Switching method of patient blood temperature prediction model based on computing power difference and related device

Abstract

The invention relates to a patient blood temperature prediction model switching method based on computing power difference and a related device. The method comprises the steps that fixed parameters and operation parameters of a hemodialysis extracorporeal circulation pipeline are obtained; based on current computing power or user preferences, the decision enables a precision mode or an efficiency mode. When the precision mode is started, a progress monitor is started to monitor the calculation duration of the complete model; if the calculation is completed within the preset duration, extracting an outlet prediction temperature from a temperature field generated by the complete model; and if the calculation is overtime, stopping the calculation and switching to the efficiency mode. When the efficiency mode is started, inputting the fixed parameters and the operation parameters into a pre-trained bridging model to calculate a group of model coefficients; and calculating to obtain the outlet predicted temperature by using the simple model and the model coefficient. The method has the advantage of balancing calculation precision and real-time responsiveness.

Description

Technical Field

[0001] This application relates to the field of dialysis equipment, and in particular to a method and related apparatus for switching a patient blood temperature prediction model based on computing power differences. Background Technology

[0002] Hemodialysis is a routine medical treatment for patients with kidney failure. The procedure involves drawing the patient's blood out of the body through an extracorporeal circulation system and passing it through a dialyzer to remove metabolic waste before returning it to the patient. A standard dialysis session typically lasts several hours. During this time, the blood continuously exchanges heat with the external environment within the extracorporeal circulation system, causing the temperature of the blood returning to the patient to deviate from its normal range, which may affect the patient's hemodynamic stability.

[0003] To accurately predict such temperature deviations, two different computational approaches exist in existing technologies. The first is a high-precision, complete model based on the partial differential equation of heat conduction, capable of providing a precise numerical solution to the spatiotemporal dynamics of blood temperature. This method is theoretically accurate, but its computational process is relatively complex, requiring significant computational resources and potentially leading to lengthy computation times. The second approach is a highly efficient simplified model, typically relying on simplified empirical formulas and pre-calibrated model coefficients. This method is computationally fast and suitable for real-time responses, but its accuracy is highly dependent on the quality and coverage of the calibration data.

[0004] In clinical scenarios requiring both real-time monitoring and high-precision analysis, computing systems need to switch between these two modes. For example, when system computing resources are strained, or when high-precision calculations time out, the system needs to degrade from precision mode to efficiency mode. This raises a technical issue: because the full and simplified models differ in their underlying construction and mathematical principles, when the system switches between these two modes, they may calculate inconsistent outlet temperature predictions for the same set of input parameters (such as the same flow rate and body temperature). This abrupt or discontinuous change in calculation results between different modes reduces the system's reliability. Summary of the Invention

[0005] In order to enable flexible switching between models of different precisions based on computing power status, it is also necessary to solve the problem of inconsistent calculation results caused by switching and ensure the smoothness and continuity of the model during the switching process. This application provides a switching method and related device for patient blood temperature prediction models based on computing power differences.

[0006] Firstly, this application provides a method for switching patient blood temperature prediction models based on differences in computing power, employing the following technical solution: A method for switching patient blood temperature prediction models based on computing power differences includes the following steps: S1. Read the pre-built complete model, simple model, and pre-trained bridging model used to associate the complete model and the simple model, and obtain the fixed parameters and operating parameters of the extracorporeal circulation tubing for hemodialysis to complete the initialization of the complete model; wherein, the fixed parameters include geometric parameters and physical parameters, and the operating parameters correspond to the hemodialysis process; S2. Make an initial mode decision based on the current computing power or user preferences, and determine whether to enable precision mode or efficiency mode; S3. When the initial mode decision is the precision mode, start the progress monitor to monitor the calculation time of the complete model processing the running parameters and the fixed parameters, and decide whether to enter S4a or S4b based on the monitoring results; S4a. When the calculation is completed within the preset time, extract the predicted outlet temperature from the temperature field generated by the complete model and return it as the final result; S4b. If the calculation time fails to be completed within the preset time, the calculation of the complete model is stopped and the system is switched to the efficiency mode; S5. When the initial mode decision is the efficiency mode, or when switching to the efficiency mode, the fixed parameters and the operating parameters are input into the pre-trained bridging model to calculate a set of model coefficients; then, based on the operating parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result.

[0007] Optionally, the training steps of the pre-trained bridging model include the following steps: The operating conditions are divided into different zones based on one or more key parameters; A transition zone is set between adjacent partitions, and the data training range of the bridging model corresponding to the partition is determined based on the partition and the transition zone. The data training range of the bridging model of the first partition extends into the transition zone of the adjacent second partition, and the data training range of the bridging model of the second partition also extends into the transition zone of the first partition, so as to form overlapping training data. Based on the data within each training range, bridging models corresponding to different partitions are trained respectively.

[0008] Optionally, the step of training bridging models corresponding to different partitions based on data within each training range includes the following sub-steps: The combination of the fixed parameters and the running parameters within the training range of the data is traversed, and for each parameter combination, the calculation of the complete model is performed to generate a corresponding reference outlet temperature as the physical true value for training. For each combination of parameters and its corresponding reference outlet temperature, the simple model is solved in reverse to determine a set of model coefficients that enable the simple model to reproduce the reference outlet temperature, and this set of model coefficients is defined as the training target. Using the parameter combination as training input and the training target as training label, the bridging model is trained using machine learning until the bridging model converges, and a nonlinear mapping relationship is established between the parameter combination and the model coefficients.

[0009] Optionally, S1 includes the following sub-steps: S11. Based on the geometric and physical parameters, establish a transient heat transfer model to characterize the spatiotemporal dynamic changes of blood temperature within the extracorporeal circulation tubing; S12. Obtain the operating parameters corresponding to the hemodialysis process, and use the operating parameters to constrain the transient heat transfer model to generate a constrained heat transfer model that reflects the heat exchange boundary conditions in the real-time dialysis scenario, as a complete model.

[0010] Optionally, S2 includes the following steps: S21. Obtain the mode preference input from the user interface, wherein the mode preference includes at least precision priority or efficiency priority; S22. Monitor the computing resources of the operating environment of the determination method in real time to generate a quantitative computing power status index; S23. Based on the mode preference and the computing power status index, execute the initial mode decision: When the mode preference is efficiency first, then the efficiency mode is enabled. When the mode preference is accuracy priority, the computing power status index is compared with a preset computing power threshold. If the computing power status indicator is greater than or equal to the computing power threshold, then the precision mode is activated. If the computing power status indicator is less than the computing power threshold, then the efficiency mode is activated.

