A data center direct current power source and load collaborative control method, device and medium

Abstract

This invention discloses a method, device, and medium for coordinated control of DC power supply, load, and energy storage in a data center, belonging to the field of DC power supply control technology. The method includes: extracting the state of charge (SOC) value at the moment of an event, the time integral of the absolute value of charging and discharging power within a preset time period before the event, and the time constant for the discharge power to rise from its initial value to its peak value after the event, establishing a three-dimensional mapping relationship; constructing an instantaneous response capability evaluation function for the energy storage system based on the three-dimensional mapping relationship, and predicting the time constant; substituting the predicted time constant, the equivalent capacitance of the bus, and the preset maximum load step amplitude into the voltage dynamic response equation to obtain a predicted voltage drop depth; when the predicted voltage drop depth exceeds the voltage collapse threshold, calculating a dynamic adjustment margin value, and invoking distributed power sources and load regulation to move the SOC value to a safe region. This invention achieves dynamic quantitative evaluation of the instantaneous response capability of the energy storage system, reducing the risk of bus voltage collapse.

Description

Technical Field

[0001] This invention relates to the field of DC power supply control technology, specifically to a method, equipment, and medium for coordinated control of DC power supply load and storage in a data center. Background Technology

[0002] As a high-energy-consuming infrastructure, the reliability of power supply is crucial for data centers. Traditional data centers use AC power supply architectures, which involve multiple energy conversion stages and significant losses. DC power supply architectures combined with energy storage systems can significantly improve efficiency, but they face the risk of voltage drops caused by load surges. Dynamic migration of virtual machines leads to sudden load changes, and the response delay of energy storage systems may cause voltage collapse, affecting business continuity. Existing technologies mainly rely on over-configuring energy storage capacity or setting conservative safety margins, resulting in low utilization of energy storage resources and inflexible adjustment strategies. In addition, energy storage characteristics change dynamically with state of charge, historical operating trajectory, and aging degree, and fixed models cannot accurately predict response capabilities. Therefore, there is an urgent need for a method for coordinated control of DC power supply load and energy storage in data centers. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention provides a method, device and medium for coordinated control of DC power supply load and storage in data centers.

[0004] Therefore, the technical problem solved by this invention is that existing data center DC power supply systems have the following problems: the instantaneous response capability of the energy storage system changes dynamically with the operating state, and the state of charge threshold alone cannot accurately reflect the system's true ability to resist voltage collapse; there is a lack of quantitative description of the relationship between the energy storage system's response time constant and its operating history and current state; the potential risks of insufficient energy storage system response capability cannot be identified in advance; and there is a lack of a source-load-storage coordinated regulation mechanism based on dynamic response capability, resulting in untimely or overly conservative voltage collapse prevention measures.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a data center DC power supply load-storage coordinated control method, which includes: acquiring DC bus voltage sequence, energy storage system charge-discharge power sequence and state of charge sequence; identifying voltage step drop events in the voltage sequence; extracting the state of charge value at the time of the event occurrence, the time integral value of the absolute value of charge-discharge power within a preset period before the event occurrence, and the time constant of the discharge power rising from the initial value to the peak value after the event occurrence for each voltage step drop event; and establishing a three-dimensional mapping relationship between the state of charge value at the time of the event occurrence, the time integral value of the absolute value of charge-discharge power within a preset period before the event occurrence, and the time constant of the discharge power rising from the initial value to the peak value after the event occurrence.

[0006] Based on the three-dimensional mapping relationship, an instantaneous response capability evaluation function for the energy storage system is constructed. The current state of charge value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset period before the current time are input into the instantaneous response capability evaluation function to obtain the predicted value of the time constant.

[0007] Substitute the predicted value of the time constant, the equivalent capacitance of the bus, and the preset maximum load step amplitude into the voltage dynamic response equation to obtain the estimated value of the voltage drop depth.

[0008] When the estimated voltage drop depth exceeds the safe threshold for voltage drop depth, the voltage dynamic response equation is solved in reverse to obtain the target value of the time constant. Based on the three-dimensional mapping relationship, the state of charge value required to reach the target value of the time constant is looked up in reverse. The difference between the state of charge value obtained in reverse and the current state of charge value is used as the dynamic adjustment margin value.

