Newborn nursing auxiliary system based on artificial intelligence

Abstract

The invention discloses a newborn nursing auxiliary system based on artificial intelligence, which belongs to the technical field of intelligent newborn nursing and comprises a data preparation module, a preprocessing module, a sleep monitoring module, a nursing strategy adjusting module and a newborn nursing auxiliary module. According to the method, sleep monitoring is carried out by adopting a parameter-optimized bidirectional long-short-term transformer model, physiological data, environmental data and historical nursing records are synthesized for analysis, multi-level and multi-dimensional features of newborn sleep can be fully captured, the training process is accelerated through an improved whale optimization algorithm, the overall performance of the model is improved, and the training efficiency is improved. Therefore, efficient and accurate sleep monitoring is realized; nursing strategy adjustment is carried out by adopting deep reinforcement learning, the nursing strategy is dynamically adjusted, more diversified neonatal nursing scenes can be adapted, neonatal nursing requirements can be better met, more accurate and efficient nursing support is provided for neonates, neonatal nursing quality is improved, workload of nursing personnel is relieved, and nursing efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of neonatal intelligent care, and specifically refers to a neonatal care assistance system based on artificial intelligence. Background Art

[0002] A neonatal care assistance system based on artificial intelligence is an intelligent system that uses artificial intelligence technology to perform real-time analysis on various data generated during the neonatal care process to optimize nursing strategies. It aims to assist nursing staff in making more scientific and effective nursing decisions, better meeting the needs of newborns and optimizing the sleep quality of newborns, thereby improving the nursing level and ensuring the healthy growth of newborns.

[0003] However, in the existing neonatal care assistance process, there are technical problems. The sleep state of newborns is comprehensively affected by factors such as indoor temperature, light, and historical nursing measures. Most of the existing sleep monitoring only considers the physiological data of newborns and fails to comprehensively consider various influencing factors, resulting in poor accuracy of sleep monitoring. There is also the technical problem that the neonatal care scenario has dynamic variability, and it is difficult for nursing staff to always pay attention to the real-time status of each newborn, resulting in untimely response to the needs of newborns. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the existing technology, the present invention provides a neonatal care assistance system based on artificial intelligence. In the existing neonatal care assistance process, there are technical problems. The sleep state of newborns is comprehensively affected by factors such as indoor temperature, light, and historical nursing measures. Most of the existing sleep monitoring only considers the physiological data of newborns and fails to comprehensively consider various influencing factors, resulting in poor accuracy of sleep monitoring. In response to this, this solution creatively uses a parameter-optimized bidirectional long short-term transformer model for sleep monitoring, analyzes comprehensive physiological data, environmental data, and historical nursing records, effectively considers the influence of various factors on the sleep state of newborns, and can fully capture the multi-level and multi-dimensional characteristics of newborn sleep. And by improving the whale optimization algorithm to accelerate the training process and improve the overall performance of the model, thereby achieving efficient and accurate sleep monitoring. In the existing neonatal care assistance process, there are technical problems. The neonatal care scenario has dynamic variability, and it is difficult for nursing staff to always pay attention to the real-time status of each newborn, resulting in untimely response to the needs of newborns. In response to this, this solution creatively uses deep reinforcement learning to adjust nursing strategies, can real-time perceive the sleep state, physiological state, and environmental changes of newborns, and dynamically adjust nursing strategies, which can not only adapt to more diverse neonatal care scenarios, but also better meet the neonatal care needs, thereby providing more accurate and efficient nursing support for newborns, improving the quality of neonatal care, and helping to reduce the workload of nursing staff and improve nursing efficiency.

[0005] The technical solution adopted by the present invention is as follows: The neonatal care assistance system based on artificial intelligence provided by the present invention includes a data preparation module, a preprocessing module, a sleep monitoring module, a nursing strategy adjustment module, and a neonatal care assistance module;

[0006] The data preparation module is used for data preparation. Through data preparation, the original neonatal care data is obtained, and the original neonatal care data is sent to the preprocessing module;

[0007] The preprocessing module is used for data preprocessing. Through data preprocessing, the optimized neonatal care data is obtained, and the optimized neonatal care data is sent to the sleep monitoring module and the nursing strategy adjustment module;

[0008] The sleep monitoring module, based on the optimized neonatal care data, uses a parameter-optimized bidirectional long short-term transformer model for sleep monitoring to obtain the neonatal sleep state, and sends the neonatal sleep state to the nursing strategy adjustment module and the neonatal care assistance module;

[0009] The nursing strategy adjustment module, based on the neonatal sleep state, the neonatal physiological data and environmental data in the optimized neonatal care data, uses deep reinforcement learning for nursing strategy adjustment to obtain the optimized neonatal care strategy, and sends the optimized neonatal care strategy to the neonatal care assistance module;

[0010] The neonatal care assistance module is used for neonatal care assistance to obtain a neonatal care recommendation plan.

