A high-voltage direct-current power supply method and system based on LSTM prediction

Abstract

The application belongs to the technical field of power supply, and relates to a high-voltage direct-current power supply method and system based on LSTM prediction. The method comprises the following steps: obtaining historical and current-day electricity price data, constructing an LSTM model, and predicting the direct power supply cost of the next time slot; establishing a joint objective function and calculating the power supply cost of each battery pack; determining the power supply source and power distribution of the current time slot based on the direct power supply cost and the power supply cost of each battery pack; controlling the on-off of the alternating current and direct current isolation of the corresponding section, establishing or disconnecting the energy channel between each battery pack and the bus, and generating a charging and discharging plan; and updating the direct power supply cost and the charging and discharging plan based on new observation data before reaching the next time slot. The technical scheme of the application realizes dynamic prediction of the target period electricity price by introducing an LSTM prediction model into the high-voltage direct-current power supply system, realizes cost quantification and forward-looking decision of different power supply paths, and makes the power supply control strategy adaptive and time-sequential.

Description

Technical Field

[0001] This invention belongs to the field of power supply technology, and specifically relates to a high-voltage DC power supply method and system based on LSTM prediction. Background Technology

[0002] In existing technologies, data center power supply typically relies on UPS (Uninterruptible Power Supply Architecture) architecture, which maintains power stability through the mains power-UPS-server link. However, under long-term operation, this UPS architecture suffers from reduced battery chemical activity due to prolonged float charging, resulting in long discharge maintenance cycles, low energy recovery efficiency, and generally insufficient power utilization. This makes it difficult to meet the comprehensive requirements of AI data centers for high power density, continuous power supply, and economical operation.

[0003] With the emergence of HVDC (High Voltage Direct Current Architecture), some systems have attempted to improve efficiency by reducing the AC / DC (Alternating Current / Direct Current) conversion stage. However, in most existing solutions, HVDC systems typically adopt fixed topologies and static charging and discharging strategies, which have limited predictive response capabilities to electricity price fluctuations and make it difficult to dynamically adjust the power supply role of energy storage units. This results in delays and redundant power supply issues when energy storage devices switch between charging and discharging.

[0004] Meanwhile, traditional electricity price forecasting or energy storage dispatching relies heavily on human experience or static models, making it difficult to capture the long-term and short-term changes in electricity price sequences. This results in dispatching results lagging behind market price signals, and energy storage devices cannot perform charging and discharging operations within the optimal electricity price window, thus restricting the system's economy and energy efficiency.

[0005] This demonstrates that existing technologies often suffer from problems such as delayed response to power supply costs, rigid energy storage dispatch, and insufficient accuracy in electricity price prediction. These are the shortcomings of existing technologies.

[0006] In view of this, it is very necessary to provide a high-voltage DC power supply method and system based on LSTM prediction to solve the above-mentioned defects in the prior art. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of the existing technologies, such as delayed response to power supply costs, rigid energy storage scheduling, and insufficient accuracy in electricity price prediction, by providing a high-voltage DC power supply method and system based on LSTM prediction to solve the aforementioned technical problems.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A high-voltage DC power supply method based on LSTM prediction includes the following steps:

[0010] Obtain historical and current electricity price data, construct an LSTM model, and use the LSTM model to predict the electricity price data for the target time period to obtain the direct power supply cost for the next time slot.

[0011] Establish a joint objective function to calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage life constraint.

[0012] Based on the direct power supply cost and the power supply cost of each battery pack, the power supply source and power allocation of the current time slot are determined in the topology of the segmented high-voltage DC bus.

[0013] Control the AC rectification and DC isolation switching of the corresponding segments, establish or disconnect the energy channels between each battery pack and the bus, and generate a charging and discharging plan;

[0014] Before reaching the next time slot, the forecast of direct power supply costs and the charge / discharge plan are updated based on new observation data.

[0015] By adopting the above technical solution, the LSTM (Long Short-Term Memory) prediction model is introduced into the high-voltage DC power supply system to dynamically predict the electricity price for the target period. This enables cost quantification and forward-looking decision-making for different power supply paths, making the power supply control strategy adaptive and forward-looking in terms of timing, meeting the needs of data center power supply cost response, flexible energy storage scheduling, and reliable prediction accuracy.

[0016] Specifically, by constructing an LSTM model using historical and current electricity price data, electricity price forecasting no longer relies on fixed assumptions and possesses the ability to separate long-term and short-term trends, providing continuous support in the time dimension for determining power supply costs. By establishing a joint objective function, charging electricity prices, battery charging and discharging efficiency, and energy storage lifespan constraints are uniformly incorporated into the optimization objective, ensuring that the calculation of power supply costs not only reflects economic factors but also takes into account the safety of energy storage recycling and power boundaries. By determining the power source and power allocation based on a comparison of direct power supply costs and battery power supply costs in the segmented high-voltage DC bus topology, the system's power supply path can achieve a balance between economic optimization and capacity constraints. By controlling the AC rectification interruption and DC isolation interruption of each segment, the energy channel between the battery bank and the bus is established or disconnected, realizing real-time switching and coordinated operation between different energy units. By performing rolling updates on the prediction model and charging and discharging plan based on new observation data before the next time slot, the control commands can continuously maintain dynamic consistency with market signals and operating status, thereby achieving a unity of economic dispatch and high-reliability power supply, meeting the needs of high-power-density data centers for low-cost, continuous, and intelligent power supply.

[0017] As a preferred approach, the inputs to the LSTM model include at least one of the following: historical electricity price data, current electricity price data, distributed power output, real-time data center load, and battery state of charge. The LSTM model introduces a gating mechanism in the hidden layer output to model long-term trends and short-term fluctuations separately.

[0018] This technical solution introduces multi-source operating parameters into the input of the LSTM model, enabling the prediction model to establish a dynamic mapping between electricity prices and system operating status within a multi-dimensional data feature space, achieving the following technical effects:

[0019] First, while receiving historical electricity price data and current electricity price data, the model takes information such as distributed power output, real-time load and battery charge status as variable inputs, so that the prediction process takes into account both time series changes and energy supply and demand characteristics, forming an overall response capability to electricity price trends, thereby maintaining prediction stability in the scenario of electricity price fluctuations.

[0020] Second, the gating mechanism separates the long-term trend and short-term fluctuations in the hidden layer output, enabling the model to maintain trend memory in the long-term series and have the ability to adjust quickly under short-term disturbances. This avoids the prediction bias and cumulative error caused by single time scale modeling and improves the adaptability of the prediction results to time series changes.

[0021] Third, the model, through the synergistic effect of multiple inputs and gating mechanisms, achieves the identification of time-series dependencies between power supply-related variables in the data center. This enables the predicted output to more accurately reflect the direction and fluctuation range of electricity price changes within future time slots, providing a high-precision reference for subsequent power supply cost calculation and scheduling decisions, and enhancing the system's response reliability and economy in dynamic market environments.

[0022] Preferably, the joint objective function is also used to calculate the available power of each battery pack, and to keep the sum of the available power of each battery pack from being lower than a preset critical value under the constraint of available power.

[0023] This technical solution incorporates available power constraints into the joint objective function, combining the battery pack's energy state with the power supply cost calculation process. This enables the optimization decision-making to possess capacity self-balancing capabilities and energy storage redundancy management characteristics, achieving the following technical effects:

[0024] First, the joint objective function outputs the available power of each battery pack while calculating the power supply cost, so that the cost optimization result not only depends on the electricity price prediction, but also is tied to the actual energy storage capacity, thereby avoiding the deep discharge and capacity imbalance problems caused by simply relying on economic judgment, and improving the long-term operational stability of the energy storage unit.

[0025] Second, by maintaining the sum of available power of each battery pack at a preset threshold under the constraint of available power, the system always has dispatchable energy storage redundancy during operation. When the main power supply fluctuates or the electricity price rises abnormally, it can still maintain the normal power supply path, thus achieving the parallel development of power supply continuity and energy security.

[0026] Third, the constraints are embedded as hard constraints in the solution of the joint objective function, so that the optimization process achieves a dynamic balance between minimizing the global cost and the energy storage safety boundary, avoiding the life decay caused by frequent charging and discharging of a single battery group, while improving the stability of power dispatch and the balanced utilization rate of the energy storage group, thereby enabling the system to maintain long-term consistency in terms of power supply economy and reliability.

[0027] As a preferred option, the power source is determined based on a comparison of the cost of direct power supply and the power supply cost of each battery pack, and the power source is selected according to the principle of minimizing cost within the current time slot.

[0028] This technical solution introduces a cost minimization principle into power supply decision-making, establishing a dynamic decision-making mechanism based on a comparison between the cost of direct power supply and the power supply costs of each battery pack. This enables the power supply path selection to have self-optimizing characteristics, achieving the following technical effects:

[0029] First, the selection of power supply sources is based on the cost parameters of the current time slot. The economics of transformer rectification power supply and multi-group battery power supply are quantitatively compared so that the power supply path is matched with the lowest cost scheme in real time, reducing unnecessary energy conversion losses and improving the overall power supply efficiency.