[0009] When the distance between the current state of charge (SOC) value of the energy storage system and the SOC boundary is less than the dynamic adjustment margin value, the distributed power source and load regulation are invoked to move the SOC value to the region that meets the requirements of the dynamic adjustment margin value.

[0010] During the adjustment process, the actual value of the time constant is monitored, the deviation between the actual value and the predicted value of the time constant is calculated, and the three-dimensional mapping relationship is updated based on the deviation using the recursive least squares method.

[0011] As a preferred embodiment of the data center DC power supply load-storage coordinated control method of the present invention, the identification of voltage step drop events in the voltage sequence includes performing a first-order time difference operation on the DC bus voltage sequence to obtain a voltage change rate sequence.

[0012] Identify the moment when the absolute value of the voltage change rate in the voltage change rate sequence exceeds a preset change rate threshold, determine that the voltage change direction at that moment is negative, and take that moment as a candidate voltage drop moment;

[0013] Extract the power reference value of the energy storage system's charging and discharging power sequence before the candidate voltage drop moment, extract the power peak value of the energy storage system's charging and discharging power sequence within a preset verification time window after the candidate voltage drop moment, and determine whether the power peak value shows a step increase relative to the power reference value. If a step increase occurs, the candidate voltage drop moment is confirmed as the voltage step drop event moment.

[0014] Based on the timestamp of the voltage step drop event, the state of charge value corresponding to the timestamp is found in the state of charge sequence as the state of charge value at the time of the voltage step drop event.

[0015] As a preferred embodiment of the data center DC power supply load-storage coordinated control method of the present invention, the establishment of a three-dimensional mapping relationship between the state of charge value at the time of the event, the time integral value of the absolute value of the charging and discharging power within a preset period before the event, and the time constant of the discharge power rising from the initial value to the peak value after the event includes establishing a three-dimensional spatial coordinate system with the state of charge value at the time of the event as the first dimension coordinate, the time integral value of the absolute value of the charging and discharging power within a preset period before the event as the second dimension coordinate, and the time constant of the discharge power rising from the initial value to the peak value after the event as the third dimension coordinate.

[0016] The state of charge value at the time of occurrence of multiple voltage step drop events, the time integral value of the absolute value of charging and discharging power within a preset period before the event occurrence, and the time constant of the discharge power rising from the initial value to the peak value after the event occurrence are used as the coordinate values ​​of sample points in the three-dimensional spatial coordinate system.

[0017] Calculate the Euclidean distance between any two sample points in the three-dimensional spatial coordinate system, identify sample point pairs whose Euclidean distance is less than a preset distance threshold, remove one sample point from the sample point pair, and obtain the remaining sample points;

[0018] Calculate the standard deviation of the remaining sample points in the first-dimensional coordinate direction, the standard deviation in the second-dimensional coordinate direction, and the standard deviation in the third-dimensional coordinate direction. Calculate the ratio of the maximum standard deviation to the minimum standard deviation among the three standard deviations, and use the ratio as the kernel width parameter of the radial basis function.

[0019] The remaining sample points are fitted with radial basis function surfaces using kernel width parameters to obtain an explicit binary function of the time constant for the discharge power to rise from the initial value to the peak value after the event. The independent variables of the explicit binary function are the state of charge value at the time of the event and the time integral value of the absolute value of the charge and discharge power within a preset period before the event. The explicit binary function is used as a three-dimensional mapping relationship.

[0020] As a preferred embodiment of the data center DC power supply load-storage coordinated control method described in this invention, the step of constructing the instantaneous response capability evaluation function of the energy storage system based on the three-dimensional mapping relationship includes: statistically analyzing the distribution of the state of charge value at the time of occurrence of multiple voltage step drop events used in establishing the three-dimensional mapping relationship and the time integral value of the absolute value of charging and discharging power within a preset period before the time of occurrence of the event in the three-dimensional space; and calculating the number density value of voltage step drop events in each local area in the three-dimensional space.

[0021] Based on the number density value, the three-dimensional space is divided into high-density regions and low-density regions. High-density regions are those with a number density value greater than a preset density threshold, while low-density regions are those with a number density value less than or equal to the preset density threshold.

