Mobile computing power vehicle scenery storage integrated container energy storage system

Abstract

The application discloses a mobile computing power vehicle wind and light storage integrated container energy storage system, belongs to the field of electric power systems, mobile computing power infrastructure and new energy storage control, and relies on industrial big data to collect computing power tasks, equipment operation, weather, energy storage and vehicle-mounted multi-source time sequence data, realizes multi-time scale computing power and wind and light power interval prediction through a time sequence convolution network, builds a multi-time scale optimization scheduling module to quantize power uncertainty and reserve energy storage standby capacity, adopts an adaptive filtering hybrid energy storage distribution strategy to separate high and low frequency power components, combines industrial real-time streaming data to roll over and correct power instructions, fuses thermal power coupling, computing power task cooperation, direct current bus virtual inertia control, and switches operation strategies according to vehicle-mounted working condition big data, and identifies and updates energy storage model parameters online. Relying on industrial big data, heterogeneous data fusion and full-process intelligent regulation and control are completed, bus voltage fluctuation is effectively inhibited, and battery high-frequency impact is reduced.

Description

Technical Field

[0001] This invention belongs to the fields of power systems, mobile computing infrastructure and new energy storage control, and specifically relates to a mobile computing vehicle integrated wind, solar and energy storage containerized energy storage system. Background Technology

[0002] Mobile computing facilities enable the rapid deployment of servers, network equipment, and data processing devices in industrial sites, temporary communication nodes, emergency support areas, remote work areas, and near edge computing nodes. These facilities typically utilize vehicle chassis, semi-trailer platforms, or detachable containers as their support base, forming temporary computing centers after parking, leveling, grounding, and interface connections. The load on mobile computing facilities is not constant. Changes in task queues, the start and stop of CPU and GPU resources, task computation phase transitions, model loading, data transmission, and the operation of liquid cooling equipment all cause power variations. Some load variations exhibit low-frequency characteristics lasting from several minutes to several hours, while others manifest as step or pulse characteristics lasting less than a few seconds.

[0003] Wind and solar power can reduce the dependence of mobile computing facilities on external power grids or fuel-fired power generation equipment, but changes in wind speed, irradiance, ambient temperature, and shading can cause fluctuations in the power output of renewable energy sources. When computing load and renewable energy power output change simultaneously, controlling battery charging and discharging solely based on the current power difference can easily lead to problems such as control lag, batteries being subjected to high-frequency power surges, supercapacitor energy deviating from its available range, and rapid changes in external power grid output. When using fixed-time scheduling for energy dispatch, load forecasting errors and power generation forecasting errors can also cause a continuous deviation between planned and actual power output.

[0004] Existing hybrid energy storage control typically utilizes low-pass filtering to allocate low-frequency power to batteries and high-frequency power to supercapacitors. A fixed cutoff frequency cannot simultaneously adapt to the power spectrum of different computing tasks, varying battery temperatures, and different supercapacitor energy storage states. When multiple high-power tasks occur consecutively, the supercapacitor may be unable to continue compensating due to insufficient energy; similarly, when the computing load drops rapidly or renewable energy generation suddenly increases, the supercapacitor may also be unable to absorb excess power due to insufficient remaining rechargeable energy. While some control schemes output prediction confidence intervals, they fail to convert prediction uncertainties into executable charging and discharging backups, resulting in data gaps between the prediction results and energy storage control.

[0005] The mobile scenario also introduces transportation, parking, deployment, and operation states. Wind power units and photovoltaic panels need to be locked during transportation, and deployment requires outrigger stabilization, vehicle tilt angle confirmation, photovoltaic panel deployment, and wind power unit unlocking. If the energy dispatch module ignores the deployment state, the dispatch result may use renewable energy power that is not yet ready for generation. Mobile relocation, changes in ambient temperature, and energy storage aging also alter the battery's equivalent internal resistance, the supercapacitor's available energy, and the bidirectional DC-DC converter's response characteristics, making it difficult to maintain control accuracy with fixed model parameters over long periods. Summary of the Invention

[0006] The purpose of this invention is to provide a mobile computing vehicle-integrated wind, solar and energy storage containerized energy storage system to solve the problem that planned power is difficult to convert into stable real-time power commands when the computing load changes rapidly, wind and solar power generation fluctuates, hybrid energy storage is limited, and mobile deployment status changes are combined.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] The mobile computing vehicle-mounted integrated wind, solar, and energy storage containerized energy storage system includes a mobile carrier component, a container, computing equipment, wind power generation modules, photovoltaic power generation modules, battery energy storage modules, supercapacitor energy storage modules, a DC bus, an external power grid interface, and an energy management main controller. The container contains at least a computing equipment compartment and an energy storage system compartment. The computing equipment compartment houses servers, network equipment, and liquid-cooled terminals. The energy storage system compartment houses the battery energy storage modules, the supercapacitor energy storage modules, the bidirectional DC-DC converter, the DC power distribution unit, the fire protection unit, and the liquid-cooling unit. The wind power generation modules and photovoltaic power generation modules remain locked or retracted during transport, undergo deployment checks during deployment, and are connected to the DC bus via corresponding power conversion units during operation.

[0009] The energy management main controller executes computing load forecasting, wind and solar power generation forecasting, multi-timescale optimized scheduling, hybrid energy storage power allocation, and rolling feedback control. The computing load forecasting module maps computing task scheduling data and server operation data to future computing device power sequences across multiple timescales. The wind and solar power generation forecasting module maps meteorological data and historical power generation data to wind power and photovoltaic power sequences with the same time index. The multi-timescale optimized scheduling module calculates the planned power of battery energy storage, the target state of charge of supercapacitors, the power of new energy utilization, and the planned power of the external power grid based on the forecast results. The hybrid energy storage power allocation module allocates battery power and supercapacitor power based on real-time net power deviation. The rolling feedback control module corrects power commands based on actual load, actual power generation, DC bus voltage, battery state of charge, and supercapacitor state of charge.

[0010] In some implementations, the computing load prediction module outputs a prediction interval, and the wind and solar power generation prediction module also outputs a prediction interval. The multi-timescale optimization scheduling module converts the prediction interval into net power uncertainty in the power shortage direction and the power surplus direction, and reserves for discharge and charging reserves. The prediction interval is no longer treated as an isolated result, but directly limits the adjustable power of energy storage and the available energy.