[0011] Optionally, S4a includes the following sub-steps: S4a1. Numerically solve the constrained heat transfer model to generate temperature field distribution data characterizing the temperature at various points inside the extracorporeal circulation pipeline under a preset dialysis duration; S4a2. Extract the temperature value corresponding to the blood outlet position of the extracorporeal circulation tubing from the temperature field distribution data, and use it as the predicted outlet temperature.

[0012] Optionally, S5 includes the following steps: S51. Based on the fixed parameters and the operating parameters, determine the position of the parameters in the partition to locate a parameter region; S52. When the parameter region is not located within the transition zone, the bridging model corresponding to the parameter region is called to calculate the model coefficients of the simple model. Based on the running parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result. S53. When the parameter region is located within the transition zone, perform the following steps: S531. Call the bridging model corresponding to the two adjacent partitions of the transition zone, and calculate the two sets of model coefficients respectively; S532. Based on the operating parameters and the two sets of model coefficients, a first intermediate temperature and a second intermediate temperature are calculated using the simple model respectively; S533. Calculate a deterministic fusion weight based on the position of the parameter within the transition band; S534. Based on the fusion weight, the first intermediate temperature and the second intermediate temperature, perform a weighted calculation to obtain the predicted outlet temperature, and return it as the final result.

[0013] Secondly, the computer device provided in this application adopts the following technical solution: A computer device comprising: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: The above-described method for switching patient blood temperature prediction models based on computational power differences is implemented.

[0014] Thirdly, this application provides a computer-readable storage medium that adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above.

[0015] The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the following: The above-mentioned method for switching patient blood temperature prediction models based on differences in computing power.

[0016] This application has at least one of the following beneficial effects: 1. This method achieves a better balance between computational accuracy and real-time responsiveness by introducing a mechanism that dynamically switches between accuracy and efficiency modes. When the system has sufficient computing power, it prioritizes the use of the high-precision complete model for calculation; while when computing power is low or calculation times out, the system can automatically switch to the high-efficiency simplified model, thereby ensuring that an outlet predicted temperature can be returned to the user in a timely manner under different operating environments.

[0017] 2. This method addresses the consistency issue between different models by constructing a pre-trained bridging model. This bridging model is trained offline using the high-precision calculation results of the full model as the physical truth. This ensures that the model coefficients relied upon by the efficiency mode are a mapping of the physical characteristics of the full model. Therefore, the calculation results of the efficiency mode can maintain a high degree of consistency with those of the precision mode, effectively avoiding the problem of abrupt changes in the predicted outlet temperature caused by differences in model principles when switching between the two modes.

[0018] 3. This method further ensures the smoothness of the efficiency model's output by introducing partitions, transition bands, and overlapping training data into the bridging model. When input parameters (such as blood flow velocity) vary between different operating condition partitions, the system uses transition bands and deterministic fusion weights to weight the model results of adjacent partitions. This avoids abrupt changes in temperature predictions that may occur due to crossing partition boundaries, improving the computational stability of the efficiency model under continuous parameter variations. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a method for switching a patient blood temperature prediction model based on computing power differences is shown in one embodiment of the present invention.

[0020] Figure 2 A schematic diagram of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0021] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application.

[0022] In the following description, numerous specific details are set forth for purposes of explanation in order to provide a thorough understanding of the inventive concept. As part of this specification, some of the accompanying drawings of this disclosure are block diagrams illustrating structures and devices to avoid complicating the disclosed principles. For clarity, not all features of the actual embodiment need to be described. Furthermore, the language used in this disclosure has been primarily chosen for readability and instructional purposes and may not have been chosen to define or limit the subject matter of the invention, thus requiring the necessary claims to determine such inventive subject matter. References to “an embodiment” or “an embodiment” in this disclosure mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment, and multiple references to “an embodiment” or “an embodiment” should not be construed as necessarily referring to the same embodiment.

[0023] Unless explicitly defined, the terms “a,” “an,” and “the” are not intended to refer to a singular entity, but rather to include a general category whose specific examples can be used for illustration. Therefore, the use of the terms “a” or “an” can mean any number of at least one, including “a,” “one or more,” “at least one,” and “one or more.” The term “or” means any of the options and any combination of the options, including all options unless explicitly indicated that the options are mutually exclusive. The phrase “at least one of” when combined with a list of items refers to a single item in the list or any combination of items in the list. The phrase does not require all items listed unless explicitly defined as such.

[0024] This application discloses a method for switching patient blood temperature prediction models based on differences in computing power, referring to... Figure 1 This includes the following steps S1-S5.

[0025] S1. Read the pre-built complete model, simple model, and pre-trained bridging model used to associate the complete model and the simple model, and obtain the fixed parameters and operating parameters of the extracorporeal circulation tubing for hemodialysis to complete the initialization of the complete model; wherein, the fixed parameters include geometric parameters and physical parameters, and the operating parameters correspond to the hemodialysis process.

[0026] During hemodialysis treatment, the hemodialysis device draws the patient's blood out of the body and flows through an extracorporeal circulation system. This extracorporeal circulation system typically consists of an arterial segment, a dialyzer, and a venous segment. The working principle is that the blood first flows through the arterial segment into the dialyzer, where it undergoes purification and heat exchange, and then returns to the patient's body via the venous segment.

[0027] A complete model is a refined model based on physical principles, such as a model built upon the partial differential equation of heat conduction. This complete model can predict the outlet temperature because it uses the geometric and physical parameters obtained in S1 to construct a mathematical description characterizing the temperature distribution inside the pipe. The model can solve for a spatially continuously varying temperature field that reflects the dynamic temperature changes of blood along its path from inlet to outlet. By extracting the temperature value corresponding to the outlet location from the final generated temperature field data, the complete model can provide a relatively accurate prediction result.

[0028] A simple model is a highly efficient empirical model. In some embodiments, the simple model can be built based on lumped parameters, such as relying on two key model coefficients: one characterizing the heat exchange efficiency of the dialyzer, and the other characterizing the heat dissipation capacity of the piping environment. The simple model is characterized by its simple structure, low computational resource requirements, high operating efficiency, and ability to quickly obtain a rough estimate, but its prediction accuracy is relatively low.

[0029] This embodiment introduces a pre-trained bridging model to address potential technical issues when switching between the full model and the simple model, two different computational paths. Because the full model and the simple model differ in their mathematical foundations and physical assumptions, switching between modes may result in two inconsistent blood outlet temperature values ​​calculated for the same set of input parameters—the same fixed and operational parameters. Such abrupt changes or discontinuities in results reduce system reliability. The bridging model establishes a correlation between the two models. During offline training, it uses the output of the full model as a benchmark, forcing the simple model to conform to the behavior of the full model during online execution, thereby ensuring consistency in the output results of the two modes.

