A remote, bidirectional, controllable, intelligent adaptive peritoneal dialysis system

Abstract

This invention discloses a remote bidirectional controllable intelligent adaptive peritoneal dialysis system, specifically relating to the field of intelligent adaptive peritoneal dialysis technology. The system includes acquiring patient monitoring data, peritoneal dialysis execution data, and medical constraint data for the current peritoneal dialysis cycle. On the patient-side edge computing side, it performs time alignment and status encoding on vital sign sequences, positional change sequences, chief complaint sequences, perfusion time, drainage time, dwell time, and alarm time, outputting a status sequence. On the patient-side edge computing side, it constructs the current cycle status sequence, generates candidate correction paths based on historical status and real-time correction results, and solves for the current cycle execution parameters. Simultaneously, it forms a cross-cycle boundary between the current cycle results and historical cycle results, which only applies to subsequent cycles. Finally, it updates the execution basis for subsequent cycles by combining the boundary correction rules remotely returned from the medical end.

Description

Technical Field

[0001] This invention relates to the field of intelligent adaptive peritoneal dialysis technology, and more specifically, to a remote bidirectional control intelligent adaptive peritoneal dialysis system. Background Technology

[0002] In existing remote peritoneal dialysis control technologies, the mainstream approach in the industry mainly addresses how to promptly detect abnormalities and adjust dialysis execution parameters based on real-time monitoring results during patients' home treatment. Typically, the patient's end collects drainage status, vital signs, equipment operating status, and chief complaint information, uploads the relevant data to the remote management side or edge control side, and then performs immediate corrections to the perfusion rate, drainage rhythm, residence time, or pause alarm strategy for the current round based on preset thresholds, single-round abnormal judgment results, or short-term trend analysis results. Alternatively, multiple rounds of abnormal records can be directly used as the basis for subsequent plan adjustments. For example, in the scenario of home-based automated peritoneal dialysis at night, the same patient needs to complete multiple rounds of perfusion and drainage without the care of others. At the same time, the patient is subject to the dynamic changes in individual peritoneal transport characteristics, frequent changes in body position, the impact of recent infection or dietary fluctuations on the tolerance boundary, network transmission delays, and the hard constraints that medical staff cannot intervene in real time round by round. Under this constraint, the mainstream approach will consistently reveal an observable and verifiable defect: on the one hand, the system tends to repeatedly perform trial-and-error corrections on the parameters of the current round based solely on the anomalies of the current round, resulting in the recurrence of similar anomalies in subsequent rounds; on the other hand, it tends to directly incorporate short-term fluctuations or occasional anomalies into the subsequent long-term control criteria, causing the patient's marginal execution boundary to fluctuate frequently in consecutive rounds of treatment. This manifests as inconsistent control results at different nights under the same conditions, and the fact that local anomalies have been resolved but subsequent plans continue to deviate. The reason for this is that the existing methods do not separate the immediate correction process that ensures the safety of the current round from the long-term adaptation process that forms individualized boundaries for multiple rounds of treatment. The technical problem this application aims to solve is: how to separate single-cycle correction and cross-cycle adaptation and establish a restricted transmission relationship during remote bidirectional control intelligent adaptive peritoneal dialysis while ensuring the safe execution of the current cycle, so as to avoid short-term abnormal repeated trial and error or contamination of the long-term control boundary. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a remote bidirectional controllable intelligent adaptive peritoneal dialysis system. This system constructs the current cycle state sequence on the patient-side edge computing side, generates candidate correction paths based on historical states and real-time correction results, and solves the execution parameters for the current cycle. Simultaneously, it forms a cross-cycle boundary between the current cycle results and historical cycle results that only applies to subsequent cycles. Then, it updates the execution basis for subsequent cycles by combining the boundary correction rules remotely returned from the medical staff end, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a remote bidirectional controllable intelligent adaptive peritoneal dialysis system, comprising a state construction module, a map generation module, a path generation module, a solution module, a boundary formation module, and an update module: The state construction module is used to acquire patient monitoring data, peritoneal dialysis execution data, and medical and nursing constraint data in the current peritoneal dialysis cycle. On the patient-side edge computing side, it performs time alignment and state encoding on the vital signs sequence, body position change sequence, chief complaint event sequence, as well as perfusion time, drainage time, dwell time, and alarm time, and outputs the state sequence. The graph generation module is used to read historical states and corresponding real-time correction results for the current state in the state sequence, calculate the state transition cost, tolerance offset and execution risk of each historical state to the current state, and perform constraint labeling in combination with medical and nursing constraint data to output a constraint decision graph. The path generation module is used to generate immediate correction paths for preceding nodes in the constraint decision graph, generate joint correction paths for multiple preceding nodes that meet the adjacent transition conditions, and determine paths that do not contain any preceding nodes as empty paths, and output a set of candidate correction paths. The solution module is used to solve the constraint strategy for each candidate correction path, calculate the risk margin, tolerance recovery amount and parameter change amount after the path is executed, and determine the candidate correction path with the first risk margin among the candidate correction paths that meet the medical and nursing constraints and whose parameter change amount is within the boundary of the current peritoneal dialysis cycle as the correction path for the current cycle, and output the execution parameters for the current cycle.

[0005] In a preferred embodiment, it further includes: The boundary formation module is used to read the results of the current peritoneal dialysis cycle after executing the parameters of the current cycle, write the cycle results and the results of the historical peritoneal dialysis cycles into the cross-cycle sequence in chronological order, calculate the repetition location of similar abnormalities, the recovery result after local correction, and the boundary offset direction of consecutive cycles, and output the cross-cycle boundary. The update module is used to limit the correction path of the current cycle to only apply to the current peritoneal dialysis cycle, and to limit the cross-cycle boundary to only provide parameter constraints for the generation process of candidate correction paths for subsequent peritoneal dialysis cycles. Then, the execution parameters of the current cycle, the cycle results, and the cross-cycle boundary are sent to the medical staff through a remote two-way communication link. The module receives the boundary correction rules returned by the medical staff, updates the cross-cycle boundary according to the boundary correction rules, and outputs the execution basis for the next peritoneal dialysis cycle.

[0006] In a preferred embodiment, the state construction module includes: Read the start and end times of perfusion, drainage, stay, and alarm times in the current peritoneal dialysis cycle, connect them in chronological order to form the cycle time sequence main chain, and divide the cycle time sequence main chain into perfusion segment, stay segment, drainage segment, and alarm segment, and output the segment benchmark sequence; Map the time of each monitoring point in the vital signs sequence, body position change sequence, and chief complaint event sequence to the corresponding segment in the segment benchmark sequence, calculate the intra-segment position value of each monitoring point relative to the starting point of its segment, the change direction value between adjacent monitoring points, and the adjacency value of the alarm time, and output the monitoring alignment sequence. The status field combination encoding is performed on each monitoring point in the monitoring alignment sequence. The segment identifier, the position value within the segment, the change direction value, the adjacent value of the alarm time, and the peritoneal dialysis execution status at the corresponding time are written into the same status unit, and the status units are arranged in chronological order to output the status sequence.

[0007] In a preferred embodiment, the map generation module includes: Read the current state, historical state, real-time correction results corresponding to each historical state, and medical constraint data from the state sequence. Use the vital signs field, body position field, chief complaint field, execution level field, and alarm field in the current state as matching fields. Perform constraint matching on each historical state and the current state, calculate the field difference group and the level consistency result, and write the historical states that meet the level consistency condition into the candidate preorder set. Output the candidate preorder set and its corresponding field difference group. For each historical state in the candidate preorder set, a transition cost function is constructed based on the field difference group and the corresponding real-time correction result. Under the parameter boundary, alarm prohibition condition and execution stage constraint of the medical and nursing constraint data, the execution constraint of the transition cost function is solved to obtain the state transition cost and feasible transition mark of each historical state to the current state, and the feasible transition edge set is output. For each feasible transfer edge in the feasible transfer edge set, calculate the tolerance offset of the current state relative to the corresponding historical state according to the changes in vital signs, chief complaints and alarms before and after the corresponding historical state. Calculate the execution risk based on the parameter boundary occupancy, alarm trigger proximity and execution segment conflict corresponding to the feasible transfer edge. Output the boundary value set, which includes the state transfer cost, tolerance offset and execution risk.

[0008] In a preferred embodiment, the map generation module further includes: Using the current state and each historical state in the candidate preorder set as nodes, and the edge value group as the edge value, perform multi-view consistency verification on the state transition cost, tolerance offset and execution risk in the edge value group. When any edge value deviates from the association direction of the other edge values, the corresponding feasible transition edge is identified as a conflict edge. The conflict edge is removed according to the edge value recalculation result or feasible transition mark, and a consistent edge set is output. Constraint marking is performed on each feasible transition edge in the consistent edge set in conjunction with medical and nursing constraint data. Feasible transition edges that meet parameter boundaries, alarm prohibition conditions, and execution segment constraints are retained as constraint edges. Each constraint edge is written between the corresponding nodes according to the state transition cost, tolerance offset, and execution risk amount to generate a constraint decision graph. Writing stops when the number of newly added constraint edges is zero, and the constraint decision graph is output.

[0009] In a preferred embodiment, the path generation module includes: Based on the state transition cost, tolerance offset, execution risk amount and constraint mark corresponding to each constraint edge, construct a single-node correction path from a single preceding node to the current node one by one, and output the set of single-node correction paths. For each single-node correction path in the single-node correction path set, adjacent transfer verification is performed according to the time order between the preceding nodes, the execution segment connection relationship, and the constraint edge compatibility relationship. Multiple preceding nodes that meet the adjacent transfer conditions are combined into a joint correction path according to the connection order. The joint correction path is then merged with the single-node correction path to output a combined correction path set. The path that does not contain any preceding node and only retains the current node is identified as an empty path. The empty paths are then combined with the combined correction path set to form a candidate correction path set.