[0011] In some implementations, the hybrid energy storage power distribution module does not use a fixed cutoff frequency. Instead, it selects the cutoff frequency for the current control window based on the battery's allowable power change rate, the remaining dischargeable energy of the supercapacitor, the remaining rechargeable energy of the supercapacitor, the allowable power of the bidirectional DC-DC converter, and the future net power deviation spectrum. After high-frequency compensation, the mobile computing vehicle-integrated wind-solar-storage containerized energy storage system generates limited supercapacitor energy recovery power, allowing the supercapacitor to return to the target state of charge range that balances power absorption and release.

[0012] In some implementations, the mobile computing vehicle-integrated wind-solar-storage containerized energy storage system determines the adjustable power domain of the computing task based on the task deadline, allowable interruption flag, checkpoint recovery time, minimum resource quota, and remaining computing power. When the energy storage backup and the available power from the external power grid cannot cover the net power gap, the mobile computing vehicle-integrated wind-solar-storage containerized energy storage system only changes the number of graphics processors, CPU quota, task concurrency, or task start time within the adjustable power domain of the computing task, without changing the minimum resource requirements of the non-adjustable tasks.

[0013] In some implementations, the mobile computing vehicle-integrated wind, solar, and energy storage containerized energy storage system extracts upcoming task start events from task scheduling messages and generates supercapacitor feedforward power before the actual power of the computing equipment increases. After task start, the mobile computing vehicle-integrated wind, solar, and energy storage containerized energy storage system uses DC bus voltage deviation and actual power deviation to correct the feedforward power and gradually transfers the continuous power component to the battery energy storage module.

[0014] In some implementations, the mobile computing vehicle integrated wind, solar and energy storage container energy storage system predicts the power of the liquid cooling system based on the predicted power of the computing equipment, the inlet temperature of the coolant, the outlet temperature of the coolant, the flow rate of the coolant and the ambient temperature, adds the power of the liquid cooling system to the total load prediction of the container, and corrects the allowable power of the battery and the attenuation cost based on the battery temperature.

[0015] In some implementations, the mobile computing vehicle-integrated wind-solar-storage containerized energy storage system manages its transportation, deployment, and operational states based on the parking brake status, outrigger position, vehicle tilt angle, and the deployment status and angle of the wind power generation module and the photovoltaic power generation module. Once in operational mode, the system performs amplitude-limited online identification, updating the energy storage equivalent internal resistance, converter response time, and available power boundary. When model effectiveness decreases, the system switches to a conservative control strategy based on DC bus voltage, battery state of charge, and supercapacitor state of charge.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] This invention incorporates computing power task scheduling data, server operation data, meteorological data, wind and solar power generation data, liquid cooling auxiliary power, and energy storage status into a unified time index, reducing data misalignment caused by different sampling periods. It separates multi-timescale planning and scheduling from real-time net power allocation; planning and scheduling are responsible for determining energy boundaries and energy storage status targets, while real-time control eliminates second-level power deviations, reducing power oscillations caused by repeated adjustments between the two control layers. Furthermore, it converts the computing power load prediction interval and the wind and solar power generation prediction interval into bidirectional reserve power and reserve energy, incorporating prediction uncertainty into executable energy storage constraints, thereby reducing load shortages and renewable energy abandonment caused by prediction errors.

[0018] This invention adjusts the power decomposition parameters based on the battery power change rate, the available energy of the supercapacitor, and the future net power deviation spectrum, and restores the supercapacitor energy after high-frequency compensation, thereby reducing the high-frequency power impact on the battery and the energy saturation of the supercapacitor.

[0019] This invention utilizes the adjustable power domain of computing power tasks constrained by task deadlines and minimum resource quotas when energy storage capacity is insufficient, and leverages task start events to drive supercapacitors in advance, thereby reducing DC bus transient deviations without compromising the resource requirements of non-adjustable tasks.

[0020] This invention incorporates mobile deployment status and online identification results into energy control, which can avoid using new energy power generation capacity that has not yet been deployed, and can adapt to changes in model parameters caused by relocation, temperature changes, and energy storage aging. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is the main flowchart of the method described in this invention.

[0023] Figure 2 This is a flowchart of the adaptive power allocation subprocess of hybrid energy storage in this invention.

[0024] Figure 3 This is a flowchart illustrating the collaborative adjustment of branches for computing power tasks in this invention.

[0025] Figure 4 This is a flowchart of the mobile deployment status management branch of the present invention. Detailed Implementation

[0026] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0027] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.

[0028] This invention discloses a mobile computing vehicle-integrated wind, solar, and energy storage containerized energy storage system, comprising a mobile carrier component, a container mounted on the mobile carrier component, computing equipment disposed within the container, a battery energy storage module, a supercapacitor energy storage module, a DC bus, an external power grid interface, and an energy management main controller, as well as a wind power generation module and a photovoltaic power generation module connected to the DC bus. The battery energy storage module and the supercapacitor energy storage module are respectively connected to the DC bus via bidirectional DC-DC converters. The energy management main controller includes a computing load prediction module, a wind and solar power generation prediction module, a multi-timescale optimized scheduling module, a hybrid energy storage power allocation module, and a rolling feedback control module.

[0029] The computing load prediction module outputs a multi-timescale power prediction sequence for computing devices based on computing task scheduling data and computing device operation data. The module constructs a temporal convolutional network. The inputs to this network include task identifier, task type, task stage, task resource quota, CPU utilization, GPU utilization, memory usage, network bandwidth usage, number of running servers, historical computing device power, and timestamps. The temporal convolutional network uses dilated causal convolution and residual connections to form a multi-scale receptive field, and outputs the median predicted power of computing devices, the lower bound of the prediction interval, and the upper bound of the prediction interval for different prediction durations through multiple output heads.