[0030] Specifically, in one embodiment, the training steps of the pre-trained bridging model include the following steps ac.

[0031] a. Divide the operating conditions into different zones based on one or more key parameters; wherein, the key parameters are operating parameters.

[0032] b. Set a transition zone between adjacent partitions, and determine the data training range of the bridging model corresponding to the partition based on the partition and the transition zone, wherein the data training range of the bridging model of the first partition extends into the transition zone of the adjacent second partition, and the data training range of the bridging model of the second partition also extends into the transition zone of the first partition, so as to form overlapping training data.

[0033] c. Based on the data within each training range, train bridging models corresponding to different partitions.

[0034] The key parameters here refer to those selected from the operating parameters that have a significant impact on the heat exchange results, and whose changes can easily distort the model's output curve. For example, blood flow rate or ambient temperature can be used as key parameters. The reason for setting up partitions is that fitting all operating conditions with a single model is difficult and may not achieve ideal accuracy under all conditions. This embodiment uses a partitioning approach to decompose the complex global fitting problem into multiple local fitting problems. The partitioning can be based on the physical thresholds of the key parameters or their statistical distribution, and the selection of partition nodes should ensure that the data characteristics within each partition are relatively consistent.

[0035] Using a partitioned model offers several advantages over a single model. First, partitioned models preserve physical interpretability. The two coefficients a simple model relies on, such as those characterizing dialyzer efficiency and those characterizing pipe heat dissipation, have clear physical meanings. If a single model is used to directly fit the final temperature, the model itself becomes a difficult-to-interpret black box, hindering debugging and validation. Second, partitioned models offer the possibility of incremental training and low-cost maintenance. When the complete model changes due to physical code updates or bug fixes, a large single model might require days of complete retraining. A partitioned model, however, only needs incremental training for the affected partitions, significantly reducing maintenance costs. Finally, partitioned models facilitate achieving high local accuracy. A partitioned expert bridging model only needs to focus on fitting data within its specific operating range, making it easier to achieve high local fitting accuracy compared to a single general model attempting to learn all operating conditions.

[0036] To put it another way, considering the various solutions tried in practice, setting a transition zone between adjacent partitions is a better approach. Without a transition zone, the bridging model invoked by the system will switch instantaneously when key parameters cross partition boundaries. Since the bridging models for different partitions are trained independently, their predicted values ​​at parameter boundaries are almost impossible to be completely continuous. This can lead to abrupt changes in the outlet predicted temperature calculated by the simple model. In the field of medical-aided computing, this inconsistency must be avoided; a deterministic computation method should guarantee that the same input parameters produce a continuous and predictable output.

[0037] This embodiment does not employ a smooth transition based on time windows to ensure the determinism of the method and avoid uncertainties in the calculation results caused by user operation history, i.e., historical correlation. The use of overlapping training data, as defined in step b, to set the transition band addresses the problem of out-of-distribution inference. If overlapping data is not used, and the model performs fusion calculations at the boundary, the low-region model will be forced to predict high-region data outside its training range, and vice versa, introducing prediction risks. By providing overlapping training data within the transition band, the predictions of two adjacent partition bridging models within this region are based on their training data, thus ensuring a natural continuity of the outputs of the two models in this boundary region. This data overlap within a small range has a controllable impact on the overall accuracy of the model, but significantly improves the stability and continuity of the model at the partition boundaries.

[0038] Specifically, in one embodiment, the step of training bridging models corresponding to different partitions based on data within each data training range includes the following sub-steps c1-c3.

[0039] c1. Iterate through the combinations of the fixed parameters and the running parameters within the training range of the data, and for each parameter combination, perform the calculation of the complete model to generate a corresponding benchmark outlet temperature as the physical true value for training.

[0040] c2. For each parameter combination and its corresponding reference outlet temperature, solve the simple model in reverse to determine a set of model coefficients that enable the simple model to reproduce the reference outlet temperature, and define this set of model coefficients as the training target.

[0041] c3. Using the parameter combination as training input and the training target as training label, perform machine learning training on the bridging model until the bridging model converges, and establish a nonlinear mapping relationship between the parameter combination and the model coefficients.

[0042] These three sub-steps describe the core principles of the offline training phase. Sub-step c1 is the process of generating high-precision benchmark data. In this step, the system performs extensive simulation calculations for a specific data training range, such as the range corresponding to a low blood flow velocity zone. For example, the system selects a set of parameter combinations within this range, such as a fixed tube length of 2000 mm, a blood flow velocity of 250 ml / min, and an ambient temperature of 22 degrees Celsius. The system then calls a high-precision complete model, i.e., a numerical solution model based on partial differential equations, to calculate this specific operating condition, ultimately obtaining a high-precision outlet temperature, such as 36.45 degrees Celsius. This 36.45 degrees Celsius is considered the physical true value or benchmark outlet temperature under this parameter combination. The system iterates through tens of thousands of parameter combinations within this data training range, repeating this process to generate a large benchmark dataset.

[0043] Sub-step c2 is the process of preparing labels for machine learning training. Continuing the example above, the system now has a set of inputs (flow rate 250, ambient temperature 22, etc.) and a corresponding standard answer (36.45 degrees Celsius). At this point, the system uses the mathematical formula of a simple model to inversely solve for a set of model coefficients, such as a dialyzer efficiency coefficient and a pipe heat dissipation coefficient, so that the simple model, when receiving the same input, will also calculate a result exactly equal to 36.45 degrees Celsius. This inverse solution process may yield a set of coefficients, for example, an efficiency coefficient of 0.265 and a heat dissipation coefficient of 0.0031. This set of coefficients (0.265, 0.0031) is defined as the training objective corresponding to this parameter combination.

[0044] Sub-step c3 is the final process of training the bridging model. The system uses the parameter combination generated in c1 as training input and the model coefficients generated in c2 as training labels. Continuing the example above, a training data pair is: input (flow rate 250, ambient temperature 22...) corresponding to labels (0.265, 0.0031). The system inputs all data pairs within this training range into a machine learning algorithm, such as a gradient boosting tree or a neural network, to train a specific bridging model for low blood flow rate zones. After the trained bridging model converges, a non-linear mapping relationship is established: given a set of parameters within the low flow rate range during runtime, it can quickly predict a corresponding set of optimal model coefficients for use by simple models.

[0045] Specifically, in one embodiment, S1 includes the following sub-steps S11-S12.