[0030] Second, under the principle of cost minimization, the charging and discharging behavior of different battery packs complements the rectification power supply strategy, enabling the system to automatically switch power supply modes under scenarios of electricity price fluctuations and load changes, avoiding redundant operation and power waste in traditional static scheduling, and improving the flexibility of system energy allocation.

[0031] Third, the dynamic decision-making of power supply sources enables the high-voltage DC bus to make rapid adjustments based on the equivalent cost of each path in the segmented topology, taking into account both power balance and economic optimization, so that energy can maintain the optimal operating state when flowing between multiple paths, thereby realizing real-time economic dispatch and high-reliability operation of the power supply system, and improving the overall energy utilization rate and dispatch intelligence level.

[0032] As a preferred option, the power allocation is determined based on the difference between the direct power supply cost and the power supply cost of each battery pack, and the power output is allocated according to the topology of the segmented high-voltage DC bus.

[0033] This technical solution introduces a weighting method based on cost differences in the power allocation stage and combines it with the topology of segmented high-voltage DC buses to allocate power output, transforming the power allocation process from static control to dynamic economic dispatch, achieving the following technical effects:

[0034] First, the cost difference is used to establish allocation weights, so that the power allocation ratio is directly related to the economy of each power supply path. This allows for the optimal distribution of energy flow while maintaining a constant total power under different power supply modes, reducing the energy share of high-cost paths and improving the overall energy efficiency of the system.

[0035] Second, by combining the topology of the segmented high-voltage DC bus to distribute power output, the segmented bus can maintain load balance under different operating conditions, avoiding voltage fluctuations and energy losses caused by centralized power supply or single bus overload, and realizing coordinated power supply between segments.

[0036] Third, the dynamic updating of power allocation weights enables the system to adjust the energy output structure according to real-time cost changes, making the bus power flow time-adjustable and spatially coordinated, thereby ensuring the balanced discharge of energy storage units and the efficient utilization of rectification paths, so that the power supply network can maintain stable, efficient and economical operation under complex electricity price environments.

[0037] As a preferred option, controlling the AC rectification switching and DC isolation switching of the corresponding segments includes: determining the switching time based on the energy remaining rate of the battery pack in each segment and the rate of change of electricity price; the determination of the switching time adopts a delayed confirmation strategy; and keeping the switching state unchanged when the difference in electricity price prediction between adjacent time slots is lower than a set threshold.

[0038] This technical solution uses both the energy surplus rate and the electricity price change rate as the on / off criteria, and sets a delayed confirmation strategy to suppress short-term fluctuations, giving the on / off control threshold memory and switching hysteresis characteristics, thus achieving the following technical effects:

[0039] First, the switching is triggered only when the energy state and price signal both reach the criterion, avoiding frequent transitions caused by a single factor, maintaining a stable on-off relationship of the power channel, and reducing conversion losses and thermal stress caused by the high-frequency operation of rectifier and isolation devices.

[0040] Second, delayed confirmation maintains the original on / off state in scenarios where the prediction difference between adjacent time slots is small, so that small prediction noise no longer causes topology oscillations, reduces repeated rises and falls of bus voltage and current, and maintains the continuity of segmented load distribution and the consistency of scheduling plan.

[0041] Third, the energy surplus rate as a criterion ensures that different segments prioritize channel stability near capacity boundaries, reducing the shallow cycle fidelity consumption caused by short-cycle switching of energy storage units. At the same time, it aligns the switching rhythm with the rolling plan, improving the predictability and economy of the entire power supply control system in multi-time slot operation.

[0042] As a preferred option, the generation of the charge and discharge plan includes setting a rotation priority sequence for each battery pack. The rotation priority sequence is determined based on a comprehensive calculation of the battery pack's cycle count, current temperature rise, and state of charge.

[0043] This technical solution constructs a rotation priority sequence based on the number of cycles, temperature rise, and state of charge, and arranges the power-on and standby order of each battery pack accordingly. This transforms the scheduling from static and fixed to adaptive rotation based on state variables, achieving the following technical effects:

[0044] First, the rotation priority sequence allows high-frequency units to give way to low-frequency units in a timely manner, the depth of discharge and cumulative cycles are distributed among the units, excessive concentrated use is suppressed, capacity decay is more balanced, the reserve availability is kept in a stable range, and the callable energy storage redundancy is maintained during abnormal fluctuations, thus enhancing the continuity of power supply and the predictability of dispatch.

[0045] Second, by using temperature rise as a key input, thermal constraints become explicit, thermal load is actively dispersed during rotation, thermal gradients and hot spots in various bus sections are suppressed, the rate of thermal stress accumulation in insulation and conductors is reduced, internal resistance changes during charging and discharging are smoother, power command response tends to be stable, voltage drop and rise amplitudes are reduced, and the frequency of thermal-related protection actions is reduced.

[0046] Third, prioritization is carried out in conjunction with the state of charge (SOC) range, so that different units can take turns to undertake power output and charging tasks around the appropriate SOC range. The ratio of deep and shallow cycles is more reasonable, the equivalent loss during energy path switching is reduced, the power allocation and capacity boundary are coordinated, the rolling adjustment of plans under price and load disturbances is smoother, and the overall operating cost and available capacity utilization rate show stable benefits.

[0047] As a preferred approach, the update involves incrementally learning the LSTM model parameters in each prediction period, continuously updating the model weights using a sliding window mechanism, and adaptively shortening the prediction period when electricity prices fluctuate drastically.

[0048] This technical solution enables the LSTM model to continuously maintain its sensitivity and generalization ability to electricity price time-series characteristics by performing incremental learning of model parameters within the prediction period and combining it with a sliding window update mechanism. It adaptively adjusts the prediction period under drastic fluctuation scenarios, achieving the following technical effects:

[0049] First, the incremental learning mechanism enables the model to update only local weights when receiving new observation data, preserving the stability of the original structure. This allows the model to maintain convergence characteristics and gradually absorb the latest price change information during long-term operation, thus continuously optimizing the prediction accuracy over time and avoiding performance degradation caused by static parameters.

[0050] Second, the sliding window mechanism ensures that the training samples remain continuously rolling in time, and the weights of new and old data are dynamically balanced. This not only prevents overfitting of recent anomalies, but also captures the trend migration process, so that the prediction output forms a stable response between short-term fluctuations and long-term trends, maintaining the smoothness and reliability of the prediction curve.

[0051] Third, when electricity prices fluctuate drastically, the forecast cycle is adaptively shortened to match the model update frequency with the pace of market changes, reducing decision bias caused by forecast lag, and ensuring that power supply dispatching plans remain timely and economical in high-speed fluctuation ranges, thereby realizing the dynamic adaptability and continuous efficient operation of the forecast model in complex market environments.

[0052] Furthermore, the present invention also provides a high-voltage DC power supply system based on LSTM prediction, comprising:

[0053] The electricity price forecasting module is used to build an LSTM model based on historical electricity price data and current electricity price data, and to forecast the electricity price data for the target time period to obtain the direct power supply cost for the next time slot.

[0054] The cost calculation module is used to establish a joint objective function to calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage lifetime constraint.

[0055] The power supply decision module is used to determine the power supply source and power allocation for the current time slot in the segmented high-voltage DC bus topology based on the direct power supply cost and the power supply cost of each battery pack.

[0056] The topology control module is used to control the AC rectification and DC isolation switching of the segmented high-voltage DC bus based on the output signal of the power supply decision module, and to establish or disconnect the energy channel between each battery pack and the bus.

[0057] The plan generation module is used to generate and store the charging and discharging plans of the battery pack. The charging and discharging plans are updated synchronously with the power supply switching instructions.

[0058] The parameter update module is used to update the electricity price forecast module and the charge / discharge plan based on new observation data before the next time slot.

[0059] By adopting the above technical solution, and by setting up functional modules such as electricity price prediction module, cost calculation module, power supply decision module, topology control module, plan generation module and parameter update module in the high voltage DC power supply system, predictive driving and multi-source coordination of power supply control are realized, which can maintain the optimal power supply path and efficient system operation under dynamic electricity price and complex load conditions.

[0060] The system comprises several modules: an electricity price prediction module that uses historical and current electricity price data to build an LSTM model, enabling the system to learn and predict electricity price trends over time, providing real-time data for determining power supply costs; a cost calculation module that uses a joint objective function to model charging prices, battery efficiency, power constraints, and lifespan parameters in a unified manner, achieving quantitative assessment and economic judgment of power supply costs; a power supply decision module that selects the optimal energy supply path and power allocation based on cost comparison results, transforming the energy scheduling process from fixed logic to real-time optimization; a topology control module that executes on / off commands on AC rectification and DC isolation channels based on decision signals, automatically reconstructing the energy flow direction of each battery bank according to the scheduling results, ensuring consistency between power transmission paths and operating strategies; a plan generation module that records charging and discharging sequences and keeps them synchronized with power supply switching commands, coordinating energy storage behavior with market price rhythms; and a parameter update module that performs rolling corrections to model parameters and scheduling plans before each time slot, ensuring that the system maintains prediction accuracy and scheduling consistency throughout long-term operation, thereby achieving comprehensive optimization of data center high-voltage DC power supply in terms of cost, efficiency, and stability.