[0030] The wind and solar power generation prediction module outputs wind power generation prediction sequences and photovoltaic power generation prediction sequences that are time-aligned with the power prediction sequence of the computing equipment, based on meteorological data and historical power generation data. Specifically, wind speed, wind direction, irradiance, ambient temperature, air pressure, humidity, historical wind power generation, and historical photovoltaic power generation are resampled according to a unified scheduling time index, and the median predicted value of wind power generation, the predicted range of wind power generation, the median predicted value of photovoltaic power generation, and the predicted range of photovoltaic power generation are output. The power prediction range of the auxiliary equipment is synchronously output by the computing load prediction module based on the historical operating data of the auxiliary equipment and the task stage information in the power prediction sequence of the computing equipment.

[0031] The multi-timescale optimization scheduling module determines the planned power of battery energy storage, the target state of charge of supercapacitors, the power of new energy utilization, and the planned power of the external power grid based on the power prediction sequences of the computing equipment, the wind power generation prediction sequences, and the photovoltaic power generation prediction sequences. The hybrid energy storage power allocation module determines the net power deviation and generates reference values ​​for battery energy storage power and supercapacitor energy storage power based on the total load power of the container, the actual wind power generation power, the actual photovoltaic power generation power, and the planned power of the external power grid.

[0032] The battery energy storage module is used to handle the low-frequency power component and continuous energy gap in the net power deviation, while the supercapacitor energy storage module is used to handle the high-frequency power component and instantaneous power gap in the net power deviation. The hybrid energy storage power distribution module performs power limiting, energy limiting, and power change rate limiting on the battery energy storage power reference value and the supercapacitor energy storage power reference value, and transfers the remaining power exceeding the allowable range of the battery energy storage module or the supercapacitor energy storage module to another energy storage module or the external power grid interface.

[0033] The rolling feedback control module corrects the reference values ​​of the battery energy storage power, the reference values ​​of the supercapacitor energy storage power, and the planned power of the external power grid based on the actual power of the computing equipment, the actual power of the auxiliary equipment, the DC bus voltage, the state of charge of the battery, and the state of charge of the supercapacitor.

[0034] The multi-timescale optimization scheduling module determines the net power uncertainty in the direction of power shortage and the net power uncertainty in the direction of power surplus based on the median predicted power value of the computing equipment, the lower bound of the power prediction interval of the computing equipment, the upper bound of the power prediction interval of the computing equipment, the power prediction interval of the auxiliary equipment, the median predicted power value of the wind power generation, the wind power prediction interval, the median predicted power value of the photovoltaic power generation, and the photovoltaic power prediction interval. It also reserves discharge reserve power, discharge reserve energy, charging reserve power, and charging reserve energy for the battery energy storage module and the supercapacitor energy storage module, respectively.

[0035] The hybrid energy storage power distribution module does not use a fixed cutoff frequency. Instead, it selects the cutoff frequency corresponding to the current rolling control window from multiple candidate cutoff frequencies based on the allowable power change rate of the battery energy storage module, the remaining dischargeable energy of the supercapacitor energy storage module, the remaining rechargeable energy of the supercapacitor energy storage module, the allowable power of the bidirectional DC-DC converter, and the predicted net power deviation spectrum. After the high-frequency power component in the net power deviation is compensated, energy recovery power is generated based on the deviation between the actual state of charge of the supercapacitor energy storage module and the target state of charge of the supercapacitor. Under the condition of satisfying the battery power change rate constraint and the DC bus power balance constraint, the supercapacitor energy storage module is restored to the range of the target state of charge of the supercapacitor.

[0036] Furthermore, the energy management main controller also includes a computing power task coordination module. This module categorizes computing power tasks into non-adjustable tasks, deferred tasks, and downgradeable tasks based on task deadlines, allowed interruption flags, checkpoint recovery times, minimum resource quotas, task priorities, and remaining computing power. It determines the lower and upper bounds of computing power task power for each scheduling cycle. When the energy storage reserve capacity and the available power of the external power grid cannot cover the net power gap, the module outputs a task resource adjustment command to the computing equipment, located between the lower and upper bounds of the computing power task power.

[0037] The computing power task collaboration module also generates an instantaneous power feedforward based on the task start time, the number of graphics processors to be activated, the task batch processing scale, the task calculation stage, and the historical power template of similar tasks. Before the task starts, the hybrid energy storage power allocation module outputs an early discharge command corresponding to the instantaneous power feedforward to the supercapacitor energy storage module. After the task starts, it corrects the supercapacitor energy storage power reference value according to the DC bus voltage deviation and the actual power deviation of the computing power device. Then, it transfers the continuous power component to the battery energy storage module according to the preset power change rate.

[0038] The energy management main controller also includes a virtual inertia and impedance reshaping module. This module calculates the target virtual inertia coefficient of the DC bus based on the rate of change of the instantaneous power feedforward, the rate of change of the actual wind power generation, and the rate of change of the actual photovoltaic power generation output by the computing power task coordination module. It then generates a reference value for the advance compensation current based on the target virtual inertia coefficient and the first derivative of the DC bus voltage deviation, and superimposes this reference value onto the supercapacitor energy storage power reference value. When the rate of change of the instantaneous power feedforward exceeds a preset rate of change threshold, the hybrid energy storage power allocation module predicts the DC bus voltage drop depth based on the advance compensation current reference value and the equivalent internal resistance of the supercapacitor. If the predicted DC bus voltage drop depth exceeds the allowable lower voltage limit, the module reconstrains the instantaneous power feedforward by limiting the transient load ramp rate of the graphics processor in the computing power device.

[0039] The energy management main controller also includes a thermoelectric coupling prediction module. This module determines a liquid cooling system power prediction sequence based on the computing equipment power prediction sequence, the server cooling circuit inlet temperature, the server cooling circuit outlet temperature, the coolant flow rate, and the ambient temperature. It then combines this liquid cooling system power prediction sequence with the computing equipment power prediction sequence to form the container's total load power prediction sequence. Finally, it adjusts the allowable charge / discharge power of the battery energy storage module and the battery energy storage degradation cost based on the battery temperature.