[0046] S11. Based on the geometric and physical parameters, establish a transient heat transfer model to characterize the spatiotemporal dynamic changes of blood temperature within the extracorporeal circulation tubing.

[0047] S12. Obtain the operating parameters corresponding to the hemodialysis process, and use the operating parameters to constrain the transient heat transfer model to generate a constrained heat transfer model that reflects the heat exchange boundary conditions in the real-time dialysis scenario, as a complete model.

[0048] To accurately describe the physical characteristics of this closed system, it is first necessary to obtain its inherent geometric and physical parameters. Geometric parameters primarily describe the macroscopic physical dimensions of the extracorporeal circulation tubing, such as the total length, inner diameter, outer diameter, and wall thickness. These parameters determine the distance blood travels within the tubing and the surface area for heat exchange with the external environment, thus affecting the spatial distribution of blood temperature. Physical parameters describe the thermodynamic properties of the various materials constituting the system, such as the thermal conductivity of the tubing materials, the density of the blood itself, its specific heat capacity, and thermal conductivity. These parameters determine the rate of heat transfer between different media and the ease with which the blood's temperature changes.

[0049] The transient heat transfer model treats temperature as a function of both spatial location and time, and mathematically integrates the aforementioned geometric and physical parameters into a unified system. This allows for the calculation of the blood temperature at any point during dialysis within the extracorporeal circulation tubing. This representation captures the dynamic evolution of temperature, rather than a static final value, facilitating subsequent accurate calculations.

[0050] Optionally, in one embodiment, the transient heat transfer model is established based on the partial differential equation of heat conduction, and the governing equation of the partial differential equation of heat conduction is: , in, For heat capacity, For heat conduction, For convective heat transfer, This refers to the heat source for external heat exchange.

[0051] The partial differential equation for heat conduction is a fundamental physical equation describing the spatiotemporal evolution of heat transfer and temperature distribution within a physical system. Essentially, it is a mathematical expression of energy conservation analysis within a small volume. The left-hand side of the equation contains the heat capacity term, where ρ is the fluid density. Specific heat capacity of the fluid The rate of change of temperature with respect to time, together with other terms, characterizes the rate at which a fluid per unit volume absorbs or releases energy due to its own temperature change. The heat conduction term on the right-hand side of the equation, where k is the thermal conductivity of the fluid, The temperature gradient term describes the conduction and diffusion of heat within the fluid medium due to temperature differences. The convective heat transfer term, where v is the fluid velocity vector, describes the process of heat being carried and transported by the macroscopic motion of the fluid. The heat source term Q for external heat exchange represents the net rate of heat exchange between the system and the external environment.

[0052] In the extracorporeal circulation process of hemodialysis, blood, as a fluid, exhibits thermodynamic behavior that conforms to the physical laws described by the partial differential equation of heat conduction. The flow of blood within the tubing corresponds to the convective heat transfer process in the equation. The heat transfer within the blood itself due to temperature differences corresponds to the heat conduction process. The heat exchange between the blood and the dialysate and surrounding air as it flows through the dialyzer and the tubing exposed to the environment corresponds to the heat source term of external heat exchange. Therefore, this equation can characterize the spatiotemporal dynamic changes in temperature caused by the combined effects of multiple physical mechanisms during the complex process of extracorporeal circulation.

[0053] The operating parameters of a hemodialysis process refer to a set of variable physical quantities used to describe the real-time operating conditions during a single dialysis treatment. These parameters mainly include the patient's body temperature, dialysis machine settings such as blood flow rate and dialysate temperature, and environmental parameters such as ambient temperature. Each operating parameter directly affects the temperature of the blood circulating outside the body. For example, the patient's body temperature determines the initial thermal state of the blood when it leaves the body; blood flow rate affects the residence time of blood in the tubing, thus changing the total amount of heat exchanged with the outside world; and dialysate temperature and ambient temperature constitute the main external heat sources or heat sinks for heat exchange in the dialyzer area and the exposed tubing area, respectively.

[0054] Constraining the transient heat transfer model using these operating parameters is achieved by transforming these parameter values ​​into the initial and boundary conditions necessary for model solving. This process is not simply substituting the parameters as variables into a fixed formula, but rather using them to define the mathematical and physical boundaries of the physical model in this specific operation. For example, the acquired real-time patient body temperature is set as the initial temperature value for the model at the tubing inlet; the dialysate temperature and ambient temperature are used to calculate the heat flux density during convective heat transfer between the dialyzer region and the tubing outer wall and the outside environment, respectively.

[0055] The generation process of the constrained heat transfer model takes as input the aforementioned general transient heat transfer model and a set of operating parameters acquired at the current moment. Its internal mechanism involves mapping and assigning values ​​to these operating parameters one-to-one with the boundary condition terms in the model, thereby transforming a universal physical equation into a mathematical model with completely defined boundaries that can accurately describe the thermodynamic system of the specific dialysis process. The output of this process is the constrained heat transfer model.

[0056] Optionally, in one embodiment, the boundary conditions used to constrain the transient heat transfer model using the operating parameters include ad.

[0057] a. At the blood inlet of the extracorporeal circulation tubing, set the inlet boundary condition to be that the blood inlet temperature is equal to the preset patient body temperature; b. In the dialyzer region of the extracorporeal circulation tubing, the boundary conditions on the dialyzer membrane surface are set to be based on convective heat transfer between blood and dialysate; c. In the non-dialyzer region of the extracorporeal circulation tubing, the boundary conditions of the tubing outer wall are set to be based on convective heat transfer between the tubing outer wall and the external environment; d. At the blood outlet of the extracorporeal circulation tubing, set the outlet boundary condition to a free outflow condition with zero temperature gradient.

[0058] These boundary conditions provide constraints for the transient heat transfer model. Specifically, the inlet boundary condition forces the temperature value at the spatial starting point of the numerical solver to be the patient's actual body temperature, providing a physiologically accurate starting state for the iterative calculation of the entire temperature field. For example, if the obtained patient body temperature is 36.8 degrees Celsius, the temperature value at the tube inlet of the model is fixed at 36.8 degrees Celsius. The boundary conditions between the dialyzer and non-dialysisr regions do not directly set temperature values, but rather dynamically calculate the heat flux density across the interface based on the real-time temperature difference between the blood and the outside environment during the model's solution process. This calculated heat flux acts as a source or sink term in the numerical algorithm, directly affecting the rate of temperature change of nodes near the boundary. For example, in the dialyzer region, the model continuously calculates the heat exchange rate based on the difference between the current blood temperature and the set dialysate temperature of 36.5 degrees Celsius. The outlet boundary condition ensures that no non-physical heat backflow or dissipation occurs at the end of the computational domain by specifying that the temperature gradient at the outlet is zero, thus guaranteeing the stability of the numerical solution.

