Insulin dose determination system and computer equipment

Abstract

The invention relates to an insulin dosage determination system and computer equipment. The determination system comprises an acquisition module and a determination module. The acquisition module is used for acquiring to-be-monitored blood glucose information; the determination module is used for inputting to-be-monitored blood glucose information into the dose prediction model to obtain a target insulin dose; wherein the dose prediction model is obtained by training an initial deep learning network according to a first blood glucose information sample and target loss, and the target loss is determined based on the corrected insulin dose; the corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network according to the residual active insulin dosage determined by the first blood glucose information sample. By using the insulin dose determination system provided by the invention, a more accurate target insulin dose can be obtained.

Description

Technical Field

[0001] The present application relates to the technical field of medical device control, and in particular to a system and computer device for determining insulin dosage. Background Art

[0002] With the change of lifestyle and diet structure, diabetes has become a global epidemic. Diabetes can cause acute and chronic complications. Currently, there is no cure for diabetes in the medical field. Insulin supplementation in vitro is still an effective means of controlling blood sugar.

[0003] A closed-loop insulin infusion system consisting of a continuous subcutaneous insulin infusion system and a continuous blood glucose monitoring system can automatically replenish extracorporeal insulin. Currently, a closed-loop control algorithm is used in the closed-loop insulin infusion system to estimate the insulin dose required based on physiological parameters and carbohydrate parameters.

[0004] However, current closed-loop insulin infusion systems suffer from low accuracy in determining insulin dosage. Summary of the invention

[0005] Based on this, it is necessary to provide an insulin dosage determination system and computer device that can improve the accuracy of the determined insulin dosage in order to solve the above technical problems.

[0006] In a first aspect, the present application provides a system for determining an insulin dose, the system comprising:

[0007] An acquisition module, used for acquiring blood sugar information to be monitored;

[0008] A determination module, used for inputting the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0009] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0010] In one embodiment, the determining system further comprises:

[0011] Building blocks for constructing the initial deep learning network;

[0012] A first sample acquisition module, used to acquire a first blood sugar information sample and determine a remaining active insulin dose according to the first blood sugar information sample;

[0013] The training module is used to train the initial deep learning network according to the first blood glucose information sample and the remaining active insulin dose to obtain a dose prediction model.

[0014] In one embodiment, the training module includes:

[0015] An input unit, used to input the first blood glucose information sample and the remaining active insulin dose into the initial deep learning network to obtain the predicted insulin dose and the target blood glucose information; the target blood glucose information is the blood glucose information corresponding to the predicted insulin dose;

[0016] a determining unit, configured to determine a target loss based on the target blood glucose information, the predicted insulin dosage, and the first blood glucose information sample;

[0017] The training unit is used to train the initial deep learning network according to the target loss to obtain a dose prediction model.

[0018] In one embodiment, the initial deep learning network includes a current value network and a simulated blood glucose metabolism network, and the input unit includes:

[0019] A first input subunit is used to input the first blood sugar information sample into the current value network for prediction to obtain a predicted insulin dose;

[0020] A correction subunit, used for correcting the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain a corrected insulin dose;

[0021] The second input subunit is used to input the corrected insulin dosage into the simulated blood glucose metabolism network to perform blood glucose simulation and obtain target blood glucose information.

[0022] In one embodiment, the initial deep learning network further includes a target value network, and the training module further includes:

[0023] A second sample acquisition unit is used to acquire a preset number of second blood glucose information samples and network parameters of the current value network at every preset time period; the second blood glucose information samples include the first blood glucose information samples, target blood glucose information, and rewards corresponding to the target blood glucose information;

[0024] The network training unit is used to train the target value network according to the second blood sugar information sample and the network parameters of the current value network to obtain the trained target value network.

[0025] In one embodiment, the initial deep learning network further includes a replay memory unit, and the second sample acquisition unit includes:

[0026] A storage subunit, used to obtain a second blood glucose information sample according to the first blood glucose information sample, target blood glucose information, and reward corresponding to the target blood glucose information during the current value network training process, and store it in a playback memory unit;

[0027] The sample acquisition subunit is used to acquire a preset number of second blood glucose information samples from the playback memory unit every preset time period.

[0028] In one embodiment, the correction subunit is specifically used for:

[0029] Correcting the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain an intermediate insulin dose;

[0030] determining whether the intermediate insulin dose is greater than a preset insulin dose threshold;

[0031] If not, the intermediate insulin dose was determined as the corrected insulin dose;

[0032] If so, the preset insulin dose threshold is determined as the corrected insulin dose.

[0033] In one embodiment, the correction subunit is further configured to:

[0034] According to the comparison result between the remaining active insulin dose and the predicted insulin dose, and the preset correction condition, a corresponding correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the preset correction condition is determined according to the first blood glucose information sample, the preset maximum blood glucose value and the preset minimum blood glucose value.

[0035] In one embodiment, the correction subunit is further configured to:

[0036] If the comparison result indicates that the predicted insulin dose is greater than the remaining active insulin dose after the first treatment, a first correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the first correction strategy includes performing a difference operation on the predicted insulin dose and the remaining active insulin dose after the first treatment;

[0037] If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the blood glucose information sample is not less than the preset minimum blood glucose value, the second correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the second correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the second processing;

[0038] If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the blood glucose information sample is between the preset minimum blood glucose value and the preset maximum blood glucose value, a third correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the third correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the third processing;

[0039] If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin after the first processing, and the preset correction conditions include that the blood glucose value in the blood glucose information sample is not greater than the preset minimum blood glucose value, the fourth correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the fourth correction strategy includes setting the predicted insulin dose to the first value.