[0011] Further, in the data preparation module, the data preprocessing is specifically to obtain the original neonatal care data by collecting neonatal physiological data, historical nursing records, and environmental data;

[0012] The neonatal physiological data includes neonatal heart rate, electroencephalogram signal, body temperature, and blood oxygen saturation;

[0013] The historical nursing records include neonatal sleep records, feeding records, and diaper change records;

[0014] The environmental data includes environmental temperature, humidity, noise level, and light intensity.

[0015] Further, in the preprocessing module, the data preprocessing is specifically to obtain the optimized neonatal care data by performing data cleaning, time alignment, and data standardization on the original neonatal care data;

[0016] The data cleaning includes operations such as outlier filtering, duplicate value removal, and missing value filling;

[0017] The time alignment specifically aligns the time axes of the neonatal physiological data, historical nursing records, and environmental data in the original neonatal care data;

[0018] The data standardization specifically performs min-max normalization on the numerical data in the original neonatal care data.

[0019] Furthermore, in the sleep monitoring module, there are a bidirectional long short-term transformer model construction unit, a parameter tuning algorithm construction unit, a sleep monitoring model construction unit, and a sleep monitoring result generation unit, including the following:

[0020] The bidirectional long short-term transformer model construction unit constructs a standard bidirectional long short-term neural network including an input layer, a forward long short-term memory layer, a backward long short-term memory layer, an activation layer, and an output layer, and constructs a standard transformer to receive the output hidden state of the standard bidirectional long short-term neural network to obtain a bidirectional long short-term transformer model;

[0021] The parameter tuning algorithm construction unit constructs an improved whale optimization algorithm, including the following:

[0022] Improved population initialization, which is used to enhance population diversity. Specifically, in the search space, the whale positions are initialized through chaotic mapping, and the calculation formula is:

[0023] ;

[0024] ;

[0025] In the formula, S k+1 is the chaotic value generated by the (k + 1)-th chaotic mapping, k is the index of the chaotic mapping times, S k is the chaotic value generated by the k-th chaotic mapping. The initial value of the chaotic value is a random number with a value range of (0, 1), sit ij is the position component of the i-th whale in the j-th dimensional search space, i is the whale index, j is the search space dimension index, lb j is the lower bound of the j-th dimensional search space, ub j is the upper bound of the j-th dimensional search space, Sit i is the position of the i-th whale, and the whale position is used to represent the model parameters to be optimized, sit i1 is the position component of the i-th whale in the 1st dimensional search space, sit i2 is the position component of the i-th whale in the 2nd dimensional search space, sit id is the position component of the i-th whale in the d-th dimensional search space, d is the search space dimension, and the search space dimension is equal to the number of model parameters to be optimized;

[0026] Evaluate the fitness value. Specifically, use the model loss function as the fitness function, and calculate the fitness value of each whale position through the fitness function;

[0027] Update the whale position. Specifically, when the absolute value of the step size adjustment factor is less than 1, perform the bubble-net attacking behavior to update the whale position; otherwise, perform the prey searching behavior to update the whale position. Replace the linear convergence factor in the whale optimization algorithm with a non-linear convergence factor to improve the algorithm convergence speed and balance;

[0028] The calculation formula of the step size adjustment factor is:

[0029] ;

[0030] In the formula, A is the step size adjustment factor, a is the non-linear convergence factor, r 1 is the first random number with a value range of [0, 1];

[0031] The calculation formula of the non-linear convergence factor is:

[0032] ;

[0033] In the formula, sin(·) is the sine function, t is the iteration number index, t max is the maximum number of iterations, π is the pi;

[0034] The calculation formula for updating the whale position by performing the bubble-net attacking behavior is:

[0035] ;

[0036] In the formula, Sit(t + 1) is the updated whale position, specifically referring to the whale position at the (t + 1)-th iteration, Sit best (t) is the optimal whale position at the t-th iteration. The optimal whale position specifically refers to the whale position with the minimum fitness value. Sit(t) is the current whale position, specifically referring to the whale position at the t-th iteration. e is the base of the natural logarithm, b is the spiral shape adjustment factor, r 2 is the second random number with a value range of [-1, 1], p is a random discrimination factor with a value range of (0, 1), which is used to determine the path of the bubble-net attacking behavior, and C is the perturbation factor;

