Digital native model energy prediction method and system of physically driven computing power center

Abstract

The application discloses a digital native model energy prediction method and system of a physically driven computing power center. The method comprises: obtaining time series mass flow, temperature data and geometric physical parameters required for the operation of a cold storage tank, setting an initial to-be-identified parameter set and prior information such as flow direction, to obtain input data and parameters; based on energy conservation, a layered ordinary differential equation model describing the temperature change of each layer is constructed by using the input data and parameters; the temperature distribution of each layer at each time point is predicted based on the layered ordinary differential equation model by using an adaptive step ODE solver according to the input data and parameters, to obtain a predicted temperature; a loss function is constructed by comparing the predicted temperature with an actual measured value; the gradient is calculated based on the loss function, and the parameters are adjusted by an optimization algorithm to minimize the loss function. By implementing the method of the application, the model dimension can be reduced and the calculation can be simplified while the main physical mechanism is retained.

Description

TECHNICAL FIELD

[0001] The present application relates to digital twinning, and more particularly to a digital native model energy prediction method and system for a physically driven computing power center. BACKGROUND

[0002] With the rapid development of artificial intelligence technology, high-performance computing facilities represented by AIDC (Artificial Intelligence Data Center) are expanding in scale. These data centers are characterized by continuous increase in rack power density and overall installed capacity, which puts higher demands on the cooling system: high cooling capacity, high reliability, and low energy consumption. However, traditional data center cooling modes, such as direct cooling methods like water chillers, cooling towers, end air conditioners, or liquid cooling, are difficult to meet the needs of peak load management, safety redundancy guarantee, and life cycle cost optimization at the same time when facing new constraints such as time-of-use electricity prices, renewable energy access, and grid demand response.

[0003] As a key equipment, the cold storage tank stores cold energy during periods of low electricity prices or high energy efficiency, and releases it during peak or failure periods, achieving time dimension peak load shifting of power load. This not only helps to reduce the pressure on the cooling machine and power distribution system, improves the flexibility of the cooling system, but also improves the proportion of local consumption of renewable energy. Therefore, in the AIDC scenario, the cold storage tank has become an indispensable important infrastructure on the cold source side. Although the application of cold storage tanks is widespread, in existing engineering practice, the capacity selection and operation strategy of cold storage tanks still mainly rely on experience and simple rules, and lack the ability to accurately depict the charging and discharging dynamic process under complex working conditions. Most methods only use rough static energy balance or black box data fitting, which cannot accurately assess the impact of these processes on unit efficiency, PUE (Power Usage Effectiveness), and cooling safety. In addition, there is currently a lack of high-fidelity, differentiable cold storage tank models and simulation tools that are deeply integrated with load forecasting, electricity / carbon price forecasting, optimal scheduling, and model predictive control algorithms, limiting the collaborative optimization potential between cold storage and chillers, new energy, and electricity prices, hindering the further cost reduction and efficiency improvement of AIDC energy systems and low-carbon transformation.

[0004] The closest existing technical solution is a layered cold storage tank physical modeling method based on PDEs (Partial Differential Equations). This method treats the cold storage tank as a one-dimensional continuous medium, using the temperature distribution T(z,t) as the state variable, and establishing PDE governing equations based on factors such as convection formed by inflow and outflow water, heat conduction between adjacent water layers, and heat exchange between the tank wall and the environment to describe the changes in the temperature field inside the cold storage tank with time and space. However, this type of method has the following shortcomings: because the model is in the form of distributed parameters, a large number of spatial nodes need to be introduced after discretization, forming a high-dimensional state and a large-scale algebraic equation system. The computational load of a single simulation is huge, making it difficult to meet the real-time requirements of long-term, multi-condition, and high-frequency simulations in AIDC scenarios. The PDE model contains many local parameters such as interlayer thermal conductivity, mixing parameters, and tank wall heat transfer coefficient. Most of these parameters rely on empirical settings, which are difficult to effectively identify and continuously correct when the number of sensors is limited, resulting in poor adaptability of the model to changes in actual operating conditions. The discrete solution process of PDE is not conducive to efficient gradient calculation within the framework of automatic differentiation, which limits its end-to-end coupling capability with algorithms such as load forecasting, electricity / carbon price forecasting, optimal scheduling, and model predictive control.

[0005] Therefore, it is necessary to design a new method that, while retaining the main physical mechanisms, reduces the model dimensionality and simplifies the calculations, so that it can efficiently simulate the charging and discharging process of cold storage tanks and be easily integrated with load forecasting and energy dispatch optimization algorithms, thereby achieving more efficient energy management and optimization. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a digital native model energy prediction method and system for physical-driven computing centers.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a digital native model energy prediction method for physically driven computing centers, comprising:

[0008] Acquire time-series mass flow rate, temperature data, and geometric physical parameters required for the operation of the cold storage tank, and set the initial set of parameters to be identified and prior information such as flow direction to obtain input data and parameters;

[0009] Based on the law of conservation of energy, a layered ordinary differential equation model describing the temperature changes of each layer is constructed using the input data and parameters.

[0010] The adaptive step-size ODE solver is used to predict the temperature distribution of each layer at each time point based on the input data and parameters according to the hierarchical ordinary differential equation model, so as to obtain the predicted temperature.