[0040] In a second aspect, an embodiment of the present application provides a computer device, including a processor, the processor performing the following steps:

[0041] Obtaining blood sugar information to be monitored;

[0042] Inputting the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0043] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0044] The above-mentioned insulin dose determination system and computer device, the determination system includes an acquisition module and a determination module. The acquisition module is used to obtain the blood sugar information to be monitored; the determination module is used to input the blood sugar information to be monitored into the dose prediction model to obtain the target insulin dose. The dose prediction model is obtained by training the initial deep learning network according to the first blood sugar information sample and the target loss, the target loss is determined based on the corrected insulin dose, and the corrected insulin dose is obtained by correcting the insulin dose predicted by the initial deep learning network according to the remaining active insulin dose determined by the first blood sugar information sample. In this embodiment, the determination module directly inputs the blood sugar information to be monitored into the trained dose prediction model to obtain the target insulin dose, without obtaining the physiological parameters of the target object corresponding to the blood sugar information to be monitored and the estimated parameters of the meal carbon water, thereby improving the accuracy and efficiency of the target insulin dose obtained. In addition, the dose prediction model used in the training process uses the remaining active insulin dose determined according to the first blood sugar sample to correct the insulin dose predicted by the initial deep learning network, so that the dose prediction model obtained by training is more accurate, thereby making the target insulin obtained by the dose prediction model more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a schematic structural diagram of a system for determining insulin dosage in one embodiment;

[0046] Figure 2 A schematic flow chart of steps performed by a system for determining insulin dosage in one embodiment;

[0047] Figure 3 A schematic diagram of a flow chart of steps for determining module execution in one embodiment;

[0048] Figure 4 is a schematic structural diagram of a system for determining insulin dosage in another embodiment;

[0049] Figure 5 A schematic flow chart of steps executed by a system for determining insulin dosage in another embodiment;

[0050] Figure 6 is a schematic diagram of the structure of a training module in one embodiment;

[0051] Figure 7 A schematic diagram of a flow chart of steps executed by a training module in one embodiment;

[0052] Figure 8 is a schematic diagram of the structure of an input unit in one embodiment;

[0053] Fig. 9 A schematic diagram of a flow chart of steps executed by an input unit in one embodiment;

[0054] Fig.10 is a schematic diagram of the structure of a training module in another embodiment;

[0055] Fig.11 A schematic flow chart of steps executed by a training module in another embodiment;

[0056] Fig.12 is a schematic structural diagram of a second sample acquisition unit in an embodiment;

[0057] Fig.13 A schematic diagram of a flow chart of steps executed by a second sample acquisition unit in one embodiment;

[0058] Fig.14 A schematic diagram of a flow chart of steps performed by a correction subunit in one embodiment;

[0059] Fig.15 A schematic flow chart of steps performed by a correction subunit in another embodiment;

[0060] Fig.16 A schematic diagram of a flow chart of steps for training a dose pre-model in one embodiment;

[0061] Fig.17 A schematic diagram of the structure of an initial deep learning network in one embodiment;

[0062] Fig.18 A schematic diagram of the structure of an initial deep learning network in another embodiment;

[0063] Fig.19 FIG. 1 is a schematic diagram of the structure of a computer device in an embodiment. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0065] The serial numbers assigned to the components in this application, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning.

[0066] First, before specifically introducing the technical solutions of the disclosed embodiments of the present application, the background technology or technical evolution context based on the embodiments of the present application is introduced. With the changes in lifestyle and dietary structure, diabetes has become a global trend. Diabetes can cause acute and chronic complications. At present, there is no treatment method for curing diabetes in the medical community, and in vitro insulin supplementation is still an effective means of controlling blood sugar. A closed-loop insulin infusion system consisting of a continuous subcutaneous insulin infusion system and a continuous blood glucose monitoring system can automatically supplement in vitro insulin so as to effectively control the blood sugar level of the target object according to in vitro insulin. At present, a closed-loop control algorithm is used in the closed-loop insulin infusion system to estimate the insulin dose to be supplemented based on physiological parameters and carbon water parameters. However, the physiological parameters of the human body, such as the insulin-carbon water ratio (ICR) and the correction factor (CF), will change with the changes in the physiological condition of the human body; the estimation of carbon water has inaccuracies, which will lead to inaccurate estimated insulin doses. In this regard, the present application provides a system for determining insulin doses that can improve the accuracy of the determined insulin doses.

[0067] The technical solution of the present application and how the technical solution of the present application solves the technical problem are described in detail below with specific embodiments.

[0068] In one embodiment, Figure 1 As shown, a system 10 for determining insulin dosage is provided, and the system comprises an acquisition module 11 and a determination module 12. Figure 2 As shown, the acquisition module 11 is used to execute the following step 200 , and the determination module 12 is used to execute the following step 210 .

[0069] Step 200: Obtain blood sugar information to be monitored.

[0070] The blood glucose information to be monitored refers to the blood glucose information of the target object, and includes the area under the blood glucose curve (AUC), the maximum blood glucose value, the minimum blood glucose value, and the time difference between the maximum blood glucose value and the minimum blood glucose value within a time interval.

[0071] The computer device obtains the blood sugar information to be monitored. The blood sugar information to be monitored can be pre-stored in a storage unit of the computer device, and the computer device directly obtains it from the storage unit when needed. The blood sugar information to be monitored can also be stored in a specific storage device, and the computer device obtains it from the specific storage device when needed. This embodiment does not limit the specific method for obtaining the blood sugar information to be monitored, as long as its function can be achieved.

[0072] Step 210: input the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0073] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0074] The dosage prediction model can be pre-stored in the computer device. After the computer device obtains the blood sugar information to be monitored, the blood sugar information to be monitored is input into the dosage prediction model, and the dosage prediction is performed through the dosage prediction model to obtain the target insulin dosage. The target insulin dosage refers to the target insulin dosage that can achieve blood sugar control by infusing insulin under the blood sugar information to be monitored.

