Patient nursing auxiliary system based on artificial intelligence

Abstract

The invention discloses a patient nursing auxiliary system based on artificial intelligence, which belongs to the technical field of intelligent medical treatment and comprises a data preparation module, a sleep stage recognition module, an intervention window recommendation module and a nursing strategy optimization module. The intervention window recommendation method combining the rhythm diagram and the interference modeling is adopted to perform intervention window recommendation, and the appropriate time period of each nursing behavior is identified on the premise of guaranteeing the sleep continuity and comfort of the patient, so that the nursing intervention time is reasonably arranged, the influence on the sleep of the patient is reduced to the maximum extent, and the sleep quality of the patient is improved. The rehabilitation of the patient is facilitated; nursing strategy optimization is conducted through multi-constraint reinforcement learning, a nursing plan conforming to the individual sleep law and physiological bearing capacity is dynamically formulated, time arrangement of various nursing behaviors is effectively coordinated, the burden of a patient is reduced to the maximum extent, and the rehabilitation effect is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent medical treatment, and particularly relates to a patient nursing assistance system based on artificial intelligence. BACKGROUND

[0002] The patient nursing assistance system based on artificial intelligence perceives and analyzes multi-channel physiological signals and nursing behavior data by using artificial intelligence technology, aims to assist nursing personnel in scientifically and reasonably arranging nursing behaviors, reduces the influence of nursing behaviors on the sleep quality of patients on the basis of guaranteeing the nursing effect, and thus improves the comfort of patients and the overall nursing quality.

[0003] However, in the existing patient nursing assistance process, there is a technical problem that the time arrangement of night nursing intervention lacks scientific basis, which leads to the fact that nursing behaviors easily disturb the sleep of patients and affect the rest quality and recovery effect of patients; there is a technical problem that nursing plan making depends on artificial experience and cannot fully consider the state change of patients, and it is difficult to simultaneously consider the nursing efficiency and the comfort of patients. SUMMARY

[0004] In view of the above situation, in order to overcome the defects of the prior art, the application provides a patient nursing assistance system based on artificial intelligence, which is aimed at the technical problem that in the existing patient nursing assistance process, the time arrangement of night nursing intervention lacks scientific basis, which leads to the fact that nursing behaviors easily disturb the sleep of patients and affect the rest quality and recovery effect of patients, and the scheme creatively adopts an intervention window recommendation method combining rhythm graph and interference modeling to recommend an intervention window, realizes the identification of a suitable time period of each nursing behavior on the premise of guaranteeing the continuity and comfort of the sleep of patients, and thus reasonably arranges the time of nursing intervention, maximally reduces the influence on the sleep of patients, and helps to promote the rehabilitation of patients; the scheme is aimed at the technical problem that in the existing patient nursing assistance process, nursing plan making depends on artificial experience and cannot fully consider the state change of patients, and it is difficult to simultaneously consider the nursing efficiency and the comfort of patients, and the scheme creatively adopts multi-constraint reinforcement learning to optimize a nursing strategy, dynamically formulates a nursing plan conforming to the sleep rhythm and physiological bearing capacity of an individual, effectively coordinates the time arrangement of each nursing behavior, maximally reduces the burden of patients, and improves the rehabilitation effect.

[0005] The technical scheme adopted by the application is as follows: the patient nursing assistance system based on artificial intelligence provided by the application comprises a data preparation module, a sleep stage identification module, an intervention window recommendation module and a nursing strategy optimization module.

[0006] The data preparation module is configured to prepare data, obtain physiological signal data of a patient and standardized nursing information data through data preparation, and send the physiological signal data of the patient to a sleep stage identification module and a nursing strategy optimization module, and send the standardized nursing information data to an intervention window recommendation module and the nursing strategy optimization module.

[0007] The sleep stage identification module is configured to identify a sleep stage, identify a sleep stage sequence by using a deep time sequence model according to the physiological signal data of the patient, and send the sleep stage sequence to the intervention window recommendation module and the nursing strategy optimization module.

[0008] The intervention window recommendation module is configured to recommend an intervention window, recommend a candidate time period set by using an intervention window recommendation method combining a rhythm graph and interference modeling according to the standardized nursing information data and the sleep stage sequence, and send the candidate time period set to the nursing strategy optimization module.