[0011] A loss function is constructed by comparing the predicted temperature with the actual measured value.

[0012] Gradients are calculated based on the loss function, and parameters are adjusted by an optimization algorithm to minimize the loss function.

[0013] Further technical solutions are that the temperature data include inlet water temperature, ambient temperature, and stratification temperature observation values of the cold storage tank at each time; and the geometric and physical parameters include layer volume, layer height, cross-sectional area, water density, and specific heat.

[0014] Further technical solutions are that the initial parameter set to be identified includes layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients, and tank body ambient heat exchange coefficients.

[0015] Further technical solutions are that the layer-specific response factor satisfies ; the thermal inertia factor satisfies ; the interlayer thermal conductivity coefficient satisfies ; and the tank body ambient heat exchange coefficient satisfies .

[0016] Further technical solutions are that based on energy conservation, the input data and parameters are used to construct a stratified ordinary differential equation model describing temperature changes of each layer, including:

[0017] The upstream temperature is determined according to the time series mass flow and its direction to determine the temperature influence;

[0018] The contribution of interlayer heat conduction to temperature change is evaluated by the second-order difference of the temperature difference between adjacent layers and the interlayer thermal conductivity coefficient;

[0019] The influence of ambient heat exchange on layer temperature is calculated using the area of the side wall of each layer, the ambient temperature difference, and the tank body ambient heat exchange coefficient;

[0020] The temperature influence, the contribution of interlayer heat conduction to temperature change, and the influence of ambient heat exchange on layer temperature are integrated to form an ordinary differential equation describing the rate of change of temperature with time;

[0021] The corresponding ordinary differential equation is established for all layers of water in the cold storage tank, and is combined into a whole stratified ODE model to obtain a stratified ordinary differential equation model.

[0022] Further technical solutions are that the ordinary differential equation is expressed as: wherein, is the upstream temperature of the i-th layer; ; is the interlayer thermal conductivity coefficient; is the area of the side wall of each layer; is the ambient temperature difference; the cold storage tank is divided into water layers along the height direction, the first layer has a temperature of , the layer volume is , and the effective heat capacity is ; wherein is the thermal inertia factor of the layer, is the water density, is the specific heat at constant pressure.

[0023] A further technical solution is that the adaptive step ODE solver is used to predict the temperature distribution of each layer at each time point based on the layered ordinary differential equation model according to input data and parameters to obtain a predicted temperature, including:

[0024] The adaptive step ODE solver is called to integrate the layered ODE from the starting time to each observation time based on the layered ordinary differential equation model according to input data and parameters to obtain the predicted temperature of each layer at each observation time.

[0025] A further technical solution is that the loss function includes , wherein is the number of observation times, is the number of layers, and a regularization term for parameter range or smoothness is added in , and is a set of physical parameters to be identified.

[0026] A further technical solution is that the gradient is calculated based on the loss function, and the parameters are adjusted through an optimization algorithm to minimize the loss function, including:

[0027] Based on the loss function, the gradient of the loss with respect to each physical parameter is calculated by automatic differentiation or adjoint extended system, and the parameters are updated according to a preset optimization algorithm.

[0028] The application also provides a digital native model energy prediction system of a physically driven computing power center, including:

[0029] An acquisition unit is configured to acquire time series mass flow, temperature data and geometric physical parameters required for operation of the cold storage tank, set an initial set of to-be-identified parameters and prior information such as flow direction, and obtain input data and parameters;

[0030] A model construction unit is configured to construct a layered ordinary differential equation model describing temperature changes of each layer based on energy conservation and using the input data and parameters;

[0031] a prediction unit configured to predict the temperature distribution of each layer at each time point based on the hierarchical ordinary differential equation model according to the input data and parameters using an adaptive step ODE solver to obtain a predicted temperature;

[0032] a function construction unit configured to construct a loss function by comparing the predicted temperature with the actual measured value;

[0033] an optimization unit configured to calculate the gradient based on the loss function and adjust the parameters by an optimization algorithm to minimize the loss function.

[0034] The present application has the following advantages compared with the prior art: the present application constructs a low-dimensional ordinary differential equation model by acquiring operation data and geometric and physical parameters and discretizing along the space, only needs to perform numerical integration on the time dimension, significantly reduces the calculation complexity, uses an adaptive step ODE solver to perform efficient temperature prediction, constructs a loss function by using the difference between the actual measured value and the predicted value, and adjusts the model parameters by an optimization algorithm to achieve accurate calibration. This method not only retains the main physical mechanism of the charging and discharging process of the cold storage tank, but also significantly reduces the model complexity and simplifies the calculation process, so that the model can not only quickly and accurately simulate the behavior of the cold storage tank, but also be easily integrated with load prediction and energy scheduling optimization algorithm, thereby effectively supporting efficient energy management and optimization strategy. This method overcomes the problems of large calculation overhead and difficulty in coupling with upper algorithms of the traditional high-dimensional PDE model, and provides a more flexible and accurate solution.

