, calls the mean change, peak difference and slope value for comparison, marks the segment where the mutation amplitude exceeds the heart rate fluctuation benchmark value, and generates a mutation interval identification sequence. The synchronous mutation identification submodule receives signals from the infusion pump, infusion monitoring, and heart rate. The sampling frequencies are once per second for the drug flow rate, once per minute for the drip rate and volume, and once per second for the heart rate monitoring waveform. To ensure time consistency, all signals are aligned to the same time base and divided into 10-second segments. Within each time segment, the average signal value is calculated, comparing it to the previous segment for significant differences. For example, if the average heart rate jumps from 76 to 84, the change is 8, exceeding the set heart rate fluctuation baseline and is marked as a mutation. Simultaneously, the difference between the maximum and minimum peak values ​​within the time segment is extracted. If this difference exceeds 10, it is considered a large fluctuation. Furthermore, linear analysis is performed on the signal curve trend within the time segment to determine the rate of increase or decrease. If the change exceeds 1.5 per second, a mutation is indicated in that signal segment. Combining these three indicators, if any two conditions meet the mutation judgment criteria, the time segment is marked as a mutation segment, and a marker sequence (0 or 1) is generated as a mutation reference.

[0028] The main channel positioning submodule calls the mutation interval identifier sequence to extract the change amplitude and direction of the device signal, filters the signal with the largest change amplitude and the highest frequency difference coefficient, and obtains the main channel device identifier value. The main channel positioning submodule extracts the infusion pump, infusion monitoring, and heart rate signals from each mutation zone based on the mutation identifier sequence. It then obtains the average difference, peak fluctuation, and direction of change for each signal type within these mutation zones. The average difference reflects whether there is a systematic change in the signal compared to the previous segment; peak fluctuation identifies extreme amplitude changes in the data waveform; and the direction of change is determined by judging the slope sign to indicate whether the signal is rising or falling. Within the mutation zone, the amplitude of change for each signal type is measured individually. The average and peak changes are combined to obtain the overall change level, and the stability of the changes is compared between different mutation zones. By calculating the consistency level of the amplitude changes of each signal in different mutation zones, i.e., judging whether the changes are balanced, if a signal shows drastic changes and large differences in all mutation zones, then that signal can be considered the main channel signal. For example, during a monitoring process, the infusion pump signal shows a higher fluctuation frequency and intensity, and the amplitude differs significantly across segments. Based on this characteristic, the infusion pump is marked as the main channel device and assigned an identifier value for subsequent module identification and processing.

[0029] The control sequence determination submodule retrieves the signal timestamps and response intervals within the corresponding time period based on the main channel device identification value, adjusts the processing order according to priority, encodes and arranges the signal sequence, and obtains the priority control sequence for the patrol node. The specific calculation formula for adjusting the processing order according to priority is as follows: ; in, This represents the signal priority adjustment value corresponding to the i-th main channel device. This represents the signal timestamp of the i-th main channel device in the j-th record. This represents the weighted value of the response interval corresponding to the i-th main channel device in the j-th record. This represents the weighted average of all response intervals for the i-th main channel device within this time period. This represents the load variation value of the i-th main channel device during this time period. This represents the total number of signal records within that time period; Parameter acquisition and value setting: According to actual monitoring, the value range is [0.5, 2.0] ms.

[0030] Statistical analysis based on device response time was obtained through the following quantification process: The maximum response time was set to 2.0ms, and the minimum to 0.5ms. The actual response time was mapped to a weighted range of [0.1, 1.0]. A smaller weight value indicates a faster response. Based on actual monitoring, the value range was [0.1, 1.0].

[0031] The value is calculated based on the load change rate of the equipment in real time. According to actual monitoring, the value range is [5, 20]%.

[0032] By calculating all The arithmetic mean is obtained.

[0033] Set to 5 times.

[0034] Specific numerical settings: The parameters for the first main channel device in the 5 records are set as follows: ; Formula calculation process: Calculate the numerator:

[0035] ; ; Calculate the denominator: ; Calculate the first term: ; Calculate the second term: ; Calculate the final result: ; Results Explanation: The result indicates that the signal priority adjustment value for the first main channel device is 1.4533. This value is used to encode and arrange the signal sequence to obtain the priority control sequence for the patrol nodes.