[0009] The nursing strategy optimization module is configured to optimize a nursing strategy, optimize an optimal nursing scheme by using multi-constraint reinforcement learning according to the physiological signal data of the patient, the standardized nursing information data, the sleep stage sequence and the candidate time period set.

[0010] Further, the data preparation specifically comprises continuously collecting multi-channel physiological signal data of the patient in a night period by using a sensor, and pre-processing the physiological signal data to obtain the physiological signal data of the patient, and collecting nursing record data of each patient from a ward management system, and standardizing the nursing record data to obtain standardized nursing information data.

[0011] The physiological signal data of the patient comprises heart rate, respiratory rate and blood oxygen saturation.

[0012] The standardized nursing information data comprises a nursing behavior type, an execution time, an execution frequency and a nursing behavior priority.

[0013] Further, the sleep stage identification specifically comprises identifying a sleep stage sequence by using a deep time sequence model according to the physiological signal data of the patient, and comprises the following steps: feature extraction and sleep stage classification.

[0014] The feature extraction specifically comprises performing time domain feature extraction on the physiological signal data of the patient in each time period, and performing frequency domain feature extraction to obtain a physiological state sequence of the patient.

[0015] The time domain feature extraction specifically comprises calculating a mean value, a standard deviation and a heart rate variability index of the heart rate, calculating a mean value and a standard deviation of the respiratory interval, and calculating a mean value, a standard deviation and a low blood oxygen duration of the blood oxygen.

[0016] The frequency domain feature extraction, specifically, is performing fast Fourier transform on the heart rate, respiratory rate and blood oxygen saturation, respectively.

[0017] The sleep stage classification, specifically, is modeling the physiological state sequence of the patient by constructing a deep time sequence model to generate a high-order time sequence feature vector, and then classifying the high-order time sequence feature vector through a full connection layer to obtain a sleep stage sequence.

[0018] Further, the intervention window recommendation is used to identify suitable nursing intervention time periods in the sleep process of the patient, specifically, according to the nursing information specification data and the sleep stage sequence, an intervention window recommendation method combining rhythm graph and interference modeling is adopted to recommend the intervention window, to obtain a candidate time period set, including the following steps: sleep rhythm graph modeling, interference degree modeling and window recommendation.

[0019] The sleep rhythm graph modeling is used to construct a graph structure of the sleep stage to mine the stability and sensitive period of the sleep rhythm, specifically, according to the sleep stage sequence, each time period is taken as a node, the edges between the nodes are established according to the sleep stage change, and the sleep stage is taken as a node feature to construct a sleep rhythm graph, and the rhythm stability degree is calculated.

[0020] The edges between the nodes are established according to the sleep stage change, specifically, if the sleep stage changes between adjacent time periods, a directed edge e ij between the node v i and the node v ii is established, otherwise a self-loop edge e i is established between the node v j ;

[0021] The interference degree modeling is used to quantify the interference influence of each nursing behavior on different sleep stages, specifically, according to the nursing information specification data and the sleep stage sequence, a standard random forest model is constructed to model the sleep stage variation amplitude of the time period before and after the nursing behavior, a nursing interference score is calculated, and then a nursing adaptation factor is constructed by combining the nursing interference score and the rhythm stability degree.

[0022] The window recommendation is used to screen a candidate time period set for each nursing behavior, specifically, for each nursing behavior, an adaptation threshold is set, all time periods with the nursing adaptation factor higher than the adaptation threshold are screened out to form a candidate time period set.

[0023] Further, the care strategy is optimized, in particular, based on the patient physiological signal data, the nursing information specification data, the sleep stage sequence and the candidate time period set, a multi-constraint reinforcement learning is used for care strategy optimization to obtain an optimal care scheme, including the following steps: constraint condition setting, bearing feature modeling, care strategy modeling and strategy optimization.

[0024] The constraint condition setting specifically sets a unique execution constraint, a care dependency constraint and a time constraint.

[0025] The unique execution constraint specifically selects at most one candidate time period for each care behavior; the care dependency constraint specifically executes a care behavior with a high priority before executing a care behavior with a low priority; and the time constraint specifically selects a candidate time period from the respective candidate time period set for each care behavior.