[0035] The present application will be further described below in conjunction with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0037] Figure 1 a flowchart of the physical-driven digital native model energy prediction method of the computing power center;

[0038] Figure 2 a sub-flowchart of the physical-driven digital native model energy prediction method of the computing power center;

[0039] Figure 3 a schematic block diagram of the physical-driven digital native model energy prediction system of the computing power center;

[0040] Figure 4A schematic block diagram of a computer device provided for an embodiment of the present application is shown in FIG. 1.

[0041] Figure 5 An ODE model schematic diagram of the digital native model energy prediction method of the physically driven computing power center provided for an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without any creative work fall within the scope of protection of the present application.

[0043] It should be understood that, when used in the specification and the appended claims, the terms “comprise” and “include” indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0044] It should also be understood that the terms used in the present application specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, unless otherwise clearly indicated by the context, the singular forms “a”, “an” and “the” are intended to include the plural forms as well.

[0045] It should be further understood that the term “and / or” used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0046] Please refer to Figure 1 , Figure 1 A flowchart schematic diagram of the digital native model energy prediction method of the physically driven computing power center provided for an embodiment of the present application. In addition, please refer to Figure 5The physical driving digital native model energy prediction method of the computing center is applied to a server. Time series mass flow, temperature data and geometric physical parameters are integrated, and an ODE model describing temperature changes of each layer is constructed by using the law of conservation of energy. An adaptive step ODE (Ordinary Differential Equation) solver is used to predict the temperature distribution, and a loss function is constructed by combining the actual measurement values to optimize the parameters. This method reduces the number and range of parameters to be identified and simplifies the calculation complexity by introducing a thermal inertia factor, thereby reducing the model dimension. As a result, the model can efficiently simulate the charging and discharging cooling process of the cold storage tank, and is convenient to integrate with load prediction and energy scheduling optimization algorithms for use, realizing more efficient energy management and optimization, and providing strong technical support for intelligent buildings and regional cooling systems.

[0047] Specifically, under the premise of ensuring energy conservation and strictly meeting the main physical mechanisms (convection, inlet mixing, interlayer heat transfer and environmental heat exchange), a continuous time, differentiable, low-dimensional, efficient and parameter interpretable layered ordinary differential equation model suitable for computing center cold storage tanks is constructed, which can be integrated with load prediction and energy scheduling optimization algorithms for high-precision prediction of cold storage behavior and digital native simulation. The model can be widely used in the design, selection, operation optimization and energy consumption evaluation of cold storage systems on the cold source side of large-scale computing centers such as artificial intelligence data centers, providing support for peak load shifting, improving the efficiency of refrigeration systems and ensuring the safe and stable operation of computing infrastructure.

[0048] Figure 1 is a flowchart of the physical driving digital native model energy prediction method of the computing center provided by the embodiment of the present application. As shown in Figure 1 , the method comprises the following steps S110 to S150.

[0049] S110, obtaining time series mass flow, temperature data and geometric physical parameters required for the operation of the cold storage tank, and setting initial to-be-identified parameter set and prior information such as flow direction to obtain input data and parameters.

[0050] In this embodiment, the input data refers to a series of actual operation data that needs to be collected and used when the method of the embodiment is implemented.

[0051] The parameters refer to the basic numerical values used to construct and calibrate the model, which can be divided into two categories: geometric physical parameters and initial to-be-identified parameter set.

[0052] The time series mass flow is ; the temperature data includes inlet water temperature , ambient temperature , and layer temperature observation value at each time ; the geometric and physical parameters include volume of each layer , layer height , cross-sectional area , density of water , specific heat .

[0053] The initial parameter set to be identified includes layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients, and tank environmental heat exchange coefficients. The initial parameter set to be identified is set , and prior information such as flow direction.

[0054] The layer-specific response factor satisfies ; the thermal inertia factor satisfies ; the interlayer thermal conductivity coefficient satisfies ; and the tank environmental heat exchange coefficient satisfies .

[0055] In this embodiment, let the layer temperature vector , and construct the initial value problem . The flow term switches the upstream temperature according to the flow direction: the remaining terms act on the interlayer temperature difference and the environmental temperature difference; the inlet mixing only takes effect in the front layer, and the weight calculation formula is as follows: ; wherein is the distance from the layer to the inlet (in the number of layers), is the mixing strength, is the attenuation coefficient.

[0056] Parameters and constraints: set the layer-specific response factor for each layer to correct the effective mass flow; the thermal inertia factor acts uniformly in the denominator of the convection term, the heat conduction term, and the environmental heat exchange term, which is equivalent to correcting the effective heat capacity of the layer to , thereby uniformly adjusting the response speed of the layer to all heat flow terms; and introduce the interlayer thermal conductivity coefficient and the environmental heat exchange coefficient to describe the interlayer heat transfer and heat dissipation to the environment. The parameter values are limited by the following physical constraint intervals: the layer-specific response factor satisfies , which is used to limit the amplification multiple of the effective mass flow of each layer; the thermal inertia factor satisfies , which is used to limit the scaling multiple of the effective heat capacity of each layer; the interlayer thermal conductivity coefficient satisfies ; and the tank environmental heat exchange coefficient satisfy .

[0057] The above value range is derived from the engineering experience of the cold storage tank, to ensure numerical stability and model physical consistency.