[0037] The calculation formula for updating the whale position by performing the prey searching behavior is:

[0038] ;

[0039] In the formula, Sit rand (t) is the randomly selected whale position at the t-th iteration;

[0040] Update and iteration, specifically, by repeatedly executing the evaluation of the fitness value and the update of the whale position, the iteration update of the whale position is carried out until the maximum number of iterations is reached, and the whale position with the minimum fitness value is used as the optimal parameters of the model;

[0041] The sleep monitoring model construction unit obtains the sleep monitoring model by training the bidirectional long short-term transformer model and tuning the parameters using the improved whale optimization algorithm;

[0042] The sleep monitoring result generation unit performs sleep monitoring through the sleep monitoring model to obtain the neonatal sleep state, and the neonatal sleep state includes sleep stages, total sleep duration, and number of awakenings.

[0043] Furthermore, in the nursing strategy adjustment module, there are an agent initialization unit, a value function initialization unit, a value update network construction unit, a nursing strategy network construction unit, a nursing strategy adjustment model training unit, and a neonatal nursing optimization strategy generation unit, including the following:

[0044] The agent initialization unit initializes the agent by defining the state space, action space, and reward function;

[0045] The state space includes neonatal sleep state, neonatal physiological data, and environmental data;

[0046] The action space includes temperature adjustment, light adjustment, and nursing frequency adjustment;

[0047] The calculation formula of the reward function is:

[0048] ;

[0049] In the formula, rf z is the reward value at the z-th time step, z is the time step index, Q z is the nursing quality at the z-th time step, Q min is the minimum standard of nursing quality to ensure that the neonatal physiological data is within the normal range, is the nursing risk value at the z-th time step, which is used to reflect the negative effects brought by the nursing strategy. When it means that there is no nursing risk;

[0050] The value function initialization unit initializes the state value function and the action value function;

[0051] The calculation formula of the state value function is:

[0052] ;

[0053] In the formula, is to adopt a nursing strategy in the current state s The cumulative value of is the state-value function is the nursing strategy is to adopt the nursing strategy The expected value function of the state-action pair when taking the nursing strategy, which is used to evaluate the cumulative value. Z is the maximum time step is the discount factor is the state-action pair generated by the nursing strategy s 1 is the initial state of the agent, and s is the current state

[0054] The calculation formula of the action-value function is

[0055] ;

[0056] In the formula is the cumulative value of executing the current action a in the current state s is the action-value function, a is the current action, a 1 is the initial action of the agent

[0057] The value update network construction unit takes the hypergraph interference network as the value update network, and the calculation formula is

[0058] ;

[0059] In the formula is the gradient of the value update network loss function is the parameter of the value update network is the gradient of the action-value function, y z is the target value at the z-th time step, which is used to represent the ideal return after selecting the current action in the current state

[0060] The nursing strategy network construction unit constructs the nursing strategy network, and the calculation formula is

[0061] ;

[0062] In the formula is the gradient of the nursing strategy network loss function is the parameter of the nursing strategy network is the current nursing strategy is the expected value function under the current nursing strategy is the gradient of the logarithm of the nursing strategy, and log is the logarithmic function is the advantage function, which is used to measure the relative value of taking the action a z in the state s z a zis the action at the z-th time step, s z is the state at the z-th time step;

[0063] The nursing strategy adjustment model training unit constructs a deep reinforcement learning model through the agent initialization unit, the value function initialization unit, the value update network construction unit, and the nursing strategy network construction unit, updates parameters by training the deep reinforcement learning model, optimizes the value update network and the nursing strategy network, and obtains a nursing strategy adjustment model;

[0064] The neonatal care optimization strategy generation unit adjusts the nursing strategy through the nursing strategy adjustment model to obtain a neonatal care optimization strategy, and the neonatal care optimization strategy includes temperature adjustment data, light adjustment data, and nursing frequency adjustment data.

[0065] Furthermore, in the neonatal care assistance module, the neonatal care assistance specifically performs neonatal care assistance based on the neonatal sleep state and the neonatal care optimization strategy to obtain a neonatal care recommendation plan, and real-time displays the changes in neonatal physiological data and neonatal sleep state for evaluating the effect after the implementation of the neonatal care recommendation plan.