[0026] The bearing feature modeling is used to model the physiological changes of the patient to each care behavior in the candidate time period, and specifically, based on the patient physiological signal data, a patient physiological change vector is constructed, and an individual bearing capacity index is calculated.

[0027] The care strategy modeling models the care strategy optimization problem as a reinforcement learning problem through action space setting, state space setting and reward function setting.

[0028] The action space setting specifically sets the care behavior set as the action space, and sets each action as a binary tuple for representing the execution of the kth care behavior in the jth time period.

[0029] The state space setting specifically concatenates the patient physiological state sequence, the care behavior execution record sequence and the time period index into a state vector; and arranges the care behavior types, execution times and execution frequencies in the nursing information specification data in chronological order to construct the care behavior execution record sequence.

[0030] The reward function setting specifically constructs a reward function in combination with the sleep stage variation amplitude, the individual bearing capacity index and the constraint condition.

[0031] The strategy optimization specifically constructs a reinforcement learning model based on the set action space and state space, and learns an optimal strategy as an optimal care scheme by maximizing the cumulative expected reward.

[0032] The above-mentioned scheme has the following beneficial effects:

[0033] (1) In view of the technical problem that in the prior patient care assistance process, the night care intervention time arrangement lacks scientific basis, leading to the nursing behavior easily disturbing the patient's sleep, affecting the patient's rest quality and recovery effect, the scheme creatively adopts the intervention window recommendation method combining rhythm diagram and interference modeling to recommend the intervention window, realizes identifying the appropriate time period of each nursing behavior under the premise of guaranteeing the continuity and comfort of the patient's sleep, thereby reasonably arranging the nursing intervention time and minimizing the impact on the patient's sleep, which helps to promote the patient's recovery;

[0034] (2) In view of the technical problem that in the prior patient care assistance process, the nursing plan formulation depends on artificial experience and fails to fully consider the patient's state change, making it difficult to balance the nursing efficiency and patient comfort, the scheme creatively adopts multi-constraint reinforcement learning for nursing strategy optimization, dynamically formulates a nursing plan that conforms to the individual sleep rhythm and physiological bearing capacity, effectively coordinates the time arrangement of each nursing behavior, maximally reduces the patient's burden, and improves the recovery effect. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The structure schematic diagram of the patient care assistance system based on artificial intelligence provided by the present application is shown in the figure;

[0036] Figure 2 The flowchart of the intervention window recommendation module is shown in the figure;

[0037] Figure 3 The flowchart of the nursing strategy optimization module is shown in the figure.

[0038] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0040] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0041] In an embodiment, referring to Figure 1 The application provides an artificial intelligence-based patient care assistance system, which comprises a data preparation module, a sleep stage identification module, an intervention window recommendation module and a care strategy optimization module.

[0042] The data preparation module is configured to prepare data, obtain physiological signal data and nursing information specification data of a patient through data preparation, and send the physiological signal data of the patient to the sleep stage identification module and the care strategy optimization module, and send the nursing information specification data to the intervention window recommendation module and the care strategy optimization module.

[0043] The sleep stage identification module is configured to identify sleep stages, identify sleep stages by using a deep time sequence model according to physiological signal data of a patient, obtain a sleep stage sequence, and send the sleep stage sequence to the intervention window recommendation module and the care strategy optimization module.

[0044] The intervention window recommendation module is configured to recommend intervention windows, recommend intervention windows by using an intervention window recommendation method combining rhythm graphs and interference modeling according to nursing information specification data and a sleep stage sequence, obtain a candidate time period set, and send the candidate time period set to the care strategy optimization module.

[0045] The care strategy optimization module is configured to optimize care strategies, optimize care strategies by using multi-constraint reinforcement learning according to physiological signal data of a patient, nursing information specification data, a sleep stage sequence and a candidate time period set, and obtain an optimal care plan.

[0046] In an embodiment, referring to Figure 1 In this embodiment, the data preparation is specifically continuous collection of multi-channel physiological signal data of a patient in a night period by a sensor and preprocessing to obtain physiological signal data of the patient, and collection of nursing record data of each patient from a ward management system and specification processing to obtain nursing information specification data.

[0047] The preprocessing includes wavelet transform denoising, sampling frequency unification and sliding time window division.

[0048] Preferably, the window length of the sliding time window division is 5 minutes, and the sliding step is 1 minute.