[0058] S120, based on energy conservation, using the input data and parameters to construct a layered ordinary differential equation model describing the temperature change of each layer.

[0059] In this embodiment, the layered ordinary differential equation model refers to establishing a mathematical model that can accurately describe the temperature change of different layers of water in the cold storage tank with time according to the actual operation data and geometric and physical parameters of the cold storage tank.

[0060] In an embodiment, referring to Figure 2 , the above step S120 can include steps S121-S125.

[0061] S121, determine the upstream temperature according to the time series mass flow and its direction to determine the temperature influence.

[0062] In this embodiment, the mass flow and its flow direction through the cold storage tank are identified, and the upstream temperature of the i-th layer is selected based on this , which is a key factor in determining the current layer temperature .

[0063] By analyzing the real-time recorded mass flow data to determine the water flow direction, the correct upstream temperature is selected as the influencing factor of the flow direction.

[0064] S122, evaluate the contribution of interlayer heat conduction to temperature change through second-order difference of adjacent layer temperature difference and interlayer thermal conductivity.

[0065] In this embodiment, the heat transfer effect between layers due to temperature difference is quantified; the second-order difference of adjacent layer temperature difference is constructed , and is used to describe the effective interlayer heat conduction, and the heat conduction contribution of each layer is calculated.

[0066] S123, calculate the influence of environmental heat exchange on layer temperature using the area of each layer side wall and the environmental temperature difference and the tank environmental heat exchange coefficient.

[0067] In this embodiment, the influence of the external environment on the internal temperature of the cold storage tank is considered.

[0068] According to the area of each layer side wall , the environmental temperature difference and the tank environmental heat exchange coefficient , the heat exchange between each layer and the environment is calculated.

[0069] S124. Integrating the effects of temperature, the contribution of interlayer heat conduction to temperature change, and the influence of environmental heat transfer on layer temperature, an ordinary differential equation describing the rate of temperature change over time is formed.

[0070] By combining all the above influencing factors, a complete mathematical expression is formed to describe the rate of change of temperature in each water layer.

[0071] Integrating the contributions from convection, inlet mixing, interlayer heat transfer, and ambient heat transfer, the final ordinary differential equation is as follows: ,in, Let be the upstream temperature of the i-th layer; ; This refers to the interlayer thermal conductivity. The area of ​​the sidewalls of each floor; This is due to the difference in ambient temperature.

[0072] S125. Establish the corresponding ordinary differential equations for all water layers in the cold storage tank, and combine them into a whole layered ODE model to obtain the layered ordinary differential equation model.

[0073] In this embodiment, the cold storage tank is divided along the height direction into... The first water layer, the... Layer temperature is The layer volume is The effective heat capacity is ;in This is the thermal inertia factor of the layer. For the density of water, It is a specific heat at constant pressure.

[0074] Specifically, for each layer inside the cold storage tank, corresponding ordinary differential equations are established, and these equations are combined to form a unified overall model.

[0075] In this embodiment, the actual water temperature inside the cold storage tank exhibits stratification, and the modeling in this embodiment closely reflects reality. For the N water layers divided along the height of the cold storage tank, ordinary differential equations as shown above are established for each layer, and these equations are combined into a system, thus obtaining the stratified ordinary differential equation model of the entire cold storage tank. The right-hand side of this model is continuously differentiable with respect to the state and parameters, making it naturally suitable for automatic differentiation and gradient-based training, prediction, and control.

[0076] Through the above steps, the embodiment successfully constructs a hierarchical ordinary differential equation model that can efficiently simulate the charging and discharging process of the cold storage tank and is convenient for integrated use with other prediction and optimization algorithms. This not only significantly reduces the model dimension and computational complexity, but also improves the fitting accuracy and robustness of the model to actual working conditions, making it an indispensable part of the energy management of the computing center.

[0077] S130, using an adaptive step ODE solver to predict the temperature distribution of each layer at each time point based on the hierarchical ordinary differential equation model according to the input data and parameters to obtain a predicted temperature.

[0078] In this embodiment, the predicted temperature refers to the temperature distribution of each layer of the cold storage tank at each observation time obtained by calling the adaptive step ODE solver and integrating the hierarchical ordinary differential equation model based on the input data and parameters.

[0079] In an embodiment, the adaptive step ODE solver is called to integrate the hierarchical ODE from the starting time to each observation time based on the hierarchical ordinary differential equation model according to the input data and parameters, to obtain the predicted temperature of each layer at each observation time.

[0080] Given the initial temperature distribution and the input sequence , the adaptive step ODE solver is called to integrate the hierarchical ODE from the starting time to each observation time , thereby obtaining the predicted temperature of each layer at each observation time . This step realizes the physical simulation of the charging and discharging process of the cold storage tank under the current parameters , and forms a differentiable forward model for subsequent gradient calculation and optimization scheduling.

[0081] S140, constructing a loss function by comparing the predicted temperature with the actual measured value.

[0082] In this embodiment, the loss function includes , where is the number of observation times, is the number of layers, and a regularization term for parameter range or smoothness is added in .