[0061] Preferably, the power supply decision module includes a power allocation unit, which calculates the allocation weight based on the difference between the direct power supply cost and the power supply cost of each battery pack, and generates a power allocation instruction based on the allocation weight.

[0062] By introducing a cost-difference-based weighting mechanism into the power allocation unit, the power command generation takes into account both economic efficiency and electrical constraint coordination, achieving the following technical effects:

[0063] First, the power distribution unit establishes a weight matrix for multiple power supply paths based on the cost difference as the decision-making factor, so that the output power of each path is directly linked to its economic contribution. This allows high-cost channels to be automatically downgraded and low-cost channels to be automatically upgraded, forming a dynamic self-balancing power distribution process that changes with market signals, reducing redundant energy flow and unnecessary power switching.

[0064] Second, the weighting generation mechanism enables the power allocation result to be synchronously matched with the bus topology. The power output ratio between different battery packs is adjusted in real time according to their respective capacity, efficiency and bus constraints, avoiding uneven energy efficiency caused by overload or idleness of a single battery pack, and improving the overall power utilization rate and bus operation stability of the system.

[0065] Third, the power allocation command is uniformly issued by the power supply decision module after real-time calculation, enabling the system to respond quickly and reconstruct the power allocation pattern in the event of sudden load changes or drastic fluctuations in electricity prices, maintain continuous output and voltage balance, and improve the adaptability and continuous economic operation capability of the high-voltage DC power supply system under complex operating conditions.

[0066] The beneficial effect of this invention is that by introducing an LSTM prediction model into the high-voltage DC power supply system to dynamically predict the electricity price for the target period, it realizes the cost quantification and forward-looking decision-making for different power supply paths, enabling the power supply control strategy to have adaptability and time-series foresight, and meeting the needs of data center power supply cost response, flexible energy storage scheduling, and reliable prediction accuracy.

[0067] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects.

[0068] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0070] Figure 1 This is a flowchart of a high-voltage DC power supply method based on LSTM prediction provided by the present invention;

[0071] Figure 2 This is a schematic diagram of a high-voltage DC power supply system based on LSTM prediction provided by the present invention.

[0072] Figure 3 This is a schematic diagram of a data center HVDC power supply architecture based on an LSTM prediction-based high-voltage DC power supply method provided by the present invention.

[0073] Figure 4 This is a schematic diagram of a segmented high-voltage DC bus structure for a data center based on an LSTM prediction-based high-voltage DC power supply method provided by the present invention.

[0074] The components include: 1. Electricity price prediction module; 2. Cost calculation module; 3. Power supply decision module; 4. Topology control module; 5. Plan generation module; 6. Parameter update module; 31. DC bus; 311. Segmented bus; 312. Bus tie switch; 313. DC / DC isolation module; 314. Data center power distribution cabinet; 315. Data center server; 316. Battery bank; 317. Supercapacitor bank; 318. AC / DC rectifier module; and 319. Transformer. Detailed Implementation

[0075] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0076] Example 1:

[0077] like Figure 1 As shown, this embodiment provides a high-voltage DC power supply method based on LSTM prediction, which includes the following steps:

[0078] Step S1: Obtain historical electricity price data and current electricity price data, construct an LSTM model, use the LSTM model to predict the electricity price data for the target time period, and obtain the direct power supply cost for the next time slot;

[0079] Step S2: Establish a joint objective function and calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage life constraint.

[0080] Step S3: Based on the direct power supply cost and the power supply cost of each battery pack, determine the power supply source and power allocation for the current time slot in the segmented high-voltage DC bus topology;

[0081] Step S4: Control the AC rectification and DC isolation switching of the corresponding segments to establish or disconnect the energy channels between each battery pack and the bus, and generate a charging and discharging plan;

[0082] Step S5: Update the forecast of direct power supply costs and the charge / discharge plan based on new observation data before reaching the next time slot.

[0083] By adopting the above technical solution, the LSTM prediction model is introduced into the high-voltage DC power supply system to dynamically predict the electricity price for the target period. This enables cost quantification and forward-looking decision-making for different power supply paths, making the power supply control strategy adaptive and forward-looking in terms of timing, meeting the needs of data centers for timely response to power supply costs, flexible energy storage scheduling, and reliable prediction accuracy.

[0084] Specifically, by constructing an LSTM model using historical and current electricity price data, electricity price forecasting no longer relies on fixed assumptions and possesses the ability to separate long-term and short-term trends, providing continuous support in the time dimension for determining power supply costs. By establishing a joint objective function, charging electricity prices, battery charging and discharging efficiency, and energy storage lifespan constraints are uniformly incorporated into the optimization objective, ensuring that the calculation of power supply costs not only reflects economic factors but also takes into account the safety of energy storage recycling and power boundaries. By determining the power source and power allocation based on a comparison of direct power supply costs and battery power supply costs in the segmented high-voltage DC bus topology, the system's energy supply path can achieve a balance between economic optimization and capacity constraints. By controlling the AC rectification interruption and DC isolation interruption of each segment, the energy channel between the battery bank and the bus is established or disconnected, realizing real-time switching and coordinated operation between different energy units. By performing rolling updates on the prediction model and charging and discharging plan based on new observation data before the next time slot, the control commands can continuously maintain dynamic consistency with market signals and operating status, thereby achieving a unity of economic dispatch and high-reliability power supply, meeting the needs of high-power-density data centers for low-cost, continuous, and intelligent power supply.

[0085] Hereinafter, steps S1 to S5 will be specifically described according to embodiments of this application.

[0086] In step S1, it is necessary to acquire historical electricity price data and current electricity price data, and construct an LSTM model for electricity price prediction in order to achieve dynamic prediction of electricity price within the target time period and quantification of the direct power supply cost in the next time slot. This step takes data centers as a typical load scenario and aims to identify electricity price change trends in advance to provide a basis for subsequent energy management and power supply path decisions.

[0087] Specifically, in this embodiment, historical electricity price data may include market settlement electricity price sequences over multiple time periods, while current day electricity price data refers to real-time electricity price information for each time slot within the current day. Furthermore, to ensure the continuity and consistency of the time series, the aforementioned data can be uniformly sampled and time-aligned.

[0088] In some embodiments of this application, during the model construction process, the LSTM model undertakes the core computational function of electricity price prediction. The model input can include historical electricity price data and current electricity price data, and can be expanded to simultaneously input any one or more of the following: distributed power output, real-time data center load, and battery state of charge. The model structure includes an input layer, several hidden layer units, and an output layer. A gating mechanism can be introduced in the hidden layer to achieve separate modeling of long-term trends and short-term fluctuations. Long-term gating is used to capture cross-day and cross-week trend changes, while short-term gating is used to describe the rapid fluctuations of peak electricity price periods. The predicted electricity price for the next time slot output by the model is used for subsequent cost calculations.

[0089] Among them, the output of distributed power sources reflects the supply capacity of local power generation, the real-time load reflects the current energy consumption level of the data center, and the state of charge of batteries reflects the dispatchable energy reserves. These features are aligned with the electricity price data sequence through a unified time index, and can be input into the model in the form of multi-dimensional tensors to improve the ability to jointly identify complex energy states and market signals. At the same time, in order to ensure the consistency of input dimensions, all feature data are normalized at the same scale.

[0090] This step introduces multi-source operating parameters into the input of the LSTM model, enabling the prediction model to establish a dynamic mapping between electricity prices and system operating status within a multi-dimensional data feature space, achieving the following technical effects:

[0091] First, while receiving historical electricity price data and current electricity price data, the model takes information such as distributed power output, real-time load and battery charge status as variable inputs, so that the prediction process takes into account both time series changes and energy supply and demand characteristics, forming an overall response capability to electricity price trends, thereby maintaining prediction stability in the scenario of electricity price fluctuations.

[0092] Second, the gating mechanism separates the long-term trend and short-term fluctuations in the hidden layer output, enabling the model to maintain trend memory in the long-term series and have the ability to adjust quickly under short-term disturbances. This avoids the prediction bias and cumulative error caused by single time scale modeling and improves the adaptability of the prediction results to time series changes.

[0093] Third, the model, through the synergistic effect of multiple inputs and gating mechanisms, achieves the identification of time-series dependencies between power supply-related variables in the data center. This enables the predicted output to more accurately reflect the direction and fluctuation range of electricity price changes within future time slots, providing a high-precision reference for subsequent power supply cost calculation and scheduling decisions, and enhancing the system's response reliability and economy in dynamic market environments.