[0066] The beneficial effects achieved by the present invention using the above solution are as follows:

[0067] (1) Aiming at the technical problem that in the existing neonatal care assistance process, the neonatal sleep state is comprehensively affected by factors such as indoor temperature, light, and historical nursing measures, and most existing sleep monitoring only considers neonatal physiological data and fails to comprehensively consider multiple influencing factors, resulting in poor sleep monitoring accuracy. This solution creatively uses a parameter-optimized bidirectional long short-term transformer model for sleep monitoring, analyzes physiological data, environmental data, and historical nursing records comprehensively, effectively considers the influence of multiple factors on the neonatal sleep state, can fully capture the multi-level and multi-dimensional characteristics of neonatal sleep, and speeds up the training process through an improved whale optimization algorithm, improving the overall performance of the model, and thus realizing efficient and accurate sleep monitoring;

[0068] (2) Aiming at the technical problem that in the existing neonatal care assistance process, the neonatal care scenario has dynamic variability, and it is difficult for nursing staff to always pay attention to the real-time state of each neonate, resulting in untimely response to neonatal needs. This solution creatively uses deep reinforcement learning for nursing strategy adjustment, can perceive the neonatal sleep state, physiological state, and environmental changes in real time, and dynamically adjust the nursing strategy, which can not only adapt to more diverse neonatal care scenarios, but also better meet the neonatal care needs, thus providing more precise and efficient nursing support for neonates, improving the quality of neonatal care, and helping to reduce the workload of nursing staff and improve nursing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic flow chart of the neonatal care assistance system based on artificial intelligence provided by the present invention;

[0070] Figure 2 It is a schematic flow chart of the sleep monitoring module;

[0071] Figure 3 It is a schematic flow chart of the parameter tuning algorithm construction unit;

[0072] Figure 4 It is a schematic flow chart of the nursing strategy adjustment module.

[0073] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0075] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. 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 invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention.

[0076] Embodiment 1, refer to Figure 1 , a neonatal care assistance system based on artificial intelligence provided by the present invention includes a data preparation module, a preprocessing module, a sleep monitoring module, a nursing strategy adjustment module, and a neonatal care assistance module;

[0077] The data preparation module is used for data preparation. Through data preparation, the original neonatal care data is obtained, and the original neonatal care data is sent to the preprocessing module;

[0078] The preprocessing module is used for data preprocessing. Through data preprocessing, the optimized neonatal care data is obtained, and the optimized neonatal care data is sent to the sleep monitoring module and the nursing strategy adjustment module;

[0079] The sleep monitoring module uses a parameter-optimized bidirectional long-short-term transformer model to perform sleep monitoring based on the neonatal care optimization data to obtain the neonatal sleep state, and sends the neonatal sleep state to the nursing strategy adjustment module and the neonatal care auxiliary module;

[0080] The nursing strategy adjustment module uses deep reinforcement learning to adjust the nursing strategy according to the neonatal sleep state, the neonatal physiological data and the environmental data in the neonatal nursing optimization data, obtains the neonatal nursing optimization strategy, and sends the neonatal nursing optimization strategy to the neonatal nursing auxiliary module;

[0081] The neonatal care assistance module is used to assist in neonatal care and obtain a neonatal care recommendation plan.

[0082] Example 2, see Figure 1 , this embodiment is based on the above embodiment, in the data preparation module, the data preprocessing is specifically to obtain the original data of neonatal care by collecting neonatal physiological data, historical nursing records and environmental data;

[0083] The neonatal physiological data includes the neonatal heart rate, electroencephalogram signal, body temperature and blood oxygen saturation;

[0084] The historical nursing records include sleeping records, feeding records, and diaper changing records for the newborn;

[0085] The environmental data includes environmental temperature, humidity, noise level and light intensity.

[0086] Example 3, see Figure 1 , this embodiment is based on the above embodiment, in the preprocessing module, the data preprocessing is specifically to obtain the neonatal care optimization data by performing data cleaning, time alignment and data standardization on the neonatal care raw data;

[0087] The data cleaning includes outlier filtering, duplicate value removal and missing value filling operations;

[0088] The time alignment specifically aligns the time axes of the neonatal physiological data, historical nursing records and environmental data in the neonatal nursing raw data;

[0089] The data standardization specifically involves performing maximum and minimum normalization processing on the numerical data in the original data of neonatal care.