[0010] In a preferred embodiment, the solving module includes: Each path action in each candidate correction path is expanded into a parameter change sequence according to the execution order, and each parameter change sequence is applied to the current state to perform path deduction, outputting the deduction state sequence corresponding to each candidate correction path; For each simulated state sequence, the constraint occupancy, alarm remaining amount, tolerance fallback amount and cumulative parameter offset are calculated according to the vital signs field, chief complaint field, alarm field and execution stage field, respectively. The path cost group is constructed with the aforementioned amounts, and the risk margin, tolerance recovery amount and parameter change amount corresponding to each candidate correction path are output. Read the historical execution results corresponding to each candidate correction path, calculate the deviation between each path cost group and the historical execution results, perform confidence updates on the risk margin, tolerance recovery amount and parameter change amount of each candidate correction path, and write back the path cost group after confidence update to the corresponding inferred state sequence, and output the confidence-corrected path set.

[0011] In a preferred embodiment, the solving module further includes: The constraint strategy is applied to the set of confidence correction paths. Candidate correction paths that meet the medical and nursing constraints and whose parameter changes are within the boundary of the current peritoneal dialysis cycle are retained as feasible paths. The feasible paths are sorted in the order of risk margin first, tolerance recovery second, and parameter change third, and the feasible path sorting results are output. The adjacent state verification is performed on the first feasible path in the feasible path ranking results. The final state after the first feasible path is applied is compared with the boundary conditions of the subsequent execution segment of the current peritoneal dialysis cycle. If they are consistent, the first feasible path is determined as the correction path of the current cycle and the corresponding parameter change sequence is extracted as the execution parameters of the current cycle. If they are inconsistent, the first feasible path is removed and the execution is returned to re-sort until the correction path of the current cycle is obtained and the execution parameters of the current cycle are output.

[0012] In a preferred embodiment, the boundary forming module includes: Read the results of the current peritoneal dialysis cycle and the results of the historical peritoneal dialysis cycles, write them into the cross-cycle sequence in chronological order, and perform corresponding associations on the abnormality category, execution segment where the abnormality is located, corresponding correction path and cycle result fields in each cycle result, and output the cross-cycle association sequence; For the same type of abnormality in the cross-cycle correlation sequence, the repetition position of each type of abnormality in different peritoneal dialysis cycles is calculated, and the recovery comparison is performed on the changes in vital signs, chief complaints and alarms after correction corresponding to each repetition position. The local correction recovery results corresponding to each type of abnormality are solved, and the abnormality recovery sequence is output. Based on the abnormal recovery sequence, the parameter boundary occupancy change, alarm proximity change and recovery result change corresponding to the same type of abnormality in consecutive peritoneal dialysis cycles are statistically analyzed in chronological order. The boundary offset direction of each type of abnormality in consecutive cycles is calculated, and cross-cycle boundaries are generated by combining the repeated positions and local correction recovery results, and the cross-cycle boundaries are output.

[0013] In a preferred embodiment, the update module includes: Write the current peritoneal dialysis cycle identifier to the current cycle correction path and generate an intra-cycle effective marker; write the subsequent peritoneal dialysis cycle identifier to the cross-cycle boundary and generate a cross-cycle constraint marker; output the intra-cycle execution set and the cross-cycle constraint set. The in-round execution set, round results, and cross-round constraint set are sent to the medical staff through a remote two-way communication link. The boundary correction rules returned by the medical staff are received, and the boundary correction rules are aligned and constraint mapped with the execution fields of each boundary item in the cross-round constraint set, and the boundary correction set is output. Based on the boundary correction set, each boundary item in the cross-cycle constraint set is replaced, retained, or deleted to generate the updated cross-cycle boundary. The updated cross-cycle boundary is then written into the candidate correction path generation process for the next peritoneal dialysis cycle, and the execution basis for the next peritoneal dialysis cycle is output.

[0014] The technical effects and advantages of this invention are as follows: 1. This scheme limits the correction path of the current cycle to only apply to the current peritoneal dialysis cycle and limits the cross-cycle boundary to only provide parameter constraints for subsequent peritoneal dialysis cycles. This separates single-cycle correction from cross-cycle adaptation and establishes a restricted transmission relationship, thereby relatively suppressing short-term abnormal repeated trial and error and its disturbance to the long-term control boundary. 2. At the edge computing side of the patient end, the execution time of patient monitoring data and peritoneal dialysis execution data are aligned and the status is encoded to form a status sequence that can be directly used for subsequent calculations. This makes the current status and historical status have a unified comparison standard, which helps to relatively reduce the impact of asynchronous data and segment mixing on the consistency of control judgment. 3. Construct a constraint decision graph based on the current state, historical state, and immediate correction results, and perform joint calculation and consistency verification on the state transition cost, tolerance offset, and execution risk, thereby relatively improving the targeting of the previous reference state screening and reducing the probability of unsuitable migration relationships entering the correction solution process. 4. Generate candidate correction paths around the constraint decision graph, and perform path deduction, path cost group calculation and confidence update for each path. This ensures that the execution parameters of the current round are determined based on the historical execution results and the current state, thereby relatively improving the convergence of real-time correction and reducing the blindness of single-round parameter adjustment. 5. Write the results of the current round and the results of the historical peritoneal dialysis rounds into the cross-round sequence, and form cross-round boundaries around the repeated locations of similar abnormalities, local correction and recovery results, and boundary offset directions. This will give the parameter constraints of subsequent rounds a continuous round basis and help the individualized execution boundary gradually converge. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the system module structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Refer to the instruction manual appendix Figure 1 The present invention discloses a remote bidirectional controllable intelligent adaptive peritoneal dialysis system, comprising a state construction module, a map generation module, a path generation module, a solution module, a boundary formation module, and an update module. The state construction module is used to acquire patient monitoring data, peritoneal dialysis execution data, and medical and nursing constraint data in the current peritoneal dialysis cycle. On the patient-side edge computing side, it performs time alignment and state encoding on the vital signs sequence, body position change sequence, chief complaint event sequence, as well as perfusion time, drainage time, dwell time, and alarm time, and outputs the state sequence. This implementation method is used to organize patient monitoring data and peritoneal dialysis execution data from different sources and arrival times in the current peritoneal dialysis cycle into a state sequence under a unified time-series caliber. This facilitates subsequent constraint matching, boundary value calculation, and path solving based on the current and historical states. The basic idea is as follows: First, a unified time skeleton for the current cycle is established based on key moments in the peritoneal dialysis execution process. Then, vital signs, positional changes, and chief complaints are mapped to their corresponding execution positions within this time skeleton. Finally, segment information, positional relationships, change relationships, alarm adjacency relationships, and the execution state at the corresponding moment are written into the state unit, forming a state sequence that can be directly read by subsequent steps. This implementation process includes the following steps: First, the start and end times of perfusion, stay, drainage, and alarm times in the current peritoneal dialysis cycle are read. The start and end times of perfusion, stay, and drainage are taken from the execution log field in the peritoneal dialysis execution data, and the alarm times are taken from the alarm field. The above times are sorted from front to back according to the timestamp, and the first and last of the sorted adjacent times are connected to form the cycle time sequence main chain. Then, the time interval corresponding to perfusion is determined according to the start and end times of perfusion, the time interval corresponding to stay is determined according to the start and end times of stay, and the time interval corresponding to drainage is determined according to the start and end times of drainage. At the same time, each alarm time is mapped to its execution interval, and the position corresponding to the alarm time is recorded as the abnormal marker position attached to the execution interval. This results in a sequentially arranged segment baseline sequence, which is written to the buffer for subsequent reading. If the start and end times are found to be missing, the parameter switching records, valve action records, or pump action records in the same cycle are read to fill in the missing times. If the missing times cannot be filled in, the current cycle is recorded as an incomplete segment cycle and subsequent state construction is stopped. After obtaining the grading baseline sequence, the monitoring point times in the vital signs sequence, body position change sequence, and chief complaint event sequence are mapped one by one to the corresponding positions in the grading baseline sequence. The vital signs sequence includes at least the collected fields of blood pressure, heart rate, respiratory rate, and body temperature; the body position change sequence includes at least the positional indicators of supine, left lateral, right lateral, and sitting-up; and the chief complaint event sequence includes at least the event indicators of abdominal distension, abdominal pain, and discomfort. When a monitoring point time falls within a certain execution interval, the monitoring point is assigned to that execution interval. When a monitoring point time is located at the boundary of an adjacent execution interval, it is assigned to the next interval based on the later arrival time; the alarm time remains unchanged. The rules for assigning a monitoring point to its original execution interval determine its unique affiliation. After affixing the monitoring point, the time difference between the monitoring point's time and the start of its execution interval is calculated as the intra-segment position value. The increase, decrease, or maintenance relationship between adjacent monitoring points in the same field is calculated as the change direction value. The preceding adjacency, following adjacency, or non-adjacency relationship between the monitoring point's time and the most recent alarm time is calculated as the alarm time adjacency value. This forms a monitoring alignment sequence and writes it to the buffer for subsequent reading. If multiple monitoring points of the same type appear at the same time, one monitoring point is retained according to the collection source priority given by the preset configuration, and the remaining monitoring points are written to the redundant record area and do not enter the monitoring alignment sequence. Furthermore, the status field combination encoding of each monitoring point in the monitoring alignment sequence is performed. The segment identifier, intra-segment position value, change direction value, alarm time adjacent value, and peritoneal dialysis execution status of the monitoring point at the corresponding time are written into the same status unit. The peritoneal dialysis execution status includes at least the current perfusion status, residence status, drainage status, alarm status, and parameter status. When writing, the field order of segment identifier, intra-segment position value, change direction value, alarm time adjacent value, vital signs field, body position field, chief complaint field, and peritoneal dialysis execution status is uniformly adopted to ensure the consistency of subsequent reading. After completing the writing of a single status unit, all status units are arranged in the order of timestamps. Status units with the same timestamp are arranged according to the order of execution status. Finally, the status sequence is output and written to the input of the subsequent atlas generation step. If a monitoring point lacks any of the vital signs, body position, or chief complaint fields, the most recent and confirmed state unit within the same execution interval is read first for filling in the missing data, and a missing marker is simultaneously written into the current state unit. If no state unit is available for filling in the missing data within the same execution interval, the state unit is recorded as a low-confidence state unit so that its participation priority is reduced during subsequent constraint matching. Taking the nighttime home peritoneal dialysis scenario as an example, when a patient changes from supine to lateral decubitus during drainage, and subsequently complains of abdominal distension and triggers an alarm, the edge computing side first determines that the time belongs to the execution interval corresponding to the drainage based on the start and end times of the drainage. Then, it calculates the intra-segment position value of the body position change time relative to the drainage start point, the change direction value relative to the previous body position record, and the adjacent value of the alarm time relative to the alarm. The above results are written together with the drainage status at that time into a state unit so that subsequent steps can directly identify that the current state is in the process of drainage, the body position has changed, and the location is close to the alarm trigger. Through the above processing, the originally scattered execution time, monitoring time and chief complaint event in the current peritoneal dialysis cycle are unified into a state sequence with clear field composition, clear attribution, and clear arrangement order. Subsequent steps can directly read the current state and historical state based on this, avoiding problems such as unclear field source, conflicting time attribution, unclear segment division or inability to unify state objects. In practical applications: After a nighttime automated peritoneal dialysis cycle begins, the patient-side edge computing first extracts the start and end times of perfusion, stay, drainage, and alarm times from the execution log to form the cycle's time sequence main chain. Then, the blood pressure, heart rate, body position changes, and abdominal distension complaints collected in this cycle are assigned to the corresponding execution positions according to the time. Subsequently, the intra-segment position value, change direction value, and alarm time adjacency value of each monitoring point are calculated, and these results and the corresponding execution status are written into the state unit. Finally, a state sequence is generated for subsequent steps to read, so that when drainage abnormalities, abdominal distension, or alarm triggering occur, constraint decision graph generation and correction path solving can be carried out directly based on a unified state object.