[0049] The specification processing includes data deduplication, timestamp alignment, outlier removal and label standardization.

[0050] The label standardization specifically converts a nursing behavior type into a standard enumeration field, and converts a nursing behavior priority into a numerical sequence, wherein the higher the numerical sequence is, the higher the priority is.

[0051] The patient physiological signal data includes heart rate, respiratory rate, and blood oxygen saturation;

[0052] The nursing information specification data includes nursing behavior type, execution time, execution frequency, and nursing behavior priority.

[0053] Embodiment three, refer to Figure 1 This embodiment is based on the above-mentioned embodiment, and the sleep stage identification is specifically identifying the sleep stage by using a deep time sequence model according to the patient physiological signal data to obtain a sleep stage sequence, including the following steps: feature extraction and sleep stage classification.

[0054] The feature extraction is specifically extracting time domain features and frequency domain features from the patient physiological signal data in each time period to obtain a patient physiological state sequence.

[0055] The time domain feature extraction is specifically calculating the mean and standard deviation of the heart rate, the heart rate variability index, the mean and standard deviation of the respiratory rate, and the mean and standard deviation of the blood oxygen.

[0056] The frequency domain feature extraction is specifically performing fast Fourier transform on the heart rate, respiratory rate, and blood oxygen saturation, respectively.

[0057] The sleep stage classification is specifically modeling the patient physiological state sequence by constructing a deep time sequence model to generate a high-order time sequence feature vector, and then classifying the high-order time sequence feature vector through a fully connected layer to obtain a sleep stage sequence.

[0058] Preferably, the deep time sequence model is a bidirectional long short-term memory network model.

[0059] Embodiment four, refer to Figure 1 and Figure 2 This embodiment is based on the above-mentioned embodiment, and the intervention window recommendation is used to identify suitable nursing intervention time periods in the patient's sleep process, to maximize the reduction of sleep interference and improve nursing efficiency and patient comfort, specifically, according to the nursing information specification data and the sleep stage sequence, an intervention window recommendation method combining rhythm graph and interference modeling is used to recommend the intervention window, to obtain a candidate time period set, including the following steps: sleep rhythm graph modeling, interference degree modeling, and window recommendation.

[0060] The sleep rhythm graph modeling is used to construct a graph structure of the sleep stage to mine the stability and sensitive period of the sleep rhythm, specifically, according to the sleep stage sequence, each time period is taken as a node, the edges between the nodes are established according to the sleep stage change, and the sleep stage is taken as a node feature to construct a sleep rhythm graph, and the rhythm stability is calculated.

[0061] The method of establishing edges between nodes based on changes in sleep stages specifically means that if the sleep stage changes between adjacent time periods, then the edges are established at node v corresponding to the current time period. i node v corresponding to the next time period j Establish a directed edge e between them ij Otherwise at node v i Establish a self-circulating edge e ii ;

[0062] The formula for calculating the rhythm stability is:

[0063] ;

[0064] ;

[0065] In the formula, R i This represents the rhythm stability of the i-th node, indicating the persistence and regularity of the sleep state, where i is the node index, representing the time period. It is the variance of the sleep stage duration of the i-th node. It is the rhythm regulation coefficient. It is the number of self-loops of the i-th node. Here, N is the total number of edges at the i-th node, N is the total number of nights collected, and n is the night index. It is the duration of the sleep stage corresponding to the i-th node on the nth night. It is the average duration of the sleep stage corresponding to the i-th node in N nights;

[0066] The interference modeling is used to quantify the interference effect of each nursing behavior on different sleep stages. Specifically, based on the nursing information standard data and sleep stage sequence, a standard random forest model is constructed to model the change in sleep stages before and after the nursing behavior, calculate the nursing interference score, and then combine the nursing interference score and rhythm stability to construct the nursing adaptation factor.