[0090] Example 4, see Figure 1 , Figure 2 and Figure 3, this embodiment is based on the above embodiment. In the sleep monitoring module, there are a bidirectional long short-term transformer model construction unit, a parameter tuning algorithm construction unit, a sleep monitoring model construction unit, and a sleep monitoring result generation unit, including the following:

[0091] The bidirectional long short-term transformer model construction unit constructs a standard bidirectional long short-term neural network including an input layer, a forward long short-term memory layer, a backward long short-term memory layer, an activation layer, and an output layer, and constructs a standard transformer to receive the output hidden state of the standard bidirectional long short-term neural network to obtain a bidirectional long short-term transformer model;

[0092] The parameter tuning algorithm construction unit constructs an improved whale optimization algorithm, including the following:

[0093] Improved population initialization, which is used to enhance population diversity. Specifically, in the search space, the whale positions are initialized through chaotic mapping, and the calculation formula is:

[0094] ;

[0095] ;

[0096] In the formula, S k+1 is the chaotic value generated by the (k + 1)-th chaotic mapping, k is the index of the chaotic mapping times, S k is the chaotic value generated by the k-th chaotic mapping. The initial value of the chaotic value is a random number with a value range of (0, 1). sit ij is the position component of the i-th whale in the j-th dimension of the search space, i is the whale index, j is the search space dimension index, lb j is the lower bound of the j-th dimension of the search space, ub j is the upper bound of the j-th dimension of the search space, Sit i is the position of the i-th whale, and the whale position is used to represent the model parameters to be optimized. sit i1 is the position component of the i-th whale in the 1st dimension of the search space, sit i2 is the position component of the i-th whale in the 2nd dimension of the search space, sit id is the position component of the i-th whale in the d-th dimension of the search space, d is the search space dimension, and the search space dimension is equal to the number of model parameters to be optimized;

[0097] Evaluate the fitness value. Specifically, the model loss function is used as the fitness function, and the fitness value of each whale position is calculated through the fitness function;

[0098] Whale position update. Specifically, when the absolute value of the step size adjustment factor is less than 1, the bubble-net attacking behavior is executed to update the whale position; otherwise, the prey searching behavior is executed to update the whale position. The linear convergence factor in the whale optimization algorithm is replaced with a non-linear convergence factor to improve the algorithm convergence speed and balance;

[0099] The calculation formula for the step size adjustment factor is:

[0100] ;

[0101] where A is the step size adjustment factor, a is the non-linear convergence factor, and r 1 is the first random number with a value range of [0, 1];

[0102] The calculation formula for the non-linear convergence factor is:

[0103] ;

[0104] where sin(·) is the sine function, t is the iteration number index, t max is the maximum number of iterations, and π is the pi;

[0105] The calculation formula for updating the whale position by executing the bubble-net attacking behavior is:

[0106] ;

[0107] where Sit(t + 1) is the updated whale position, specifically the whale position at the (t + 1)-th iteration, Sit best (t) is the optimal whale position at the t-th iteration. The optimal whale position specifically refers to the whale position with the minimum fitness value. Sit(t) is the current whale position, specifically the whale position at the t-th iteration. e is the base of the natural logarithm, b is the spiral shape adjustment factor, and r 2 is the second random number with a value range of [-1, 1], and p is a random discrimination factor with a value range of (0, 1) used to determine the bubble-net attacking behavior path, and C is the perturbation factor;

[0108] The calculation formula for the perturbation factor is:

[0109] ;

[0110] where r 3 is the third random number with a value range of [0, 1];

[0111] The calculation formula for updating the whale position by executing the prey searching behavior is:

[0112] ;

[0113] where Sit rand (t) is the position of the randomly selected whale at the t-th iteration;

[0114] Update the iteration. Specifically, by repeatedly performing the evaluation of the fitness value and the update of the whale position, the iteration update of the whale position is carried out until the maximum number of iterations is reached, and the whale position with the minimum fitness value is used as the optimal parameters of the model;

[0115] The sleep monitoring model construction unit obtains a sleep monitoring model by training a bidirectional long short-term transformer model and tuning parameters using an improved whale optimization algorithm;

[0116] The sleep monitoring result generation unit performs sleep monitoring through the sleep monitoring model to obtain the neonatal sleep state, where the neonatal sleep state includes sleep stages, total sleep duration, and number of awakenings;

[0117] By performing the above operations, in view of the technical problem that in the existing neonatal care assistance process, the neonatal sleep state is comprehensively affected by factors such as indoor temperature, light, and historical care measures, and most of the existing sleep monitoring only considers neonatal physiological data and fails to comprehensively consider various influencing factors, resulting in poor sleep monitoring accuracy. This solution creatively uses a parameter-optimized bidirectional long short-term transformer model for sleep monitoring, analyzes physiological data, environmental data, and historical care records comprehensively, effectively considers the influence of various factors on the neonatal sleep state, can fully capture the multi-level and multi-dimensional characteristics of neonatal sleep, and speeds up the training process through an improved whale optimization algorithm, improving the overall performance of the model, and thus realizing efficient and accurate sleep monitoring.