[0018] The graph generation module is used to read historical states and corresponding real-time correction results for the current state in the state sequence, calculate the state transition cost, tolerance offset and execution risk of each historical state to the current state, and perform constraint labeling in combination with medical and nursing constraint data to output a constraint decision graph. This implementation method organizes the current and historical states in a state sequence into a constraint decision graph that can be directly invoked for subsequent path generation and constraint strategy solving. Its purpose is to filter out historical states with reference value to the current state from the already occurred state changes and immediate correction results, further solve for the cost, tolerance offset, and execution risk of each historical state migrating to the current state, and remove conflicting migration relationships under the constraints of medical and nursing data, retaining constraint edges that can enter the subsequent path generation process. Its basic principle is as follows: First, historical state screening is performed around the current state to form a candidate preorder set; then, a state transition cost function is constructed for each historical state in the candidate preorder set and constraint solving is performed to obtain feasible transition edges; subsequently, the tolerance offset and execution risk of each feasible transition edge are calculated to form an edge value group; based on this, multi-view consistency verification is performed and conflicting edges are removed; finally, constraint labeling is performed on the consistent edge set in conjunction with medical and nursing constraint data to generate a constraint decision graph. This implementation process includes the following steps: First, comparable historical states are screened out based on the current state, and a unified difference object is prepared for subsequent state transition cost calculation. The input quantities are the current state, historical states, the instantaneous correction results corresponding to each historical state, and medical constraint data in the state sequence. The current state is the latest state unit written in the state sequence that triggers the correction judgment condition, and the historical states are the state units that have been encoded before the current state. The vital signs field, body position field, chief complaint field, execution level field, and alarm field in the current state are used as matching fields. Constraint matching is performed on each historical state and the current state one by one. The execution level field is used as a hard matching field, and the vital signs field, body position field, chief complaint field, and alarm field are used as difference fields. For historical states that meet the same execution segment or preset connectable execution segments, calculate the differences in vital signs, body position, chief complaint, and alarm values ​​respectively to form a field difference group, and record the corresponding historical state as a segment-consistent state and write it into the candidate preorder set; historical states that do not meet the segment consistency condition are directly removed and do not enter the subsequent calculation; finally, the candidate preorder set and its corresponding field difference group are output and written into the cache of the graph generation module for subsequent reading; if the current state is missing any of the vital signs, body position, chief complaint, or alarm fields, first read the lowest confidence state unit within the same execution segment and perform the supplementation, and the historical states that are still missing after supplementation do not participate in the generation of the candidate preorder set; Subsequently, each historical state in the candidate preorder set is transformed into a feasible transfer object that can be determined whether migration to the current state is allowed. The inputs are the candidate preorder set, field difference groups, the real-time correction results corresponding to each historical state, and medical constraint data. For each historical state in the candidate preorder set, a transfer cost function is constructed based on the field difference groups and the corresponding real-time correction results. The transfer cost function consists of at least the state field difference cost, the real-time correction result deviation cost, the execution stage jump cost, and the alarm adjacency cost. The state field difference cost is used to characterize the degree of difference between the historical state and the current state. The real-time correction result deviation cost is used to characterize the degree of inheritance deviation between the previous correction action and the current state. The execution stage jump cost is used to characterize the additional cost brought about by cross-stage migration. The alarm adjacency cost is used to characterize the additional cost of alarm neighboring states entering the migration path. After construction, the parameter boundaries, alarm prohibition conditions, and execution segment constraints in the medical and nursing constraint data are read, and the transition cost function is solved for execution constraints. The parameter boundaries are used to limit the range of parameter values ​​allowed to enter the subsequent correction path deduction. The alarm prohibition conditions are used to limit the transition relationships that are prohibited from being retained when a specified alarm state occurs. The execution segment constraints are used to limit the segment combinations that are allowed to migrate. For the transition relationship from the historical state to the current state that satisfies all constraints, the state transition cost and feasible transition mark are output, and a feasible transition edge set is formed and written to the buffer for subsequent reading. If the real-time correction result corresponding to a certain historical state is missing, the execution parameter result finally adopted in the corresponding round of the historical state is used to replace the real-time correction result in the construction. If it still cannot be replaced, the historical state is deleted from the candidate preorder set. Furthermore, tolerance and risk attributes are added to each feasible transfer edge to provide a complete boundary value basis for subsequent consistency verification and path generation. The input quantities are the set of feasible transfer edges, the historical state corresponding to each feasible transfer edge, the current state, and the parameter boundaries, alarm prohibition conditions, and execution segment constraints in the medical and nursing constraint data. For each feasible transfer edge in the feasible transfer edge set, the changes in vital signs, chief complaints, and alarms before and after the correction of the corresponding historical state are read first. The tolerance offset of the current state relative to the historical state is calculated. Among them, the change in vital signs is used to characterize the change direction of physiological response after correction, the change in chief complaints is used to characterize the change direction of the patient's subjective feelings, and the change in alarms is used to characterize the change direction of abnormal triggering state. The three together determine the degree of tolerance deviation of the current state relative to the historical state. After obtaining the tolerance offset, the execution risk is calculated based on the parameter boundary occupancy, alarm trigger proximity, and execution segment conflict corresponding to the feasible transition edge. The parameter boundary occupancy is given by the degree of occupancy of the parameter boundary by the correction result corresponding to the historical state. The alarm trigger proximity is given by the degree of adjacency between the current state and the alarm prohibition condition. The execution segment conflict is given by the segment connection relationship between the current state and the historical state. Finally, the state transition cost, tolerance offset, and execution risk are written into the same edge value group, and the edge value group is associated with the corresponding feasible transition edge for storage, for subsequent consistency verification. If any change record is missing in the changes in vital signs, chief complaints, and alarms before and after the correction, the recovery result of the same type of abnormality is read from the corresponding historical round result and the record is filled in. If the record is still missing after filling in the record, the feasible transition edge is recorded as a low-confidence transition edge and its retention priority is reduced in subsequent consistency verification. After forming the boundary value group, conflicting edges that are inconsistent with the directions of other migration relationships are eliminated through multi-view consistency verification to avoid introducing contradictory precedent references in subsequent path generation. The input quantities are the current state, each historical state in the candidate precedent set, each feasible transition edge and its corresponding boundary value group. Using the current state and each historical state in the candidate precedent set as nodes, and the boundary value group as the boundary value between nodes, multi-view consistency verification is performed on the state transition cost, tolerance offset and execution risk in the boundary value group. The state transition cost perspective is used to determine the degree of consistency between the migration relationship and other migration relationships in the state proximity direction, the tolerance offset perspective is used to determine the degree of consistency between the migration relationship and other migration relationships in the tolerance change direction, and the execution risk perspective is used to determine the degree of consistency between the migration relationship and other migration relationships in the risk change direction. When a feasible transition edge deviates from the common association direction of the other edge values ​​in any of the three perspectives mentioned above, the feasible transition edge is identified as a conflict edge. For feasible transition edges identified as conflict edges, the edge value group is recalculated based on the corresponding historical state, current state, and medical constraint data. If the consistency condition is still not met after recalculation, the feasible transition edge is removed based on its feasible transition mark. Finally, a consistent edge set is output and written to the buffer for subsequent reading. If all feasible transition edges are removed after consistency verification, the process returns to the previous step to broaden the participation range of non-hard matching fields in the field difference group and re-executes the candidate predecessor set filtering. If the re-filtering is still empty, the current state is recorded as a state without predecessor reference, so that an empty path can be directly generated in the subsequent path generation process. Finally, the consistent edge set is transformed into a constraint decision graph that can be directly called by the subsequent path generation module, and it is determined which migration relationships are allowed to continue into the path correction generation process. The inputs are the consistent edge set and medical constraint data. Each feasible transition edge in the consistent edge set is constrained by combining the medical constraint data. It is determined whether the feasible transition edge satisfies the parameter boundary, alarm prohibition condition and execution stage constraint. Feasible transition edges that satisfy the above constraints are retained as constraint edges, and feasible transition edges that do not satisfy any constraint are deleted. Then, each constraint edge is written between the corresponding nodes according to the state transition cost, tolerance offset and execution risk, forming a constraint decision graph with the current state as the center and the historical states in the candidate preorder set as the preorder nodes. After each round of constraint edge writing is completed, the number of newly added constraint edges is counted. If the number of newly added constraint edges is greater than zero, the next round of constraint edge writing continues based on the updated node connection relationship. If the number of newly added constraint edges is zero, writing stops and the constraint decision graph is output. The constraint decision graph is written to the input of the path generation module for subsequent reading. Taking the nighttime home automatic peritoneal dialysis scenario as an example, when the current state is in the process of drainage and both abdominal distension complaints and proximity alarms occur, the system first filters out historical states that belong to the same drainage execution position and have similar complaints and alarm characteristics from the historical states. Then, it calculates the transfer cost, tolerance offset, and execution risk of these historical states to the current state. If a certain historical state has a low transfer cost, but its corresponding immediate correction result is close to the parameter boundary set by the medical staff, and the alarm trigger proximity is high, then the feasible transfer edge corresponding to the historical state is removed during the consistency verification or constraint marking process and does not enter the constraint decision graph. Finally, the constraint edges that are retained only correspond to the migration relationship that is both close to the current state and meets the medical staff constraint data limit. Through the above processing, the referable relationships between the current state and the historical state are organized into a constraint decision graph with clear nodes, clear boundary values, and clear constraints. The subsequent path generation module can then directly generate immediate correction paths and joint correction paths around the preceding nodes, avoiding problems such as unclear historical state selection boundaries, undefined boundary value sources, unresolved conflict migration relationships, or unfulfilled constraints. At the same time, by fixing the field difference group, transition cost function, tolerance offset, execution risk amount, multi-view consistency verification, and stop write entry point one by one, the constraint decision graph generation process has an executable, reproducible, and verifiable implementation foundation. In practical applications: After reading the current drainage status unit, the patient-side edge computing side first retrieves historical states in the same segment with consistent alarm adjacency in the status sequence. It then calculates the differences in vital signs, body position, chief complaints, and alarms between these historical states and the current state. Combining the corresponding real-time correction results and medical constraint data, it solves the state transition cost and feasible transition edges. Subsequently, it further calculates the tolerable offset and execution risk of each feasible transition edge, performs multi-view consistency verification and conflict edge removal on all edges, and finally writes feasible transition edges that meet the parameter boundaries, alarm prohibition conditions, and execution segment constraints into the constraint decision graph for subsequent generation of candidate correction paths around the current state.