[0059] In hemodialysis, the operating parameters involved can be categorized into fixed parameters and variable parameters. This distinction is made because the external environment and the patient's condition can actually change during the several-hour dialysis process, and these changes continuously affect the constrained heat transfer model. For example, the ambient temperature in the treatment room may fluctuate periodically due to the start and stop of the air conditioning system, and the patient's body temperature may also undergo slight changes due to physiological regulation. When the ambient temperature decreases, the heat exchange rate described by the boundary conditions on the outer wall of the tubing increases, leading to increased heat dissipation from the blood; when the patient's body temperature rises, the initial temperature value set by the inlet boundary conditions will also increase accordingly, thereby altering the temperature distribution baseline throughout the tubing.

[0060] Therefore, the processing of operating parameters can differ in different specific embodiments. For devices with limited computing power, or in application scenarios where high-precision temperature prediction is not required, operating parameters can be treated as fixed parameters. In this case, the method acquires all operating parameters at once at the start of dialysis, for example, by reading standard values ​​preset in a database or manually input values, and uses this set of fixed parameters to constrain the model once. When computing power is sufficient, the operating parameters can be dynamically adjusted according to changes in the external environment and the patient's condition. In this implementation, the method periodically or in real-time acquires the latest parameter values ​​from devices such as sensors, and continuously updates the constrained heat transfer model with these dynamically changing parameters. That is, what is acquired at this time are the operating parameters dynamically adjusted for the real-time hemodialysis process to achieve higher-precision temperature prediction.

[0061] S2. Make an initial mode decision based on the current computing power or user preferences, and determine whether to enable the precision mode or the efficiency mode.

[0062] This step is the decision-making step for this method to achieve a balance between computing resource optimization and response speed. The system will first make an initial mode decision each time a computing request is executed.

[0063] Current computing power refers to the computing resource status of the computer device executing the determined method at the current moment. This can be a quantitative metric, for example, by comprehensively assessing factors such as real-time monitoring of processor (CPU) utilization, available memory size, or whether other high-load computing tasks are being processed in parallel. A state of sufficient computing power means that the device is capable of completing complex calculations in precision mode within a reasonable timeframe.

[0064] User preferences refer to configuration options that can be preset by the user. For example, a user can choose between accuracy priority or efficiency priority based on the usage scenario. Accuracy priority indicates that the user wants to obtain the most accurate calculation results of the complete model, while efficiency priority indicates that the user wants the system to always provide the fastest instant response. In this case, the system can be configured to always enable efficiency mode.

[0065] It should be noted that this decision is not a one-time event. After switching to efficiency mode, the system still has the opportunity to revert to precision mode. In an optional embodiment, the system can be configured to periodically or continuously monitor the computing power status while executing efficiency mode. When the system detects that the low computing power state that previously caused the switch has recovered, for example, when the CPU utilization rate falls below a preset threshold, the system can re-execute the S2 decision at the start of the next computing task and determine whether to return to enabling precision mode based on the recovered sufficient computing power.

[0066] Specifically, in one embodiment, S2 includes the following steps S21-S23.

[0067] S21. Obtain the mode preference input by the user interface, wherein the mode preference includes at least precision priority or efficiency priority.

[0068] S22. Monitor the computing resources of the operating environment of the determination method in real time to generate a quantitative computing power status index.

[0069] S23. Based on the mode preference and the computing power status index, execute the initial mode decision: When the mode preference is efficiency first, then the efficiency mode is enabled. When the mode preference is accuracy priority, the computing power status index is compared with a preset computing power threshold. If the computing power status indicator is greater than or equal to the computing power threshold, then the precision mode is activated. If the computing power status indicator is less than the computing power threshold, then the efficiency mode is activated.

[0070] These three sub-steps detail the specific implementation of the S2 decision-making step. In S21, mode preference is a configuration option provided to the user, allowing them to select based on the priority of their use case. For example, when conducting clinical research or validating device parameters, the user might choose accuracy priority to obtain high-precision data calculated by the full model. In a typical real-time monitoring scenario, the user might choose efficiency priority to ensure the system always provides the fastest computational response.

[0071] The computing resources of the operating environment can include the current utilization of the central processing unit (CPU), the amount of available physical memory, or the busy / idle status of dedicated computing units. A quantified computing power status metric is a numerical value used for decision-making. For example, the system can set a CPU utilization monitoring benchmark. If the current CPU utilization is 30%, the computing power status metric can be quantified as a value indicating ample resources, such as 70; if the utilization is 90%, the metric can be quantified as a value indicating low resources, such as 10.

[0072] When the user selects efficiency priority in S21, the system will unconditionally determine to enable efficiency mode, i.e., execute step S5. When the user selects precision priority, the system will perform a judgment: it will compare the computing power status index generated in S22 with a preset computing power threshold. For example, suppose the computing power threshold is set to 50. If the index generated in S22 is 70, since 70 is greater than 50, it indicates that the computing power is sufficient, and the system will determine to enable precision mode, i.e., execute step S3.

[0073] One preferred scenario that requires special explanation is if the user selects precision priority, but the computing power status index generated by S22 is 30. Since 30 is less than 50, the system will determine that the current computing power is insufficient to support smooth operation in precision mode. In this case, the decision logic of S23 will determine to still enable efficiency mode. This design is a proactive degradation strategy that avoids the system forcibly entering precision mode when computing power is insufficient, which would then immediately trigger an S4b timeout due to slow computation and switch back to efficiency mode. This ensures that even if the user prefers precision, the system can still provide a stable computing service without timeouts in low computing power environments.

[0074] S3. When the initial mode decision is the precision mode, start the progress monitor to monitor the calculation time of the complete model processing the running parameters and the fixed parameters, and decide whether to enter S4a or S4b based on the monitoring results.

[0075] The progress monitor is a software mechanism that tracks in real time the time consumed by the complete model computation task initiated by S3, i.e., the computation duration.

[0076] A specific example of the monitoring method is as follows: After step S2 determines that precision mode is enabled, the system records a start timestamp while starting to execute the full model. The progress monitor periodically checks the difference between the current timestamp and the start timestamp to obtain a real-time calculation duration. Simultaneously, the system presets a maximum allowable calculation time, i.e., a preset duration, such as 5 seconds. The progress monitor continuously compares the real-time calculation duration with this preset duration. If the full model is calculated and returns a success flag within 5 seconds, the monitoring result is calculation completion, and the system proceeds to S4a. If the real-time calculation duration reaches 5 seconds, but the full model calculation has not yet returned a success flag, the monitoring result is calculation timeout, and the system triggers S4b.