[0036] Please see Figure 2 The task hierarchical execution module includes: The task extraction submodule retrieves the function response waiting time and function category based on the priority control sequence of the round of visits node. It filters operation items whose function category does not belong to the core monitoring task, judges whether the waiting time exceeds the benchmark value of the round of visits response, obtains operation items that have timed out and have no core task association, and generates a set of non-core operations. The task extraction submodule, based on the priority control sequence of the rounds nodes, first needs to establish an urgency score for each rounds node. This score is calculated by weighting information such as the number of historical alarms, alarm severity level, and the degree of dependence on medical orders. Each information item is assigned a value, and the scores are accumulated proportionally to form the priority score. For example, if a node has 30 alarms, an alarm level set to medium, and depends on 2 medical orders, it can be assigned 6, 3, and 2 points respectively, ultimately totaling 11 points. Based on this score, all rounds nodes are sorted to generate a priority queue. Subsequently, the response waiting time and functional classification information corresponding to each operation item are retrieved. The response waiting time can be calculated by extracting the difference between the operation request time and the first system response time. For example, if the request time is 14:00 and the response time is 14:05, the waiting time is 5 minutes. Functional classification information can be extracted from the system task definition. Each task has a preset category identifier. The system can set a set of core task types, such as monitoring vital signs and handling emergency calls. Any category identifier not included in the core task set is considered a non-core task. Then, the response waiting time of each of these non-core task operations is checked to see if it exceeds the patrol response benchmark value. This benchmark value can be set based on the median of past system response data or a specified standard, for example, 3 minutes. If the waiting time of a task exceeds this value, it is recorded as a response timeout operation. By screening all non-core tasks one by one, a set of operations containing all response timeouts that do not belong to core tasks is formed. This set can be used for subsequent analysis and processing. In a real scenario, if a nursing system still has a response delay of more than 5 minutes for the curtain control task during off-peak hours, this task will be included in this set for the system to further analyze task efficiency or resource allocation.

[0037] The type aggregation submodule calls the set of non-core operations, classifies them according to the task behavior tags under the function category, merges operation combinations with related tags, counts the number of times operations co-occur, identifies frequently co-occurring task clusters, obtains the classification segment corresponding to the task tag, and generates the task type integration segment. The type aggregation submodule loads the non-core operation set and processes each operation task according to its functional category. Each category has a corresponding behavior label; for example, lighting control is categorized as visual adjustment, radio volume adjustment as auditory adjustment, and curtain operation as occlusion tasks. After each task is labeled with a behavior label, they are aggregated into a task combination under the same label. The co-occurrence frequency of these combinations is then analyzed in the task records of each round of visits. If two tasks frequently appear together in a group of visits, they can be identified as frequently co-occurring combinations. For example, if lighting control and curtain opening / closing have co-occurred more than 30 times in the past 100 rounds, a significant co-occurrence relationship can be determined. Further statistical analysis is then performed on all... The co-occurrence frequency of task combinations is determined by setting a baseline threshold, which can be based on the average of all co-occurrences. For example, if each combination co-occurs an average of 12 times, then combinations with more than 20 co-occurrences are considered frequent itemsets. The behavioral labels corresponding to these frequent combinations are further mapped to the corresponding functional classification segments, such as mapping visual adjustment to the visual segment and auditory adjustment to the audio segment, forming a classification set of task type integration segments. This set provides a clustering basis for the subsequent construction of task sequence structures. In practical applications, if curtain and lighting control tasks are found to co-occur highly, they can be uniformly classified into the visual task segment and given priority in the system scheduling for combined operation plans.

[0038] The sequence construction submodule integrates segments according to task type, extracts task trigger identifiers and execution time order, sorts them by priority identifier values ​​between tasks, identifies the task with the lowest priority identifier as the starting node, connects tasks according to the arrangement order to form a chain structure, and generates a hierarchical map of the round-trip execution strategy. The specific formula for calculating the task trigger identifier and execution time sequence is as follows: ; in, Represents the eigenvalues ​​of order stability. Representing the The trigger identifier value for each task. This represents the average of all task trigger flag values. Representing the The execution timestamp value of each task. This represents the earliest execution timestamp value among all tasks. Representing the The priority identifier integer value for each task. Represents the total number of tasks. Representing the The execution time sequence number of each task. This represents the average of the sequential execution times of all tasks; Detailed explanation of the formula and its calculation derivation: Methods for obtaining and quantizing each parameter: It is obtained through the quantization processing of state signals in an event-triggered control system.

[0039] By calculating all The arithmetic mean is obtained.

[0040] Obtained through the actual execution time recorded by the task scheduling system.

[0041] By comparing all The value determines the minimum value.