[0040] The container is equipped with a heat exchange isolation device, through which the liquid cooling circuit of the battery energy storage module is coupled to the server cooling circuit. The thermoelectric coupling prediction module also extracts the local task load rate of multiple server racks within the container to construct a spatial heat source distribution matrix. Based on this matrix, it calculates the independent proportional control valve opening command for each server cooling circuit and predicts the transient hydraulic load power based on the system's comprehensive hydraulic damping coefficient and the circulating pump motor model corresponding to the independent proportional control valve opening command. The thermoelectric coupling prediction module feeds this transient hydraulic load power forward as a high-frequency disturbance to the hybrid energy storage power allocation module, so that when the computing power and circulating pump power increase simultaneously, the supercapacitor energy storage module can synchronously compensate for the superimposed transient power gap of the computing equipment and the liquid cooling system. The thermoelectric coupling prediction module outputs a global heat dissipation power command based on the heat generation power of the battery energy storage module, the heat generation power of the supercapacitor energy storage module, and the spatial heat source distribution matrix to adjust the chiller unit power.

[0041] The energy management main controller collects data on the parking brake status, outrigger position, vehicle tilt angle, wind power module deployment status, and photovoltaic power module deployment angle through onboard sensors. Based on these data, it determines the transportation, deployment, or operation status and limits the allowable power output of wind power and photovoltaic power generation. Upon entering the operation status, the energy management main controller sends amplitude-limited online identification power sequences to the battery energy storage module and the supercapacitor energy storage module. It updates the energy storage equivalent internal resistance, converter response time, and available power boundary based on the DC bus voltage response and energy storage current response. When the prediction error or model residual continuously exceeds the effective range, it switches to a conservative control strategy based on the DC bus voltage, battery state of charge, and supercapacitor state of charge.

[0042] The details of the above technical solution and its beneficial effects compared with the prior art are further illustrated below by means of specific embodiments and comparative examples.

[0043] Example 1: In this example, after the mobile computing vehicle completes parking and grounding, four wind power generation devices are deployed, and the photovoltaic panels are extended to 30°. The computing equipment cabin houses ten server racks with a total rated power of 400kW. The liquid cooling system consists of a coolant distribution unit, a circulating pump, and an air-cooled chiller unit. Computing equipment operation data is collected every 1 second, meteorological data every 1 minute, scheduling updates every 5 minutes, and rolling feedback control is executed every 1 second.

[0044] The computing load prediction module reads data from the past 24 hours and outputs a series of predicted computing device power for the next 5 minutes, 15 minutes, 1 hour, and 4 hours.

[0045] The median predicted power of computing equipment over a 5-minute time period is 320kW, with a lower bound of 305kW and an upper bound of 342kW. When the thermoelectric coupling prediction module is not enabled, the planned power of auxiliary equipment is the average of the most recent 15 minutes, which is 42kW. The median predicted power of wind power is 24kW, the median predicted power of photovoltaic power is 16kW, and the planned power of the external power grid is 140kW.

[0046] Within one real-time control cycle, the actual power of the computing equipment is 320kW, the actual power of the auxiliary equipment is 42kW, the actual power of wind power generation is 24kW, and the actual power of photovoltaic power generation is 16kW. The net power deviation is calculated using the following formula:

[0047]

[0048] In the formula, For the first Net power deviation per real-time control cycle This refers to the actual power of the computing equipment. The actual power of the auxiliary equipment, This represents the actual power output of wind power generation. This represents the actual power output of photovoltaic power generation. Planned power for the external power grid.

[0049] Substituting these values, the net power deviation is 182kW. The output of a fourth-order Butterworth low-pass filter with a fixed cutoff frequency of 0.01Hz is 138kW, forming an initial power reference value of 138kW for battery energy storage and 44kW for supercapacitor energy storage. The bidirectional DC-DC converter connected to the battery energy storage module has a rated power of 250kW, and the current allowable battery discharge power is 220kW, which is within limits; the bidirectional DC-DC converter connected to the supercapacitor energy storage module has an allowable discharge power of 600kW, which is also within limits.

[0050] The rolling feedback control module detects that the DC bus voltage drops from 800V to 792V, forming a negative 8V deviation. After control solution, it increases the reference value of the supercapacitor energy storage power by 6kW and the external grid power by 2kW. In the next control cycle, the DC bus voltage rises back to 798V, and the rolling feedback control module continues to reduce the correction amount.

[0051] Example 2: This example adds bidirectional backup power, adaptive cutoff frequency and supercapacitor energy recovery to Example 1, and sets a comparative example to illustrate the technical effects.

[0052] During a specific planned scheduling period, the predicted lower limit for total container load is 340kW, the median predicted value is 362kW, and the upper limit is 385kW; the predicted lower limit for wind power generation is 18kW, the median predicted value is 24kW, and the upper limit is 29kW; the predicted lower limit for photovoltaic power generation is 12kW, the median predicted value is 16kW, and the upper limit is 19kW; the planned power of the external power grid is 140kW. Calculate the net power deviation and reserve power using the following formula:

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] In the formula, For the planned scheduling time period index, This is the upper limit for the total container load forecast. This is the median forecast for total container load. This is the lower bound for forecasting total container load. This is the upper limit for predicting wind power generation capacity. This represents the median forecast value for wind power generation. This is the lower bound for predicting wind power generation capacity. This is the upper limit for predicting photovoltaic power generation. This represents the median forecast value for photovoltaic power generation. This is the lower bound for predicting photovoltaic power generation. This represents the upper limit of the net power deviation. This is the median value of the net power deviation. This is the lower bound of the net power deviation. Reserved power for the discharge direction. Reserved power for charging direction.

[0059] The system obtains a reserve power of 33kW for discharge and 30kW for charging. The short-time response portion is allocated to the supercapacitor energy storage module, and the continuous portion is allocated to the battery energy storage module. Scheduling does not utilize the reserved reserves.

[0060] The current state of charge (SOC) of the supercapacitor is 46%, and the SOC of the battery is 68%, with an allowable power change rate of 40 kW / s. The net power deviation over the next 60 seconds includes two rapid increases and one rapid decrease. The hybrid energy storage power allocation module performs low-pass filtering and evaluation on each candidate cutoff frequency, eliminating those that cause exceedances, and finally selects 0.008 Hz. After the rapid power event ends, the SOC of the supercapacitor is 37%. Due to the high reserve in the future discharge window, the target SOC of the supercapacitor is set to 55%, generating energy recovery power to restore the SOC of the supercapacitor to the target range using a limiting ramp method.