[0075] The first blood glucose information sample includes multiple blood glucose information, each of which includes AUC, maximum blood glucose value, minimum blood glucose value and time difference between the maximum blood glucose value and the minimum blood glucose value. The multiple blood glucose information can be the blood glucose information of the target object at multiple times.

[0076] The initial deep learning network can be any one of a radial basis function neural network, a convolutional neural network, a long short-term memory network, and a deep neural network. The target loss is the loss used in the process of training the initial deep learning network. The target loss can be any one of a regression loss and a classification loss. This embodiment does not limit the type of the initial deep learning network and the type of the target loss, and the user can choose according to the actual application environment.

[0077] Active insulin refers to the insulin that has not been consumed and will take effect. In the process of training the initial deep learning network using the first blood sugar information sample and the target loss, the initial deep learning network predicts the dosage of the first blood sugar information and can obtain the predicted insulin dosage; calculate the remaining active insulin dosage according to the first blood sugar information sample, that is, the remaining active insulin dosage of the target object at the moment of obtaining the blood sugar information sample; use the remaining active insulin dosage to correct the insulin dosage predicted by the initial deep learning network to obtain the corrected insulin dosage; determine the target loss according to the corrected insulin dosage, and use the target loss to update the network parameters of the initial deep learning network to realize the training of the initial deep learning network.

[0078] The insulin dosage determination system provided in the embodiment of the present application includes an acquisition module and a determination module, wherein the acquisition module is used to obtain the blood sugar information to be monitored, and the determination module is used to input the blood sugar information to be monitored into the dosage prediction model to obtain the target insulin dosage. The dosage prediction model is obtained by training the initial deep learning network according to the first blood sugar information sample and the target loss, and the target loss is determined based on the modified insulin dosage, and the modified insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network according to the residual active insulin dosage determined by the first blood sugar information sample. In this embodiment, the determination module directly inputs the blood sugar information to be monitored into the trained dosage prediction model, and the target insulin dosage can be obtained without obtaining the physiological parameters of the target object corresponding to the blood sugar information to be monitored and the estimated parameters of the meal carbon water, thereby improving the accuracy and efficiency of the target insulin dosage obtained. In addition, the dosage prediction model used in the training process uses the residual active insulin dosage determined according to the first blood sugar sample to correct the insulin dosage predicted by the initial deep learning network, so that the dosage prediction model obtained by training is more accurate, so that the target insulin obtained by the dosage prediction model is more accurate, and then the user can control the blood sugar level of the target object more accurately according to the target insulin dosage.

[0079] In an optional embodiment, if Figure 3 As shown, the determination module 12 is specifically used to perform the following steps 300 to 320:

[0080] Step 300: input the blood glucose information to be monitored into the dosage prediction model to obtain the initial insulin dosage;

[0081] After obtaining the blood sugar information to be monitored, the computer device directly inputs the blood sugar information to be monitored into the dosage prediction model, and obtains the initial insulin dosage through prediction by the dosage prediction model.

[0082] Step 310: Determine the target remaining active insulin dose according to the blood glucose information to be monitored.

[0083] The computer device can determine the remaining active insulin decay curve corresponding to the blood sugar information to be monitored based on the acquired blood sugar information to be monitored, and can determine the target remaining active insulin dose based on the remaining active insulin decay curve.

[0084] Step 320: Correct the initial insulin dose according to the target remaining active insulin dose to obtain a target insulin dose.

[0085] After obtaining the target remaining active insulin dose, the computer device corrects the initial insulin dose predicted by the dose prediction model according to the target remaining active insulin dose to obtain the target insulin dose. Optionally, the computer device can determine the target insulin dose by calculating the difference between the remaining active insulin dose and the initial insulin dose.

[0086] In this embodiment, after the initial insulin dose is obtained through the dose prediction model, the target remaining active insulin dose is used to correct the initial insulin dose to obtain the target insulin dose, so that the target insulin dose obtained is more accurate.

[0087] In one embodiment, Figure 4 As shown, the insulin dosage determination system 10 further includes a construction module 13, a first sample acquisition module 14 and a training module 15. Figure 5 As shown, the construction module 13 is used to execute the following step 500 , the first sample acquisition module 14 is used to execute the following step 510 , and the training module 15 is used to execute the following step 520 .

[0088] Step 500: construct an initial deep learning network.

[0089] The computer device constructs an initial deep learning network, that is, obtains a required deep learning network, initializes the network parameters of the deep learning network, and obtains the initial deep learning network.

[0090] Step 510: Obtain a first blood glucose information sample, and determine a remaining active insulin dose according to the first blood glucose information sample.

[0091] After obtaining the first blood sugar information sample, the computer device calculates the corresponding remaining active insulin dose according to the first blood sugar information sample. The computer device may first obtain the corresponding remaining active insulin decay curve according to the first blood sugar information sample, and calculate the remaining active insulin dose according to the remaining active insulin decay curve.

[0092] In an optional embodiment, it is assumed that the residual active insulin decay curve corresponding to the blood glucose information sample can be expressed as in, a, b, c, d and e are conventional coefficients, a = -0.995, b = 0.04, c = 0.84, d = 30, e = 12, t4 = 2h (hours), t5 = 60 (minutes). The remaining active insulin dose at time i can be expressed as u IOB,i , τ represents the insulin infusion time, which can be taken as a time interval of t1 (5min). represents the length of the residual active insulin decay curve, represents the length of the remaining active insulin decay curve corresponding to the first blood glucose sample, θ i Represents a value containing the length of the remaining active insulin decay curve Array of θ i[j] express The length value of the i-th curve in the array; θ MEAL represents an array containing the decay curve of the remaining active insulin at t2(4h), ∧ BAEAL，i represents the historical insulin basal rate at time i, t3 (8h), ∧ MEAL，i Represents the historical insulin dose at time t3 (8h).