[0118] Example Five, refer to Figure 1 and Figure 4 , based on the above example, in the nursing strategy adjustment module, there are an agent initialization unit, a value function initialization unit, a value update network construction unit, a nursing strategy network construction unit, a nursing strategy adjustment model training unit, and a neonatal care optimization strategy generation unit, including the following contents:

[0119] The agent initialization unit initializes the agent by defining the state space, action space, and reward function;

[0120] The state space includes the neonatal sleep state, neonatal physiological data, and environmental data;

[0121] The action space includes temperature adjustment, light adjustment, and nursing frequency adjustment;

[0122] The calculation formula of the reward function is:

[0123] ;

[0124] wherein, rf z is the reward value at the z-th time step, z is the time step index, and Q z is the nursing quality at the z-th time step, and Q min is the minimum standard of nursing quality, which is used to ensure that the physiological data of the newborn is within the normal range. is the nursing risk value at the z-th time step, which is used to reflect the negative effect brought by the nursing strategy. When it means that there is no nursing risk.

[0125] The value function initialization unit initializes the state value function and the action value function;

[0126] The calculation formula of the state value function is:

[0127] ;

[0128] wherein, is the cumulative value of taking the nursing strategy in the current state s, is the state value function, is the nursing strategy, is the expected value function of the state-action pair when taking the nursing strategy for evaluating the cumulative value, Z is the maximum time step, is the discount factor, is the state-action pair generated by the nursing strategy , and s 1 is the initial state of the agent, and s is the current state;

[0129] The calculation formula of the action value function is:

[0130] ;

[0131] wherein, is the cumulative value of executing the current action a in the current state s, is the action value function, a is the current action, and a 1 is the initial action of the agent;

[0132] The value update network construction unit uses the hypergraph interference network as the value update network, and the calculation formula is:

[0133] ;

[0134] wherein, is the gradient of the value update network loss function, is the parameter of the value update network, is the gradient of the action value function, y z is the target value at the z-th time step, which is used to represent the ideal return after selecting the current action in the current state;

[0135] The nursing strategy network construction unit constructs the nursing strategy network, and the calculation formula is:

[0136] ;

[0137] In the formula, is the gradient of the nursing strategy network loss function, are the parameters of the nursing strategy network, is the current nursing strategy, is the expected value function under the current nursing strategy, is the gradient of the logarithm of the nursing strategy, and log is the logarithmic function, is the advantage function, which is used to measure the relative value of taking action a z in state s z The relative value of, a z is the action at the z-th time step, s z is the state at the z-th time step;

[0138] The nursing strategy adjustment model training unit constructs a deep reinforcement learning model through the agent initialization unit, the value function initialization unit, the value update network construction unit, and the nursing strategy network construction unit, and updates the parameters by training the deep reinforcement learning model to optimize the value update network and the nursing strategy network to obtain a nursing strategy adjustment model;

[0139] The neonatal nursing optimization strategy generation unit adjusts the nursing strategy through the nursing strategy adjustment model to obtain a neonatal nursing optimization strategy, and the neonatal nursing optimization strategy includes temperature adjustment data, light adjustment data, and nursing frequency adjustment data;

[0140] By performing the above operations, in view of the technical problem that in the existing neonatal nursing assistance process, the neonatal nursing scenario is dynamically changing, and it is difficult for nursing staff to always pay attention to the real-time state of each neonate, resulting in untimely response to the needs of neonates, this solution creatively uses deep reinforcement learning for nursing strategy adjustment, which can real-time perceive the sleep state, physiological state and environmental changes of neonates, and dynamically adjust the nursing strategy, which can not only adapt to more diverse neonatal nursing scenarios, but also better meet the neonatal nursing needs, thereby providing more accurate and efficient nursing support for neonates, improving the quality of neonatal nursing, and helping to reduce the workload of nursing staff and improve the nursing efficiency.

[0141] Example 6, refer to Figure 1, based on the above embodiment, in the neonatal care assistance module, the neonatal care assistance specifically refers to performing neonatal care assistance according to the neonatal sleep state and neonatal care optimization strategy, obtaining a neonatal care recommendation plan, and real-time displaying the changes in neonatal physiological data and neonatal sleep state for evaluating the effect after the implementation of the neonatal care recommendation plan.

[0142] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0143] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention.