[0061] Comparative Example 1: This comparative example uses a low-pass filter hybrid energy storage control scheme with a fixed cutoff frequency of 0.01Hz. The remaining control architecture, initial state, and boundary conditions are exactly the same as in Example 2. The same 180s net power deviation sequence as in Example 2 was used for testing.

[0062] The test results are as follows: the root mean square value of the battery power is 142kW, the maximum power change rate of the battery is 96kW / s, the state of charge of the supercapacitor is 12% to 93%, three energy storage state limits occurred, and the maximum voltage deviation of the DC bus is 4.8% of the rated voltage.

[0063] Comparative Example 2: This comparative example does not reserve bidirectional backup power; the remaining control architecture, initial state, and boundary conditions are exactly the same as in Example 2. The same 4-hour prediction sequence and actual deviation sequence as in Example 2 were used for testing.

[0064] The test results showed that the load shortage was 5.6 kWh, the abandoned renewable energy was 3.9 kWh, and the external power grid exceeded its limits for four periods.

[0065] After adopting the adaptive cutoff frequency and bidirectional backup of Example 2, the root mean square value of the battery power is 128kW, the maximum power change rate is 43kW / s, the state of charge of the supercapacitor is 28% to 78%, no state limiting occurs, and the maximum voltage deviation of the DC bus is 1.9%; the load shortage energy is reduced to 0.8kWh, the energy abandoned by new energy sources is reduced to 1.2kWh, and the power of the external grid does not exceed the limit.

[0066] Example 3: This example adds adjustable power domain for computing power tasks, task event feedforward control, and thermoelectric coupling prediction to Example 2.

[0067] The computing power task queue includes Task A, Task B, and Task C. Task A is a real-time inference task, which cannot be interrupted, with a minimum GPU quota of eight, and is classified as a non-adjustable task. Task B is a batch training task, which allows interruption after checkpoint saving, with a deadline of 6 hours, and is classified as a deferred task. Task C is a data preprocessing task, which allows reducing the CPU concurrency, with a minimum quota of 50% of the current quota, and is classified as a downgradeable task. Therefore, the upper and lower power limits for Task A are both 120kW, for Task B the upper limit is 160kW and the lower limit is 0, and for Task C the upper limit is 80kW and the lower limit is 40kW, with a total adjustable power range of 200kW.

[0068] When a 70kW power supply gap is predicted, and the combined remaining capacity of energy storage and the external power grid is insufficient, the computing power task coordination module keeps the unadjustable task unchanged and reduces the power of task C from 80kW to 70kW to eliminate the gap.

[0069] Subsequently, the task scheduling system issues task D, which will activate sixteen graphics processors after 2 seconds. Based on the historical template of task D, the initialization phase of task D will generate an instantaneous power increment of 80kW. The computing power task coordination module generates an 80kW feedforward, and the hybrid energy storage power allocation module outputs a 70kW pre-discharge command to the supercapacitor energy storage module before task startup. After task startup, the actual additional power of task D is 86kW, the DC bus voltage drops by 6V, and the rolling feedback control module increases the reference value of the supercapacitor energy storage power by 16kW, matching the total additional power to 86kW. After entering the stable calculation phase, the battery power increases to 65kW at a slope of 10kW / s, and the reference value of the supercapacitor energy storage power synchronously decreases to 0, achieving smooth power transfer.

[0070] The thermoelectric coupling prediction module reads the predicted power of the computing equipment as 360kW, inlet temperature as 25℃, outlet temperature as 32℃, total coolant flow rate as 48m³ / h, and ambient temperature as 35℃. Based on the performance table, the circulating pump power is 12kW, the chiller unit power is 36kW, and the total liquid cooling auxiliary power is 48kW. The predicted total load of the container is revised to 408kW. When the average battery temperature rises from 38℃ to 46℃, the allowable continuous discharge power decreases from 220kW to 180kW. The boundary of the allowable continuous discharge power and the battery energy storage degradation cost are written into the multi-timescale optimization scheduling module, prompting the scheduling scheme to increase external grid planning and retain more instantaneous power from the supercapacitor.

[0071] Example 4: This example adds mobile deployment status management, online identification after deployment, and model failure safety rollback to the existing features in example 3.

[0072] In transport mode, the parking brake signal is invalid, the outriggers are not extended, the photovoltaic panel deployment angle is 0°, and the wind power generation device lock signal is valid. The energy management main controller sets both the wind power generation allowable power and the photovoltaic power generation allowable power to 0, prohibiting the power converter from supplying power to the DC bus. The battery energy storage module only maintains auxiliary power supply for the controller, communication, and safety monitoring. Upon reaching the deployment position, it enters the deployment state, the parking brake is valid, the outriggers are extended and the force difference is within the allowable range, the longitudinal tilt angle is 0.4°, the lateral tilt angle is 0.3°, and the protective grounding test passes. During the photovoltaic panel deployment process, the allowable power is released in stages according to the angle; the wind power generation device can only enter the power generation state after mechanical position confirmation, vibration sensor self-test, and wind speed range check pass.

[0073] Upon entering operational status, online identification is performed. When the load is stable and reserves are sufficient, a positive and negative step power sequence with an amplitude of 5% of the rated power and a duration of 4 seconds is sent to the bidirectional DC-DC converter connected to the battery energy storage module, and a positive and negative step power sequence with an amplitude of 3% of the rated power and a duration of 2 seconds is sent to the bidirectional DC-DC converter connected to the supercapacitor energy storage module. Based on the voltage and current responses, the battery equivalent internal resistance is updated to 0.1Ω, the response time constant is updated to 0.8s, and the supercapacitor equivalent internal resistance and converter response time are also updated.

[0074] The energy management main controller continuously monitors the residuals of each module. When the DC bus model residuals exceed the effective range for 10 consecutive control cycles, it switches to a conservative control strategy. Under this conservative control strategy, the supercapacitor energy storage module directly generates rapid power based on the DC bus voltage deviation, the battery energy storage module follows the continuous power gap at a maximum change rate of 20 kW / s, the external power grid adjusts its power at a maximum change rate of 5 kW / s, and delayed tasks are paused. Model predictive control resumes after the model is re-identified and passes residual verification for three consecutive verification windows.