[0093] Step 520: Train the initial deep learning network according to the first blood glucose information sample and the remaining active insulin dose to obtain a dose prediction model.

[0094] After obtaining the remaining active insulin dose, the computer device trains the initial deep learning network according to the first blood glucose information sample and the remaining active insulin dose to obtain a dose prediction model.

[0095] In an optional embodiment, the computer device can input the first blood glucose information sample into the initial deep learning network to obtain a predicted insulin dose; correct the predicted insulin dose according to the remaining active insulin dose to obtain a corrected insulin dose; determine the target loss of the initial deep learning network according to the corrected insulin dose; update the network parameters of the initial deep learning network according to the target loss to obtain a dose prediction model.

[0096] In this embodiment, the remaining active insulin dose is determined by the first blood glucose information sample, and the initial deep learning network is trained according to the first blood glucose information sample and the remaining active insulin dose, so that the dose prediction model obtained is more accurate, thereby making the target insulin dose obtained by the dose prediction model more accurate.

[0097] In one embodiment, Figure 6 As shown, the training module 15 includes an input unit 151, a determination unit 152 and a training unit 153. Figure 7 As shown, the input unit 151 is used to execute the following step 700 , the determination unit 152 is used to execute the following step 710 , and the training unit 153 is used to execute the following step 720 .

[0098] Step 700: Input the first blood glucose information sample and the remaining active insulin dose into the initial deep learning network to obtain the predicted insulin dose and the target blood glucose information; the target blood glucose information is the blood glucose information corresponding to the predicted insulin dose.

[0099] When the computer device trains the initial deep learning network, the first blood sugar information sample and the remaining active insulin dose corresponding to the first blood sugar information sample are input into the initial deep learning network, and the initial deep learning network is used to predict, so as to obtain the predicted insulin dose and the blood sugar information (target blood sugar information) corresponding to the predicted insulin dose. The predicted insulin dose refers to the predicted insulin dose that needs to be infused in the case of the first blood sugar information sample; the target blood sugar information refers to the blood sugar information after the predicted insulin dose is infused, that is, the next blood sugar information corresponding to the first blood sugar information sample.

[0100] Assume that the blood sugar information of a day is divided into 4 states, namely 4 hours before breakfast, 4 hours before lunch, 4 hours before dinner and 4 hours after dinner. If the first blood sugar information sample is 4 hours before breakfast, the target blood sugar information is the predicted 4 hours before lunch; if the first blood sugar information sample is 4 hours before lunch, the target blood sugar information is the predicted 4 hours before dinner.

[0101] Step 710: Determine target loss based on target blood glucose information, predicted insulin dosage and first blood glucose information sample.

[0102] After obtaining the target blood sugar information and the predicted insulin dose, the computer device calculates the target loss using the target blood sugar information, the predicted insulin dose and the first blood sugar information. Specifically, the computer device may first calculate the reward of the target blood sugar information, and calculate the target loss based on the reward of the target blood sugar information, the first blood sugar information sample and the predicted insulin dose.

[0103] In an optional embodiment, the computer device calculates the reward of the target blood glucose information according to a reward function. The reward function is used to evaluate the target blood glucose information, and if the blood glucose value in the target blood glucose information is within the target range, a higher reward is assigned, and if the blood glucose value in the target blood glucose information is outside the target range, a lower reward is assigned. The reward function can be as follows:

[0104]

[0105] in, represents the upper limit of blood sugar information at time i (i.e., the maximum blood sugar value), represents the lower limit of blood sugar information at time i (i.e., the minimum blood sugar value). From the above reward function, we can see that when blood sugar information is not Within this range, blood sugar information is divided into the risk of high blood sugar and the risk of low blood sugar, and the reward for low blood sugar is smaller than the reward for high blood sugar.

[0106] Step 720: Train the initial deep learning network according to the target loss to obtain a dose prediction model.

[0107] After obtaining the target loss, the computer device updates the network parameters of the initial deep learning network according to the target loss until the training end condition is reached to obtain the dose prediction model. The training end condition can be the convergence of the target loss (reaching a preset threshold) or the number of iterations reaching a preset threshold.

[0108] In this embodiment, the target loss is calculated by using the first blood sugar information sample, the predicted insulin dose and the target blood sugar information output by the initial deep learning network, and the network parameters of the initial deep learning network are updated according to the target loss to train the initial deep learning network and obtain a dose prediction model. This method for determining the dose prediction model is accurate and fast.

[0109] In one embodiment, the initial deep learning network includes a current value network and a simulated blood glucose metabolism network. The current value network is used to predict insulin dosage, and the simulated blood glucose metabolism network is used to simulate the blood glucose metabolism environment of the human body.

[0110] In this case, if Figure 8 As shown, the input unit 151 includes a first input subunit 1511, a correction subunit 1512, and a second input subunit 1513. Fig. 9 As shown, the first input subunit 1511 is used to execute the following step 900 , the correction subunit 1512 is used to execute the following step 910 , and the second input subunit 1513 is used to execute the following step 920 .

[0111] Step 900: Input the first blood sugar information sample into the current value network for prediction to obtain a predicted insulin dose.

[0112] The computer device inputs the first blood sugar information sample into the current value network, predicts the first blood sugar information sample through the current value network, and obtains the predicted insulin dosage.

[0113] Step 910: Correct the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain a corrected insulin dose.