[0094] The predicted electricity price output by the model is mapped to the direct power supply cost in the next time slot through parameters such as rectification efficiency and line loss. The mapped cost is used for subsequent energy supply strategy selection and power allocation decisions. At the same time, the prediction results can be compared and archived with the actual electricity price for subsequent accuracy evaluation and model updates.

[0095] For example, the input window length can be set to 96 time steps, corresponding to a one-day cycle at a 15-minute resolution; the number of hidden layer units can be 64 to balance computational overhead and feature representation capability; the electricity price input can be standardized in the [0,1] interval.

[0096] At this point, step S1 completes the collection, processing, and prediction of electricity price data, and obtains the direct power supply cost for the next time slot, providing basic data for subsequent energy allocation and optimization control.

[0097] In step S2, the core task is to establish a joint objective function to calculate the power supply cost of each battery pack. By combining parameters such as charging price, battery charging and discharging efficiency, bus power constraints, and energy storage life constraints, an optimization model that can quantify the power supply cost of batteries is constructed, providing a basis for subsequent energy allocation and scheduling decisions.

[0098] Specifically, in the embodiments of this application, the joint objective function aims to minimize the energy supply cost per unit time. It considers factors such as the discharge power of the battery, the electricity price signal, and the energy efficiency parameters. The charging electricity price cost of each battery pack consists of two parts: the direct cost corresponding to the electricity price factor and the implicit cost corresponding to energy loss and lifespan reduction. The output power of each battery pack is used as the decision variable. By finding the optimal solution of the joint objective function, a balance between cost and constraints is achieved.

[0099] In the cost modeling stage, the energy loss during the discharge process is determined by the battery's charging and discharging efficiency. Confirmed, when At the same time, the equivalent energy consumed during discharge increases, thus driving up the unit charging cost; simultaneously, if the battery pack is in a high-rate charge-discharge state for a long time, its lifespan will decrease with the number of cycles. Therefore, a lifespan decay factor is introduced into the joint objective function. The joint objective function is used to characterize the long-term economic losses caused by over-discharge. Thus, the joint objective function can be expressed as a comprehensive mapping relationship between cost and constraints. Its logic is to minimize the weighted power supply cost of battery operation while ensuring the balance between load power demand and bus power.

[0100] In embodiments of this application, the parameters of the joint objective function may further include at least one of battery charge / discharge efficiency, bus power constraint, and energy storage lifetime constraint. Each parameter is derived from real-time or historical monitoring data to ensure the model's responsiveness to changes in operating conditions. The charge / discharge efficiency is determined by the battery model or operating curve. The bus power constraint reflects the limitations of the data center's total load and upstream power supply capacity. The energy storage lifetime constraint is used to suppress high-rate discharge. For example, when the upper limit of the bus power is... At any given time slot, the sum of the battery output power and the direct power supply must not exceed [a certain value]. When the life constraint coefficient When an empirical value of 0.01 to 0.05 is used, a balance can be achieved between lifespan loss and energy cost.

[0101] Furthermore, to ensure the computability of the modeling process, the cost calculation for each battery pack is performed on a time-slice basis. Within each prediction time slot, the current electricity price, load demand, and battery operating status are read, and the initial discharge power of each battery pack is allocated. This allocation is then iteratively corrected based on the objective function. Each iteration checks the feasibility of the bus power constraint and the remaining energy storage capacity constraint. If any constraint is triggered, the power allocation is readjusted until the conditions are met. This process ensures that the solution to the joint objective function satisfies both energy balance and economic efficiency.

[0102] In some embodiments of this application, the joint objective function can also be used to calculate the available power of each battery pack. Available power is defined as the amount of energy that can be discharged from the battery under its current state of charge, influenced by the state-of-charge limit, discharge efficiency, and temperature correction coefficient. The model incorporates available power constraints into the optimization process to ensure that the sum of the available power of each battery pack is not lower than a preset critical value, thereby guaranteeing that the system still has sufficient energy reserves during peak loads. This constraint can be achieved by introducing a safety margin coefficient. Achieve, ensure total available energy ,in This is a preset critical threshold. For example, if the system contains three sets of batteries, two of which are used for main power support and one for peak shaving, the model can ensure that the total energy reserve is not lower than the preset critical threshold by constraining the available power during optimization.

[0103] This step, by introducing available power constraints into the joint objective function, combines the energy state of the battery pack with the power supply cost calculation process, enabling the optimization decision to possess capacity self-balancing capability and energy storage redundancy management characteristics, and achieving the following technical effects:

[0104] First, the joint objective function outputs the available power of each battery pack while calculating the power supply cost, so that the cost optimization result not only depends on the electricity price prediction, but also is tied to the actual energy storage capacity, thereby avoiding the deep discharge and capacity imbalance problems caused by simply relying on economic judgment, and improving the long-term operational stability of the energy storage unit.

[0105] Second, by maintaining the sum of available power of each battery pack at a preset threshold under the constraint of available power, the system always has dispatchable energy storage redundancy during operation. When the main power supply fluctuates or the electricity price rises abnormally, it can still maintain the normal power supply path, thus achieving the parallel development of power supply continuity and energy security.

[0106] Third, the constraints are embedded as hard constraints in the solution of the joint objective function, so that the optimization process achieves a dynamic balance between minimizing the global cost and the energy storage safety boundary, avoiding the life decay caused by frequent charging and discharging of a single battery group, while improving the stability of power dispatch and the balanced utilization rate of the energy storage group, thereby enabling the system to maintain long-term consistency in terms of power supply economy and reliability.

[0107] Furthermore, to reduce local optima, the optimization of the joint objective function can employ gradient descent, quasi-Newton methods, or policy search algorithms based on reinforcement learning. The optimizer fine-tunes the discharge power of each battery pack during each time slot, gradually converging the gradient of the objective function. When the bus power is detected to be approaching the constraint boundary, the constraint weights can be dynamically adjusted to prioritize ensuring continuous power supply.

[0108] In other embodiments of this application, the joint objective function may also take into account the performance differences of different types of battery packs. For example, lithium-ion batteries have high efficiency and long cycle life, but are more expensive; lead-carbon batteries have lower cost and faster response speed, but have limited energy density. By introducing weighting coefficients to differentiate and correct different battery packs, synergistic optimization of multiple types of energy storage units can be achieved.

[0109] Thus, step S2 achieves the quantification and constraint coordination from electricity price signals to energy storage operating costs, enabling the system to maximize economic efficiency while ensuring power stability. At the same time, the constructed joint objective function has scalability and differentiability, providing a continuous optimization foundation for subsequent energy allocation and scheduling.

[0110] In step S3, based on the direct power supply cost obtained in step S1 and the power supply cost of each battery pack obtained in step S2, it is necessary to determine the power supply source and power allocation scheme for the current time slot. The core is to take the actual topology of the segmented high-voltage DC bus as the basis, comprehensively consider cost, energy constraints and power supply continuity, and achieve coordinated control of different power supply paths so that the system can complete the load power supply at the lowest comprehensive cost in the current time slot.

[0111] like Figure 3 and Figure 4 As shown in the embodiments of this application, the high-voltage DC power supply system for the data center mainly includes a transformer 319, several AC / DC rectifier modules 318, a high-voltage DC bus 31, multiple segmented busbars 311, a bus tie switch 312, a battery pack 316, a supercapacitor pack 317, a DC / DC isolation module 313, a data center power distribution cabinet 314, and a data center server 315.

[0112] The high-voltage AC power output from the substation (e.g., 10kV) is stepped down by transformer 319 to an AC voltage (e.g., 1kV) suitable for the data center side. The stepped-down AC power is then rectified by AC / DC rectifier module 318 to form high-voltage DC power (e.g., 800V) and connected to the high-voltage DC bus 31. The high-voltage DC bus 31, as the main DC combiner node of the system, is divided into multiple segmented busbars 311 along its length. Each segmented busbar 311 can be connected to or isolated from the DC bus 31 through a bus tie switch 312. This allows for the connection or disconnection of the corresponding segmented busbar 311 from the DC bus 31 when power dispatching, capacity expansion, maintenance, or fault clearing is required. At the same time, adjacent segmented busbars 311 can be electrically isolated through the bus tie switch 312 to achieve regional power supply autonomy and fault limitation during abnormal periods.

[0113] Battery banks 316 and supercapacitor banks 317 with a capacity ratio (e.g., 10:1) matching the battery banks 316 can be arranged on the DC bus 31 or the segment bus 311. The battery banks 316 provide continuous power supply and can directly support the segment within the allowable voltage range of the segment bus 311 to ensure power supply continuity and bus voltage stability. The supercapacitor banks 317 support short-term loads and provide transient power support during load surges or rectifier fluctuations to balance the bus voltage. Simultaneously, an energy exchange device can be installed between the battery banks 316 and the supercapacitor banks 317 to float charge the battery banks 316 when the supercapacitor banks 317 have residual energy, thereby improving the overall cycle efficiency of the energy storage system.