[0019] The path generation module is used to generate immediate correction paths for preceding nodes in the constraint decision graph, generate joint correction paths for multiple preceding nodes that meet the adjacent transition conditions, and determine paths that do not contain any preceding nodes as empty paths, and output a set of candidate correction paths. This implementation method is used to organize the node relationships in the constraint decision graph into a set of candidate correction paths that can be directly called by the subsequent solution module. Its purpose is to transform the constraint edges retained by the graph generation module into path objects with clear predecessor sources, clear connection order, and clear retention boundaries, so that subsequent path deduction, path cost calculation, and feasible path ranking all revolve around a unified path structure. Its basic principle is: first, based on each constraint edge, generate single-node correction paths pointing from a single predecessor node to the current node; then, perform adjacent transition checks on the connection relationships between multiple single-node correction paths, retaining path combinations that can be continuously connected according to time and segment order to form joint correction paths; finally, introduce empty paths that do not reference any predecessor nodes as baseline paths in scenarios without predecessor references, and combine them with the aforementioned paths to form a set of candidate correction paths. This implementation process includes the following steps: First, each constraint edge in the constraint decision graph is transformed into a single-node correction path that can independently participate in the solution process, and the basic field structure of the path is fixed. The inputs are the current node, each predecessor node, and the constraint edges between each predecessor node and the current node in the constraint decision graph. The constraint edges must contain at least the state transition cost, tolerance offset, execution risk, and constraint label. For each constraint edge between the predecessor node and the current node, the edge value and constraint label are read one by one. Using the predecessor node as the path start point and the current node as the path end point, a single-node correction path is constructed in the order of predecessor node, constraint edge, and current node. The state transition cost, tolerance offset, execution risk, and constraint label corresponding to the constraint edge are written into the path attribute field of the single-node correction path. During construction, the path action corresponding to the previous node entering the current node is extracted based on the constraint mark corresponding to the constraint edge. The path action includes at least the parameter adjustment direction, parameter action object and action order to ensure that the subsequent solution module can directly expand the parameter change sequence from the single-node correction path. After all constraint edges are processed, the single-node correction path set is output and written to the path generation module buffer for the next step to read. If a constraint edge is missing any field of state transition cost, tolerance offset, execution risk amount or constraint mark, the corresponding edge value group in the constraint decision graph is read back and the missing edge is filled in. If the missing constraint edge is still missing after filling in the missing edge, a single-node correction path is not generated. Subsequently, consecutively connected preceding reference links are identified from the single-node correction path set, forming a joint correction path that contains more historical references than the single-node correction path. The inputs are the single-node correction path set, the preceding node attributes and constraint edge attributes corresponding to each single-node correction path. For each single-node correction path in the single-node correction path set, the time position, execution segment, and constraint edge attributes of its corresponding preceding node are read pairwise. Adjacent transfer verification is performed according to the time order, execution segment connection relationship, and constraint edge compatibility relationship between preceding nodes. The time order is used to determine whether two preceding nodes satisfy the sequential continuity relationship, the execution segment connection relationship is used to determine whether the execution positions of two preceding nodes satisfy the preset connectable conditions, and the constraint edge compatibility relationship is used to determine whether there are parameter conflicts, alarm conflicts, or segment conflicts between the path actions corresponding to two single-node correction paths. Only when all three conditions above are met simultaneously will the corresponding multiple preceding nodes be connected in chronological order to generate a joint correction path, and the node sequence, edge sequence, and path action sequence in the joint correction path will be written into the same path object in the order of connection; if any condition is not met, path combination will not be performed, and the original single-node correction path will remain unchanged; after all verifications are completed, the generated joint correction path will be merged with the original single-node correction path, the combined correction path set will be output and written to the buffer for the next step of reading; if a preceding node is missing a time position field or an execution segment field, the segment position value and segment identifier in the corresponding state unit will be read first to perform the supplementation. If the adjacency relationship cannot be determined after supplementation, the preceding node will only be retained in the original single-node correction path and will not enter the joint correction path generation process; Furthermore, a baseline path is reserved for scenarios without a preceding reference, and all path objects that can enter subsequent solutions are uniformly organized into a candidate correction path set; the input is the combined correction path set and the current node; firstly, a path object containing only the current node and not any preceding node or constraint edge is constructed, and this path object is determined as an empty path, where an empty path means that in the current state, no historical state transition relationship is referenced, and only the current execution boundary is maintained to enter the baseline path for subsequent solutions; Subsequently, empty paths and combined correction path sets are aggregated and all path objects are arranged in ascending order of path length, consistent path endpoints, and forward order of path start time to form a candidate correction path set. This set is then written into the input of the solver module for subsequent path deduction and constraint strategy solving. If the combined correction path set is empty, only empty paths are retained in the candidate correction path set. If there are path objects in the combined correction path set with the same node sequence but different path action sequences, those with more compatible constraint edges are retained first. If the number of compatible edges is the same, those with lower total execution risk are retained. The remaining path objects are written to the redundant path record area and do not enter the candidate correction path set. Taking the scenario of nighttime home-based automated peritoneal dialysis as an example, when the current node corresponds to the abdominal distension alarm state during the drainage process, the system first generates multiple single-node correction paths from different historical drainage states based on the constraint edges retained in the constraint decision graph. Then, it performs time sequence and segment connection verification on these single-node correction paths. If two historical drainage states are sequential in time and both belong to connectable drainage execution positions, and there is no parameter conflict between the corresponding path actions, then the two are combined into a joint correction path. If the current node has not found a predecessor node that meets the connection conditions before this round, the system still generates an empty path, so that the subsequent solution module can directly perform a conservative solution on the current state without historical link reference. Through the above processing, the node relationships in the constraint decision graph are uniformly transformed into a set of candidate correction paths with clear structure and order, which can directly participate in subsequent solutions. Based on this, the subsequent solution module can directly unfold path actions, deduce parameter changes, and calculate path costs, avoiding problems such as unclear path object sources, vacant joint path connection conditions, or inability to enter the solution without prior reference scenarios. At the same time, by fixing the source of path actions, adjacent transition conditions, joint path retention rules, and empty path action scope one by one, the path generation process has an executable, reproducible, and verifiable implementation foundation. In practical applications: After obtaining the constraint decision graph, the patient-side edge computing side first converts each constraint edge pointing from the historical drainage state to the current drainage state into a single-node correction path, and writes the state transition cost, tolerance offset, execution risk amount, and constraint mark on the edge into the path attributes; then, these single-node correction paths are verified according to the time sequence of the preceding nodes, the connection relationship of the execution position, and the compatibility relationship of the path actions, and multiple historical drainage states that can be continuously connected are connected into a joint correction path; finally, an empty path that only retains the current node is added, which together with the aforementioned paths form a candidate correction path set, which is used by the solution module to continue to perform path deduction, risk margin calculation, and determination of the correction path for the current round.