[0022] For high-density areas, a first grid step size is used for spatial grid division, and for low-density areas, a second grid step size is used for spatial grid division. The first grid step size is smaller than the second grid step size.

[0023] For each spatial grid, the first and second dimension coordinates of the center point of the spatial grid are substituted into the three-dimensional mapping relationship to calculate the pre-stored value of the time constant corresponding to the spatial grid.

[0024] When receiving the current state of charge (SOC) value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset time period prior to the current time, the spatial grid to which the SOC value and the time integral value belong are determined, the time constant pre-stored value corresponding to the spatial grid is obtained, and the time constant pre-stored value is output as the predicted value of the time constant.

[0025] In a preferred embodiment of the data center DC power supply load-storage coordinated control method described in this invention, the voltage dynamic response equation is expressed as follows:

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] in, This represents the equivalent capacitance of the busbar. Indicates the DC bus voltage. This indicates the time from when the energy storage system begins to respond. This represents the first derivative of the DC bus voltage with respect to time. Indicates the current of the energy storage system. Indicates the load current. Indicates the peak current of the energy storage system. Represents the dynamic response factor. This represents the reduction coefficient. The predicted value representing the time constant. This indicates the rated voltage of the DC bus. Indicates the equivalent resistance of the load. Represents the natural exponential function;

[0032] The peak current of the energy storage system is determined based on the preset maximum load step amplitude and the rated DC bus voltage. The equivalent resistance of the load is determined based on the load power and the rated DC bus voltage. The equivalent capacitance of the bus, the predicted value of the time constant, the peak current of the energy storage system, the equivalent resistance of the load, and the rated DC bus voltage are substituted into the voltage dynamic response equation. The rated DC bus voltage is used as the initial voltage value. The fourth-order Runge-Kutta method is used to solve the voltage dynamic response equation to obtain the numerical solution sequence of the DC bus voltage.

[0033] The minimum DC bus voltage is identified by traversing the DC bus voltage numerical solution sequence. The difference between the rated DC bus voltage and the minimum DC bus voltage is calculated as the estimated voltage drop depth.

[0034] As a preferred embodiment of the DC power supply load-storage coordinated control method for data centers described in this invention, the following steps are taken: The target value of the time constant is obtained by inversely solving the voltage dynamic response equation; the state of charge value required to reach the target value of the time constant is retrieved based on the three-dimensional mapping relationship; the difference between the retrieved state of charge value and the current state of charge value is used as a dynamic adjustment margin value; the target value of the time constant is obtained by inversely solving the voltage dynamic response equation using a bisection method; the time constant search interval is initialized to a preset minimum time constant to a preset maximum time constant using the bisection method; in each iteration, the midpoint value of the search interval is substituted into the voltage dynamic response equation to obtain the voltage drop depth calculation value; the search interval is updated according to the relationship between the voltage drop depth calculation value and the voltage drop depth safety threshold; the iteration continues until the absolute value of the difference between the voltage drop depth calculation value and the voltage drop depth safety threshold is less than a preset error threshold.

[0035] Based on the three-dimensional mapping relationship, the contour lines of the target value whose time constant is equal to the time constant are identified in the two-dimensional space formed by the time integral of the absolute value of charging and discharging power during a preset period before the current time of the energy storage system.

[0036] The target state point is the point on the contour line corresponding to the minimum Euclidean distance from the current state point, which is calculated by the time integral of the current state value of the energy storage system and the absolute value of the charging and discharging power during a preset period before the current state.

[0037] Extract the coordinates of the state of charge (SOC) value of the target state point as the target SOC value, and calculate the difference between the target SOC value and the current SOC value of the energy storage system as the dynamic adjustment margin value.

[0038] As a preferred embodiment of the data center DC power supply load-storage coordinated control method described in this invention, the step of calling distributed power supply and load regulation to move the state of charge value to a region that meets the requirements of the dynamic regulation margin when the distance between the current state of charge value of the energy storage system and the state of charge boundary is less than the dynamic adjustment margin value includes calculating the state of charge value adjustment amount required for the current state of charge value of the energy storage system to move to a position that is equal to the dynamic adjustment margin value from the state of charge boundary.

[0039] Calculate the charging and discharging power adjustment of the energy storage system based on the state of charge adjustment amount and the energy storage system capacity.