[0042] Obtained through the task priority settings in the task management system.

[0043] This is obtained by counting the number of tasks in the current task list.

[0044] Obtained by the execution sequence number of the task in the task scheduling system.

[0045] By calculating all The arithmetic mean is obtained.

[0046] Specific numerical settings and calculation process: Suppose there are 5 tasks with the following parameters: Tasks numbered 1 to 5 The values ​​are 12, 15, 14, 13, and 16 respectively; The values ​​are 10 seconds, 20 seconds, 15 seconds, 25 seconds, and 30 seconds, respectively. The values ​​are 2, 1, 3, 2, 1 respectively; The values ​​are 1, 2, 3, 4, and 5 respectively.

[0047] Calculate intermediate parameters: ; ; ; Calculate the numerator: ; = ; ≈ ; ≈ ; ≈ ; Calculate the denominator: ; = ; = ; = ; = ; = ; Final calculation: ; The results show that the sequential stability eigenvalue is 3.4042. The higher the value, the worse the stability of the task execution order. Further optimization of task scheduling is needed to improve system efficiency.

[0048] Please see Figure 2 The control command scheduling module includes: The instruction information parsing submodule retrieves control instructions for bed adjustment, voice prompts, and moving mechanisms based on the hierarchical map of the round-trip execution strategy. It reads the start time and operation sequence of the instructions, extracts instructions within the same time period and divides them into time interval groups, and generates operation instruction time interval groups. The instruction information parsing submodule uses the ward round strategy hierarchy map as its core, retrieving control instructions such as bed adjustment, voice prompts, and movement mechanisms. During parsing, it first loads the pre-defined strategy nodes from the map. Each node includes execution conditions, controlled objects, mode types, and operation delay times. For example, the bed adjustment node might be set to "Upon reaching the bed area, if the current bed height is below 60 cm and the angle is zero, execute a raising operation, with a target height of 70 cm and a 2-second delay." Voice prompts might specify "Play a greeting voice message '001' at a distance of 1 meter from the bed, with a 1-second delay." Movement mechanism instructions might be set to "Forward," with a target distance of 1.2 meters, a speed of 0.4 meters per second, and a 3-second delay. After strategy parsing, the current system status is read by the device status acquisition module, including equipment... Parameters such as position, attitude, historical execution records, and current instruction cache are used. All extracted control instructions are uniformly numbered according to the reading time and written into the buffer. Then, they are sorted by the start time field. Each instruction carries an execution time tag, such as voice prompts as 12:01:01, bed adjustment as 12:01:01, and movement operation as 12:01:05. After sorting, instructions with adjacent times and related execution logic are merged and classified using a time interval clustering strategy. For example, using the standard of no more than 2 seconds per group, instructions between 12:01:01 and 12:01:03 are merged into the first group, and operations after 12:01:04 form the second group. Through this logic, the operation instructions are decomposed into a series of time interval sets that can be processed in parallel. Synchronous control can be achieved within each group, ultimately forming multiple operation instruction time interval groups as the basis for subsequent priority processing.

[0049] The priority determination submodule extracts the start time, instruction type, function classification, and task level based on the operation instruction time interval group, compares the task level with the core identifier value, selects the instruction with the highest level value, and generates a priority instruction identifier sequence. The priority determination submodule extracts parameters such as start time, category, functional module type, and task level for each instruction in the operation instruction time interval group. The task level is derived from the map setting and is usually represented by integers from 1 to 5, with higher values ​​indicating higher levels. For example, voice prompts are level 1, bed adjustment is level 3, and movement control is level 2. In addition, the function classification value is a quantitative setting parameter, usually set according to the function complexity. For example, voice prompts are 0.3, bed adjustment is 0.7, and movement mechanism is 0.5. The setting level weight is 0.6, and the function classification weight is 0.4. Each instruction is calculated by multiplying the task level by the level weight and adding the function classification by its corresponding weight to obtain the final priority value. For example, if the bed adjustment level is 3 and the function value is 0.7, the priority value is calculated as 3 multiplied by 0.6 plus 0.7 multiplied by 0.4, which gives 2.66. The instruction with the largest value in the same group is determined as the priority instruction, and its identifier is written into the identifier sequence. This process is repeated for each interval group, ultimately resulting in multiple priority instruction identifiers.