[0020] The solution module is used to solve the constraint strategy for each candidate correction path, calculate the risk margin, tolerance recovery amount and parameter change amount after the path is executed, and determine the candidate correction path with the first risk margin among the candidate correction paths that meet the medical and nursing constraint data and whose parameter change amount is within the boundary of the current peritoneal dialysis round as the correction path for the current round, and output the execution parameters for the current round. This implementation method is used to perform deduction, measurement, correction, screening, and verification on a set of candidate correction paths to determine the current correction path and execution parameters for the current peritoneal dialysis cycle. Its purpose is to transform the path objects output by the path generation module into directly implementable parameter adjustment results, ensuring that the current state, under the premise of meeting medical and nursing constraints and the boundaries of the current peritoneal dialysis cycle, obtains a unique correction output that can enter the subsequent execution process. Its basic principle is as follows: First, the path actions in each candidate correction path are expanded into a sequence of parameter changes and applied to the current state to form a deduced state sequence. Then, based on the deduced state sequence, risk margin, tolerance recovery, and parameter change are calculated to form a path cost group. Subsequently, the path cost group is updated with confidence using historical execution results to correct the cost evaluation of each candidate correction path. Based on this, feasible paths are screened out according to medical and nursing constraints and the boundaries of the current peritoneal dialysis cycle and ranked. Finally, the consistency of subsequent execution segments of the ranked first path is verified, and the current cycle correction path and execution parameters are output. This implementation process includes the following steps: First, the candidate correction paths are converted into parameter change processes that can be gradually applied to the current state, forming a deduced state sequence that can be directly read for subsequent calculations. The inputs are the candidate correction path set and the current state. The candidate correction paths include at least a node sequence, an edge sequence, and a path action sequence. The current state is taken from the state unit that was last written and triggered the correction judgment condition in the state sequence. Each path action in each candidate correction path is unfolded in a predetermined order within the path and mapped to a parameter change sequence. The parameter change sequence includes at least the parameter object, parameter adjustment direction, parameter adjustment step size, and parameter activation order. The parameter object includes one or more of the following: infusion parameters, drainage parameters, dwell parameters, and alarm processing parameters. The parameter adjustment step size is taken from the parameter boundary field in the medical constraint data or the current round boundary field. The parameter activation order is taken from the position of the corresponding path action in the path action sequence. After obtaining the parameter change sequence, the current state is used as the starting point for deduction. Each parameter change item is applied sequentially, and the vital signs field, chief complaint field, alarm field, and execution stage field are updated item by item to obtain the stage deduction state corresponding to each parameter change item. All stage deduction states are arranged in the order of parameter activation to form the deduction state sequence of the corresponding candidate correction path and written to the buffer for subsequent reading. If a certain path action is missing the parameter action object or parameter adjustment direction, the constraint mark in the corresponding constraint edge is read back and the missing parameter is filled in. If the missing parameter is still missing after filling in the missing parameter, the candidate correction path is recorded as an incomplete path and removed from the subsequent deduction process. Subsequently, the simulated state sequence is quantified into comparable path costs for subsequent confidence correction and feasible path ranking. The inputs are the simulated state sequence corresponding to each candidate correction path, medical and nursing constraint data, and the current peritoneal dialysis cycle boundary. For each simulated state sequence, the vital signs field, chief complaint field, alarm field, and execution stage field of each stage of the simulated state are read, and the constraint occupancy, alarm remaining amount, tolerance reduction amount, and cumulative parameter offset are calculated. The constraint occupancy is used to characterize the degree of occupancy of the parameter boundaries in the medical and nursing constraint data by the current simulated state, and is calculated item by item according to the distance relationship between the current value of each parameter and the corresponding parameter boundary. The alarm remaining amount is used to characterize the remaining space of the current simulated state from the alarm prohibition condition, and is calculated according to the difference relationship between the current alarm field and the alarm prohibition condition. The tolerance reduction amount is used to characterize the degree of abnormal reduction of the current simulated state relative to the current state in the vital signs field, chief complaint field, and alarm field, and is calculated according to the direction and magnitude of the difference between the current state and the simulated state at each stage. The cumulative parameter offset is used to characterize the cumulative offset of the parameter change sequence relative to the initial execution parameters of the current round. It is calculated by accumulating the absolute offsets of each parameter change item in order. After obtaining the aforementioned quantities, a path cost group is constructed, and the risk margin, tolerance recovery amount, and parameter change amount corresponding to each candidate correction path are solved. The risk margin is determined by the constraint occupancy amount and the alarm remaining amount, the tolerance recovery amount is determined by the tolerance fallback amount, and the parameter change amount is determined by the cumulative parameter offset. Finally, the above results are written into the cost field of the corresponding candidate correction path for subsequent reading. If a certain simulation state sequence is missing any field in the vital signs field or chief complaint field, it is first written in the previous stage of the simulation state of the same path. If it still cannot be written, the simulation state sequence only participates in the calculation of parameter change amount, does not participate in the calculation of tolerance recovery amount, and is reduced in retention priority during subsequent sorting. Furthermore, the cost evaluation of each candidate correction path is revised using the actual execution results from completed rounds, reducing the deviation based solely on the current simulation. The inputs are the path cost group corresponding to each candidate correction path and the historical execution results corresponding to each candidate correction path. The historical execution results include at least the round results, recovery results, and alarm results corresponding to the same path action sequence or the same node combination in the historical rounds. For each candidate correction path, its path cost group and corresponding historical execution results are read, and the deviation between the path cost group and the historical execution results is calculated. The deviation includes at least the risk deviation, recovery deviation, and parameter deviation. The risk deviation is used to characterize the difference between the current risk margin and the historical actual alarm results. The recovery deviation is used to characterize the difference between the current tolerable recovery amount and the historical recovery results. The parameter deviation is used to characterize the difference between the current parameter change amount and the historical execution parameter results. After obtaining the deviation, confidence updates are performed on the risk margin, tolerance recovery, and parameter change of each candidate correction path. The update direction is determined by the consistency between the deviation and historical results. A high degree of consistency increases the confidence level of the corresponding cost value, while a low degree of consistency decreases the confidence level of the corresponding cost value. After the update is completed, the path cost set with the updated confidence is written back to the corresponding inferred state sequence, and the confidence-corrected path set is output for subsequent constraint strategy solutions to read. If a candidate correction path does not have a corresponding historical execution result, the path cost set of the candidate correction path is retained as the initial value, and a "no historical reference" mark is written, so that it will only participate in the solution based on the current inferred result in the future. After confidence correction, feasible candidate correction paths are screened based on medical and nursing constraint data and the current peritoneal dialysis cycle boundary, and a unique ranking result is formed. The inputs are the confidence correction path set, medical and nursing constraint data, and the current peritoneal dialysis cycle boundary. Constraint strategies are solved for each candidate correction path in the confidence correction path set one by one. It is determined whether the risk margin, tolerance recovery amount, and parameter change amount meet the parameter boundary, alarm prohibition condition, and execution stage constraint in the medical and nursing constraint data. It is also determined whether the parameter change amount is within the current peritoneal dialysis cycle boundary. Candidate correction paths that meet all the above conditions are retained as feasible paths, and candidate correction paths that do not meet any of the conditions are removed. Then, all feasible paths are sorted in the following order: risk margin first, tolerance recovery second, and parameter change third. Risk margin is listed first for comparison. If risk margins are the same, tolerance recovery is compared; if tolerance recovery is the same, parameter change is compared; if parameter change is still the same, shorter path length is prioritized; if path length is the same, the path whose preceding node time is closer to the current state is prioritized. After sorting, the feasible path sorting results are output and written to a buffer for subsequent review. If none of the candidate correction paths meet the constraints, an empty path is retained as the only feasible path. If no empty path exists in the candidate correction path set, the process returns to the path generation module to fill in the empty path and then repeats this step. Finally, the consistency of subsequent execution segments is checked on the first-ranked path to ensure that the output execution parameters of the current round can continue to hold during the remaining execution of the current round. The inputs are the feasible path ranking results, the boundary conditions of the subsequent execution segments of the current peritoneal dialysis round, and the deduced state sequence corresponding to each feasible path. The first feasible path in the feasible path ranking results is read, the final state after the first feasible path is applied is extracted, and the final state is compared with the boundary conditions of the subsequent execution segments of the current peritoneal dialysis round for consistency. The boundary conditions of the subsequent execution segments include at least the allowed parameter range of the subsequent segment, the alarm prohibition condition of the subsequent segment, and the execution segment connection condition of the subsequent segment. When the final state meets the above boundary conditions, the first feasible path is determined as the correction path of the current round, and the parameter change sequence corresponding to the first feasible path is extracted as the execution parameters of the current round and written to the input of the execution module. When the final state does not meet any boundary condition, the first feasible path is removed from the feasible path ranking results, the remaining feasible paths are re-ranked and the consistency comparison is repeated until the correction path of the current round that meets the boundary conditions of the subsequent execution segments is obtained. If all feasible paths are removed, the empty path is determined as the correction path for the current round, and the execution parameters for the current round are kept at the initial execution parameters for the current round. Taking the scenario of nighttime home-based automated peritoneal dialysis as an example, when the current state is in the process of drainage and abdominal distension and proximity alarm occur, the system first unfolds the parameter change sequence of each candidate correction path from multiple historical drainage states and infers its impact on the current state. Then, it calculates the risk margin, tolerance recovery amount and parameter change amount of each path, and performs confidence correction in combination with the execution results in similar historical rounds. Subsequently, feasible paths that meet the medical and nursing constraints and the current round boundary are screened out and sorted. If the first path in the sort can reduce the risk of triggering the current alarm, but its final state does not meet the parameter entry range of the corresponding execution position for subsequent stays, the system removes the path and continues to check the next path until the current round correction path and the current round execution parameters that meet both the current correction requirements and the boundary conditions of the subsequent execution segment are obtained. Through the above processing, the candidate correction path set is organized into an executable path set with inference results, cost evaluation, confidence correction results, and subsequent execution segment verification results. Ultimately, it can output the correction path and execution parameters of the current round, which can be directly called by the subsequent boundary formation and update process. At the same time, by fixing the parameter change sequence, inference state sequence, constraint occupancy, alarm remaining amount, tolerance fallback amount, cumulative parameter offset, confidence update, parallel resolution, and reordering termination conditions one by one, the solution process has an executable, reproducible, and verifiable implementation basis, avoiding problems such as path actions not being able to be implemented, sorting results not being uniquely determined, or the first path not being connected with the subsequent execution segment. In practical applications: After obtaining the candidate correction path set, the patient-side edge computing side first expands the path actions in each path into a sequence of parameter changes for perfusion parameters, drainage parameters, or dwell parameters. Then, these parameter change sequences are applied to the current drainage state to generate the corresponding inferred state sequence. Subsequently, the constraint occupancy, alarm remaining amount, tolerance reduction amount, and cumulative parameter offset of each path are calculated, and the risk margin, tolerance recovery amount, and parameter change amount are solved. Then, confidence updates are performed in conjunction with the actual execution results in previous similar drainage abnormality rounds. After that, feasible paths that meet the medical and nursing constraint data and the current round boundary are screened out and sorted. Finally, consistency comparison is performed between the first ranked path and the boundary conditions of the subsequent execution segments, and the correction path and execution parameters of the current round are output for parameter adjustment in the current round.