[0067] The formula for calculating the nursing interference score is as follows:

[0068] ;

[0069] In the formula, It is the nursing interference score, E[·] is the expected value operator, a k This is the k-th nursing action, where k is the nursing action index, and s j This refers to the sleep stage in the j-th time period, where j is the time period index. It represents the magnitude of sleep stage variation after the nursing intervention is performed in the j-th time period;

[0070] The care adaptation factor is used for evaluating the comprehensive suitability of the care behavior, and the calculation formula is:

[0071] ;

[0072] In the formula, is the care adaptation factor of the kth care behavior in the jth time period, which is used for evaluating whether the jth time period is suitable for performing the kth care behavior, and the higher the care adaptation factor indicates that the rhythm stability is less disturbed, R j is the rhythm stability degree of the jth time period;

[0073] The window recommendation is used for screening a candidate time period set for each care behavior, and specifically, for each care behavior, an adaptation threshold is set, and time periods with a care adaptation factor higher than the adaptation threshold are screened to form a candidate time period set;

[0074] The calculation formula for screening the time periods with a care adaptation factor higher than the adaptation threshold is:

[0075] ;

[0076] In the formula, W k is the candidate time period set of the kth care behavior, is the adaptation threshold of the kth care behavior.

[0077] By performing the above operations, in order to solve the technical problem that in the existing patient care assistance process, the night care intervention time arrangement lacks scientific basis, which easily disturbs the patient's sleep and affects the patient's rest quality and recovery effect, the present scheme creatively adopts an intervention window recommendation method combining rhythm graph and interference modeling to recommend the intervention window, realizes the identification of suitable time periods for each care behavior under the premise of guaranteeing the continuity and comfort of the patient's sleep, and thus reasonably arranges the care intervention time and maximally reduces the influence on the patient's sleep, which is helpful for promoting the patient's recovery.

[0078] In the embodiment, the care strategy optimization is specifically as follows: Figure 1 and Figure 3 The embodiment is based on the above-mentioned embodiments, and the care strategy optimization is specifically as follows: according to the patient physiological signal data, the care information specification data, the sleep stage sequence and the candidate time period set, a multi-constraint reinforcement learning is adopted to optimize the care strategy, and an optimal care scheme is obtained, including the following steps: constraint condition setting, bearing feature modeling, care strategy modeling and strategy optimization.

[0079] The constraint condition setting is specifically setting a unique execution constraint, a care dependency constraint and a time constraint.

[0080] The unique execution constraint is specifically selecting one candidate time period for each nursing behavior at most; the nursing dependency constraint is specifically executing a nursing behavior with high priority before executing a nursing behavior with low priority; and the time constraint is specifically selecting a candidate time period from a respective candidate time period set for each nursing behavior;

[0081] The bearing feature modeling is used for modeling physiological changes of the patient in the candidate time period for each nursing behavior, and specifically includes constructing a patient physiological change vector and calculating an individual bearing capacity index according to patient physiological signal data;

[0082] The calculation formula of the patient physiological change vector is:

[0083] ;

[0084] In the formula, r k,j is a patient physiological change vector after performing the kth nursing behavior in the jth time period, is a heart rate change amount after performing the nursing behavior in the jth time period, is a respiration rate change amount after performing the nursing behavior in the jth time period, is a blood oxygen saturation change amount after performing the nursing behavior in the jth time period;

[0085] The calculation formula of the individual bearing capacity index is:

[0086] ;

[0087] In the formula, is an individual bearing capacity index of the kth nursing behavior, is a candidate time period set size, and ||·||2 is an L2 norm operator;

[0088] The nursing strategy modeling models the nursing strategy optimization problem as a reinforcement learning problem through action space setting, state space setting, and reward function setting;

[0089] The action space setting specifically includes taking a nursing behavior set as an action space, and taking each action as a binary tuple for representing performing the kth nursing behavior in the jth time period, and the calculation formula is:

[0090] A j =(a k ,j);

[0091] In the formula, A j represents performing the kth nursing behavior in the jth time period;

[0092] The state space is set, specifically, the patient physiological state sequence, the nursing behavior execution record sequence and the time period index are spliced into a state vector; the nursing behavior type, the execution time and the execution frequency in the nursing information specification data are arranged in time sequence to construct the nursing behavior execution record sequence;

[0093] The reward function is set, specifically, the reward function is constructed by combining the sleep stage variation amplitude, the individual bearing capacity index and the constraint condition, and the calculation formula is:

[0094] ;

[0095] In the formula, f j is a reward function, is a sleep interference weight, is an individual bearing weight, is a penalty term weight, P1 is a penalty value for violating the unique execution constraint, P2 is a penalty value for violating the nursing dependence constraint, and P3 is a penalty value for violating the time constraint;

[0096] The strategy optimization is specifically constructing a reinforcement learning model based on the set action space and state space, learning the optimal strategy as the optimal nursing scheme by maximizing the cumulative expected reward;

[0097] The calculation formula of the maximum cumulative expected reward is:

[0098] ;

[0099] In the formula, is an optimal strategy, is a strategy that maximizes the cumulative expected reward, is a strategy function, J is the total number of night time periods, is a discount factor with a value range of (0, 1].