[0077] It's important to note that the preset duration (e.g., 5 seconds) is reasonable because the calculation frequency of blood outlet temperature doesn't need to be extremely high real-time. This is because the corresponding operating parameters, such as changes in patient body temperature or ambient temperature, are inherently a slow process. Therefore, this method can be configured to execute at a relatively low frequency, such as updating the predicted outlet temperature every 10 seconds. With this configuration, the 5-second preset duration ensures sufficient computation time in accuracy mode while also allowing for a quick switch to efficiency mode in case of timeout, thus ensuring the user always receives a calculation result within the overall 10-second cycle. It should also be noted that in some application scenarios, high-speed, high-precision prediction is required, making the switching between efficiency and accuracy modes a practical necessity.

[0078] The monitoring results are primarily influenced by two factors. The first is the complexity of the computational task itself, determined by the fixed and runtime parameters acquired in S1. For example, the longer the user-defined preset dialysis duration (a runtime parameter), or the finer the discretized mesh of the pipeline space defined in the fixed parameters, the more iterations are required for the numerical solution of the complete model, naturally resulting in a longer computation time. The second factor is the current computing power status assessed in step S2. Even if S2 considers the computing power sufficient, the computing resources of the operating environment may dynamically change during the computation in S3. For instance, other system processes might consume processor resources, leading to a reduction in resources allocated to the complete model computation, thus extending its actual computation time and increasing the likelihood of triggering an S4b timeout.

[0079] S4a. When the calculation time is completed within the preset time, extract the predicted outlet temperature from the temperature field generated by the complete model and return it as the final result.

[0080] The temperature field here is generated by the complete model through a high-precision numerical solution process. It is a dataset that characterizes the temperature distribution of blood at all spatial nodes from the inlet to the outlet inside the extracorporeal circulation tubing.

[0081] The extraction operation refers to locating and reading a specific temperature value corresponding to the blood outlet location of the extracorporeal circulation tubing from the temperature field data set. This temperature value is considered the predicted outlet temperature for this calculation. Since this temperature value is calculated using a high-precision, complete model, it is directly returned to the user or system as the final result, thus completing the calculation cycle for this precision mode.

[0082] Specifically, S4a includes the following sub-steps S4a1-S4a2.

[0083] S4a1. Numerically solve the constrained heat transfer model to generate temperature field distribution data characterizing the temperature at various points inside the extracorporeal circulation pipeline under a preset dialysis duration.

[0084] The process of numerically solving the constrained heat transfer model is to transform a physical problem that is continuous in time and space into a discrete computational process that can be executed by a computing device. This solution process may specifically include the following steps S4a11-S4a15.

[0085] S4a11. Perform spatial and temporal discretization. Spatial discretization decomposes the continuous physical length of the extracorporeal circulation tubing into a finite number of discrete spatial nodes, while temporal discretization divides the preset total dialysis duration into a series of discrete time steps.

[0086] S4a12. Set initial conditions. At time zero, when the calculation begins, initialize the temperature values ​​on all spatial nodes to a known state, typically the patient's body temperature.

[0087] S4a13. Perform iterative calculations over time. Starting from the initial moment, within the first time step, solve for the new temperature values ​​of all spatial nodes at the end of the time step based on the algebraic equations transformed by the constrained heat transfer model.

[0088] S4a14. Use this newly calculated temperature distribution as the initial condition for the next time step, and repeat the same solution process.

[0089] S4a15. Continue the iterative process, advancing step by step over time until the preset total dialysis time is calculated.

[0090] Through the above solution process, a temperature field distribution data is finally generated. This data is one or more data structures that store the temperature value corresponding to each spatial node at the end of dialysis. For example, if the tubing is discretized into N nodes along its length, the generated temperature field distribution data can be a one-dimensional array containing N elements. A specific example of this array could be: [36.80, 36.78, 36.75, ..., 36.55, ..., 36.42], where the first element, 36.80 degrees Celsius, corresponds to the temperature of the tubing inlet node, the last element, 36.42 degrees Celsius, corresponds to the temperature of the tubing outlet node, and the middle elements correspond to the temperatures of the intermediate nodes within the tubing. This data visually presents the complete spatial distribution of blood temperature along the tubing from inlet to outlet at the end of dialysis.

[0091] S4a2. Extract the temperature value corresponding to the blood outlet position of the extracorporeal circulation tubing from the temperature field distribution data, and use it as the predicted outlet temperature.

[0092] Extracting the temperature value corresponding to the blood outlet location from the temperature field distribution data is a direct data query operation. In S4a1, the continuous space of the extracorporeal circulation tubing is discretized into a series of ordered spatial nodes, and the generated temperature field distribution data is a set of data storing the temperature values ​​corresponding to these nodes one-to-one at the end of dialysis. Since the blood outlet location corresponds to the last calculation node in the spatial discretization process, the extraction operation involves locating and reading the temperature value corresponding to the last node from this data set.

[0093] The reason for using the method of first calculating the complete temperature field distribution data and then extracting the outlet temperature value from it, rather than directly calculating only the temperature at the outlet, is that in this model, the temperature at any point in the pipeline is the cumulative result of the heat exchange history of all upstream points. During the numerical solution process, the temperature value of each spatial node depends on the temperature value of its adjacent node as the calculation input. This computational dependency constitutes a complete chain from the pipeline inlet to the outlet.

[0094] Therefore, to obtain an accurate outlet temperature, the temperature evolution must be calculated node by node along the entire pipeline length, starting from the inlet. Only by fully simulating the transfer and distribution of heat throughout the entire spatial domain can the final outlet temperature be guaranteed.

[0095] S4b. If the calculation time fails to be completed within the preset time, the calculation of the complete model is stopped and the system is switched to the efficiency mode.

[0096] This step is a timeout handling path in precision mode. This path is triggered by the progress monitor when it detects that the computation time has reached the preset time, but the numerical solution task has not yet been completed.

[0097] The switching process involves two actions. First, the system suspends the calculation of the complete model. This action involves the progress monitor sending a termination signal to the computational process or thread currently performing numerical solutions. This termination signal interrupts the computational task, immediately releasing the computing resources it occupies and preventing it from continuing to consume system performance.

[0098] Secondly, the system switches to efficiency mode. After the computation task is aborted, the control flow of this method will no longer wait for the return result of S4a, but will be redirected to S5 to immediately begin the computation process in efficiency mode. The fixed parameters and operating parameters obtained in step S1 will be passed to S5 for use. This switching mechanism ensures that even if the user prefers high precision and has sufficient computing power, but S3 times out due to excessive task complexity, the system can still provide the user with an outlet predicted temperature calculated in efficiency mode within a predetermined total cycle, such as the 10-second update cycle described in section S3, thus guaranteeing the responsiveness and availability of the system.