[0021] The boundary formation module is used to read the results of the current peritoneal dialysis cycle after executing the parameters of the current cycle, write the cycle results and the results of the historical peritoneal dialysis cycles into the cross-cycle sequence in chronological order, calculate the repetition location of similar abnormalities, the recovery result after local correction, and the boundary offset direction of consecutive cycles, and output the cross-cycle boundary. This implementation method is used to extract cross-cycle inheritable boundary convergence information from the execution results of the current peritoneal dialysis cycle and the execution results of historical cycles after the current cycle is completed, forming cross-cycle boundaries that only apply to subsequent peritoneal dialysis cycles. Its purpose is to link the local recovery status after a single-cycle correction with similar abnormalities that recur in multiple cycles, identifying which abnormalities are short-term fluctuations and which have shown a stable boundary shift trend in consecutive cycles, thereby providing clear parameter constraints for the generation process of candidate correction paths for subsequent cycles. Its basic principle is: first, the current peritoneal... The results of each dialysis cycle and the results of historical peritoneal dialysis cycles are written into a cross-cycle sequence in chronological order, and a correspondence is established between the abnormality category, the execution segment where the abnormality is located, the corresponding correction path, and the cycle result fields. Then, the recurrence positions of similar abnormalities in different peritoneal dialysis cycles are calculated, and the recovery status after corresponding corrections is compared to obtain the local correction recovery results. Finally, based on the changes in parameter boundary occupancy, alarm proximity changes, and recovery result changes in consecutive cycles, the boundary offset direction is calculated, and cross-cycle boundaries are generated. This implementation process includes the following steps: First, the results of peritoneal dialysis cycles scattered across different cycles are organized into cross-cycle associated objects that can be continuously tracked according to abnormality categories. The input consists of the results of the current peritoneal dialysis cycle and the results of historical peritoneal dialysis cycles. The cycle results include at least the abnormality category, the execution stage where the abnormality occurred, the corresponding correction path, the vital signs result field, the chief complaint result field, the alarm result field, and the execution parameter result of the current cycle. The results of the current peritoneal dialysis cycle and the results of historical peritoneal dialysis cycles are written into the cross-cycle sequence from front to back according to the cycle completion time, and the abnormality category, the execution stage where the abnormality occurred, the corresponding correction path, and the cycle result fields in each cycle result are associated accordingly. The anomaly category is determined based on a preset anomaly tag library. The execution segment where the anomaly is located is taken from the execution position that triggered the correction judgment in that round. The corresponding correction path is taken from the correction path of the current round output by the solving module. The round result fields are taken from the vital signs result field, chief complaint result field, alarm result field, and current round execution parameter result written at the end of the round. After completing the above corresponding association, a cross-round association sequence is formed and written to the boundary to form the module cache area for subsequent reading. If a historical peritoneal dialysis round lacks a corresponding correction path, the current round execution parameter result and the reconstruction path identifier of the execution segment where the anomaly is located are read first. The reconstruction path identifier is used to replace the corresponding correction path in the writing. If it still cannot be reconstructed, the round result is kept in the cross-round sequence but not entered into the cross-round association sequence. Subsequently, cross-round records with the same anomaly category are screened out from the cross-round association sequence, and the local correction and recovery results at each repetition position are solved; the input is the cross-round association sequence; for each anomaly category in the cross-round association sequence, all round records with the same anomaly category identifier and the same execution segment position or satisfying the preset comparable execution segment conditions are retrieved in round order. The round position of these round records in the cross-round sequence and the execution segment position in the corresponding round are recorded together as the repetition position of the same type of anomaly. The preset comparable execution segment conditions are taken from rule constraints and are used to limit the comparison of the same type of anomaly only between the same execution position or the preset connectable execution position. After obtaining the duplicate locations, the changes in vital signs, chief complaints, and alarms corresponding to each duplicate location after correction are read, and a recovery comparison is performed. The changes in vital signs are used to determine whether the vital signs have returned to the range defined by the medical and nursing constraints after correction. The changes in chief complaints are used to determine whether the chief complaints have been relieved or reduced to an acceptable state after correction. The changes in alarms are used to determine whether the alarms have been relieved after correction. When the changes in vital signs, chief complaints, and alarms all meet the preset recovery conditions, the local correction recovery result corresponding to the duplicate location is recorded as successful recovery. When any one of them is not met, it is recorded as unsuccessful recovery. Finally, an abnormal recovery sequence is formed and written to the cache for subsequent reading. If any field of vital signs, chief complaints, or alarms is missing at a certain duplicate location, the most recent confirmed result before the end of the corresponding round is read from the results of the round and the missing field is filled in. If the missing field is still missing after filling in the missing field, the duplicate location is recorded as a location to be determined for recovery, and its reference priority is reduced when calculating the boundary offset direction in the subsequent calculation. Furthermore, based on the continuous performance of similar abnormalities in consecutive peritoneal dialysis cycles, cross-cycle boundaries for subsequent peritoneal dialysis cycles are generated. The input quantities are the parameter boundaries and alarm prohibition conditions in the abnormal recovery sequence, cross-cycle association sequence, and medical constraint data. For each similar abnormality in the abnormal recovery sequence, the parameter boundary occupancy change, alarm proximity change, and recovery result change corresponding to that similar abnormality in consecutive peritoneal dialysis cycles are read in chronological order. The parameter boundary occupancy change is given by the change in the degree of occupancy of the execution parameter results of each cycle relative to the corresponding parameter boundary. The alarm proximity change is given by the change in the degree of proximity of the alarm result field of each cycle relative to the alarm prohibition condition. The recovery result change is given by the change in the establishment and non-establishment of the local correction recovery results of each cycle in consecutive cycles. After obtaining the above three types of changes, the boundary offset direction of the same type of anomaly in consecutive rounds is calculated. When the parameter boundary occupancy level continuously increases, the alarm proximity level continuously increases, and the number of successful recoverys continuously decreases, the boundary offset direction is determined as the tightening direction. When the parameter boundary occupancy level continuously decreases, the alarm proximity level continuously decreases, and the number of successful recoverys continuously increases, the boundary offset direction is determined as the loosening direction. In other cases, it is determined as the maintaining direction. After determining the boundary offset direction, the cross-round boundary is generated by combining the repeated positions of the same type of anomaly and the local correction and recovery results. The cross-round boundary includes at least the anomaly category identifier, applicable execution segment, parameter boundary item, alarm constraint item, boundary offset direction, and effective round range. The cross-round boundary is then output to the input of the update module for subsequent reading. If the number of records of a certain type of abnormality in consecutive rounds does not reach the preset round number threshold, a new boundary offset direction will not be generated, and the cross-round boundary confirmed in the previous round will be directly inherited. The preset round number threshold is taken from the preset configuration and is used to limit the formation of a new cross-round boundary only when there are at least several consecutive rounds of records. Taking the nighttime home automatic peritoneal dialysis scenario as an example, when the patient repeatedly experiences the same type of abnormality with abdominal distension alarm in multiple consecutive drainage corresponding execution positions, the system first locates the round position and execution position of these similar abnormalities in the cross-round association sequence, and then compares whether the changes in vital signs, chief complaints and alarm changes after the corresponding correction have recovered. If the parameter boundary occupancy degree of the same type of abnormality increases round by round in multiple consecutive rounds, the alarm proximity degree increases round by round, and the local correction recovery result gradually changes from established to not established, then the system determines the boundary offset direction of the same type of abnormality as the tightening direction and generates a cross-round boundary applicable to the subsequent drainage corresponding execution position to limit the parameter range that can be entered when the next round of candidate correction path generation is generated. Through the above processing, the results of the current peritoneal dialysis cycle are incorporated into the continuous analysis process from a cross-cycle perspective. The local recovery situation after single-cycle correction does not directly rewrite the long-term boundary. Instead, it is first identified by the identification of the repetitive locations of similar anomalies, the solution of local correction and recovery results, and the statistical analysis of continuous cycle changes before forming a cross-cycle boundary that only applies to subsequent peritoneal dialysis cycles. This avoids short-term anomalies directly contaminating the boundary constraints of subsequent cycles. At the same time, by fixing the composition of cycle result fields, the criteria for judging similar anomalies, the definition of repetitive locations, the judgment rules for local correction and recovery results, the solution conditions for boundary offset direction, and the composition of cross-cycle boundary items one by one, the boundary formation process has an executable, reproducible, and verifiable implementation basis. In practical applications: After the current peritoneal dialysis cycle ends, the patient-side edge computing side first writes the abnormality category, execution segment where the abnormality is located, the current cycle correction path, vital sign result field, chief complaint result field, alarm result field, and the current cycle execution parameter result together with the historical peritoneal dialysis cycle results into the cross-cycle sequence, forming a cross-cycle association sequence; then, focusing on the abdominal distension proximity alarm, it identifies its repeated positions in multiple drainage corresponding execution positions, and compares the changes in vital signs, chief complaints, and alarms after correction at each repeated position to solve the local correction recovery result; then, it statistically analyzes the parameter boundary occupancy changes, alarm proximity changes, and recovery result changes in these consecutive cycles, calculates the boundary offset direction of this type of abnormality in the drainage corresponding execution position, and generates the cross-cycle boundary for the candidate correction path generation process of the next peritoneal dialysis cycle.