[0114] Furthermore, each segment bus 311 is connected to at least two sets of DC / DC isolation modules 313. The DC / DC isolation modules 313 are used to achieve isolation and voltage matching between DC power of different voltage levels, enabling the server side to obtain a suitable operating voltage while the high-voltage DC bus 31 remains stable. At the same time, the output of the DC / DC isolation modules 313 is connected to the data center power distribution cabinet 314, which distributes power to multiple server racks and load nodes of the data center server 315, realizing power distribution within the system.

[0115] In some embodiments of this application, the high-voltage DC bus 31 and its segmented bus 311 adopt a modular structure design. When the power consumption of the data center increases, the capacity can be expanded by adding AC / DC rectifier modules 318, battery packs 316, supercapacitor packs 317, or DC / DC isolation modules 313 in parallel to the segmented bus 311 without changing the original system topology. For example, when the server deployment scale expands from the original capacity to 1.1 times or 1.2 times, the expansion can be completed simply by connecting additional DC / DC isolation modules 313 and battery packs 316 to the corresponding segmented bus 311, thereby improving the scalability and evolution capability of the system.

[0116] At the start of the current time slot, the direct power supply cost and the power supply cost of each battery pack obtained in the previous step can be read first. The numerical relationship between the two can be compared, and the power supply source can be selected according to the principle of cost minimization. If the direct power supply cost is lower than the power supply cost of all battery packs, it is determined that the internal network rectifier link will be the main power supply for this time slot. If the power supply cost of at least one battery pack is lower than the direct power supply cost, the corresponding battery pack is selected as the main power supply source, while the rectifier link is reserved as an auxiliary power supply path to cope with power fluctuations or bus voltage deviations.

[0117] Based on this, to avoid bus fluctuations caused by frequent switching, a small cost difference threshold can be set. When the cost difference between the two power supply paths is less than this threshold, the power supply mode of the previous time slot remains unchanged to improve system stability. For example, if the difference between the cost of direct power supply and the cost of the lowest battery power supply is less than 0.02 yuan / kWh, they can be considered to be cost-equivalent, and the original power supply source is maintained.

[0118] This step introduces the principle of cost minimization into the power supply decision-making process. A dynamic decision-making mechanism is established based on a comparison between the cost of direct power supply and the power supply costs of each battery pack. This enables the power supply path selection to have self-optimizing characteristics, achieving the following technical effects:

[0119] First, the selection of power supply sources is based on the cost parameters of the current time slot. The economics of transformer rectification power supply and multi-group battery power supply are quantitatively compared so that the power supply path is matched with the lowest cost scheme in real time, reducing unnecessary energy conversion losses and improving the overall power supply efficiency.

[0120] Second, under the principle of cost minimization, the charging and discharging behavior of different battery packs complements the rectification power supply strategy, enabling the system to automatically switch power supply modes under scenarios of electricity price fluctuations and load changes, avoiding redundant operation and power waste in traditional static scheduling, and improving the flexibility of system energy allocation.

[0121] Third, the dynamic decision-making of power supply sources enables the high-voltage DC bus to make rapid adjustments based on the equivalent cost of each path in the segmented topology, taking into account both power balance and economic optimization, so that energy can maintain the optimal operating state when flowing between multiple paths, thereby realizing real-time economic dispatch and high-reliability operation of the power supply system, and improving the overall energy utilization rate and dispatch intelligence level.

[0122] In this embodiment, after determining the power supply source, the power allocation stage begins. Allocation weights are established based on cost differences, ensuring that lower-cost sources receive a higher power output share, while higher-cost sources have a reduced power supply share, thereby minimizing overall operating costs. Power allocation is simultaneously constrained by bus topology, available battery capacity, and load distribution across different sections.

[0123] For example, the total load power of the current time slot is denoted as The unit cost of direct power supply is , No. The unit power supply cost of a single battery pack is The cost difference between each power supply path is defined as:

[0124]

[0125] when When, it indicates the first A single battery pack provides more economical power, and its output should be increased; when When this is the case, it indicates that direct power supply is more economical, and the battery should be kept at low power or in standby mode.

[0126] Therefore, the allocation ratio can be determined based on the cost difference:

[0127]

[0128] in, For the first The load power of each battery pack This is the cumulative cost difference of all economically viable sources. If only one battery pack meets the economic requirements, then that pack will provide the main power supply, and the remaining sources will supplement the supply based on their remaining power capacity.

[0129] Furthermore, the topological characteristics of the segmented high-voltage DC bus must be considered during power allocation. The DC bus is usually divided into several segments, each of which is independently connected to the corresponding server array and energy storage branch. The segments maintain a consistent potential through DC isolation or voltage equalization links. Therefore, power allocation not only determines the total output, but also needs to determine the allocation ratio of each segment under topological constraints.

[0130] In bus power distribution, the load power of each segment can be expressed as: The total requirements for each segment are satisfied:

[0131]

[0132] Based on this, the power allocation target for each segment is calculated according to the real-time load ratio of each segment. Then, according to the connection relationship and allowable output of each battery pack, the target power allocation is mapped to the corresponding power source. For example, when a battery pack is only connected to the first and second segments, its allocated power is limited to the allocation within these two segments; if a segment is undergoing maintenance or electrical isolation in that time slot, its power supply path is shielded and it does not participate in the allocation.

[0133] This step introduces a weighting method based on cost differences in the power allocation stage, and combines it with the topology of segmented high-voltage DC buses to allocate power output, transforming the power allocation process from static control to dynamic economic dispatch, achieving the following technical effects:

[0134] First, the cost difference is used to establish allocation weights, so that the power allocation ratio is directly related to the economy of each power supply path. This allows for the optimal distribution of energy flow while maintaining a constant total power under different power supply modes, reducing the energy share of high-cost paths and improving the overall energy efficiency of the system.

[0135] Second, by combining the topology of the segmented high-voltage DC bus to distribute power output, the segmented bus can maintain load balance under different operating conditions, avoiding voltage fluctuations and energy losses caused by centralized power supply or single bus overload, and realizing coordinated power supply between segments.

[0136] Third, the dynamic updating of power allocation weights enables the system to adjust the energy output structure according to real-time cost changes, making the bus power flow time-adjustable and spatially coordinated, thereby ensuring the balanced discharge of energy storage units and the efficient utilization of rectification paths, so that the power supply network can maintain stable, efficient and economical operation under complex electricity price environments.

[0137] In some embodiments of this application, based on topology allocation, the feasibility of bus constraints and energy storage constraints can be further verified. The output of each power supply path should be less than its maximum allowable output. After allocation, the available power of each battery pack should still meet the minimum energy storage requirements of the next time slot to prevent energy overflow. When the bus power is close to the capacity limit or the available power of the battery is insufficient, the output of that source is automatically reduced and the proportion of other sources is adjusted to maintain power balance and constraint safety.

[0138] For example, if the current total load power is 1200 kilowatts, the direct power supply cost is... The cost of power supply for two battery banks is 0.85 yuan / kWh. and With prices of 0.78 yuan / kWh and 0.82 yuan / kWh respectively, two battery packs are selected as the main power sources, and the power ratio is calculated to be approximately 7:3 based on the cost difference. If the first segment of the load accounts for 40% of the total load and the second segment accounts for 60%, the power is further distributed to the corresponding bus according to the topology mapping. The supercapacitor is used to provide short-term support during segment voltage fluctuations, startup shocks, or power surges.

[0139] For example, if the direct supply cost is lower than any battery pack, the rectifier direct supply is maintained and the charging sequence and duration of the batteries are arranged according to the predicted curve during low-price periods to achieve "more charging at low prices and less charging at high prices"; if the direct supply cost is higher than all battery packs, the corresponding segment is closed and isolated, and the batteries are discharged to the grid. The load is preferentially borne by the lowest cost group, and the insufficient part is supplemented by the second lowest cost group; if only individual battery packs have a cost lower than the direct supply, only that group is discharged to the grid in its connected segment, and the other groups are on standby and will be recharged when the rectifier capacity allows; in any case, the available power of each battery pack is not lower than the set threshold value to ensure that power supply can still be maintained for a period of time when the upstream power supply fails.

[0140] Thus, through step S3, dynamic power supply switching and segmented power allocation based on cost signals are realized in each time slot, enabling the power supply mode of the high-voltage DC bus to be adaptively adjusted according to market electricity prices and energy storage status. The output of this step is the set of power supply sources for the current time slot and the power allocation instructions of each source in each segment, providing accurate input for the scheduling execution in the next stage.

[0141] In step S4, based on the operating topology of the segmented high-voltage DC bus and the power allocation results of step S3, the control of AC rectification switching and DC isolation switching is executed to establish or disconnect the energy channel between each battery pack and the bus, and generate corresponding charging and discharging plans within the current and subsequent time slots.