[0114] After obtaining the predicted insulin dosage, the computer device corrects the predicted insulin dosage according to the calculated remaining active insulin dosage and the first blood glucose information sample to obtain a corrected insulin dosage.

[0115] Step 920: Input the corrected insulin dose into the simulated blood glucose metabolism network to perform blood glucose simulation and obtain target blood glucose information.

[0116] After obtaining the corrected insulin dose, the computer device inputs the corrected insulin dose into the simulated blood glucose metabolism network for blood glucose simulation to obtain target blood glucose information; that is, to simulate the blood glucose information of the human body after the corrected insulin dose is infused into the human metabolic environment.

[0117] In this embodiment, the structure of the initial deep learning network includes a current value network and a simulated blood glucose metabolism network. The predicted insulin dosage is determined by the current value network, and the target blood glucose information is determined by simulating the blood glucose metabolism environment. This method of determining the predicted insulin dosage and target blood glucose information is quick and easy to implement.

[0118] In one embodiment, the initial deep learning network also includes a target value network, and the structure of the target value network is the same as that of the current value network. Fig.10 As shown, the training module 15 also includes a second sample acquisition unit 154 and a network training unit 155. Fig.11 As shown, the second sample acquisition unit 154 is used to execute the following step 111, and the network training unit 155 is used to execute the following step 112.

[0119] Step 111, obtaining a preset number of second blood glucose information samples and network parameters of the current value network at every preset time period; the second blood glucose information samples include the first blood glucose information samples, target blood glucose information, and rewards corresponding to the target blood glucose information.

[0120] The preset time period is pre-set by the user and stored in the computer device. During the training of the current value network, the computer device will obtain a preset number of second blood glucose information samples and the network parameters of the current value network every preset time period. That is, after each preset time period of training the current value network, the network parameters of the trained current value network are obtained. The second blood glucose information sample includes the first blood glucose information sample, the target blood glucose information, and the reward corresponding to the target blood glucose information. The second blood glucose information sample is a blood glucose information sample randomly selected from the first blood glucose information sample, and the target blood glucose information refers to the blood glucose information after the infusion of the corrected insulin dose. For the description of the reward corresponding to the target blood glucose information, reference can be made to the specific description in the above embodiment, which will not be repeated here.

[0121] Step 112: training the target value network according to the second blood glucose information sample and the network parameters of the current value network.

[0122] The network parameters of the target value network before training are initialization parameters, that is, the computer device initializes the network parameters of the target value network before training the target value network. The initialization parameters of the target value network parameters can be the same as the initialization parameters of the current value network. The initialization parameters of the current value network are the parameters that are initialized before training the current value network.

[0123] After obtaining the network parameters of the current value network, the computer device uses the network parameters to replace the initialization parameters of the target value network, inputs the second blood glucose information sample into the target value network after the network parameters are replaced, and trains the target value network.

[0124] Specifically, the computer device inputs the first blood sugar information sample and the target blood sugar information in the second blood sugar information sample into the target value network after the network parameters are replaced, and calculates and outputs the maximum value through the target value network. The current value network will also output multiple values ​​during the training process. The loss function is calculated through the multiple values ​​output by the current value, the reward of the target blood sugar information stored in the playback memory unit, and the maximum value output by the target value network; the network parameters of the current value network are updated according to the loss function until the training end condition is reached. The training end condition can be the convergence of the loss function.

[0125] In an optional embodiment, the loss function can be expressed as: If l+1 is the last second blood sugar information sample, then y l =r l ; if s l+1 If it is not the last second blood sugar information sample, then Where m represents the number of second blood glucose information samples, γ represents the discount factor, Q(s l ,a l |w) represents the value obtained by network training corresponding to the current value network of the lth second blood glucose information sample under the condition of network parameter w, r l represents the reward corresponding to the lth second blood glucose information sample, Indicates the value obtained by training the target network corresponding to the second blood glucose information sample (l+1) under the network parameter w′. By solving the gradient of the loss function Update the network parameters w.

[0126] In one embodiment, the initial deep learning network further includes a replay memory unit, and the second blood glucose information of the training target value network can be stored in the replay memory unit. Fig.12 As shown, the second sample acquisition unit 154 includes a storage subunit 1541 and a sample acquisition subunit 1542. Fig.13 As shown, the storage subunit 1541 is used to execute the following step 130, and the sample acquisition subunit 1542 is used to execute the following step 131.

[0127] Step 130: construct a second blood glucose information sample based on the first blood glucose information sample, target blood glucose information, and reward corresponding to the target blood glucose information during the current value network training process, and store it in a playback memory unit.

[0128] During the training of the current value network, the parameters used in the training process: the first blood glucose information sample, the target blood glucose information and the reward corresponding to the target blood glucose information are used as the second blood glucose information sample and stored in the playback memory unit.

[0129] In an optional embodiment, the corrected insulin dosage corresponding to the first blood glucose information sample may also be stored in the playback memory unit.

[0130] Step 131: Acquire a preset number of second blood glucose information samples from the playback memory unit every preset time period.

[0131] The computer device obtains a preset number of second blood glucose information samples from the playback memory unit every preset time period. The number of the obtained second blood glucose information samples is less than the number of the first blood glucose information samples, that is, the preset number is less than the number of the first blood glucose information samples.

[0132] In this embodiment, the relevant parameters in the process of training the current value network are constructed into a second blood glucose information sample and stored in the playback memory unit; when training the target value network, they are directly obtained from the playback memory unit; and when training the target value network, the network parameters of the current value network are directly used to replace the initial network parameters of the target value network. This can improve the efficiency of training the target value network, thereby improving the efficiency of training the dose prediction model.

[0133] In one embodiment, Fig.14 As shown, the correction subunit is specifically used to execute the following steps 140 to 143.