[0075] Example 5: Based on Example 4, this example further introduces a collaborative mechanism that includes the coupling of electrical transient micro-control and fluid dynamics, and details the complete control process of the virtual inertia and impedance reshaping module, the spatial heat source distribution matrix, and the superimposed transient power gap.

[0076] Mobile computing vehicles perform multimodal inference tasks in remote work areas. These tasks are characterized by power nonlinearity; when the graphics processing unit cluster receives large-scale tensor computation requests, the transistors flip synchronously, causing the power consumption of the computing equipment to surge at a rate of hundreds of kilowatts per second. Conventional proportional-integral control and basic feedforward control are limited by the current sensor sampling period, which can create control blind spots during sudden load changes, leading to a deep drop in DC bus voltage.

[0077] The virtual inertia and impedance reshaping module inside the energy management main controller initiates high-frequency timing analysis. This module analyzes the absolute amplitude of the instantaneous power feedforward emitted by the computing power task coordination module and extracts the time derivative characteristics of the instantaneous power feedforward. The wind power generation module is affected by natural wind turbulence, and the photovoltaic power generation module is obstructed by local cloud movement; both exhibit wide-frequency fluctuations in their actual power generation. The virtual inertia and impedance reshaping module orthogonally synthesizes the power derivatives from multiple sources to calculate the target virtual inertia coefficient of the DC bus.

[0078] The above calculation process is quantified based on the following equation:

[0079]

[0080] In the formula, The target virtual inertia coefficient, The fundamental inertia coefficient provided for the system's inherent static capacitance. The gain factor is the rate of change of computing power. The instantaneous power feedforward is the output of the computing power task coordination module. For time variables, For the gain factor of new energy power generation fluctuation, The actual wind power generation capacity is mentioned. The actual photovoltaic power generation is denoted as .

[0081] Once the target virtual inertia coefficient is determined, the control flow enters the impedance reshaping stage. In addition to receiving the conventional supercapacitor energy storage power reference value, the bidirectional DC-DC converter matched to the supercapacitor energy storage module also simultaneously receives an advance compensation current reference value derived from the target virtual inertia coefficient. This advance compensation current reference value forces the bidirectional DC-DC converter to actively inject compensation energy based on the trend of small voltage fluctuations before a large-scale voltage drop actually occurs. The physical meaning of this advance compensation current reference value is equivalent to connecting a dynamically variable virtual capacitor with no actual volume in parallel across the DC bus.

[0082] The equation for solving the reference value of the lead compensation current is defined as follows:

[0083]

[0084] In the formula, This is the reference value for the advance compensation current. This is the nominal value of the physical distributed capacitance of the DC bus. The target virtual inertia coefficient, The DC bus voltage is [value missing].

[0085] When the rate of change of the instantaneous power feedforward exceeds the preset rate of change threshold, the mobile computing vehicle-integrated wind-solar-storage containerized energy storage system no longer faces a simple power replenishment problem, but rather a matching problem between the physical release limit of the energy storage devices and the bus impedance. At this time, the hybrid energy storage power distribution module intervenes, predicting the upcoming DC bus voltage drop depth in advance based on the advance compensation current reference value and the equivalent internal resistance of the supercapacitor.

[0086] The nonlinear prediction model for fall depth is expressed as follows:

[0087]

[0088] In the formula, To predict the depth of DC bus voltage sag, This is the nominal value of the physical distributed capacitance of the DC bus. The target virtual inertia coefficient, At the initial moment of load mutation, This is the moment when the load current climbs to its peak value. The transient current of the total load of the container, including the computing equipment and auxiliary equipment. This refers to the transient power supply current at the source end, which includes wind and solar power generation and external grid support. This represents the maximum unbalanced current difference within the abrupt change interval. To identify the updated equivalent internal resistance of supercapacitors online.

[0089] The hybrid energy storage power distribution module compares the predicted DC bus voltage drop depth with the allowable lower voltage limit set by the mobile computing vehicle's integrated wind-solar-storage containerized energy storage system. If the predicted DC bus voltage drop depth exceeds the hard lower limit boundary of 750V, the energy management main controller triggers a reverse task intervention mechanism. The computing power task coordination module forcibly intervenes in the underlying task distribution queue of the computing power device, directly limiting the transient loading ramp rate of the graphics processor in the computing power device by sending a current limiting command to the unified computing resource scheduling platform. This forces the original step-type task scheduling to be decomposed into micro-batch scheduling with millisecond-level time intervals, cutting off the physical cause of the extreme voltage drop at its source.

[0090] Electrical energy conversion is accompanied by high heat flux density. When hundreds of kilowatts of electrical energy are converted into waste heat in a tiny volume, the lag in the response of the liquid cooling system will expose server chips to the risk of thermal penetration. Traditional control often assumes that the power of auxiliary equipment is quasi-static constant, but under extremely high task change rates, the dynamic response of the liquid cooling system will exert a secondary power surge on the DC bus.

[0091] The thermoelectric coupling prediction module breaks away from the global average temperature monitoring mode, treating the ten server racks inside the container as independent thermodynamic nodes. The module extracts the local task load rate of the central processing unit, graphics processing unit, and network switching chip within each rack, constructing a spatial heat source distribution matrix that includes spatial location and heat flux magnitude. This spatial heat source distribution matrix is ​​a ten-dimensional multi-scale dynamic tensor, reflecting the non-uniform distribution of the thermal field inside the computing vehicle.

[0092] For the heat flux hotspots presented in the spatial heat source distribution matrix, the thermoelectric coupling prediction module calculates the opening commands of the independent proportional control valves for each server cooling loop in real time. The rapid changes in the opening commands of these independent proportional control valves directly alter the hydrodynamic topology of the entire microchannel liquid cooling network.

[0093] The relationship between the overall hydraulic damping coefficient of the system and the opening command of the independent proportional control valve follows the laws of fluid continuity and conservation of momentum:

[0094]

[0095] In the formula, The overall hydraulic damping coefficient of the system is given. This represents the total number of server cooling loops connected in parallel. For loop index, This is the friction factor along the pipeline. For the first The equivalent length of the pipe for each loop. For the first The inner diameter of the pipe in each loop, Pi In order to be with the first The independent proportional control valve opening command A dynamic function of strongly correlated local resistance coefficients.