[0142] Specifically, at the entry point of on / off control, the consistency of the operating status, power reference, and safety boundary of each segment needs to be checked. This can include whether the target power of the segment is less than the upper limit allowed for that segment, whether the source output capacity covers the reference value, whether the pre-charge conditions are met, and whether the bus voltage deviation is within the allowable bandwidth. After verification, the on / off decision is made. On / off on the rectifier side is used to establish or release the AC rectification channel, while on / off on the DC side is used to establish or disconnect the energy channel between the segment and each battery pack. To reduce impact, the on / off action can be executed in a "soft first, hard later" order, i.e., pre-charge first, then close the isolation, and finally release the current limit; when exiting, the action is reversed, first reducing the power to the zero power window, then opening the isolation, and finally exiting rectification.

[0143] In some embodiments of this application, the determination of the switching time follows a dual-signal drive of energy and price, and a delayed confirmation strategy is adopted to avoid high-frequency jitter. Each battery pack uses its remaining energy rate as the first trigger value, and the remaining energy rate of that pack in the current time slot is recorded as... The electricity price signal uses the predicted rate of change of electricity prices in adjacent time slots as the second trigger, denoted as... In any given segment, a switch from standby to grid connection or vice versa is triggered only if the following conditions are met: 1) the source corresponding to that segment obtains a positive power reference in step S3 and has sufficient interface capacity; 2) the trigger combination of energy and price reaches a set threshold; and 3) the delayed confirmation timer arrives. The core of the delayed confirmation strategy is to maintain the shortest possible confirmation time window for the trigger conditions. If the price prediction difference between adjacent time slots is lower than a set threshold within this time window, the original on / off state remains unchanged, and no switch is performed. This maintains electrical topology stability during periods of low price signal disturbance, reducing unnecessary thermal stress and contactor wear.

[0144] This step combines the energy surplus rate and the electricity price change rate as the on / off criteria, and sets a delayed confirmation strategy to suppress short-term fluctuations, enabling the on / off control to have threshold memory and switching hysteresis characteristics, thus achieving the following technical effects:

[0145] First, the switching is triggered only when the energy state and price signal both reach the criterion, avoiding frequent transitions caused by a single factor, maintaining a stable on-off relationship of the power channel, and reducing conversion losses and thermal stress caused by the high-frequency operation of rectifier and isolation devices.

[0146] Second, delayed confirmation maintains the original on / off state in scenarios where the prediction difference between adjacent time slots is small, so that small prediction noise no longer causes topology oscillations, reduces repeated rises and falls of bus voltage and current, and maintains the continuity of segmented load distribution and the consistency of scheduling plan.

[0147] Third, the energy surplus rate as a criterion ensures that different segments prioritize channel stability near capacity boundaries, reducing the shallow cycle fidelity consumption caused by short-cycle switching of energy storage units. At the same time, it aligns the switching rhythm with the rolling plan, improving the predictability and economy of the entire power supply control system in multi-time slot operation.

[0148] In the energy-triggered logic, the priority condition for discharge grid connection is that the energy surplus rate of the group is higher than the limit threshold and the temperature rise and health status are permissible; the priority condition for charging grid connection is that the energy surplus rate of the group is lower than the replenishment threshold and the electricity price signal or segmented power reference allows charging replenishment. The price-triggered logic is used to accelerate or delay the switchover: when the price trend is obvious and the group undertakes the discharge power in step S3, the delay time can be shortened; when the price trend is obvious and the group is used for charging replenishment, the delay time can be shortened; in other cases, the nominal delay is maintained. For example, the discharge threshold and replenishment threshold of the energy surplus rate can be taken as the boundary of the target range given by the operating strategy.

[0149] Meanwhile, the delayed confirmation adopts a three-stage process: "start-confirmation-execution". The start point is the moment when the trigger combination is first met; the confirmation point is the first expiration check point after the start of the process; and the execution point is the on / off action window entered after confirmation. If the difference in electricity price forecast between adjacent time slots is detected to be lower than the threshold at the confirmation point, the timer is reset to zero and the original state is retained.

[0150] For rectifier channels and DC isolation, the execution follows the principle of segment priority and cross-segment restriction. The rectifier channel of each segment is only set for the target power of that segment. When the target power of a segment is zero and there is no cross-segment support task, the rectification of that segment remains disconnected. The isolation channel of each battery pack is only closed within the segment it is connected to, and the power reference of that segment is the upper limit. When the same battery pack has the connection conditions across two or more segments, the power reference of this segment is met first, and then it participates in the support of adjacent segments according to the remaining capacity. In addition, all closing actions are performed after the pre-charge is completed. Isolation closure is only allowed after the pre-charge criterion is met. The exit action is executed in the order of "power reduction - isolation disconnection - pre-charge release - rectification exit", and before exiting, it is checked whether the bus voltage drop has entered the dead zone.

[0151] In the embodiments of this application, after completing the on / off control, it is necessary to generate a charging and discharging plan for the current and several subsequent time slots within the rolling window. The plan is generated with segmented power reference, available power of each battery pack, health status and environmental thermal boundary as inputs, and output as time slot-level power reference trajectory and grid-connected / off-grid flag, along with a rotation priority sequence, to distribute the cycle consumption and heat load when multiple groups have equivalent economics, so as to delay the life degradation of individual battery packs.

[0152] Furthermore, the priority sequence calculation combines three dimensions: the number of iterations, the current temperature rise, and the state of charge (SBC), resulting in a ranking order tailored to the time slot application. A higher iteration count places the battery pack further down the sequence; a higher temperature rise places it further down the sequence; a lower SBC position places it further down the sequence when it's in a high-risk zone (too high or too low); and a higher SBC position places it earlier when it's in the target return zone. Based on this ranking, under scenarios involving equivalent cost and equivalent topology reachability, the battery pack with the highest SBC position is prioritized for discharging or charging in the current time slot, while the remaining battery packs participate according to their remaining power quota.

[0153] This step involves constructing a rotation priority sequence based on the number of cycles, temperature rise, and state of charge, and arranging the power-on and standby order of each battery pack accordingly. This transforms the scheduling from a static, fixed process to an adaptive rotation based on state variables, achieving the following technical effects:

[0154] First, the rotation priority sequence allows high-frequency units to give way to low-frequency units in a timely manner, the depth of discharge and cumulative cycles are distributed among the units, excessive concentrated use is suppressed, capacity decay is more balanced, the reserve availability is kept in a stable range, and the callable energy storage redundancy is maintained during abnormal fluctuations, thus enhancing the continuity of power supply and the predictability of dispatch.

[0155] Second, by using temperature rise as a key input, thermal constraints become explicit, thermal load is actively dispersed during rotation, thermal gradients and hot spots in various bus sections are suppressed, the rate of thermal stress accumulation in insulation and conductors is reduced, internal resistance changes during charging and discharging are smoother, power command response tends to be stable, voltage drop and rise amplitudes are reduced, and the frequency of thermal-related protection actions is reduced.

[0156] Third, prioritization is carried out in conjunction with the state of charge (SOC) range, so that different units can take turns to undertake power output and charging tasks around the appropriate SOC range. The ratio of deep and shallow cycles is more reasonable, the equivalent loss during energy path switching is reduced, power allocation and capacity boundaries are coordinated, and the rolling adjustment of plans under price and load disturbances is smoother. The overall operating cost and available capacity utilization rate show stable benefits.

[0157] In some embodiments of this application, to ensure the executability of the plan, the plan generation may include two layers of constraint checks. The first layer checks time consistency, meaning that within any time slot within the window, the grid connection status of the plan does not contradict the on / off status of the energy channel, and the planned power does not exceed the interface capacity and segment upper limit. The second layer checks energy continuity, meaning that rolling energy balancing does not cause the energy surplus rate to fall below the limit threshold or exceed the upper limit, and necessary control margin is reserved for the next time slot at the end of any time slot. If either check fails, adjustments are made according to the strategy of "smoothing first, then reducing the rate": the power trajectory is smoothed on the time axis first, and if it still does not meet the requirements, the rate is reduced uniformly. In unavoidable scenarios, the priority sequence is adjusted, and the task allocation of high-cycle and high-temperature battery packs is postponed.

[0158] Meanwhile, to maintain interface consistency with the power allocation in step S3, the power reference for each segment and each source in the plan can use the result of step S3 as the upper bound, without expanding the command value of any source in this step. When the delayed confirmation trigger maintains the original state, the plan remains consistent with the previous time slot; when the confirmation triggers a switch, the plan sets a "zero-power transition segment" within the switching window to complete pre-charging and isolation closure; for cases requiring cross-segment support, the plan sets transition segments simultaneously in both segments, and ensures that the voltage deviation monitored on both sides is lower than the allowable bandwidth before releasing the current limit. For example, a single switching window can contain consecutive segments of pre-charging, closure, and current limit release, with a total duration less than the minimum control cycle allowed for the segment.