[0099] S5. When the initial mode decision is the efficiency mode, or when switching to the efficiency mode, the fixed parameters and the operating parameters are input into the pre-trained bridging model to calculate a set of model coefficients; then, based on the operating parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result.

[0100] The bridging model works by receiving fixed and runtime parameters acquired in S1 during execution in S5. Internally, it establishes a nonlinear mapping from input parameters, such as blood flow rate, ambient temperature, and tubing geometry, to model coefficients. It processes these input fixed and runtime parameters through this mapping and quickly calculates the corresponding model coefficients for a simple set of output models.

[0101] In a specific embodiment, the simple model can be constructed as a two-stage temperature drop calculation model. In this case, the model coefficients output by the bridging model correspond to the key coefficients of these two stages, such as a first coefficient α and a second coefficient β. These two coefficients have clear physical meanings: the first coefficient α characterizes the heat exchange efficiency of blood flowing through the dialyzer, determining the amount of heat exchanged between the blood and the dialysate within the dialyzer due to the temperature difference. The second coefficient β characterizes the ability of blood to dissipate heat to the surrounding environment as it flows through the arterial and venous segments of the tubing.

[0102] Therefore, in step S5, the specific process of calculating the predicted outlet temperature using the simple model is as follows: First, using the first coefficient α and the patient's body temperature and dialysate temperature in the operating parameters, the first intermediate temperature when the blood leaves the dialyzer is calculated; then, using the second coefficient β, the first intermediate temperature, and the ambient temperature and dialysis duration in the operating parameters, the final temperature of the blood after further heat dissipation in the tubing is calculated, and this final temperature is the predicted outlet temperature of S5.

[0103] Specifically, S5 includes the following steps S51-S53.

[0104] S51. Based on the fixed parameters and the operating parameters, determine the position of the parameters in the partition to locate a parameter region.

[0105] The system first extracts one or more parameter values ​​selected as key parameters during the offline training phase. These parameter values ​​are derived from fixed parameters and / or operational parameters acquired by S1. For example, if blood flow velocity is defined as a key parameter, the system reads the current blood flow velocity value. Subsequently, the system compares this value with preset partition boundary points to determine the parameter region to which the current operating condition belongs.

[0106] S52. When the parameter region is not located within the transition zone, the bridging model corresponding to the parameter region is called to calculate the model coefficients of the simple model. Based on the running parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result.

[0107] S52 describes the computation process within the non-transition zone. This step is activated when the parameter region determined by S51 explicitly falls within the core range of a partition, rather than at the boundary between two partitions, i.e., within the transition zone.

[0108] Continuing with the example in the previous c1-c3 sub-steps, assume that the system has been divided into low flow rate zones (e.g., below 290 ml / min) and medium flow rate zones, with a transition zone (e.g., 290 to 310 ml / min) set between them.

[0109] In step S51, the system obtains the current operating parameters, including a blood flow rate of 250 ml / min. The system compares this with the zone boundary and determines that the parameter area (250 ml / min) is located in the low flow rate zone and is not located in the transition zone (290 to 310 ml / min).

[0110] Therefore, the system executes step S52. It invokes a specific bridging model, trained offline, specifically for the low-flow-rate zone. This bridging model receives all the fixed and operational parameters acquired in S1 (including flow rate 250, ambient temperature 22, etc.) and quickly calculates and outputs a set of optimal model coefficients (e.g., efficiency coefficient 0.265 and heat dissipation coefficient 0.0031). The system then substitutes these coefficients along with the operational parameters (such as patient temperature, dialysate temperature, etc.) into the calculation formula of the simple model to calculate the final predicted outlet temperature, for example, 36.45 degrees Celsius. This temperature value is directly returned as the final result of S5.

[0111] S53. When the parameter region is located within the transition zone, perform the following steps S531-S534.

[0112] S531. Call the bridging model corresponding to the two adjacent partitions of the transition zone and calculate the two sets of model coefficients respectively.

[0113] S532. Based on the operating parameters and the two sets of model coefficients, a first intermediate temperature and a second intermediate temperature are calculated using the simple model respectively.

[0114] S533. Calculate a deterministic fusion weight based on the position of the parameter within the transition zone.

[0115] S534. Based on the fusion weight, the first intermediate temperature and the second intermediate temperature, perform a weighted calculation to obtain the predicted outlet temperature, and return it as the final result.

[0116] These four sub-steps S531 to S534 define in detail the smooth fusion calculation process within the parameter transition band, which is crucial to ensuring the continuity of the efficiency mode output results.

[0117] Continuing with the example in S52, the low flow rate zone is below 290 ml / min, and the transition zone is 290 to 310 ml / min. In step S51, the system obtains the current blood flow rate as 300 ml / min. S51 determines that this parameter region is within the transition zone, thus triggering the execution of S53.

[0118] In S531, the system simultaneously invokes two bridging models for adjacent zones in the transition zone: the bridging model for the low-flow-rate zone and the bridging model for the medium-flow-rate zone. Due to the use of overlapping data training, both models are capable of handling an input of 300 ml / min. The low-flow-rate bridging model calculates and outputs the first set of model coefficients (e.g., efficiency coefficient 0.260, heat dissipation coefficient 0.0030) based on all current fixed and operating parameters. Simultaneously, the medium-flow-rate bridging model calculates and outputs the second set of model coefficients (e.g., efficiency coefficient 0.250, heat dissipation coefficient 0.0028) based on the same input parameters.

[0119] In S532, the system uses the two sets of model coefficients obtained in S531 to perform two simple model calculations. First, the system uses the first set of model coefficients (0.260, 0.0030) and operating parameters to calculate the first intermediate temperature, for example, 36.40 degrees Celsius. Then, the system uses the second set of model coefficients (0.250, 0.0028) and the same operating parameters to calculate the second intermediate temperature, for example, 36.36 degrees Celsius.

[0120] In S533, the system calculates a deterministic fusion weight based on the specific location of the parameter within the transition zone. This calculation method ensures the determinism of the result and avoids inconsistencies introduced by operational history. For example, using linear interpolation, the fusion weight can be determined based on the relative position of the current flow velocity (300) within the transition zone interval (290 to 310). The calculation formula can be (current value - interval start point) / (interval end point - interval start point). In this example, the fusion weight is (300 - 290) / (310 - 290), i.e., 10 / 20, resulting in a deterministic fusion weight of 0.5.