[0040] Determine the sign of the energy storage system's charge and discharge power adjustment amount. When the energy storage system's charge and discharge power adjustment amount is positive, it is determined that the power flowing into the energy storage system needs to be increased. When the energy storage system's charge and discharge power adjustment amount is negative, it is determined that the power flowing into the energy storage system needs to be reduced.

[0041] When it is necessary to increase the power flowing into the energy storage system, the distributed power source is called to increase the output power, and the load regulation is called to reduce the data center load power.

[0042] When it is necessary to reduce the power flowing into the energy storage system, distributed power sources are called to reduce output power, and load regulation is called to increase the load power of the data center.

[0043] As a preferred embodiment of the data center DC power supply load-storage coordinated control method described in this invention, the following steps are included: during the adjustment process, monitoring the actual value of the time constant, calculating the deviation between the actual value and the predicted value of the time constant, and updating the three-dimensional mapping relationship based on the deviation using the recursive least squares method, including detecting the charging and discharging power response process of the energy storage system triggered by the call to distributed power supply and load adjustment, and extracting the measured value of the time constant reflecting the response speed of the energy storage system from the charging and discharging power response process as the actual value of the time constant;

[0044] The state of charge (SOC) value at the corresponding moment of the energy storage system response process and the time integral value of the absolute value of the charging and discharging power within a preset period before the corresponding moment are obtained. The SOC value and the time integral value are input into a three-dimensional mapping relationship to calculate the predicted value of the time constant.

[0045] The difference between the actual value and the predicted value of the time constant is calculated as the prediction bias;

[0046] The cumulative number of times the energy storage system's charging and discharging power response process is triggered within a preset statistical time window is counted. The operating intensity of the energy storage system is determined based on the cumulative number of times. The forgetting factor of the recursive least squares method is adjusted based on the operating intensity. The higher the operating intensity, the smaller the forgetting factor.

[0047] The radial basis function value vector is calculated by inputting the state of charge value and time integral value into the radial basis function in the three-dimensional mapping relationship. The weight coefficient vector of the radial basis function in the three-dimensional mapping relationship is updated by the recursive least squares method based on the radial basis function value vector, prediction bias and forgetting factor, so that the three-dimensional mapping relationship can adapt to the changes in the characteristics of the energy storage system.

[0048] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the data center DC power supply load-storage coordinated control method.

[0049] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the aforementioned data center DC power supply load-storage coordinated control method.

[0050] The beneficial effects of this invention are as follows: This invention establishes a three-dimensional mapping relationship between the response time constant of an energy storage system and its state of charge and historical charge-discharge integrals, thereby achieving a dynamic quantitative assessment of the instantaneous response capability of the energy storage system. Based on this mapping relationship, an evaluation function is constructed to predict the response speed of the energy storage system under the current state, and the voltage drop depth during load surges is estimated through the voltage dynamic response equation. This overcomes the shortcomings of existing technologies that rely solely on static state of charge thresholds and cannot accurately reflect the dynamic response characteristics of the system.

[0051] By introducing the concept of dynamic adjustment margin, the operating boundary of the energy storage system is transformed from a fixed state-of-charge threshold to a dynamic constraint based on actual response capability, solving the problems of delayed or overly conservative adjustment timing in traditional methods. When the energy storage system's response capability is detected to be insufficient to withstand a preset load step, distributed power sources or loads are invoked in advance for coordinated adjustment, shifting the energy storage system's state of charge to a safe area that meets the dynamic response requirements, thereby reducing the risk of bus voltage collapse.

[0052] By employing a recursive least squares method to continuously correct the three-dimensional mapping relationship based on the actual monitored response time constant, the evaluation model tracks the characteristic drift of the energy storage system caused by aging and temperature changes, maintaining prediction accuracy. This invention improves the ability of data center DC power supply systems to withstand load surges and reduces the probability of voltage drop accidents. Attached Figure Description

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a flowchart illustrating a data center DC power supply load-storage coordinated control method according to an embodiment of the present invention. Detailed Implementation

[0055] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0056] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a data center DC power supply load-storage coordinated control method, including:

[0057] Step 1: Collect the DC bus voltage sequence, the energy storage system charge / discharge power sequence, and the state of charge sequence. Identify voltage step drop events in the voltage sequence. For each voltage step drop event, extract the state of charge value at the time of the event, the time integral value of the absolute value of the charge / discharge power within a preset period before the event, and the time constant for the discharge power to rise from the initial value to the peak value after the event. Establish a three-dimensional mapping relationship between the state of charge value at the time of the event, the time integral value of the absolute value of the charge / discharge power within a preset period before the event, and the time constant for the discharge power to rise from the initial value to the peak value after the event.