[0096] As the opening command of the independent proportional regulating valve rapidly adjusts to match the speed of the task hotspot, the circulating pump system faces a sudden increase in back pressure. To ensure that the coolant flow rate in the hotspot area meets the standard, the circulating pump motor must instantly output a large electromagnetic torque to overcome the sharply increased hydraulic damping. This transforms what was originally a single grid load step triggered solely by the computing equipment into a superimposed transient power gap between the computing equipment and the liquid cooling equipment.

[0097] The mathematical derivation of transient hydraulic load power is as follows:

[0098]

[0099] In the formula, The transient hydraulic load power, For coolant density, It is the acceleration due to gravity. To meet the total flow command required for heat dissipation of the current spatial heat source distribution matrix, The overall hydraulic damping coefficient of the system is given. For the static head of the pipeline network, For the hydraulic efficiency of the circulating pump, The electromechanical conversion efficiency of the pump drive motor.

[0100] The thermoelectric coupling prediction module treats the transient hydraulic load power as a high-frequency disturbance and feeds it strictly forward to the hybrid energy storage power distribution module. While coping with the surge in computing load, the supercapacitor energy storage module simultaneously releases the superimposed transient starting power of the circulating pump motor, completely eliminating the bus secondary voltage drop caused by the mechanical inertia of the hydraulic equipment and the frequency converter response delay.

[0101] The battery energy storage module and the supercapacitor energy storage module also generate considerable internal Joule heat under high-rate charge and discharge. The liquid cooling circuit of the battery energy storage module and the server cooling circuit are physically decoupled through the heat exchange isolation device, but tightly coupled in terms of energy management. The thermoelectric coupling prediction module collects the heat generation power of the battery energy storage module and the supercapacitor energy storage module in the energy storage compartment, links them with the spatial heat source distribution matrix of the computing compartment, and outputs a global heat dissipation power command that coordinates global thermal balance to guide the variable frequency operation of the external chiller compressor.

[0102] Table 1 is a comparison table of various performance indicators under transient load change conditions:

[0103] The computing power load jumps dramatically (from 40kW to 280kW, with a ramp rate of 150kW / s), while the wind power generation drops sharply (from 20kW to 5kW). Maximum voltage drop depth of DC bus 42.6V 14.2V Same as above Time required to recover to steady-state voltage 1.82s 0.35s Same as above Peak power lag time of circulating pump 2.15s -0.05s (Advance compensation takes effect) Same as above Battery energy storage module withstands maximum transient current surge 115A 28A Same as above The percentage of tasks slowed down due to thermal throttling of computing devices 14.5% 1.2%

[0104] In the test, an extremely harsh operating condition was simulated, with the total load of the container at a base value of 80kW and wind and solar power undergoing drastic fluctuations. At a certain moment, an emergency image rendering array task was received, and the power requirement of the computing equipment surged to 280kW within 1.5 seconds. In the basic feedforward scheme without applying the spatial thermoelectric synergy and virtual inertia mechanism of this embodiment, due to the coarse logic of the circulating pump adjusting according to the total return water temperature, the local cabinet temperature approached the thermal wall within 3 seconds, triggering the graphics processor's underlying self-frequency reduction protection, resulting in 14.5% of the tasks failing to run at full load. At the same time, the delayed start-up of the liquid cooling pump power and the computing power peak severely overlapped in time, causing the bus voltage to drop by 42.6V.

[0105] After activating the complete mechanism described in this embodiment, the virtual inertia and impedance reshaping module senses the first derivative of the task power and completes the reshaping of the target virtual inertia coefficient before the load surge. The supercapacitor energy storage module releases the energy corresponding to the advance compensation current reference value before the load fully bursts. The spatial heat source distribution matrix accurately locks the three core load racks at the same time as the graphics processor starts, and the independent proportional regulating valve opening command is fully opened in advance. The thermoelectric coupling prediction module pre-programs the predicted additional power surge of the 6kW hydraulic pump into the discharge command of the supercapacitor energy storage module.

[0106] Alignment of multiple physical quantities in both the time and spatial domains was achieved. The maximum DC bus voltage drop was reduced to 14.2V, far from the device reset threshold. The circulating pump achieved an advanced response to thermal flux surges through feedforward drive. The battery energy storage module was completely isolated from high-frequency surges by the flexible electromagnetic impedance barrier constructed by the supercapacitor energy storage module, reducing the maximum transient current it could withstand from 115A to 28A, significantly mitigating the irreversible loss of battery active materials. No thermal protection frequency limiting occurred throughout the task computation, ensuring reliable output of the mobile edge computing node under the dual pressure of harsh external natural environment and extreme task load.

[0107] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A mobile computing vehicle-integrated wind, solar, and energy storage containerized energy storage system, comprising a mobile carrier component, a container mounted on the mobile carrier component, computing equipment installed within the container, a battery energy storage module, a supercapacitor energy storage module, a DC bus, an external power grid interface, and an energy management main controller, as well as a wind power generation module and a photovoltaic power generation module connected to the DC bus; the battery energy storage module and the supercapacitor energy storage module are respectively connected to the DC bus via bidirectional DC-DC converters; characterized in that: The energy management main controller includes a computing load prediction module, a wind and solar power prediction module, a multi-timescale optimization scheduling module, a hybrid energy storage power allocation module, and a rolling feedback control module. The computing load prediction module outputs multiple time-scale computing device power prediction sequences based on computing task scheduling data and computing device operation data. The wind and solar power prediction module outputs wind power prediction sequences and photovoltaic power prediction sequences that are time-aligned with the computing device power prediction sequences based on meteorological data and historical power generation data. The multi-timescale optimization scheduling module determines the planned power of battery energy storage, the target state of charge of supercapacitors, the power of new energy utilization, and the planned power of the external power grid based on the power prediction sequence of the computing equipment, the power prediction sequence of wind power generation, and the power prediction sequence of photovoltaic power generation. The hybrid energy storage power allocation module determines the net power deviation and generates reference values ​​for battery energy storage power and supercapacitor energy storage power based on the total load power of the container, the actual power of wind power generation, the actual power of photovoltaic power generation, and the planned power of the external power grid. The rolling feedback control module corrects the reference values ​​for battery energy storage power, supercapacitor energy storage power, and the planned power of the external power grid based on the actual power of the computing equipment, the actual power of the auxiliary equipment, the DC bus voltage, the state of charge of the battery, and the state of charge of the supercapacitor.