[0144] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. The artificial intelligence-based neonatal care assistance system is characterized by: It includes data preparation module, preprocessing module, sleep monitoring module, nursing strategy adjustment module and neonatal nursing assistance module; The data preparation module is used for data preparation, obtains raw data of neonatal care through data preparation, and sends the raw data of neonatal care to the preprocessing module; The preprocessing module is used for data preprocessing, and obtains neonatal care optimization data through data preprocessing, and sends the neonatal care optimization data to the sleep monitoring module and the nursing strategy adjustment module; The sleep monitoring module uses a parameter-optimized bidirectional long-short-term transformer model to perform sleep monitoring based on the neonatal care optimization data to obtain the neonatal sleep state, and sends the neonatal sleep state to the nursing strategy adjustment module and the neonatal care auxiliary module; The nursing strategy adjustment module uses deep reinforcement learning to adjust the nursing strategy according to the neonatal sleep state, the neonatal physiological data and the environmental data in the neonatal nursing optimization data, obtains the neonatal nursing optimization strategy, and sends the neonatal nursing optimization strategy to the neonatal nursing auxiliary module; The neonatal care assistance module is used to assist in neonatal care and obtain a neonatal care recommendation plan.

2. The artificial intelligence-based neonatal care assistance system according to claim 1, characterized in that: In the sleep monitoring module, there are a bidirectional long-term and short-term transformer model building unit, a parameter tuning algorithm building unit, a sleep monitoring model building unit and a sleep monitoring result generation unit, including the following contents: A bidirectional long-short-term transformer model construction unit constructs a standard bidirectional long-short-term neural network including an input layer, a forward long-short-term memory layer, a backward long-short-term memory layer, an activation layer, and an output layer, and constructs a standard transformer to receive the output hidden state of the standard bidirectional long-short-term neural network to obtain a bidirectional long-short-term transformer model; Parameter tuning algorithm construction unit, building an improved whale optimization algorithm; The sleep monitoring model building unit obtains the sleep monitoring model by training the bidirectional long-short term transformer model and using the improved whale optimization algorithm to perform parameter tuning; The sleep monitoring result generating unit performs sleep monitoring through a sleep monitoring model to obtain the sleeping state of the newborn.

3. The artificial intelligence-based neonatal care assistance system according to claim 2, characterized in that: In the parameter tuning algorithm construction unit, the construction of the improved whale optimization algorithm includes the following contents: Improved population initialization is used to improve population diversity. Specifically, in the search space, the position of the whale is initialized by chaotic mapping. The calculation formula is: ； ； In the formula, S k+1 is the chaotic value generated by the k+1th chaotic mapping, k is the chaotic mapping index, S k is the chaotic value generated by the kth chaotic mapping. The initial value of the chaotic value is a random number in the range of (0,1). ij is the position component of the ith whale in the jth dimensional search space, i is the whale index, j is the search space dimension index, and lb j is the lower bound of the j-th dimension search space, ub j is the upper bound of the j-th dimension search space, Sit i is the i-th whale position, which is used to represent the model parameters to be optimized. i1 is the position component of the ith whale in the first-dimensional search space, sit i2 is the position component of the ith whale in the second-dimensional search space, sit id is the position component of the ith whale in the d-dimensional search space, where d is the search space dimension, which is equal to the number of model parameters to be optimized; Evaluate the fitness value, specifically, use the model loss function as the fitness function, and calculate the fitness value of each whale position through the fitness function; Whale position update: when the absolute value of the step adjustment factor is less than 1, the bubble attack behavior is executed to update the whale position; otherwise, the prey search behavior is executed to update the whale position; the linear convergence factor in the whale optimization algorithm is replaced with a nonlinear convergence factor to improve the convergence speed and balance of the algorithm; The calculation formula of the step size adjustment factor is: ； Where A is the step size adjustment factor, a is the nonlinear convergence factor, and r1 is the first random number in the range of [0,1]; The calculation formula of the nonlinear convergence factor is: ； Where sin(·) is the sine function, t is the iteration index, and t max is the maximum number of iterations, π is the circumference of a circle; The calculation formula for updating the whale position by executing the bubble attack behavior is: ； Where Sit(t+1) is the updated whale position, specifically the whale position at the t+1th iteration, and Sit best (t) is the optimal whale position at the t-th iteration, and the optimal whale position specifically refers to the whale position with the smallest fitness value. Sit(t) is the current whale position, specifically the whale position at the t-th iteration. e is the base of the natural logarithm, b is the spiral shape adjustment factor, r2 is the second random number in the range of [-1,1], p is a random discriminant factor in the range of (0,1), which is used to determine the path of the bubble attack behavior, and C is the disturbance factor. The calculation formula for updating the whale's position by executing the prey-seeking behavior is: ； In the formula, Sit rand (t) is the randomly selected whale position at the tth iteration; The updating iteration is specifically to iteratively update the whale position by repeatedly executing the fitness value evaluation and the whale position update until the maximum number of iterations is reached, and the whale position with the smallest fitness value is used as the optimal parameter of the model.