[0083] Specifically, at the observation points with measured data, the predicted temperature obtained by forward solving is compared with the corresponding measured temperature to construct a loss function. Preferably, the mean absolute error (MAE) is used, which has the form: , where is the number of observation times, is the number of layers, and a regularization term for parameter range or smoothness can also be added in The regularization term of the superposition pair parameter range or smoothness provides constraints for the automatic calibration process, The set of physical parameters to be identified.

[0084] S150, calculate the gradient based on the loss function, and adjust the parameters through the optimization algorithm to minimize the loss function.

[0085] In this embodiment, based on the loss function, the layered ordinary differential equation model is calculated by automatic differentiation or adjoint expansion system, the gradient of the loss with respect to each physical parameter is calculated, and the parameters are updated according to the preset optimization algorithm.

[0086] Based on the loss function , the layered ODE is calculated by automatic differentiation or adjoint expansion system, and the gradient of the loss with respect to each physical parameter is calculated . Then update the parameters according to the preset optimization algorithm (such as gradient descent or adaptive learning rate algorithm): ; Wherein, is the learning rate. If necessary, project or clip the updated parameters to keep them within the pre-set physically reasonable range.

[0087] Using JAX to automatically differentiate the discrete time step equation and using Adam to optimize the parameters can realize the present application. After the model converges, the obtained parameter set can be fixed for use in the prediction phase

[0088] By cyclically performing the forward prediction process of steps S110-S130 and the parameter calibration process of steps S130-S150 within the same differentiable hierarchical model framework, the present application creates a continuous closed loop of “prediction-calibration-re-prediction-re-calibration” in the time dimension. On the one hand, given the control inputs such as unit output, inlet and outlet water temperature, and environmental working conditions, steps S110-S130 can perform high-precision dynamic prediction of the temperature changes of each layer of the cold storage tank, the available cold energy, and the overall cold storage state under future working conditions, providing complete state trajectories for subsequent optimization and control. On the other hand, steps S130-S150 construct a loss function based on measured data and automatically adjust physical parameters including layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients, and tank body environmental heat transfer coefficients, so that the state evolution equation and the parameter update rule form a unified integrated iteration mechanism. This allows the model to gradually approach the dynamic characteristics of the real cold storage device while continuously absorbing new data, while maintaining energy conservation and good generalization ability and adaptive convergence characteristics under the premise of maintaining the main physical mechanism. Based on this unified differentiable structure, the response of the cold storage tank to any control input sequence can be converted into a differentiable objective function and constraint condition in the high-level optimization and control problem, and its gradient can be efficiently calculated by automatic differentiation or adjoint methods. This allows key performance indicators such as peak shaving degree, unit load smoothing degree, cooling redundancy, and cold energy utilization efficiency to be realized in the same solution framework as the cold load prediction, energy optimization scheduling, model predictive control, and other algorithms in the algorithm center, realizing the deep integration and end-to-end joint optimization of “physical model + data-driven”.

[0089] In this embodiment, based on the differentiable ordinary differential equation (ODE) forward operator, a loss function (such as the weighted mean absolute error MAE by layer and by time) for the hierarchical temperature observation data of the cold storage tank is introduced, and the gradient of the loss function with respect to each physical parameter is calculated by automatic differentiation or adjoint sensitivity method. This method has constant memory complexity O(1) and does not require storage of the entire forward trajectory. Combined with gradient descent or adaptive learning rate optimization algorithms, the present application can periodically or online update key physical parameters including layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients, and tank body environmental heat transfer coefficients, automatically calibrate the model within the preset physical constraint range, and ensure that it remains consistent with the real device in long-term operation and working condition changes.

[0090] The cold storage tank is modeled as a set of continuous-time ordinary differential equations, which simultaneously contain convection terms, inlet mixing terms, interlayer heat transfer terms, and environmental heat exchange terms in the same framework. By integrating the model using an adaptive step ODE solver, the charging and discharging process of the cold storage tank can be regarded as a differentiable forward operator. This allows the loss function to be directly inserted at observation points with unequal time intervals, providing a foundation for subsequent gradient-based optimization and control. The differentiable cold storage tank ODE model serves as a cold source side operator, integrated and coupled with upper-level algorithms such as IT load forecasting, electricity / carbon price forecasting, and energy scheduling optimization in AIDC. The model forward integration output of the future period's layered temperature distribution, available cold quantity, or cold storage state indicators provides accurate and differentiable physical constraints and prediction capabilities for peak shaving strategies, emergency cooling support, and overall energy consumption / carbon emission optimization.

[0091] The method of the embodiment preserves the main physical mechanisms of the cold storage tank, converts the one-dimensional convection-diffusion PDE model into a set of lower-dimensional ordinary differential equations through spatial discretization, and calibrates the parameters with physical meanings under the automatic differentiation framework. The resulting model can efficiently simulate the charging and discharging process of the cold storage tank and is easy to integrate with load forecasting and energy scheduling optimization algorithms. The invention significantly reduces model dimension and computational complexity, supports fast simulation and batch scenario calculation; by reasonably designing an interpretable and physically constrained parameter structure, and combining historical operation data for parameter identification and continuous optimization under the automatic differentiation framework, the model's fitting accuracy and robustness to actual working conditions are improved; at the same time, the constructed ODE model is deeply integrated with cold load forecasting, electricity price strategy and scheduling optimization, model predictive control, etc. algorithms, realizing fine prediction and collaborative optimization of the charging and discharging behavior of the cold storage tank, and better supporting the peak shaving, cooling safety guarantee and energy saving and carbon reduction goals of the computing center.