[0022] The update module is used to limit the correction path of the current cycle to only be effective for the current peritoneal dialysis cycle, and to limit the cross-cycle boundary to only provide parameter constraints for the generation process of candidate correction paths for subsequent peritoneal dialysis cycles. Then, the execution parameters of the current cycle, the cycle results, and the cross-cycle boundary are sent to the medical staff through a remote two-way communication link. The module receives the boundary correction rules returned by the medical staff, updates the cross-cycle boundary according to the boundary correction rules, and outputs the execution basis for the next peritoneal dialysis cycle. This implementation method is used to separate and manage the intra-cycle correction results that only apply to the current peritoneal dialysis cycle from the cross-cycle boundaries that only apply to subsequent peritoneal dialysis cycles after the current cycle ends. It then combines the boundary correction rules returned by the healthcare provider to form the execution basis for the next peritoneal dialysis cycle. The purpose of this process is to prevent the correction path of the current cycle from being directly inherited across cycles, and to prevent cross-cycle boundaries from directly entering the next cycle without correction by the healthcare provider, thereby ensuring that intra-cycle correction and cross-cycle constraints function within their respective effective ranges. Its basic principle is: first, the correction path of the current cycle... Different round identifiers and activation markers are written to the round and cross-round boundaries respectively, forming the intra-round execution set and the cross-round constraint set; then, the intra-round execution set, round results, and cross-round constraint set are sent to the medical staff through a remote two-way communication link, and the boundary correction rules returned by the medical staff are received to complete the field alignment and constraint mapping at the boundary item level; finally, according to the boundary correction set, each boundary item in the cross-round constraint set is replaced, retained, or deleted to generate the updated cross-round boundary, and it is written into the candidate correction path generation process of the next peritoneal dialysis round; this implementation process includes the following steps: First, the intra-cycle correction results and cross-cycle constraint results are separated into two independent objects according to their effective range to prevent confusion between the current cycle correction path and the constraint boundary of subsequent cycles. The inputs are the current cycle correction path, the current cycle execution parameters, the cycle result, and the cross-cycle boundary. The current cycle correction path is taken from the output of the solver module, and the cross-cycle boundary is taken from the output of the boundary formation module. The current cycle correction path is written with the current peritoneal dialysis cycle identifier, and an intra-cycle effectiveness marker is generated. The current peritoneal dialysis cycle identifier is used to limit the path to be valid only in the execution process corresponding to perfusion, stay, and drainage in this cycle, and the intra-cycle effectiveness marker is used to indicate that the path must not enter the candidate correction path generation process of subsequent peritoneal dialysis cycles. Simultaneously, the subsequent peritoneal dialysis cycle identifier is written into each boundary item in the cross-cycle boundary, and cross-cycle constraint markers are generated. The subsequent peritoneal dialysis cycle identifier is used to limit the boundary item to take effect after the end of the current peritoneal dialysis cycle. The cross-cycle constraint marker is used to indicate that the boundary item can only be used as a constraint for the candidate correction path generation of the next peritoneal dialysis cycle and its subsequent peritoneal dialysis cycles, and does not directly replace the execution parameters of the current cycle. After writing is completed, the correction path of the current cycle with the execution parameters of the current cycle is organized into an intra-cycle execution set, and the cross-cycle boundary with each boundary item is organized into a cross-cycle constraint set, and written into the update module cache for subsequent reading. If the correction path of the current cycle is missing the current peritoneal dialysis cycle identifier, it is supplemented by the cycle number in the cycle result. If the cross-cycle boundary is missing the applicable cycle range, it is assumed to take effect from the next peritoneal dialysis cycle, and a default effective marker is written into the corresponding boundary item. Subsequently, the healthcare system enables manual correction of the constraint boundaries for subsequent rounds, and accurately maps the correction results from the healthcare system to the corresponding boundary items in the cross-round constraint set. The inputs are the in-round execution set, round results, cross-round constraint set, and boundary correction rules returned by the remote two-way communication link. First, the in-round execution set, round results, and cross-round constraint set are sent to the healthcare system through the remote two-way communication link. The round results include at least the abnormality category, the execution segment where the abnormality is located, the execution parameter results of the current round, the vital signs result field, the chief complaint result field, and the alarm result field, so that the healthcare system can determine whether the cross-round boundary needs to be corrected based on the execution results of this round. After the medical staff returns the boundary correction rule, the system reads the exception category field, applicable execution segment field, correction action type field, and corrected boundary field from the boundary correction rule, and aligns and maps them with the execution fields of each boundary item in the cross-cycle constraint set. The field alignment matches the exception category and applicable execution segment item by item, and the constraint mapping corresponds to the parameter boundary item, alarm constraint item, and effective round range item by item. For boundary items that simultaneously satisfy the consistency of exception category and applicable execution segment, a one-to-one boundary correction relationship is generated, and the boundary correction set is output and written to the buffer for subsequent reading. If the remote two-way communication link does not receive the boundary correction rule within the preset waiting time, the cross-cycle constraint set is used as the default retention object and the system directly proceeds to the next step. The preset waiting time is taken from the preset configuration. If the received boundary correction rule has a missing exception category field or applicable execution segment field, the mapping for that boundary correction rule is not performed, and it is only written to the correction rule confirmation area for the medical staff to resend later. Furthermore, the cross-cycle boundary is updated based on the boundary correction set returned by the medical staff, forming the execution basis for direct reading in the next peritoneal dialysis cycle. The inputs are the cross-cycle constraint set and the boundary correction set. For each correction relationship in the boundary correction set, its correction action type is read item by item, and the corresponding boundary item in the cross-cycle constraint set is replaced, retained, or deleted according to the correction action type. When the correction action type is replacement, the corresponding field in the original boundary item is overwritten with the corrected boundary field. When the correction action type is retention, the original boundary item remains unchanged. When the correction action type is deletion, the corresponding boundary item is removed from the cross-cycle constraint set. After all correction relationships are processed, the updated cross-cycle boundary is generated and written into the candidate correction path generation process of the next peritoneal dialysis cycle. Specifically, the writing location is the constraint input end of the transfer cost function constraint solution step in the atlas generation module and the feasible path screening step in the solution module, so that the next peritoneal dialysis cycle is directly constrained by the updated cross-cycle boundary when generating candidate correction paths and screening feasible paths. Subsequently, the updated cross-cycle boundary, the parameter boundary items applicable to the next peritoneal dialysis cycle, the alarm constraint items, and the effective cycle range are compiled together as the execution basis for the next peritoneal dialysis cycle and output. If the same boundary item receives multiple boundary correction rules with different correction action types at the same time, the boundary correction rule with the latest return time from the medical staff is executed first. If the updated cross-cycle boundary is empty, the cross-cycle boundary of the previous confirmed cycle is used as the temporary execution basis until the medical staff returns a new boundary correction rule. Taking the nighttime home automatic peritoneal dialysis scenario as an example, when the current cycle has a similar abnormality with an abdominal distension alarm at the corresponding execution position of the drainage, the system first writes the current cycle correction path into the current peritoneal dialysis cycle identifier to form an intra-cycle execution set, and then writes the cross-cycle boundary generated for the corresponding execution position of the drainage into the identifier of the subsequent peritoneal dialysis cycle to form a cross-cycle constraint set. The intra-round execution set, round results, and inter-round constraint set are then sent to the medical staff. If the medical staff determines that the same type of abnormality needs to be further tightened in the next peritoneal dialysis round, a boundary correction rule containing the abnormality category, applicable execution segment, correction action type, and corrected boundary fields is returned. The system then maps this rule to the corresponding boundary item in the inter-round constraint set, replaces the original boundary item, forms the updated inter-round boundary, and writes it into the candidate correction path generation process for the next peritoneal dialysis round, so that the next round can directly avoid the parameter range tightened by the medical staff when generating the path. Through the above processing, the correction path for the current cycle is limited to be effective within the current peritoneal dialysis cycle, and the cross-cycle boundary is limited to providing parameter constraints only for the generation process of candidate correction paths for subsequent peritoneal dialysis cycles. The effective scope and target of the two are clearly separated. At the same time, by introducing boundary correction rules from the medical staff through a remote two-way communication link, and performing field alignment, constraint mapping, and replacement, retention, and deletion processing on boundary items, the update module can stably output the execution basis for the next peritoneal dialysis cycle, avoiding problems such as cross-cycle contamination of intra-cycle correction paths, cross-cycle boundaries taking effect directly without confirmation, or medical staff correction rules failing to be applied to specific boundary items. In practical applications: After the current peritoneal dialysis cycle ends, the patient-side edge computing side first organizes the current cycle correction path and execution parameters into an intra-cycle execution set, and then organizes the cross-cycle boundaries generated by the boundary formation module into a cross-cycle constraint set. Subsequently, the intra-cycle execution set, cycle results, and cross-cycle constraint set are sent to the medical staff end, and the boundary correction rules returned by the medical staff end for the boundary items corresponding to the abdominal distension proximity alarm are received. Then, the boundary correction rules are mapped to the corresponding boundary items in the cross-cycle constraint set according to the abnormality category and applicable execution segment. The boundary items are replaced, retained, or deleted to generate updated cross-cycle boundaries. The updated cross-cycle boundaries are written into the candidate correction path generation process of the next peritoneal dialysis cycle, and finally the execution basis of the next peritoneal dialysis cycle is output.

[0023] Working principle: This solution first receives patient monitoring data, peritoneal dialysis execution data, and medical constraint data in real time at the patient's edge computing side. These data from different sources are then organized into a state sequence in chronological order. Next, reference states that are close to the current state are identified from historical states, generating multiple correction paths. The risks, recovery outcomes, and parameter changes after each path are executed are compared to select the appropriate execution parameters for the current round. After the current round ends, the system analyzes the results of this round together with those of historical rounds to determine whether similar abnormalities are recurring, gradually worsening, or gradually mitigating. Based on this, the cross-round boundaries to be followed in the next round are established. Simultaneously, the results of this round and the cross-round boundaries are sent to the medical staff, who remotely correct the data and write it back to the system as the basis for the next round's execution. For example, if a patient experiences abdominal distension and triggers an alarm during the drainage phase while undergoing peritoneal dialysis at home at night, the system will first determine which past situations this abnormality is similar to. Then, it will select a more suitable corrective path from the corresponding historical handling methods and automatically adjust the parameters for this round to ensure the safe completion of this round. If this abnormality occurs in similar locations for several consecutive nights, the system will further tighten the parameter boundaries for the next round and synchronize this change with the medical staff for confirmation. In this way, the system can not only handle the sudden problems of the current round, but also gradually adjust the execution scope of subsequent rounds based on the results of multiple consecutive rounds.