4. The artificial intelligence-based neonatal care assistance system according to claim 3, characterized in that: In the nursing strategy adjustment module, there are an agent initialization unit, a value function initialization unit, a value update network construction unit, a nursing strategy network construction unit, a nursing strategy adjustment model training unit and a neonatal nursing optimization strategy generation unit, including the following: The agent initialization unit initializes the agent by defining the state space, action space and reward function; The state space includes the neonatal sleep state, neonatal physiological data and environmental data; The action space includes temperature adjustment, light adjustment and care frequency adjustment; The calculation formula of the reward function is: ； In the formula, rf z is the reward value at the zth time step, z is the time step index, Q z is the quality of care at the zth time step, Q min It is the minimum standard of nursing quality, used to ensure that the physiological data of newborns are within the normal range. is the nursing risk value at the zth time step, which is used to reflect the negative effects of the nursing strategy. When , it means there is no nursing risk; Value function initialization unit, initializing state value function and action value function; The calculation formula of the state value function is: ； In the formula, Is to adopt nursing strategies under the current status The cumulative value of is the state value function, It is a nursing strategy. Is to adopt nursing strategies The expected value function of the state-action pair at time is used to evaluate the cumulative value, Z is the maximum time step, is the discount factor, By nursing strategy The generated state-action pair, s1 is the initial state of the agent, and s is the current state; The calculation formula of the action value function is: ； In the formula, is the cumulative value of executing the current action a in the current state s, is the action value function, a is the current action, and a1 is the initial action of the agent; The value update network construction unit uses the hypergraph interference network as the value update network, and the calculation formula is: ； In the formula, is the value update network loss function gradient, is the value update network parameter, is the action value function gradient, y z is the target value at the zth time step, which is used to represent the ideal reward after selecting the current action in the current state; Nursing strategy network construction unit, to construct the nursing strategy network, the calculation formula is: ； In the formula, is the gradient of the loss function of the nursing strategy network, is the nursing strategy network parameter, is the current nursing strategy, is the expected value function under the current nursing strategy, is the gradient of the logarithm of the nursing strategy, log is a logarithmic function, is the advantage function, used to measure the z Take action a z The relative value of a z is the action at the zth time step, s z is the state at the zth time step; The nursing strategy adjustment model training unit constructs a deep reinforcement learning model through the agent initialization unit, the value function initialization unit, the value update network construction unit and the nursing strategy network construction unit, updates parameters by training the deep reinforcement learning model, optimizes the value update network and the nursing strategy network, and obtains the nursing strategy adjustment model; The neonatal nursing optimization strategy generating unit adjusts the nursing strategy through the nursing strategy adjustment model to obtain the neonatal nursing optimization strategy, wherein the neonatal nursing optimization strategy includes temperature adjustment data, light adjustment data and nursing frequency adjustment data.

5. The artificial intelligence-based neonatal care assistance system according to claim 4, characterized in that: In the neonatal care assistance module, the neonatal care assistance is specifically to perform neonatal care assistance based on the neonatal sleep status and neonatal care optimization strategy, obtain a neonatal care recommendation plan, and display the changes in the neonatal physiological data and the neonatal sleep status in real time, which is used to evaluate the effect of the neonatal care recommendation plan after implementation.

6. The artificial intelligence-based neonatal care assistance system according to claim 5, characterized in that: In the preprocessing module, the data preprocessing specifically includes obtaining the neonatal care optimization data by performing data cleaning, time alignment and data standardization on the neonatal care raw data; The data cleaning includes outlier filtering, duplicate value removal and missing value filling operations; The time alignment specifically aligns the time axes of the neonatal physiological data, historical nursing records and environmental data in the neonatal nursing raw data; The data standardization specifically involves performing maximum and minimum normalization processing on the numerical data in the original data of neonatal care.

7. The artificial intelligence-based neonatal care assistance system according to claim 6, characterized in that: In the data preparation module, the data preprocessing is specifically to obtain the original data of neonatal care by collecting neonatal physiological data, historical care records and environmental data.

Images