[0092] A layered ordinary differential equation (ODE) model is constructed for the cold storage tank in the cold source system of the computing center. The model divides the cold storage tank into several layer units along the height direction and describes the temperature change of each layer of water in each time-continuous state variable, considering the convection transfer caused by inlet flow and temperature, cold and hot water mixing, interlayer heat conduction, and heat exchange between the tank wall and the environment. By uniformly expressing the above physical mechanisms as a set of ODE equations with continuous and differentiable state and parameters, and combining automatic differentiation and optimization algorithms, the key parameters can be automatically calibrated while ensuring physical reasonableness, thereby achieving high-precision prediction of the charging process. The parameter set has a clear physical meaning and is constrained by a preset interval to ensure numerical stability and energy conservation. For observation data with unequal intervals and different steps, the actual time stamp and layer position are aligned with the model output during training, and the adaptive step ODE solver is used to realize end-to-end differentiable optimization with the adjoint sensitivity method.

[0093] The existing PDE scheme needs to subdivide the space, form a high-dimensional state and a large-scale equation set, and has large simulation overhead. The application adopts a hierarchical ODE to reduce the modeling of the cold storage tank, uses a limited number of layer temperature states to simultaneously depict main mechanisms such as convection, inlet mixing, interlayer heat transfer and environmental heat exchange, significantly reduces the calculation complexity, is more suitable for AIDC long-time, multi-condition, high-frequency simulation and online optimization, and is also more convenient for engineering implementation and deployment.

[0094] The PDE discrete solving process is difficult to efficiently solve the gradient in the automatic differentiation framework, which limits the coupling with the upper algorithm. The application models the cold storage tank as a continuous-time hierarchical ODE, and the right end function is continuously differentiable with respect to the state and the parameter, which can be directly embedded as a “differentiable forward operator” in the automatic differentiation framework, and the gradient propagation link can be established end to end with the cold load prediction, electricity / carbon price prediction, model predictive control, joint scheduling optimization, reinforcement learning and other algorithms.

[0095] The PDE scheme usually contains a large number of distributed local parameters, which are difficult to identify as a whole under the condition of limited measurement points, and mostly rely on one-time empirical setting. The application designs a limited dimension, learnable parameter (such as layer-specific response factor, thermal inertia factor, interlayer heat transfer coefficient, and tank environmental heat transfer coefficient) with clear physical meaning for convection, inlet mixing, interlayer heat transfer and environmental heat exchange, and applies physical constraints such as non-negativity, upper and lower bounds, and monotonicity. Combined with automatic differentiation or adjoint method, periodic or online parameter updating is performed to realize automatic calibration based on measured data, so that the model can still be long-term fitted to the real device as the device ages and the working condition changes.

[0096] The PDE model is mostly used for offline fine simulation and is difficult to be directly integrated into the business and energy integration decision of the power center. The ODE model of the application has a compact structure and efficient solving, and can be coordinated with IT load prediction, new energy output, and business side signals such as electricity / carbon price to output future layered temperature, available cold quantity (cold storage state) and other indicators, providing real-time available and differentiable physical constraints for peak shaving strategy making, emergency cold supply guarantee, energy consumption and carbon emission optimization, and significantly improving the value of the model in the practical application of AIDC.

[0097] In an embodiment, the high-dimensional mechanism model based on PDE is continued and discretely solved by numerical difference; in addition, a completely black-box data-driven model (such as a deep neural network) can also be used to directly fit the input-output relationship of the cold storage tank.

[0098] The digital native model energy prediction method of the physical driven computing center has the advantages that: by acquiring operation data and geometric physical parameters, a low-dimensional ordinary differential equation model is constructed along spatial discretization, only numerical integration is required for time dimension, the calculation complexity is significantly reduced, an adaptive step ODE solver is used for efficient temperature prediction, a loss function is constructed by using the difference between the actual measured value and the predicted value, and the model parameters are adjusted by an optimization algorithm to realize accurate calibration. While the main physical mechanism of the charging and discharging cold process of the cold storage tank is retained, the model complexity is significantly reduced, and the calculation process is simplified, so that the model can not only quickly and accurately simulate the behavior of the cold storage tank, but also be easily integrated with load prediction and energy scheduling optimization algorithm, thereby effectively supporting efficient energy management and optimization strategy. This method overcomes the problems of large calculation overhead and difficulty in coupling with upper algorithms of the traditional high-dimensional PDE model, and provides a more flexible and accurate solution.