[0024] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A remote bidirectional controllable intelligent adaptive peritoneal dialysis system, comprising a state construction module, a map generation module, a path generation module, a solution module, a boundary formation module, and an update module, characterized in that: The state construction module is used to acquire patient monitoring data, peritoneal dialysis execution data, and medical and nursing constraint data in the current peritoneal dialysis cycle. On the patient-side edge computing side, it performs time alignment and state encoding on the vital signs sequence, body position change sequence, chief complaint event sequence, as well as perfusion time, drainage time, dwell time, and alarm time, and outputs the state sequence. The graph generation module is used to read historical states and corresponding real-time correction results for the current state in the state sequence, calculate the state transition cost, tolerance offset and execution risk of each historical state to the current state, and perform constraint labeling in combination with medical and nursing constraint data to output a constraint decision graph. The path generation module is used to generate immediate correction paths for preceding nodes in the constraint decision graph, generate joint correction paths for multiple preceding nodes that meet the adjacent transition conditions, and determine paths that do not contain any preceding nodes as empty paths, and output a set of candidate correction paths. The solution module is used to solve the constraint strategy for each candidate correction path, calculate the risk margin, tolerance recovery amount and parameter change amount after the path is executed, and determine the candidate correction path with the first risk margin among the candidate correction paths that meet the medical and nursing constraint data and whose parameter change amount is within the boundary of the current peritoneal dialysis round as the correction path for the current round, and output the execution parameters for the current round. The boundary formation module is used to read the results of the current peritoneal dialysis cycle after executing the parameters of the current cycle, write the cycle results and the results of the historical peritoneal dialysis cycles into the cross-cycle sequence in chronological order, calculate the repetition location of similar abnormalities, the recovery result after local correction, and the boundary offset direction of consecutive cycles, and output the cross-cycle boundary. The update module is used to limit the correction path of the current cycle to only apply to the current peritoneal dialysis cycle, and to limit the cross-cycle boundary to only provide parameter constraints for the generation process of candidate correction paths for subsequent peritoneal dialysis cycles. Then, the execution parameters of the current cycle, the cycle results, and the cross-cycle boundary are sent to the medical staff through a remote two-way communication link. The module receives the boundary correction rules returned by the medical staff, updates the cross-cycle boundary according to the boundary correction rules, and outputs the execution basis for the next peritoneal dialysis cycle.

2. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 1, characterized in that: The state construction module includes: Read the start and end times of perfusion, drainage, stay, and alarm times in the current peritoneal dialysis cycle, connect them in chronological order to form the cycle time sequence main chain, and divide the cycle time sequence main chain into perfusion segment, stay segment, drainage segment, and alarm segment, and output the segment benchmark sequence; Map the time of each monitoring point in the vital signs sequence, body position change sequence, and chief complaint event sequence to the corresponding segment in the segment benchmark sequence, calculate the intra-segment position value of each monitoring point relative to the starting point of its segment, the change direction value between adjacent monitoring points, and the adjacency value of the alarm time, and output the monitoring alignment sequence. The status field combination encoding is performed on each monitoring point in the monitoring alignment sequence. The segment identifier, the position value within the segment, the change direction value, the adjacent value of the alarm time, and the peritoneal dialysis execution status at the corresponding time are written into the same status unit, and the status units are arranged in chronological order to output the status sequence.

3. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 2, characterized in that: The map generation module includes: Read the current state, historical state, real-time correction results corresponding to each historical state, and medical constraint data from the state sequence. Use the vital signs field, body position field, chief complaint field, execution level field, and alarm field in the current state as matching fields. Perform constraint matching on each historical state and the current state, calculate the field difference group and the level consistency result, and write the historical states that meet the level consistency condition into the candidate preorder set. Output the candidate preorder set and its corresponding field difference group. For each historical state in the candidate preorder set, a transition cost function is constructed based on the field difference group and the corresponding real-time correction result. Under the parameter boundary, alarm prohibition condition and execution stage constraint of the medical and nursing constraint data, the execution constraint of the transition cost function is solved to obtain the state transition cost and feasible transition mark of each historical state to the current state, and the feasible transition edge set is output. For each feasible transfer edge in the feasible transfer edge set, the tolerance offset of the current state relative to the corresponding historical state is calculated according to the changes in vital signs, chief complaints and alarms before and after the corresponding historical state. The execution risk is calculated based on the parameter boundary occupancy, alarm trigger proximity and execution segment conflict corresponding to the feasible transfer edge. The boundary value set is output, which includes the state transfer cost, tolerance offset and execution risk.

4. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 3, characterized in that: The map generation module also includes: Using the current state and each historical state in the candidate preorder set as nodes, and the edge value group as the edge value, perform multi-view consistency verification on the state transition cost, tolerance offset and execution risk in the edge value group. When any edge value deviates from the association direction of the other edge values, the corresponding feasible transition edge is identified as a conflict edge. The conflict edge is removed according to the edge value recalculation result or feasible transition mark, and a consistent edge set is output. Constraint marking is performed on each feasible transition edge in the consistent edge set in conjunction with medical and nursing constraint data. Feasible transition edges that meet parameter boundaries, alarm prohibition conditions, and execution segment constraints are retained as constraint edges. Each constraint edge is written between the corresponding nodes according to the state transition cost, tolerance offset, and execution risk amount to generate a constraint decision graph. Writing stops when the number of newly added constraint edges is zero, and the constraint decision graph is output.

5. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 4, characterized in that: The path generation module includes: Based on the state transition cost, tolerance offset, execution risk amount and constraint mark corresponding to each constraint edge, construct a single-node correction path from a single preceding node to the current node one by one, and output the set of single-node correction paths. For each single-node correction path in the single-node correction path set, adjacent transfer verification is performed according to the time order between the preceding nodes, the execution segment connection relationship, and the constraint edge compatibility relationship. Multiple preceding nodes that meet the adjacent transfer conditions are combined into a joint correction path according to the connection order. The joint correction path is then merged with the single-node correction path to output a combined correction path set. The path that does not contain any preceding node and only retains the current node is identified as an empty path. The empty paths are then combined with the combined correction path set to form a candidate correction path set.

6. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 5, characterized in that: The solution module includes: Each path action in each candidate correction path is expanded into a parameter change sequence according to the execution order, and each parameter change sequence is applied to the current state to perform path deduction, outputting the deduction state sequence corresponding to each candidate correction path; For each simulated state sequence, the constraint occupancy, alarm remaining amount, tolerance fallback amount and cumulative parameter offset are calculated according to the vital signs field, chief complaint field, alarm field and execution stage field, respectively. The path cost group is constructed with the aforementioned amounts, and the risk margin, tolerance recovery amount and parameter change amount corresponding to each candidate correction path are output. Read the historical execution results corresponding to each candidate correction path, calculate the deviation between each path cost group and the historical execution results, perform confidence updates on the risk margin, tolerance recovery amount and parameter change amount of each candidate correction path, and write back the path cost group after confidence update to the corresponding inferred state sequence, and output the confidence-corrected path set.

7. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 6, characterized in that: The solution module also includes: The constraint strategy is applied to the set of confidence correction paths. Candidate correction paths that meet the medical and nursing constraints and whose parameter changes are within the boundary of the current peritoneal dialysis cycle are retained as feasible paths. The feasible paths are sorted in the order of risk margin first, tolerance recovery second, and parameter change third, and the feasible path sorting results are output. The adjacent state verification is performed on the first feasible path in the feasible path ranking results. The final state after the first feasible path is applied is compared with the boundary conditions of the subsequent execution segment of the current peritoneal dialysis cycle. If they are consistent, the first feasible path is determined as the correction path of the current cycle and the corresponding parameter change sequence is extracted as the execution parameters of the current cycle. If they are inconsistent, the first feasible path is removed and the execution is returned to re-sort until the correction path of the current cycle is obtained and the execution parameters of the current cycle are output.

8. The remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 7, characterized in that: The boundary forming module includes: Read the results of the current peritoneal dialysis cycle and the results of the historical peritoneal dialysis cycles, write them into the cross-cycle sequence in chronological order, and perform corresponding associations on the abnormality category, execution segment where the abnormality is located, corresponding correction path and cycle result fields in each cycle result, and output the cross-cycle association sequence; For the same type of abnormality in the cross-cycle correlation sequence, the repetition position of each type of abnormality in different peritoneal dialysis cycles is calculated, and the recovery comparison is performed on the changes in vital signs, chief complaints and alarms after correction corresponding to each repetition position. The local correction recovery results corresponding to each type of abnormality are solved, and the abnormality recovery sequence is output. Based on the abnormal recovery sequence, the parameter boundary occupancy change, alarm proximity change and recovery result change corresponding to the same type of abnormality in consecutive peritoneal dialysis cycles are statistically analyzed in chronological order. The boundary offset direction of each type of abnormality in consecutive cycles is calculated, and cross-cycle boundaries are generated by combining the repeated positions and local correction recovery results, and the cross-cycle boundaries are output.

9. A remote bidirectional controllable intelligent adaptive peritoneal dialysis system according to claim 8, characterized in that: The update module includes: Write the current peritoneal dialysis cycle identifier to the current cycle correction path and generate an intra-cycle effective marker; write the subsequent peritoneal dialysis cycle identifier to the cross-cycle boundary and generate a cross-cycle constraint marker; output the intra-cycle execution set and the cross-cycle constraint set. The in-round execution set, round results, and cross-round constraint set are sent to the medical staff through a remote two-way communication link. The boundary correction rules returned by the medical staff are received, and the boundary correction rules are aligned and constraint mapped with the execution fields of each boundary item in the cross-round constraint set, and the boundary correction set is output. Based on the boundary correction set, each boundary item in the cross-cycle constraint set is replaced, retained, or deleted to generate the updated cross-cycle boundary. The updated cross-cycle boundary is then written into the candidate correction path generation process for the next peritoneal dialysis cycle, and the execution basis for the next peritoneal dialysis cycle is output.

Images