[0100] By performing the above operations, in view of the technical problems that in the existing patient nursing assistance process, the nursing plan is formulated depending on artificial experience, the patient state change is not fully considered, and it is difficult to simultaneously consider the nursing efficiency and patient comfort, the present scheme creatively uses multi-constraint reinforcement learning to optimize the nursing strategy, dynamically formulates a nursing plan that conforms to the individual sleep rule and physiological bearing capacity, effectively coordinates the time arrangement of various nursing behaviors, maximally reduces the patient burden, and improves the rehabilitation effect.

[0101] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other presenters can develop. It is also possible, however, that only a single element can be present. Terms such as "a", "an", and "the" are not intended to refer to only a singular form but include the plural form as well. "Including" and like terms are used to indicate the inclusion of one or more elements, but not to the exclusion of other elements. Further, it is to be understood that terms such as "including", "comprising", and "having" are meant to be interpreted dynamically to mean that the terms encompass the item recited and equivalents thereof.

[0102] While the embodiments of the application have been shown and described, it is to be understood that the embodiments described are only by way of example and that modifications, changes, substitutions, and variations can be made to the disclosed embodiments without departing from the spirit of the application.

[0103] The above description of the application and its embodiments is not restrictive, and the embodiments shown in the drawings are only one of the embodiments of the application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired by it, without departing from the purpose of the application, without creative design, similar structure and embodiments of the technical solution can be developed, which should belong to the protection scope of the application.

Claims

1. An artificial intelligence-based patient care assistance system, characterized by: The system comprises a data preparation module, a sleep stage identification module, an intervention window recommendation module, and a care strategy optimization module. The data preparation module is configured to prepare data, obtain physiological signal data and standardized care information data of a patient, and send the physiological signal data to the sleep stage identification module and the care strategy optimization module, and send the standardized care information data to the intervention window recommendation module and the care strategy optimization module. The sleep stage identification module is configured to identify sleep stages, identify sleep stages based on the physiological signal data of the patient by using a deep time sequence model, obtain a sleep stage sequence, and send the sleep stage sequence to the intervention window recommendation module and the care strategy optimization module. The intervention window recommendation module is configured to recommend intervention windows, recommend intervention windows based on the standardized care information data and the sleep stage sequence by using an intervention window recommendation method combining rhythm graphs and interference modeling, obtain a candidate time period set, and send the candidate time period set to the care strategy optimization module. The care strategy optimization module is configured to optimize care strategies, optimize care strategies based on the physiological signal data, the standardized care information data, the sleep stage sequence, and the candidate time period set by using multi-constraint reinforcement learning, and obtain an optimal care plan.

2. The artificial intelligence-based patient care assistant system according to claim 1, characterized by: The intervention window recommendation is configured to identify suitable care intervention time periods in a sleep process of a patient, specifically, recommend intervention windows based on the standardized care information data and the sleep stage sequence by using an intervention window recommendation method combining rhythm graphs and interference modeling, obtain a candidate time period set, and include the following steps: sleep rhythm graph modeling, interference degree modeling, and window recommendation. The sleep rhythm graph modeling is configured to construct a graph structure of sleep stages to mine stability and sensitive time periods of sleep rhythm, specifically, construct a sleep rhythm graph by taking each time period as a node, establishing edges between nodes based on sleep stage changes, and taking sleep stages as node features, and calculate a rhythm stability degree. The edges between nodes are established according to sleep stage changes, specifically, if the sleep stage between adjacent time periods changes, a directed edge e between the nodes v corresponding to the current time period and the node v corresponding to the latter time period is established i , otherwise a self-loop edge e is established between the nodes v i ; j ij ii ;​​ The interference degree modeling is configured to quantify interference effects of each care behavior on different sleep stages, specifically, model sleep stage variation amplitudes of time periods before and after a care behavior by constructing a standard random forest model based on the standardized care information data and the sleep stage sequence, calculate a care interference score, and then construct a care adaptation factor by combining the care interference score and the rhythm stability degree. The window recommendation is configured to screen a candidate time period set for each care behavior, specifically, set an adaptation threshold for each care behavior, screen time periods with adaptation factors higher than the adaptation threshold to form a candidate time period set.