[0099] Figure 3 is a schematic block diagram of a physical driven computing center digital native model energy prediction system 300 provided by an embodiment of the present application. As shown in Figure 3 corresponding to the above-mentioned physical driven computing center digital native model energy prediction method, the present application also provides a physical driven computing center digital native model energy prediction system 300. The physical driven computing center digital native model energy prediction system 300 includes units for executing the above-mentioned physical driven computing center digital native model energy prediction method, and the system can be configured in a server. Specifically, referring to Figure 3 , the physical driven computing center digital native model energy prediction system 300 includes an acquisition unit 301, a model construction unit 302, a prediction unit 303, a function construction unit 304, and an optimization unit 305.

[0100] The acquisition unit 301 is configured to acquire time series mass flow and temperature data required for the operation of the cold storage tank and geometric physical parameters, and set initial to-be-identified parameter sets and prior information such as flow direction, so as to obtain input data and parameters. The model construction unit 302 is configured to construct a layered ordinary differential equation model describing the temperature change of each layer based on energy conservation and using the input data and parameters. The prediction unit 303 is configured to use an adaptive step ODE solver to predict the temperature distribution of each layer at each time point based on the layered ordinary differential equation model according to the input data and parameters, so as to obtain predicted temperature. The function construction unit 304 is configured to construct a loss function by comparing the predicted temperature with the actual measured value. The optimization unit 305 is configured to calculate the gradient based on the loss function, and adjust the parameters by an optimization algorithm to minimize the loss function.

[0101] In an embodiment, the model construction unit 302 includes:

[0102] a temperature influence determining subunit for determining an upstream temperature according to the time series mass flow and its direction to determine the temperature influence; an interlayer heat conduction influence subunit for evaluating the contribution of interlayer heat conduction to temperature change by the second-order difference of the temperature difference of adjacent layers and the interlayer heat conduction coefficient; an environmental heat exchange influence subunit for calculating the influence of environmental heat exchange on layer temperature by using the area of the side wall of each layer and the temperature difference with the environment and the environmental heat exchange coefficient of the tank body; an integration subunit for integrating the temperature influence, the contribution of interlayer heat conduction to temperature change, and the influence of environmental heat exchange on layer temperature to form a first-order ordinary differential equation describing the rate of change of temperature with time; and a combination subunit for establishing a corresponding first-order ordinary differential equation for each layer of water of the cold storage tank and combining them into an overall layered ODE model to obtain a layered first-order ordinary differential equation model.

[0103] In an embodiment, the prediction unit 303 is configured to call an adaptive step ODE solver to integrate the layered ODE from a starting time to each observation time based on the layered first-order ordinary differential equation model according to the input data and parameters to obtain the predicted temperature of each layer at each observation time.

[0104] In an embodiment, the tuning unit 305 is configured to calculate the gradient of the loss with respect to each physical parameter by automatic differentiation or adjoint extended system based on the loss function, and update the parameters according to a preset optimization algorithm.

[0105] It should be noted that the specific implementation process of the above-mentioned physically driven computing power center digital native model energy prediction system 300 and each unit can be clearly understood by those skilled in the art, which can be referred to the corresponding description in the foregoing method embodiments. For the convenience and brevity of description, it will not be described here.

[0106] The above-mentioned physically driven computing power center digital native model energy prediction system 300 can be implemented in the form of a computer program, which can run on a computer device as shown in Figure 4 .

[0107] Please refer to Figure 4 , Figure 4 is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server, wherein the server can be a stand-alone server or a server cluster composed of multiple servers.

[0108] Please refer to Figure 4 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, wherein the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0109] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform the physical-driven digital native model energy prediction method of the computing power center.

[0110] The processor 502 is configured to provide computing and control capabilities to support the operation of the entire computer device 500.

[0111] The memory 504 provides an environment for the computer program 5032 in the non-volatile storage medium 503 to run, and the computer program 5032, when executed by the processor 502, causes the processor 502 to perform the physical-driven digital native model energy prediction method of the computing power center.

[0112] The network interface 505 is configured to communicate with other devices via a network. Those skilled in the art can understand that the network interface 505 can be configured to support wired communication or wireless communication or both wired and wireless communication. Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device 500 to which the scheme of the present application is applied. The specific computer device 500 can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0113] The processor 502 is configured to run the computer program 5032 stored in the memory to implement all steps of the physical-driven digital native model energy prediction method of the computing power center.

[0114] It should be understood that, in the embodiments of the present application, the processor 502 can be a central processing unit (CPU), and the processor 502 can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0115] Those skilled in the art can understand that all or part of the processes in the method of the above embodiments can be completed by instructing the relevant hardware by a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above method embodiments.

[0116] Therefore, the application also provides a storage medium. The storage medium can be a computer readable storage medium. The storage medium stores a computer program, wherein the computer program is executed by a processor to make the processor execute all steps of the digital native model energy prediction method of the physical driven computing power center.

[0117] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk, and various computer readable storage media that can store program codes.

[0118] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0119] In several embodiments provided by the application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed.

[0120] The steps in the method of the embodiments of the application can be adjusted, combined and deleted in sequence according to actual needs. The units in the system of the embodiments of the application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0121] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a storage medium. Based on such an understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application.