[0058] Step 2: Construct an instantaneous response capability evaluation function for the energy storage system based on the three-dimensional mapping relationship. Input the current state of charge value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset period before the current time into the instantaneous response capability evaluation function to obtain the predicted value of the time constant.

[0059] Step 3: Substitute the predicted value of the time constant, the equivalent capacitance of the bus, and the preset maximum load step amplitude into the voltage dynamic response equation to obtain the estimated value of the voltage drop depth.

[0060] Step 4: When the estimated voltage drop depth exceeds the safe threshold for voltage drop depth, the voltage dynamic response equation is solved in reverse to obtain the target value of the time constant. Based on the three-dimensional mapping relationship, the state of charge value required to reach the target value of the time constant is looked up. The difference between the state of charge value obtained in reverse and the current state of charge value is used as the dynamic adjustment margin value.

[0061] Step 5: When the distance between the current state of charge (SOC) value of the energy storage system and the SOC boundary is less than the dynamic adjustment margin value, the distributed power source and load regulation are invoked to move the SOC value to the region that meets the requirements of the dynamic adjustment margin value.

[0062] Step 6: During the adjustment process, monitor the actual value of the time constant, calculate the deviation between the actual value and the predicted value of the time constant, and update the three-dimensional mapping relationship based on the deviation using the recursive least squares method.

[0063] In step 1, identifying voltage step drop events in the voltage sequence includes performing a first-order time difference operation on the DC bus voltage sequence to obtain a voltage change rate sequence.

[0064] Identify the moment when the absolute value of the voltage change rate in the voltage change rate sequence exceeds a preset change rate threshold, determine that the voltage change direction at that moment is negative, and take that moment as a candidate voltage drop moment;

[0065] Extract the power reference value of the energy storage system's charging and discharging power sequence before the candidate voltage drop moment, extract the power peak value of the energy storage system's charging and discharging power sequence within a preset verification time window after the candidate voltage drop moment, and determine whether the power peak value shows a step increase relative to the power reference value. If a step increase occurs, the candidate voltage drop moment is confirmed as the voltage step drop event moment.

[0066] Based on the timestamp of the voltage step drop event, the state of charge value corresponding to the timestamp is found in the state of charge sequence as the state of charge value at the time of the voltage step drop event.

[0067] It should be noted that the preset rate of change threshold is determined based on the voltage fluctuation characteristics of the data center's DC bus during normal operation. Specifically, the DC bus voltage sequence under normal data center operation is collected, and a first-order time-difference operation is performed on the voltage sequence to obtain the normal operation voltage rate of change sequence. The standard deviation of the normal operation voltage rate of change sequence is calculated, and the preset rate of change threshold is set to 3 to 10 times the standard deviation. In this embodiment, 5 times can be used based on experience. This threshold setting method is based on the statistical 3σ principle and can effectively distinguish between normal voltage fluctuations and abnormal voltage drops caused by virtual machine migration.

[0068] For example, in a test at a data center, the standard deviation of the normal operating voltage change rate was calculated to be 0.8V / ms, and the preset change rate threshold was set to 4.0V / ms (5 times the standard deviation). The preset change rate threshold can effectively detect virtual machine migration events.

[0069] The step of determining whether the peak power value has a step increase relative to the power reference value includes: calculating the power change amplitude, wherein the power change amplitude is equal to the difference between the peak power value and the power reference value;

[0070] Calculate the power change rate, which is equal to the power change amplitude divided by the time interval from the moment the candidate voltage drops to the moment the power peak occurs;

[0071] Determine whether the power change amplitude exceeds a first threshold and whether the power change rate exceeds a second threshold. If both conditions are met, it is determined to be a step increase.