[0050] The execution sequence filtering submodule rearranges the start time according to the priority instruction identifier sequence, filters the instruction set whose time does not overlap, combines them according to the sequential numbering, and generates an instruction collaborative distribution control table; The execution sequence filtering submodule sorts and compares all instruction times based on the start time of the priority instructions in the above-mentioned identifier sequence. First, it sets the start and end times for each instruction. For example, instruction A starts at 12:01:01, lasts for 2 seconds, and ends at 12:01:03. Instruction B starts at 12:01:02, lasts for 3 seconds, and ends at 12:01:05. Because the time periods overlap, the system will remove items that overlap with instructions with higher priority values ​​in terms of time, and only retain the set of instructions whose times do not overlap. Assuming that the priority value of instruction A is higher than that of instruction B, instruction A is retained and B is removed. Then, the next instruction that does not overlap with the time of A is selected to form a set. This process is repeated until the filtering is completed. The final filtered sets are numbered from high to low priority values. For example, combination 1 contains instructions A and C, and combination 2 contains instructions D and E. Each combination generates a distribution control command to form a collaborative control table for coordinating the execution of each module.

[0051] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A mobile medical robot based on deep learning algorithms and big data analysis models, characterized in that, The mobile clinic robot is equipped with a control system, which includes: The physiological data integration module is used to acquire tidal volume, saturation, and heart rate sequences from ventilators, pulse oximeters, and electrocardiogram devices, organize the signals according to the push time sequence, identify and group sequence segments with time overlap, determine the density of data fluctuations within the group, and generate signal synchronous fluctuation segments. The behavior state recognition module is used to call patient posture, hand and chest fluctuation trajectory data based on the synchronous fluctuation segment of the signal, compare the consistency of the direction and time of change of physiological signal and movement trajectory, extract matching segments and classify them into the same group, and generate a linkage action feature set. The equipment conflict control module is used to extract synchronous mutation segments from the infusion pump, infusion monitoring and heart rate signals according to the linkage action feature set, compare the change amplitude of equipment signals, determine the equipment corresponding to the main channel, rearrange the data processing order, and generate a priority control sequence for the patrol node. The task hierarchical execution module is used to retrieve the function response waiting time and function classification based on the priority control sequence of the patrol nodes, identify operations that do not belong to the core monitoring tasks, reorganize them by type, establish the task call order, and generate a patrol execution strategy hierarchy map.

2. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 1, characterized in that, The signal synchronization fluctuation segment includes tidal volume sequence, saturation sequence, and heart rate sequence; the linkage action feature set includes posture trajectory, hand trajectory, and chest rise and fall trajectory; the priority control sequence of the rounds node includes infusion pump signal, infusion monitoring signal, and main channel equipment; and the rounds execution strategy hierarchy map includes function response waiting time, function classification, and task call order.

3. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 1, characterized in that, The physiological data integration module includes: The sequence acquisition submodule is used to acquire the tidal volume sequence, saturation sequence, and heart rate sequence output by the ventilator, pulse oximeter, and electrocardiogram device, classify them according to the push time, call the timestamp field in the classification result, arrange the same type of sequences in chronological order, and generate a multi-source sequence time mapping value set. The overlap identification submodule is used to detect the start and end times of continuous time periods between each group of sequences based on the multi-source sequence time mapping value set, call the start and end time values ​​of adjacent sequence segments, combine and group sequence segments with time intersection relationships, filter the intersection time periods and calculate the duration length value of the intersection interval, and generate a set of intersection sequence time intervals. The fluctuation density judgment submodule is used to determine the frequency of numerical changes per unit time by calling the corresponding interval segments in the tidal volume sequence, saturation sequence, and heart rate sequence based on the cross sequence time interval set, calculating the time density coefficient of the change amplitude, filtering cross segments whose density coefficient exceeds the density judgment threshold, and obtaining the signal synchronous fluctuation segment.

4. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 1, characterized in that, The behavior state recognition module includes: The posture trajectory recognition submodule is used to acquire the patient's posture trajectory, hand trajectory, and chest fluctuation trajectory based on the synchronous fluctuation section of the signal. According to the start and end times of the tidal volume sequence, saturation sequence, and heart rate sequence in the synchronous fluctuation section of the signal, the duration of the trajectory data is filtered. The amplitude of the trajectory change and the frequency of fluctuation within the synchronous section are aggregated. The trajectory information is divided by type and the section is calibrated to generate the average range of trajectory fluctuation. The action consistency comparison submodule is used to call the trajectory change information in the average range of trajectory fluctuations, match the direction of change with the trend of tidal volume, saturation and heart rate signal sequences, compare the degree of fit between the trend of trajectory segment change and the trend of signal sequence change, and filter according to the consistency judgment criteria to obtain the segment where the trajectory and signal direction are consistent. The linkage feature collection submodule is used to call the segments where the trajectory and signal are consistent in direction, identify data segments that meet the consistency of direction and have time alignment features based on the overlap range of the trajectory and signal time periods, classify and integrate them according to the trajectory start and end time range and the overlap ratio, and generate a linkage action feature set.

5. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 1, characterized in that, The equipment conflict control module includes: The synchronous mutation identification submodule is used to acquire the infusion pump, infusion monitoring and heart rate signals in the linkage action feature set, divide the signal segments according to the time window, call the mean change, peak difference and slope value for comparison, mark the segment where the mutation amplitude exceeds the heart rate fluctuation benchmark value, and generate a mutation interval identification sequence. The main channel positioning submodule is used to call the mutation interval identifier sequence, extract the change amplitude and direction of the device signal, filter the signal with the largest change amplitude and the highest frequency difference coefficient, and obtain the main channel device identifier value. The control sequence determination submodule is used to retrieve the signal timestamp and response interval within the corresponding time period based on the main channel device identification value, adjust the processing order according to priority, encode and arrange the signal sequence, and obtain the priority control sequence of the patrol node.

6. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 5, characterized in that, The specific calculation formula for adjusting the processing order according to priority is as follows: ； in, This represents the signal priority adjustment value corresponding to the i-th main channel device. This represents the signal timestamp of the i-th main channel device in the j-th record. This represents the weighted value of the response interval corresponding to the i-th main channel device in the j-th record. This represents the weighted average of all response intervals for the i-th main channel device within this time period. This represents the load variation value of the i-th main channel device during this time period. This represents the total number of signal records within that time period.

7. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 5, characterized in that, The task hierarchical execution module includes: The task extraction submodule is used to retrieve the function response waiting time and function category based on the priority control sequence of the patrol node, filter the operation items whose function category does not belong to the core monitoring task, judge whether the waiting time exceeds the patrol response benchmark value, obtain the operation items that have timed out and have no core task association, and generate a non-core operation set. The type aggregation submodule is used to call the non-core operation set, classify it according to the task behavior tags under the functional category, merge operation combinations with related tags, count the number of times operations co-occur, identify frequently co-occurring task clusters, obtain the classification segment corresponding to the task tag, and generate the task type integration segment. The sequence construction submodule is used to integrate segments according to the task type, extract task trigger identifiers and execution time order, sort them by priority identifier values ​​between tasks, identify the task with the lowest priority identifier as the starting node, connect tasks in the order of arrangement to form a chain structure, and generate a hierarchical map of the round-trip execution strategy.

8. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 7, characterized in that, The specific formula for calculating the extraction task trigger identifier and execution time sequence is as follows: ； in, Represents the eigenvalues ​​of order stability. Representing the The trigger identifier value for each task. This represents the average of all task trigger flag values. Representing the The execution timestamp value of each task. This represents the earliest execution timestamp value among all tasks. Representing the The priority identifier integer value for each task. Represents the total number of tasks. Representing the The execution time sequence number of each task. This represents the average of the sequential numbers of all task execution times.

9. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 1, characterized in that, The system also includes: The control command scheduling module is used to retrieve control commands for bed adjustment, voice prompts, and moving mechanisms according to the hierarchical map of the round-trip execution strategy, read the command start time and execution order, compare the urgency of operation commands within the same time period, select priority commands to enter the execution process, and generate a command collaborative distribution control table. The instruction coordination and distribution control table includes bed adjustment instructions, voice prompt instructions, and movement mechanism instructions.

10. The mobile clinic robot based on deep learning algorithms and big data analysis models according to claim 9, characterized in that, The control command scheduling module includes: The instruction information parsing submodule is used to retrieve control instructions for bed adjustment, voice prompts, and moving mechanisms based on the hierarchical map of the round-trip execution strategy, read the start time and operation sequence of the instructions, extract instructions within the same time period and divide them into time interval groups, and generate operation instruction time interval groups. The priority determination submodule is used to extract the start time, instruction type, function classification, and task level according to the operation instruction time interval group, compare the task level with the core identifier value, select the instruction with the highest level value, and generate a priority instruction identifier sequence. The execution sequence filtering submodule is used to rearrange the start time according to the priority instruction identifier sequence, filter the instruction set whose time does not overlap, combine them according to the sequential numbering, and generate an instruction collaborative distribution control table.

Images