[0159] Furthermore, regarding safety and protection, the following bottom lines must be met during on / off and planned execution: AC side faults or over-temperature alarms take priority, forcibly exiting the branch and clearing the power plan for that source in this time slot; DC side overcurrent, bus overvoltage, or undervoltage triggers immediately derated and enter protection timing, and if it is not restored by the end of the timing period, it will enter the exit process; before any grid connection closing action, the pre-charge current and bus voltage difference must fall within the specified range, and if they are not met, the closing will be postponed until the window ends; after all protection actions are completed, the monitoring side records the mismatch power and the source of the gap in this time slot to provide a basis for subsequent statistics and model correction.

[0160] To implement the delayed confirmation strategy, a threshold for the difference in electricity price predictions between adjacent time slots can be set to determine whether to maintain the on / off state. The threshold can be configured as a fixed constant or adaptively updated according to the recent electricity price prediction error distribution. When the difference between adjacent time slots is less than the threshold and the current topology is stable, the existing on / off state is retained. When the difference exceeds the threshold and energy and temperature rise conditions allow, a switching timer is initiated. Simultaneously, the delay time window can be implemented on the equipment side at the millisecond to second level and on the scheduling side at the time slot granularity; the two are combined to form a hardware and software dual-layer jitter reduction mechanism. For example, the electricity price threshold for delayed confirmation can be on the order of the root mean square error of recent price predictions.

[0161] Furthermore, the generation of the rotation priority sequence can be completed internally by the planner, without exposing weight details. Only the order and effective time period are output. In time slots with equivalent cost and abundant resources, battery packs later in the sequence automatically enter the "maintenance time slot" to maintain their state of charge within the allowable range and reduce temperature rise through low-power charging. In time slots with scarce resources, battery packs later in the sequence only participate under necessary conditions, and their participation power limit is lower than that of the groups earlier in the sequence to control cycle accumulation. For example, several maintenance time slots can be configured within a rolling window, prioritizing battery packs with higher cycle counts and temperature rise indicators to enter this time slot, gradually reducing their stress exposure. For example, the rotation priority sequence is updated at least once a working day, and updated on a shorter cycle during periods of high price volatility.

[0162] At this point, step S4 transforms the power allocation in step S3 into physically executable on / off actions and time-slot-level charging and discharging trajectories. This not only constrains the switching frequency but also provides a rotation mechanism for the lifespan and thermal management of multiple battery groups, thereby reducing the long-term costs and risks caused by invalid switching and unbalanced cycles while maintaining power supply continuity.

[0163] In step S5, the core task is to complete two types of updates before entering the next time slot: refresh the prediction results of direct power supply cost based on new observation data; and perform rolling correction on the charging and discharging plan generated in the previous time slot so that the power reference to be issued is consistent with the price signal, load changes and energy storage status.

[0164] Specifically, the data entry covers the latest electricity price sequence of the day, distributed power output, real-time load, battery state of charge and temperature rise, rectifier and DC / DC efficiency estimation, online quota of segmented buses, and other observations. All data are aligned under a unified timestamp, and missing data is first filled in and anomalies are suppressed. Peak spikes are retained but their training weights are reduced to ensure that short-term drastic changes can be identified by the model, while avoiding excessive disturbance to the parameters. Then, the latest window data drives the price forecaster to obtain the electricity price forecast for the next time slot and synchronously map it to the new direct power supply cost. If changes in rectifier efficiency or line loss estimation are detected during operation, the equivalent parameters are replaced first in the cost mapping to keep the prediction caliber consistent with the cost caliber.

[0165] In some embodiments of this application, the model can employ a combination of incremental learning and a sliding window mechanism. The window slides forward one time slot, introducing the latest samples and discarding the oldest, performing small-step updates on the LSTM model weights. Abnormal segments participate in the update by reducing their weights, avoiding one-time extreme prices from skewing the long-term trend. Furthermore, when the rate of change in electricity prices exceeds a preset threshold, the prediction period can be adaptively shortened, and the output frequency increased to reflect market turning points more quickly; when fluctuations subside, the prediction period returns to the normal granularity. During incremental learning, the previous weights and indicators are retained. If the verification error deteriorates beyond the limit, it immediately reverts to a stable version and records the sample segments of the failed update to prevent the spread of erroneous updates. For example, when prices enter a rapidly rising range, the prediction period is shortened from the normal granularity to half-granularity.

[0166] This step involves incremental learning of model parameters within the prediction period, combined with a sliding window update mechanism. This enables the LSTM model to continuously maintain its sensitivity and generalization ability to electricity price time-series characteristics, and to adaptively adjust the prediction period under drastic fluctuation scenarios, achieving the following technical effects:

[0167] First, the incremental learning mechanism enables the model to update only local weights when receiving new observation data, preserving the stability of the original structure. This allows the model to maintain convergence characteristics and gradually absorb the latest price change information during long-term operation, thus continuously optimizing the prediction accuracy over time and avoiding performance degradation caused by static parameters.

[0168] Second, the sliding window mechanism ensures that the training samples remain continuously rolling in time, and the weights of new and old data are dynamically balanced. This not only prevents overfitting of recent anomalies, but also captures the trend migration process, so that the prediction output forms a stable response between short-term fluctuations and long-term trends, maintaining the smoothness and reliability of the prediction curve.

[0169] Third, when electricity prices fluctuate drastically, the forecast cycle is adaptively shortened to match the model update frequency with the pace of market changes, reducing decision bias caused by forecast lag, and ensuring that power supply dispatching plans remain timely and economical in high-speed fluctuation ranges, thereby realizing the dynamic adaptability and continuous efficient operation of the forecast model in complex market environments.

[0170] The rolling correction on the planning side uses the source cost relationship given in step S3 and the predetermined on / off state in step S4 as the boundary. It recalculates the economic ranking of the next time slot using the newly predicted direct power supply cost and the power supply cost of each battery pack, but does not directly trigger the on / off switching. Only when the delay confirmation in step S4 has been met, and the energy surplus rate, temperature rise, and interface capacity are all allowed, will it be reflected as a change in the grid-connected / off-grid flag at the planning level. Subsequently, without exceeding the segment limit and interface upper limit, the power reference of each group is slightly redistributed, prioritizing the allocation of the newly added economic advantages to battery packs with lower costs and lower temperature rise. If the available power of any group is close to the safety lower limit, its discharge reference is reduced and a recharge time slot is arranged in the plan to ensure that there is still sufficient reserve in the next time slot.

[0171] Furthermore, the rotation priority sequence can be evaluated synchronously with each rollout. Battery packs with higher cycle counts, higher current temperature rises, or near-edge states of charge are automatically moved backward to reduce high-stress exposure; battery packs in the target range are moved forward to take on the task of this time slot. The resulting new sequence only takes effect within the range of cost equivalence and topology accessibility equivalence, avoiding reversal of the established economic dominance. If the difference between the new forecast and the previous forecast is less than the delay confirmation threshold, the plan remains unchanged, ensuring continuous power trajectory and controlled switching frequency.

[0172] In other embodiments of this application, to ensure consistency, timestamps and signatures can be implemented on the plan version, and any changes are accompanied by the source, window range, and rollback point. Two rapid checks are performed before the plan is implemented: an energy continuity check to confirm that no state of charge will exceed limits within the rolling window; and an electrical safety check to confirm that the sequence of pre-charging, current limiting, isolation, and rectification can still be completed within the switching window. If the check fails, it is processed in the order of "prioritizing smoothing, then derating, and retaining the original plan if necessary," and the mismatch information is written to the operation log for subsequent parameter tuning.

[0173] For example, after the direct power supply cost is increased, the plan is to increase the discharge ratio of the more economical battery packs without changing the confirmed on / off conditions, and to arrange maintenance slots for high-temperature battery packs; if the subsequent two short-cycle forecasts show that the price has fallen and the difference is below the threshold again, the regular cycle will be restored, and the current plan will be frozen to avoid repeated rewriting near the boundary.

[0174] Thus, through the processing in step S5, the direct power supply cost and charging / discharging plan are updated to be consistent before the start of the next time slot. This ensures both synchronization with new observation data and adherence to delayed confirmation and rotation mechanisms, guaranteeing a smooth plan, controllable switching, and sufficient reserves, providing a stable entry point for the next cycle.

[0175] In summary, this method, by establishing an electricity price prediction model, constructing a joint objective function, and executing power switching and power allocation under a segmented high-voltage DC bus topology, achieves coordinated operation of dynamic electricity price response, optimal energy storage cost allocation, and adaptive energy channel control. It can maintain system economy and power supply continuity under load fluctuations and electricity price changes, enabling DC power supply systems for data centers or high-power loads to have comprehensive optimization capabilities in energy management, including predictive driving, cost adaptation, and lifetime balancing.