[0134] Step 140: Correct the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain an intermediate insulin dose.

[0135] The computer device corrects the predicted insulin dose based on the remaining active insulin dose and the first blood glucose information sample, and uses the corrected insulin dose as the intermediate insulin dose.

[0136] In one embodiment, when determining the intermediate insulin dose, the correction subunit is further configured to perform the following steps:

[0137] According to the comparison result between the remaining active insulin dose and the predicted insulin dose, and the preset correction condition, a corresponding correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the preset correction condition is determined according to the first blood glucose information sample, the preset maximum blood glucose value and the preset minimum blood glucose value.

[0138] The preset maximum blood glucose value and the preset minimum blood glucose value may be pre-stored in the computer device by the user, and the preset maximum blood glucose value and the preset minimum blood glucose value constitute a preset blood glucose interval. The preset maximum blood glucose value is the same as the upper limit of the blood glucose information described in the above embodiment, and the preset minimum blood glucose value is the same as the lower limit of the blood glucose information described in the above embodiment.

[0139] In an optional embodiment, the user sets different preset blood sugar intervals for different time periods. Select τ1 = 20 o'clock, τ2 = 22 o'clock, τ3 = 14 o'clock, and τ4 = 16 o'clock in a day for segmentation. 16 o'clock to 20 o'clock is one time period, and the rest of the time is one time period. The preset maximum blood sugar value in the preset blood sugar interval It can be expressed as: Preset minimum blood sugar value It can be expressed as: Among them, f1, f2, k1 and k2 are preset parameters. When f1=140, f2=80, k1=160, k2=90, the preset blood glucose range in the time period from 16:00 to 20:00 is 70mg / dL~120mg / dL, the preset maximum blood glucose value is 120mg / dL, and the preset minimum blood glucose value is 70mg / dL; the preset blood glucose range in the time period at other times is 80mg / dL~140mg / dL, the preset maximum blood glucose value is 140mg / dL, and the preset minimum blood glucose value is 80mg / dL.

[0140] The computer device compares the remaining active insulin dose obtained with the predicted insulin dose to obtain a comparison result. Different comparison results and prediction correction conditions correspond to different correction strategies. The corresponding correction strategy is selected according to the comparison result and the preset correction condition, and the predicted insulin dose is corrected using the correction strategy to obtain an intermediate insulin dose.

[0141] Step 141: Determine whether the intermediate insulin dose is greater than a preset insulin dose threshold.

[0142] The preset insulin dose threshold is set by the user and pre-stored in the computer device in order to reduce the risk of hypoglycemia caused by excessive insulin dose infusion. After obtaining the intermediate insulin dose, the computer device compares the intermediate insulin dose with the preset insulin dose threshold, and determines whether the intermediate insulin dose is greater than the preset insulin dose based on the comparison result.

[0143] In an optional embodiment, it is assumed that the preset insulin dose threshold at time i is represented by u i ,u i It can be calculated using the following formula: Among them, 1320, 1440 and 240 represent the number of minutes in a day, and L is the proportional coefficient.

[0144] Step 142: If not, the intermediate insulin dose is determined as the corrected insulin dose.

[0145] If the computer device determines that the intermediate insulin dose is not greater than the preset insulin dose threshold, indicating that there is no risk of hypoglycemia when infusing the intermediate insulin dose, the intermediate insulin dose is determined as the final corrected insulin dose.

[0146] Step 143: If yes, the preset insulin dosage threshold is determined as the corrected insulin dosage.

[0147] If the computer device determines that the intermediate insulin dose is greater than the revised insulin dose threshold, indicating that infusing the intermediate insulin dose may result in a risk of hypoglycemia, the preset insulin dose threshold is determined as the final revised insulin dose.

[0148] In this embodiment, after the predicted insulin dose is corrected using the remaining active insulin dose and the first blood glucose information sample to obtain an intermediate insulin dose, the intermediate insulin dose is compared with the preset insulin dose threshold; if the intermediate insulin dose is greater than the preset insulin dose threshold, the preset insulin dose threshold is determined as the corrected insulin dose; if the intermediate insulin dose is not greater than the preset insulin dose threshold, the intermediate insulin dose is determined as the corrected insulin dose. This ensures that the final corrected insulin dose is more accurate, thereby making the dose prediction model obtained by training more accurate, and further improving the accuracy of the determined target insulin dose.

[0149] In one embodiment, Fig.15 As shown, the correction subunit is specifically further used to execute the following steps 150 to 153.

[0150] Step 150: If the comparison result indicates that the predicted insulin dose is greater than the remaining active insulin dose after the first treatment, select a first correction strategy to correct the predicted insulin dose to obtain an intermediate insulin dose; the first correction strategy includes performing a difference operation on the predicted insulin dose and the remaining active insulin dose after the first treatment.

[0151] When comparing the remaining active insulin dose with the predicted insulin dose, the computer device first performs a first process on the remaining active insulin dose to obtain the remaining active insulin dose after the first process, and then determines whether the predicted insulin dose is greater than the remaining active insulin dose after the first process. The first process may be calculating the ratio of the remaining active insulin dose to the first proportionality coefficient.

[0152] If the computer device determines that the first preset insulin dose is greater than the remaining active insulin dose after the first treatment, the difference between the predicted insulin dose and the remaining active insulin dose after the first treatment is calculated (first correction strategy), that is, the predicted insulin dose minus the remaining active insulin dose after the first treatment to obtain the intermediate insulin dose.

[0153] Step 151: If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is not less than the preset minimum blood glucose value, select the second correction strategy to correct the predicted insulin dose to obtain an intermediate insulin dose; the second correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the second processing.