2. The mobile computing vehicle-integrated wind, solar, and energy storage container energy storage system according to claim 1, characterized in that: the computing load prediction module constructs a temporal convolutional network, the input of the temporal convolutional network includes task identifier, task type, task stage, task resource quota, central processing unit utilization, graphics processing unit utilization, memory occupancy, network bandwidth occupancy, number of running servers, historical computing power, and time stamp; the temporal convolutional network uses dilated causal convolution and residual connections to form a multi-scale receptive field, and outputs the median predicted value of computing power, the lower bound of the prediction interval, and the upper bound of the prediction interval corresponding to different prediction durations through multiple output heads.

3. The mobile computing vehicle-integrated wind, solar and energy storage container energy storage system according to claim 1, characterized in that: the wind and solar power generation prediction module resamples wind speed, wind direction, irradiance, ambient temperature, air pressure, humidity, historical wind power generation and historical photovoltaic power generation according to a unified scheduling time index to form wind power generation input sequence and photovoltaic power generation input sequence respectively, and outputs the median predicted value of wind power generation, the predicted range of wind power generation, the median predicted value of photovoltaic power generation and the predicted range of photovoltaic power generation.

4. The mobile computing vehicle integrated wind, solar, and energy storage container energy storage system according to claim 1, characterized in that: the multi-time-scale optimization scheduling module takes the external power grid supply, the amount of new energy abandoned and the cost of battery energy storage degradation as optimization objectives, and takes the DC bus power balance, the battery state of charge range, the supercapacitor state of charge range, the battery energy storage module charging and discharging power range, the supercapacitor energy storage module charging and discharging power range, the bidirectional DC converter power range and the external power grid interface power range as constraints to obtain multiple candidate scheduling schemes, and selects the execution scheduling scheme from the multiple candidate scheduling schemes according to the target priority corresponding to the operation mode.

5. The mobile computing vehicle integrated wind, solar, and energy storage containerized energy storage system according to claim 1, characterized in that: the battery energy storage module is used to bear the low-frequency power component and continuous energy gap in the net power deviation, and the supercapacitor energy storage module is used to bear the high-frequency power component and instantaneous power gap in the net power deviation; the hybrid energy storage power distribution module performs power limiting, energy limiting, and power change rate limiting on the battery energy storage power reference value and the supercapacitor energy storage power reference value, and transfers the remaining power exceeding the allowable range of the battery energy storage module or the allowable range of the supercapacitor energy storage module to another energy storage module or the external power grid interface.

6. The mobile computing vehicle wind-solar-storage integrated container energy storage system according to claim 5, characterized in that: the hybrid energy storage power allocation module uses the net power deviation obtained by subtracting the actual wind power generation power, the actual photovoltaic power generation power and the planned power of the external power grid from the total load power of the container as the net power deviation, performs low-pass filtering on the net power deviation to obtain the initial power reference value of battery energy storage, uses the difference between the net power deviation and the initial power reference value of battery energy storage as the initial power reference value of supercapacitor energy storage, and corrects the initial power reference value of battery energy storage and the initial power reference value of supercapacitor energy storage according to the allowable charge and discharge power and allowable state of charge of the battery energy storage module and the supercapacitor energy storage module.

7. The mobile computing vehicle integrated wind, solar and energy storage container energy storage system according to claim 2 or 3, characterized in that: the multi-timescale optimization scheduling module determines the net power uncertainty in the power shortage direction and the net power uncertainty in the power surplus direction based on the median predicted power value of the computing equipment, the lower bound of the power prediction interval of the computing equipment, the upper bound of the power prediction interval of the computing equipment, the power prediction interval of auxiliary equipment, the median predicted power value of the wind power generation, the wind power prediction interval, the median predicted power value of the photovoltaic power generation and the photovoltaic power prediction interval, and reserves discharge reserve power, discharge reserve energy, charging reserve power and charging reserve energy for the battery energy storage module and the supercapacitor energy storage module respectively.

8. The mobile computing vehicle wind-solar-storage integrated container energy storage system according to claim 6, characterized in that: the hybrid energy storage power allocation module selects the cutoff frequency corresponding to the current rolling control window from multiple candidate cutoff frequencies based on the allowable power change rate of the battery energy storage module, the remaining dischargeable energy of the supercapacitor energy storage module, the remaining rechargeable energy of the supercapacitor energy storage module, the allowable power of the bidirectional DC converter, and the predicted net power deviation spectrum; after the high-frequency power component in the net power deviation is compensated, energy recovery power is generated based on the deviation between the actual state of charge of the supercapacitor energy storage module and the target state of charge of the supercapacitor, and the supercapacitor energy storage module is restored to the range of the target state of charge of the supercapacitor under the conditions of satisfying the battery power change rate constraint and the DC bus power balance constraint.

9. The mobile computing vehicle integrated wind, solar, and energy storage container energy storage system according to claim 1, characterized in that: the energy management main controller further includes a computing task coordination module, which divides computing tasks into non-adjustable tasks, deferred tasks, and downgradeable tasks according to task deadline, allowed interruption flag, checkpoint recovery time, minimum resource quota, task priority, and remaining computing power, determines the lower limit and upper limit of computing task power in each scheduling cycle, and outputs a task resource adjustment instruction between the lower limit and upper limit of computing task power to the computing device when the energy storage backup capacity and the available power of the external power grid cannot cover the net power gap.

10. The mobile computing vehicle integrated wind, solar and energy storage container energy storage system according to claim 9, characterized in that: the computing task coordination module generates an instantaneous power feedforward based on the task start time, the number of graphics processors to be activated, the task batch processing scale, the task calculation stage and the historical power template of similar tasks; the hybrid energy storage power allocation module outputs an early discharge command corresponding to the instantaneous power feedforward to the supercapacitor energy storage module before the task starts, and corrects the supercapacitor energy storage power reference value according to the DC bus voltage deviation and the actual power deviation of the computing equipment after the task starts, and then transfers the continuous power component to the battery energy storage module according to the preset power change rate.

Images