[0072] The first threshold ranges from 10% to 30% of the rated power of the energy storage system; in this embodiment, 20% can be used based on experience. The second threshold ranges from 50% to 200% of the rated power of the energy storage system per second; in this embodiment, 100% per second can be used based on experience. This dual-condition judgment can effectively distinguish between a step response and a gradual response with slowly increasing load.

[0073] For example, for an energy storage system with a rated power of 500kW, the first threshold is set to 100kW and the second threshold is set to 500kW / s. In a certain event, the power increases from 50kW to 180kW in a time interval of 150ms. The calculated power change amplitude of 130kW is greater than 100kW, and the power change rate of 867kW / s is greater than 500kW / s, which is judged as a step increase.

[0074] Furthermore, in step 1, establishing a three-dimensional mapping relationship between the state of charge value at the moment of the event, the time integral value of the absolute value of the charging and discharging power during the preset period before the moment of the event, and the time constant of the discharge power rising from the initial value to the peak value after the event includes establishing a three-dimensional spatial coordinate system with the state of charge value at the moment of the event as the first coordinate, the time integral value of the absolute value of the charging and discharging power during the preset period before the moment of the event as the second coordinate, and the time constant of the discharge power rising from the initial value to the peak value after the event as the third coordinate.

[0075] The state of charge value at the time of occurrence of multiple voltage step drop events, the time integral value of the absolute value of charging and discharging power within a preset period before the event occurrence, and the time constant of the discharge power rising from the initial value to the peak value after the event occurrence are used as the coordinate values ​​of sample points in the three-dimensional spatial coordinate system.

[0076] Calculate the Euclidean distance between any two sample points in the three-dimensional spatial coordinate system, identify sample point pairs whose Euclidean distance is less than a preset distance threshold, remove one sample point from the sample point pair, and obtain the remaining sample points;

[0077] Calculate the standard deviation of the remaining sample points in the first-dimensional coordinate direction, the standard deviation in the second-dimensional coordinate direction, and the standard deviation in the third-dimensional coordinate direction. Calculate the ratio of the maximum standard deviation to the minimum standard deviation among the three standard deviations, and use the ratio as the kernel width parameter of the radial basis function.

[0078] A radial basis function surface fitting was performed on the remaining sample points using the kernel width parameter to obtain an explicit binary function representing the time constant for the discharge power to rise from its initial value to its peak value after the event. The independent variables of the explicit binary function are the state of charge value at the moment of the event and the time integral value of the absolute value of the charge and discharge power within a preset period before the moment of the event. The explicit binary function is used as a three-dimensional mapping relationship. It should be noted that the preset period before the moment of the event is based on the historical event time, and the preset period before the current moment of the energy storage system is based on the current moment. The time lengths are the same, but the time reference points are different.

[0079] It should be noted that the preset time period before the event occurs is determined based on the thermal time constant of the individual battery cells in the energy storage system, reflecting the impact of the temperature accumulation effect caused by historical charging and discharging on the instantaneous response capability. The preset time period ranges from 0.5 to 2 times the thermal time constant of the individual battery cells in the energy storage system; in this embodiment, 1 times is used based on experience. The thermal time constant of the individual battery cells in the energy storage system is obtained through thermal characteristic testing, with typical values ​​ranging from 300 seconds to 1800 seconds. For example, if a data center energy storage system uses lithium iron phosphate batteries with a thermal time constant of 600 seconds, then the preset time period is set to 600 seconds.

[0080] When calculating the time integral of the absolute value of the charging and discharging power within a preset period before the event occurs, the trapezoidal integral method is used:

[0081] ;

[0082] in, The time integral value of the absolute value of the charging and discharging power within a preset time period before the event occurs; Indicates the serial number of the historical charge / discharge power sampling point; Indicates the number of sampling points within the preset time period; Indicates the first The charging and discharging power of the energy storage system at each sampling point; Indicates the first The charging and discharging power of the energy storage system at each sampling point; Indicates the sampling time interval.

[0083] For example, a data center energy storage system has a sampling interval of 1 second and a preset time period of 600 seconds. The average absolute value of the energy storage system's charging and discharging power in the 600 seconds prior to the event is 80kW. Therefore, the time integral value of the absolute value of the charging and discharging power in the preset time period prior to the event is approximately... .