[0154] If the computer device determines that the predicted insulin dose is less than the remaining active insulin after the first processing, and the blood glucose value in the first blood glucose information sample is not less than the preset minimum blood glucose value, the predicted insulin dose is first subjected to the second processing to obtain the predicted insulin dose after the second processing, and then the difference between the predicted insulin dose and the predicted insulin dose after the second processing is calculated (the second correction strategy), that is, the first preset insulin dose minus the predicted insulin dose after the second processing, to obtain the intermediate insulin dose. The second processing may be to calculate the product between the predicted insulin dose and the second proportionality coefficient.

[0155] Step 152: If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is between the preset minimum blood glucose value and the preset maximum blood glucose value, select the third correction strategy to correct the predicted insulin dose to obtain an intermediate insulin dose; the third correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the third processing.

[0156] If the computer device determines that the predicted insulin dose is less than the remaining active insulin after the first processing, and the blood glucose value in the first blood glucose information sample is greater than or equal to the preset minimum blood glucose value, and less than or equal to the preset maximum blood glucose value, the predicted insulin dose is first subjected to the third processing to obtain the predicted insulin dose after the third processing, and then the difference between the predicted insulin dose and the predicted insulin dose after the third processing is calculated (third correction strategy), that is, the predicted insulin dose minus the predicted insulin dose after the third processing to obtain the intermediate insulin dose. The third processing may be to calculate the product between the predicted insulin dose and the third proportional coefficient.

[0157] Step 153: If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is not greater than the preset minimum blood glucose value, select the fourth correction strategy to correct the predicted insulin dose to obtain an intermediate insulin dose; the fourth correction strategy includes setting the predicted insulin dose to the first value.

[0158] If the computer device determines that the first preset insulin dose is less than the remaining active insulin dose after the first processing, and the blood glucose value in the first blood glucose information sample is less than or equal to the preset minimum blood glucose value, the directly predicted insulin dose is set to the first value (the fourth correction strategy), that is, the intermediate insulin dose is the first value. The first value is zero.

[0159] Assume that the remaining active insulin dose at time i is represented by u IOB,i , the predicted insulin dose at time i is represented by u DQN,i , the intermediate insulin dose at time i is represented by u bolus,i , the first proportional coefficient is K, the second proportional coefficient is K1, the third proportional coefficient is K2, and the first blood glucose sample information at the i-th moment is G i The intermediate insulin dose at time i can be expressed as follows:

[0160]

[0161] In this embodiment, the comparison result is different from the preset correction condition satisfied, and different correction strategies are used to correct the first preset insulin dose, so that a more accurate intermediate insulin dose can be obtained, so that an accurate dose prediction model can be trained, and then an accurate target insulin dose can be obtained through the dose prediction model.

[0162] See also Fig.16 An embodiment of the present application provides a dose prediction model training method, the method comprising the steps of:

[0163] Step 160: construct an initial deep learning network and obtain a first blood glucose information sample; the initial deep learning network includes a current value network, a simulated blood glucose metabolism network, a target value network and a playback memory unit; the structure of the initial deep learning network is as follows: Fig.17 shown.

[0164] Step 161: Input the first blood sugar information sample into the current value network to obtain a first predicted insulin dose and a plurality of corresponding values;

[0165] Step 162: input the predicted insulin dose into the simulated blood glucose metabolism network to obtain target blood glucose information, and determine the reward for the target blood glucose information; the target blood glucose information is the blood glucose information corresponding to the predicted insulin dose;

[0166] Step 163: construct the first blood glucose information sample, the target blood glucose information, and the reward of the target blood glucose information into a second blood glucose information sample, and store it in a playback memory unit;

[0167] Step 164: acquiring a second blood glucose information sample from the playback memory unit at every preset time period, and training the target value network according to the second blood glucose information sample to obtain the maximum value output by the target value network;

[0168] Step 165: Calculate the loss function based on the multiple values ​​output by the current value network and the maximum value output by the target value, and update the network parameters of the current value network based on the loss function to obtain a dose prediction model.

[0169] In an optional embodiment, assuming that the first blood glucose information sample is used to form a state space, the state space expression corresponding to the blood glucose information sample in the time interval t can be expressed as s t ={AUC t ,maxG t ,minG t ,T t}, where AUC t represents the area under the blood glucose curve in the t time region, maxG t Indicates the maximum blood sugar value in the blood sugar information sample within the t time zone; minG t represents the minimum blood sugar value in the blood sugar information sample within the time zone t, T t Indicates the time difference between the maximum blood sugar value and the minimum blood sugar value.

[0170] After the first blood glucose information sample is input into the initial deep learning network, the initial deep learning network determines the predicted insulin dose corresponding to the first blood glucose information sample according to the first blood glucose information sample and the predefined action space (i.e., predicted insulin dose). The classification of the action space can be expressed by the following formula:

[0171]

[0172] Among them, A, B and C represent three action spaces, each of which includes F actions. Fig.18 As shown, the initial deep learning network may include an input layer, a hidden layer and an output layer, and the hidden layer may include at least three fully connected hidden layers. In the case where the hidden layer includes three fully connected hidden layers, the three hidden layers are used for the above-mentioned three action spaces A, B and C respectively.

[0173] In an optional embodiment, for the current value network, an action can be randomly selected with probability ε and a predicted insulin dose can be selected with probability 1-ε. The end condition for training the initial deep learning network can include an exponential decay detection rate equal to

[0174] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0175] In one embodiment, a computer device is provided, whose internal structure diagram can be as follows: Fig.19 shown.

[0176] The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected to a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, each step described in the above embodiment is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the computer device housing, or an external keyboard, touch pad or mouse, etc.