3. The artificial intelligence-based patient care assistance system according to claim 2, characterized by: The care strategy optimization is configured to optimize care strategies, specifically, optimize care strategies based on the physiological signal data, the standardized care information data, the sleep stage sequence, and the candidate time period set by using multi-constraint reinforcement learning, and obtain an optimal care plan, and include the following steps: constraint condition setting, bearing feature modeling, care strategy modeling, and strategy optimization. The constraint condition setting is configured to set a unique execution constraint, a care dependency constraint, and a time constraint. The bearing feature is modeled, and is used for modeling physiological changes of a patient to each nursing behavior in a candidate time period, specifically, constructing a patient physiological change vector according to patient physiological signal data, and calculating an individual bearing capacity index; The nursing strategy is modeled, and the nursing strategy optimization problem is modeled as a reinforcement learning problem through action space setting, state space setting and reward function setting; The strategy optimization is specifically constructing a reinforcement learning model on the basis of the set action space and state space, learning an optimal strategy as an optimal nursing scheme by maximizing cumulative expected reward.

4. The artificial intelligence-based patient nursing assistance system according to claim 3, characterized in that: The unique execution constraint is specifically that each nursing behavior is executed in at most one candidate time period; the nursing dependency constraint is specifically that a nursing behavior with high priority must be executed before a nursing behavior with low priority; and the time constraint is specifically that each nursing behavior can only be selected from a respective candidate time period set.

5. The artificial intelligence-based patient care assistance system according to claim 4, characterized by: The action space setting is specifically that a nursing behavior set is taken as an action space, and each action is set as a binary tuple for representing that a kth nursing behavior is executed in a jth time period; The state space setting is specifically that a patient physiological state sequence, a nursing behavior execution record sequence and a time period index are spliced into a state vector; and the nursing behavior execution record sequence is constructed by arranging nursing behavior types, execution times and execution frequencies in the nursing information specification data in time sequence. The reward function setting is specifically that a reward function is constructed in combination with a sleep stage variation amplitude, an individual bearing capacity index and a constraint condition.

6. The artificial intelligence-based patient care assistance system according to claim 5, characterized by: The sleep stage identification is specifically that a deep time sequence model is used to identify a sleep stage according to patient physiological signal data, and a sleep stage sequence is obtained, including the following steps: feature extraction and sleep stage classification; The feature extraction is specifically that time domain feature extraction and frequency domain feature extraction are performed on patient physiological signal data in each time period, and a patient physiological state sequence is obtained; The time domain feature extraction is specifically that a heart rate mean value, a standard deviation and a heart rate variation index are calculated, and a respiration rate mean value and a respiration interval standard deviation are calculated, and a blood oxygen mean value, a standard deviation and a low blood oxygen duration are calculated; The frequency domain feature extraction is specifically that fast Fourier transform is performed on a heart rate, a respiration rate and a blood oxygen saturation, respectively; The sleep stage classification is specifically that a deep time sequence model is constructed to model the patient physiological state sequence, and a high-order time sequence feature vector is generated, and then the high-order time sequence feature vector is classified through a full connection layer to obtain a sleep stage sequence.

7. The artificial intelligence-based patient care assistance system according to claim 6, characterized by: The data preparation is specifically that multi-channel physiological signal data of a patient is continuously collected in a night period through a sensor, and preprocessed to obtain patient physiological signal data, and nursing record data of each patient is collected from a ward management system, and standardized to obtain nursing information specification data; The patient physiological signal data includes a heart rate, a respiration rate and a blood oxygen saturation. The nursing information specification data includes a nursing behavior type, an execution time, an execution frequency, and a nursing behavior priority. The nursing information specification data includes a nursing behavior type, an execution time, an execution frequency, and a nursing behavior priority.

Images