[0122] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A digital native model energy prediction method for physically driven compute centers, characterized in that, The application comprises: acquiring time-series mass flow rate, temperature data and geometric and physical parameters required for the operation of the cold storage tank, setting an initial set of to-be-identified parameters and a flow direction as prior information to obtain input data and parameters; wherein the initial set of to-be-identified parameters comprises layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients and tank body environmental heat exchange coefficients; based on energy conservation, constructing a layered ordinary differential equation model describing temperature changes of each layer by using the input data and parameters; wherein the construction of the layered ordinary differential equation model describing temperature changes of each layer comprises: integrating temperature influences, contributions of interlayer heat conduction to temperature changes and influences of environmental heat exchange on layer temperatures to form an ordinary differential equation describing a rate of change of temperature with time; establishing a corresponding ordinary differential equation for each layer of water in the cold storage tank and combining them into an overall layered ODE model to obtain the layered ordinary differential equation model; using an adaptive step ODE solver to predict temperature distribution of each layer at each time point based on the layered ordinary differential equation model according to the input data and parameters to obtain predicted temperatures; constructing a loss function by comparing the predicted temperatures with actual measured values; calculating gradients based on the loss function and adjusting parameters by an optimization algorithm to minimize the loss function.

2. The physically driven digital native model energy prediction method of a compute center of claim 1, wherein, The temperature data comprises inlet water temperature, environmental temperature and observed values of layered temperatures in the cold storage tank at each time; the geometric and physical parameters comprise volumes, layer heights, cross-sectional areas, densities and specific heats of water of each layer.

3. The physically driven digital native model energy prediction method of a compute center of claim 2, wherein, The layer-specific response factor satisfies ; the thermal inertia factor satisfies ; the interlayer thermal conductivity satisfies ; the tank body ambient heat exchange coefficient satisfies .

4. The physically driven digital native model energy prediction method of a compute center of claim 3, wherein, The construction of the layered ordinary differential equation model describing temperature changes of each layer based on energy conservation by using the input data and parameters further comprises: determining an upstream temperature according to time-series mass flow rate and its direction to determine temperature influences; evaluating contributions of interlayer heat conduction to temperature changes by second-order differences of temperature differences between adjacent layers and the interlayer thermal conductivity coefficients; calculating influences of environmental heat exchange on layer temperatures by using areas of side walls of each layer, temperature differences between the environment and each layer and tank body environmental heat exchange coefficients.

5. The physically driven digital native model energy prediction method of a compute center of claim 4, wherein, The ordinary differential equation is expressed as: wherein, is the temperature of the first i layer upstream temperature; ; is the interlayer thermal conductivity; is the area of the side wall of each layer; is the ambient temperature difference; let the cold storage tank be divided into water layers along the height direction, the temperature of the first layer is , the volume of the layer is , and the effective heat capacity is ; wherein is the thermal inertia factor of the layer, is the water density, is the specific heat at constant pressure.

6. The physically driven digital native model energy prediction method of a compute center of claim 1, wherein, The prediction of temperature distribution of each layer at each time point based on the layered ordinary differential equation model by using the adaptive step ODE solver according to the input data and parameters to obtain predicted temperatures comprises: calling the adaptive step ODE solver to integrate the layered ODE from a starting time to each observed time based on the layered ordinary differential equation model according to the input data and parameters to obtain predicted temperatures of each observed time and each layer position.

7. The physically driven digital native model energy prediction method of a compute center of claim 1, wherein, The loss function comprises wherein, is the number of observation instants, is the number of layers, and a regular term is added to the parameter range or smoothness in is the set of physical parameters to be identified.

8. The physically driven digital native model energy prediction method of a compute center of claim 7, wherein, The calculation of gradients based on the loss function and the adjustment of parameters by the optimization algorithm to minimize the loss function comprises: based on the loss function, performing back propagation calculation on the layered ordinary differential equation model by automatic differentiation or adjoint extended systems to calculate gradients of the loss with respect to each physical parameter and updating parameters according to a preset optimization algorithm.

9. A digital-native model energy prediction system with a physically driven computing center, characterized in that, The application comprises: An acquisition unit is configured to acquire time-series mass flow, temperature data and geometric and physical parameters required for operation of the cold storage tank, set an initial set of to-be-identified parameters and a flow direction as prior information, and obtain input data and parameters; the initial set of to-be-identified parameters includes layer-specific response factors, thermal inertia factors, interlayer thermal conductivity coefficients and tank body environmental heat exchange coefficients; A model construction unit is configured to construct a layered ordinary differential equation model describing temperature changes of each layer based on energy conservation and using the input data and parameters; the construction of the layered ordinary differential equation model describing temperature changes of each layer includes integrating temperature influences, contributions of interlayer heat conduction to temperature changes and influences of environmental heat exchange on layer temperatures to form an ordinary differential equation describing a rate of change of temperature with time; a corresponding ordinary differential equation is established for each layer of water of the cold storage tank, and is combined into an overall layered ODE model to obtain the layered ordinary differential equation model; A prediction unit is configured to predict temperature distributions of each layer at each time point based on the layered ordinary differential equation model using an adaptive step ODE solver according to the input data and parameters, and obtain predicted temperatures; A function construction unit is configured to construct a loss function by comparing the predicted temperatures with actual measured values; An optimization unit is configured to calculate a gradient based on the loss function, and adjust parameters through an optimization algorithm to minimize the loss function.

Images