[0176] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S1 to S5 are described sequentially, but this does not mean that steps S1 to S5 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S1 to S5 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S1 to S5 can be appropriately adjusted according to actual needs.

[0177] Example 2:

[0178] like Figure 2 As shown, this embodiment provides a high-voltage DC power supply system based on LSTM prediction, comprising:

[0179] Electricity price prediction module 1 is used to build an LSTM model based on historical electricity price data and current electricity price data, and to predict the electricity price data for the target time period to obtain the direct power supply cost for the next time slot;

[0180] Cost calculation module 2 is used to establish a joint objective function to calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage lifetime constraint.

[0181] Power supply decision module 3 is used to determine the power supply source and power allocation of the current time slot in the topology of the segmented high-voltage DC bus based on the direct power supply cost and the power supply cost of each battery pack.

[0182] Topology control module 4 is used to control the AC rectification and DC isolation switching of the segmented high-voltage DC bus based on the output signal of power supply decision module 3, and to establish or disconnect the energy channel between each battery pack and the bus.

[0183] The plan generation module 5 is used to generate and store the charging and discharging plan of the battery pack. The charging and discharging plan is updated synchronously with the power supply switching instructions.

[0184] The parameter update module 6 is used to update the electricity price prediction module 1 and the charging and discharging plan based on new observation data before the next time slot.

[0185] By adopting the above technical solution, and by setting up functional modules such as electricity price prediction module 1, cost calculation module 2, power supply decision module 3, topology control module 4, plan generation module 5 and parameter update module 6 in the high voltage DC power supply system, predictive driving and multi-source coordination of power supply control are realized, which can maintain the optimal power supply path and efficient system operation under dynamic electricity price and complex load conditions.

[0186] Specifically, the electricity price prediction module 1 uses historical and current electricity price data to construct an LSTM model, enabling the system to learn and predict electricity price trends over time, providing real-time basis for subsequent power supply cost determination; the cost calculation module 2 uses a joint objective function to model charging prices, battery efficiency, power constraints, and lifespan parameters in a unified manner, achieving quantitative assessment and economic judgment of power supply costs; the power supply decision module 3 selects the optimal energy supply path and power allocation based on cost comparison results, transforming the energy scheduling process from fixed logic to real-time optimization; the topology control module 4 executes on / off commands on AC rectification and DC isolation channels based on decision signals, automatically reconstructing the energy flow direction of each battery pack according to the scheduling results, ensuring that the power transmission path is consistent with the operating strategy; the plan generation module 5 records the charging and discharging sequence and keeps it synchronized with the power supply switching command, coordinating energy storage behavior with market price rhythm; the parameter update module 6 performs rolling corrections on model parameters and scheduling plans before each time slot, enabling the system to continuously maintain prediction accuracy and scheduling consistency in long-term operation, thereby achieving comprehensive optimization of data center high-voltage DC power supply in terms of cost, efficiency, and stability.

[0187] In this embodiment of the application, the power supply decision module 3 may include a power allocation unit, which is used to calculate the allocation weight based on the difference between the direct power supply cost and the power supply cost of each battery pack, and generate a power allocation instruction based on the allocation weight.

[0188] In the power supply decision module 3, by introducing a weighting mechanism based on cost differences in the power allocation unit, the generation of power commands can take into account both economic efficiency and electrical constraint coordination, achieving the following technical effects:

[0189] First, the power distribution unit establishes a weight matrix for multiple power supply paths based on the cost difference as the decision-making factor, so that the output power of each path is directly linked to its economic contribution. This allows high-cost channels to be automatically downgraded and low-cost channels to be automatically upgraded, forming a dynamic self-balancing power distribution process that changes with market signals, reducing redundant energy flow and unnecessary power switching.

[0190] Second, the weighting generation mechanism enables the power allocation result to be synchronously matched with the bus topology. The power output ratio between different battery packs is adjusted in real time according to their respective capacity, efficiency and bus constraints, avoiding uneven energy efficiency caused by overload or idleness of a single battery pack, and improving the overall power utilization rate and bus operation stability of the system.

[0191] Third, the power allocation command is uniformly issued by the power supply decision module 3 after real-time calculation, enabling the system to respond quickly and reconstruct the power allocation pattern in the event of sudden load changes or drastic fluctuations in electricity prices, maintain continuous output and voltage balance, and improve the adaptability and continuous economic operation capability of the high voltage DC power supply system under complex operating conditions.

[0192] In summary, by integrating collaborative modules such as electricity price prediction module 1, cost calculation module 2, power supply decision module 3, topology control module 4, plan generation module 5, and parameter update module 6 into the high-voltage DC power supply architecture, this system achieves end-to-end adaptive control from electricity price prediction to power allocation. It can dynamically optimize the power supply path in the context of multiple power sources operating in parallel and fluctuating electricity prices, taking into account both power supply economy and operational stability, and ensuring that data centers achieve efficient, low-cost, and intelligent continuous power supply in high-power-density scenarios.

[0193] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A high-voltage DC power supply method based on LSTM prediction, characterized in that, Includes the following steps: Historical electricity price data and current electricity price data are obtained, an LSTM model is constructed, and the LSTM model is used to predict the electricity price data for the target time period to obtain the direct power supply cost for the next time slot. A joint objective function is established to calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage life constraint. Based on the direct power supply cost and the power supply cost of each battery pack, the power supply source and power allocation of the current time slot are determined in the topology of the segmented high-voltage DC bus. Control the AC rectification and DC isolation switching of the corresponding segments, establish or disconnect the energy channels between each battery pack and the bus, and generate a charging and discharging plan; The control of the AC rectification switching and DC isolation switching of the corresponding segments includes: determining the switching time based on the energy remaining rate of the battery pack of each segment and the rate of change of electricity price. The determination of the switching time adopts a delayed confirmation strategy, and the on / off state remains unchanged when the difference in electricity price prediction between adjacent time slots is lower than a set threshold. The generation of the charging and discharging plan includes setting a rotation priority sequence for each of the battery packs. The rotation priority sequence is determined based on a comprehensive calculation of the number of cycles, current temperature rise, and state of charge of the battery pack. Before reaching the next time slot, the prediction of the direct power supply cost and the charging / discharging plan are updated based on new observation data; The update is achieved by performing incremental learning on the LSTM model parameters in each prediction period, continuously updating the model weights using a sliding window mechanism, and adaptively shortening the prediction period when electricity prices fluctuate drastically.

2. The high-voltage DC power supply method based on LSTM prediction as described in claim 1, characterized in that, The input to the LSTM model includes at least one of historical electricity price data, current electricity price data, distributed power output, real-time data center load, and battery state of charge. The LSTM model introduces a gating mechanism in the hidden layer output to model long-term trends and short-term fluctuations respectively.

3. The high-voltage DC power supply method based on LSTM prediction as described in claim 1, characterized in that, The joint objective function is also used to calculate the available power of each of the battery packs, and to maintain the sum of the available power of each battery pack at a preset threshold value under the constraint of the available power.

4. The high-voltage DC power supply method based on LSTM prediction as described in claim 1, characterized in that, The power source is determined based on a comparison between the cost of direct power supply and the cost of power supply for each battery pack. Within the current time slot, the power source is selected according to the principle of minimizing cost.

5. The high-voltage DC power supply method based on LSTM prediction as described in claim 4, characterized in that, The power allocation is determined based on the difference between the direct power supply cost and the power supply cost of each battery pack, and the power output is allocated according to the topology of the segmented high-voltage DC bus.

6. A high-voltage DC power supply system based on LSTM prediction, used to implement the high-voltage DC power supply method based on LSTM prediction as described in claim 1, characterized in that, include: The electricity price forecasting module is used to build an LSTM model based on historical electricity price data and current electricity price data, and to forecast the electricity price data for the target time period to obtain the direct power supply cost for the next time slot. The cost calculation module is used to establish a joint objective function to calculate the power supply cost of each battery pack. The parameters of the joint objective function include at least one of the following: charging price, battery charging and discharging efficiency, bus power constraint, and energy storage lifetime constraint. The power supply decision module is used to determine the power supply source and power allocation for the current time slot in the topology of the segmented high-voltage DC bus based on the direct power supply cost and the power supply cost of each battery pack. The topology control module is used to control the AC rectification and DC isolation switching of the segmented high-voltage DC bus according to the output signal of the power supply decision module, and to establish or disconnect the energy channel between each of the battery packs and the bus. A plan generation module is used to generate and store the charging and discharging plan of the battery pack, and the charging and discharging plan is updated synchronously with the switching command of the power supply source; The parameter update module is used to update the electricity price prediction module and the charging and discharging plan based on new observation data before reaching the next time slot.

7. A high-voltage DC power supply system based on LSTM prediction as described in claim 6, characterized in that, The power supply decision module includes a power allocation unit, which is used to calculate the allocation weight based on the difference between the direct power supply cost and the power supply cost of each battery pack, and generate a power allocation instruction based on the allocation weight.

Images