[0177] Those skilled in the art will understand that Fig.19 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0178] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0179] Obtaining blood sugar information to be monitored;

[0180] Inputting the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0181] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0182] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0183] Obtaining blood sugar information to be monitored;

[0184] Inputting the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0185] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0186] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0187] Obtaining blood sugar information to be monitored;

[0188] Inputting the blood glucose information to be monitored into the dosage prediction model to obtain the target insulin dosage;

[0189] Among them, the dosage prediction model is obtained by training the initial deep learning network based on the first blood glucose information sample and the target loss. The target loss is determined based on the corrected insulin dosage. The corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network based on the remaining active insulin dosage determined by the first blood glucose information sample.

[0190] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0191] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0192] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A system for determining insulin dosage, characterized in that: The determination system comprises: An acquisition module, used for acquiring blood sugar information to be monitored; A determination module, used for inputting the blood glucose information to be monitored into a dosage prediction model to obtain a target insulin dosage; Among them, the dosage prediction model is obtained by training the initial deep learning network according to the first blood glucose information sample and the target loss, the target loss is determined based on the corrected insulin dosage, and the corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network according to the remaining active insulin dosage determined according to the first blood glucose information sample.

2. The determination system according to claim 1, characterized in that The determination system further comprises: A construction module, used to construct the initial deep learning network; A first sample acquisition module, used for acquiring the first blood sugar information sample, and determining the remaining active insulin dose according to the first blood sugar information sample; A training module is used to train the initial deep learning network according to the first blood glucose information sample and the remaining active insulin dose to obtain the dose prediction model.

3. The determination system according to claim 2, characterized in that: The training module includes: An input unit, used to input the first blood glucose information sample and the remaining active insulin dose into the initial deep learning network to obtain predicted insulin dose and target blood glucose information; the target blood glucose information is the blood glucose information corresponding to the predicted insulin dose; a determining unit, configured to determine the target loss according to the target blood glucose information, the predicted insulin dosage and the first blood glucose information sample; A training unit is used to train the initial deep learning network according to the target loss to obtain the dose prediction model.

4. The determination system according to claim 3, characterized in that: The initial deep learning network includes a current value network and a simulated blood glucose metabolism network, and the input unit includes: A first input subunit, used for inputting the first blood glucose information sample into the current value network for prediction to obtain the predicted insulin dosage; a correction subunit, configured to correct the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain the corrected insulin dose; The second input subunit is used to input the corrected insulin dosage into the simulated blood glucose metabolism network to perform blood glucose simulation to obtain the target blood glucose information.

5. The determination system according to claim 4, characterized in that: The initial deep learning network also includes a target value network, and the training module also includes: A second sample acquisition unit is used to acquire a preset number of second blood glucose information samples and network parameters of the current value network at every preset time period; the second blood glucose information sample includes the first blood glucose information sample, the target blood glucose information, and a reward corresponding to the target blood glucose information; A network training unit is used to train the target value network according to the second blood glucose information sample and the network parameters of the current value network.

6. The determination system according to claim 5, characterized in that: The initial deep learning network further includes a playback memory unit, and the second sample acquisition unit includes: a storage subunit, configured to obtain the second blood glucose information sample according to the first blood glucose information sample, the target blood glucose information, and the reward corresponding to the target blood glucose information during the current value network training process, and store the second blood glucose information sample in the playback memory unit; The sample acquisition subunit is used to acquire a preset number of the second blood glucose information samples from the playback memory unit every preset time period.

7. The determination system according to claim 4, characterized in that: The correction subunit is specifically used for: Correcting the predicted insulin dose according to the remaining active insulin dose and the first blood glucose information sample to obtain an intermediate insulin dose; determining whether the intermediate insulin dose is greater than a preset insulin dose threshold; If not, determining the intermediate insulin dose as the corrected insulin dose; If so, the preset insulin dose threshold is determined as the corrected insulin dose.

8. The determination system according to claim 7, characterized in that: The correction subunit is further specifically used for: According to the comparison result between the remaining active insulin dose and the predicted insulin dose, and the preset correction condition, a corresponding correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; The preset correction condition is determined according to the first blood glucose information sample, a preset maximum blood glucose value and a preset minimum blood glucose value.

9. The determination system according to claim 8, characterized in that: The correction subunit is further specifically used for: If the comparison result indicates that the predicted insulin dose is greater than the remaining active insulin dose after the first processing, selecting a first correction strategy to correct the predicted insulin dose to obtain an intermediate insulin dose; the first correction strategy includes performing a difference operation on the predicted insulin dose and the remaining active insulin dose after the first processing; If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is not less than the preset minimum blood glucose value, a second correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the second correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the second processing; If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin dose after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is between the preset minimum blood glucose value and the preset maximum blood glucose value, a third correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the third correction strategy includes performing a difference operation on the predicted insulin dose and the predicted insulin dose after the third processing; If the comparison result indicates that the predicted insulin dose is less than the remaining active insulin after the first processing, and the preset correction condition includes that the blood glucose value in the first blood glucose information sample is not greater than the preset minimum blood glucose value, the fourth correction strategy is selected to correct the predicted insulin dose to obtain an intermediate insulin dose; the fourth correction strategy includes setting the predicted insulin dose to the first value.

10. A computer device comprising a processor, characterized in that: The processor performs the following steps: Obtaining blood sugar information to be monitored; Inputting the blood glucose information to be monitored into a dosage prediction model to obtain a target insulin dosage; Among them, the dosage prediction model is obtained by training the initial deep learning network according to the first blood glucose information sample and the target loss, the target loss is determined based on the corrected insulin dosage, and the corrected insulin dosage is obtained by correcting the insulin dosage predicted by the initial deep learning network according to the remaining active insulin dosage determined according to the first blood glucose information sample.

Images