[0121] In S534, the system performs the final weighted calculation. The system uses the fusion weight (0.5) calculated in S533 to perform a weighted average of the two intermediate temperatures obtained in S532. For example, the calculation process is (1-0.5)*36.40 degrees Celsius + 0.5*36.36 degrees Celsius, ultimately yielding a predicted outlet temperature of 36.38 degrees Celsius. This temperature value is returned as the final result of S5. This smooth fusion process ensures that when the blood flow rate continuously changes from 290 ml / min to 310 ml / min, the output temperature also changes smoothly and continuously, avoiding abrupt changes in the result.

[0122] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0123] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 2 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database contains data related to a switching method for patient blood temperature prediction models based on computational power differences. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a switching method for patient blood temperature prediction models based on computational power differences.

[0124] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the switching method of the patient blood temperature prediction model based on the difference in computing power described in the above embodiment.

[0125] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the switching method for the patient blood temperature prediction model based on computing power differences described in the above embodiment.

[0126] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments of this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0127] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for switching patient blood temperature prediction models based on differences in computing power, characterized in that, Includes the following steps: S1. Read the pre-built complete model, simple model, and pre-trained bridging model used to associate the complete model and the simple model, and obtain the fixed parameters and operating parameters of the extracorporeal circulation tubing for hemodialysis to complete the initialization of the complete model; wherein, the fixed parameters include geometric parameters and physical parameters, and the operating parameters correspond to the hemodialysis process; S2. Make an initial mode decision based on the current computing power or user preferences, and determine whether to enable precision mode or efficiency mode; S3. When the initial mode decision is the precision mode, start the progress monitor to monitor the calculation time of the complete model processing the running parameters and the fixed parameters, and decide whether to enter S4a or S4b based on the monitoring results; S4a. When the calculation is completed within the preset time, extract the predicted outlet temperature from the temperature field generated by the complete model and return it as the final result; S4b. If the calculation time fails to be completed within the preset time, the calculation of the complete model is stopped and the system is switched to the efficiency mode; S5. When the initial mode decision is the efficiency mode, or when switching to the efficiency mode, the fixed parameters and the operating parameters are input into the pre-trained bridging model to calculate a set of model coefficients; then, based on the operating parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result.

2. The method for switching patient blood temperature prediction models based on computing power differences according to claim 1, characterized in that, The training steps for the pre-trained bridging model include the following: The operating conditions are divided into different zones based on one or more key parameters; wherein, the key parameters are operating parameters. A transition zone is set between adjacent partitions, and the data training range of the bridging model corresponding to the partition is determined based on the partition and the transition zone. The data training range of the bridging model of the first partition extends into the transition zone of the adjacent second partition, and the data training range of the bridging model of the second partition also extends into the transition zone of the first partition, so as to form overlapping training data. Based on the data within each training range, bridging models corresponding to different partitions are trained respectively.

3. The method for switching patient blood temperature prediction models based on computing power differences according to claim 2, characterized in that, The step of training bridging models corresponding to different partitions based on data within each training range includes the following sub-steps: The combination of the fixed parameters and the running parameters within the training range of the data is traversed, and for each parameter combination, the calculation of the complete model is performed to generate a corresponding reference outlet temperature as the physical true value for training. For each combination of parameters and its corresponding reference outlet temperature, the simple model is solved in reverse to determine a set of model coefficients that enable the simple model to reproduce the reference outlet temperature, and this set of model coefficients is defined as the training target. Using the parameter combination as training input and the training target as training label, the bridging model is trained using machine learning until the bridging model converges, and a nonlinear mapping relationship is established between the parameter combination and the model coefficients.

4. The method for switching patient blood temperature prediction models based on computing power differences according to claim 1, characterized in that, S1 includes the following sub-steps: S11. Based on the geometric and physical parameters, establish a transient heat transfer model to characterize the spatiotemporal dynamic changes of blood temperature within the extracorporeal circulation tubing; S12. Obtain the operating parameters corresponding to the hemodialysis process, and use the operating parameters to constrain the transient heat transfer model to generate a constrained heat transfer model that reflects the heat exchange boundary conditions in the real-time dialysis scenario, as a complete model.

5. The method for switching patient blood temperature prediction models based on computing power differences according to claim 1, characterized in that, S2 includes the following steps: S21. Obtain the mode preference input from the user interface, wherein the mode preference includes at least precision priority or efficiency priority; S22. Monitor the computing resources of the operating environment of the determination method in real time to generate a quantitative computing power status index; S23. Based on the mode preference and the computing power status index, execute the initial mode decision: When the mode preference is efficiency first, then the efficiency mode is enabled. When the mode preference is accuracy priority, the computing power status index is compared with a preset computing power threshold. If the computing power status indicator is greater than or equal to the computing power threshold, then the precision mode is activated. If the computing power status indicator is less than the computing power threshold, then the efficiency mode is activated.

6. The method for switching patient blood temperature prediction models based on computing power differences according to claim 4, characterized in that, S4a includes the following sub-steps: S4a1. Numerically solve the constrained heat transfer model to generate temperature field distribution data characterizing the temperature at various points inside the extracorporeal circulation pipeline under a preset dialysis duration; S4a2. Extract the temperature value corresponding to the blood outlet position of the extracorporeal circulation tubing from the temperature field distribution data, and use it as the predicted outlet temperature.

7. The method for switching patient blood temperature prediction models based on computing power differences according to claim 2, characterized in that, S5 includes the following steps: S51. Based on the fixed parameters and the operating parameters, determine the position of the parameters in the partition to locate a parameter region; S52. When the parameter region is not located within the transition zone, the bridging model corresponding to the parameter region is called to calculate the model coefficients of the simple model. Based on the running parameters and the model coefficients, the predicted outlet temperature is calculated using the simple model and returned as the final result. S53. When the parameter region is located within the transition zone, perform the following steps: S531. Call the bridging model corresponding to the two adjacent partitions of the transition zone, and calculate the two sets of model coefficients respectively; S532. Based on the operating parameters and the two sets of model coefficients, a first intermediate temperature and a second intermediate temperature are calculated using the simple model respectively; S533. Calculate a deterministic fusion weight based on the position of the parameter within the transition band; S534. Based on the fusion weight, the first intermediate temperature and the second intermediate temperature, perform a weighted calculation to obtain the predicted outlet temperature, and return it as the final result.

8. A computer device, characterized in that, It includes: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: execute the switching method for a patient blood temperature prediction model based on computing power differences according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement: the switching method for patient blood temperature prediction models based on computing power differences as described in any one of claims 1 to 7.

Images