[0084] The preset distance threshold is adaptively determined based on the spatial distribution of sample points. The Euclidean distances between all pairs of sample points in the three-dimensional coordinate system are calculated to obtain a distance set. This distance set is then sorted in ascending order, and the 5th percentile is calculated. and the 95th percentile Set the preset distance threshold as follows:

[0085] ;

[0086] in, The adjustment coefficient ranges from 0.05 to 0.2. In this embodiment, based on experience, 0.1 can be used. Indicates the 5th percentile; This represents the 95th percentile.

[0087] This method is based on the physical nature of the response characteristics of data center energy storage systems: when an energy storage system responds multiple times under similar operating conditions, its time constant should remain stable. Therefore, sample points that are spatially close actually reflect the same physical state. The method automatically identifies the degree of clustering of data center operating conditions using percentile methods, avoiding excessive retention of redundant data or accidental deletion of valid data due to manually set fixed thresholds.

[0088] When calculating the Euclidean distance between any two sample points, since the three physical quantities—the state of charge value at the moment of the event, the time integral of the absolute value of the charging and discharging power during the preset period before the event, and the time constant for the discharge power to rise from its initial value to its peak value after the event—have different dimensions, it is necessary to first normalize the coordinates of each dimension.

[0089] ;

[0090] in, Represents the original coordinate values; This represents the normalized coordinate values; This represents the minimum value of the coordinate in that dimension; This represents the maximum value of the coordinate in that dimension.

[0091] The formula for calculating the normalized Euclidean distance is:

[0092] ;

[0093] in, Represents sample points and sample points The Euclidean distance between them; Represents sample points Normalized state of charge value at the moment the event occurs; Represents sample points The time integral of the absolute value of charging and discharging power within a preset time period before the event occurs, after normalization; Represents sample points The normalized time constant for the discharge power to rise from its initial value to its peak value after the event occurs; Represents sample points Normalized state of charge value at the moment the event occurs; Represents sample points The time integral of the absolute value of charging and discharging power within a preset time period before the event occurs, after normalization; Represents sample points The normalized time constant for the discharge power to rise from its initial value to its peak value after the event occurs.

[0094] For example, a data center collects 50 sample points of voltage step drop events and calculates the distance set. , ,set up The preset distance threshold is 1.17. After removing redundant samples using this threshold, 42 sample points are retained.

[0095] Identify sample point pairs whose Euclidean distance is less than a preset distance threshold, and remove one sample point from each pair. Calculate the absolute value of the numerical difference between the two sample points in the third-dimensional coordinate direction. Compare the absolute value of the difference with a preset difference threshold. The size relationship. If If two sample points represent the same system state, retain the sample point with the smaller time constant for the discharge power to rise from the initial value to the peak value after the event, and discard the sample point with the larger time constant; if To determine whether two sample points represent different system states, neither sample point is discarded. This represents the absolute value of the difference in the time constant between two sample points when the discharge power rises from its initial value to its peak value after the event occurs. This indicates the preset difference threshold.

[0096] The physical basis for retaining sample points with smaller time constants is that, under the same state of charge and historical load conditions, a fast response speed (small time constant) in a data center energy storage system represents the optimal operating state, such as good consistency of individual battery cells, low internal resistance, and suitable temperature. Conversely, a slow response speed may be affected by measurement noise or transient disturbances. Retaining sample points of the optimal state can improve the accuracy of the three-dimensional mapping relationship in predicting the true capability of the energy storage system.

[0097] The preset difference threshold is determined based on the measurement accuracy of the time constant, and the value ranges from 5% to 15% of the median of the time constant. In this embodiment, 10% can be used based on experience.

[0098] For example, the median time constant of a data center energy storage system is 20ms, and a preset difference threshold of 2ms is set. Two sample points are detected where the state of charge (SBC) at the time of the event is 60%, and the time integral of the absolute value of the charging and discharging power within a preset time period before the event is 15kWh. However, the time constants for the discharge power to rise from its initial value to its peak value after the event are 18ms and 19.5ms respectively. The difference of 1.5ms is less than 2ms, so the sample point corresponding to 18ms is retained.

[0099] Calculate the standard deviation of the remaining sample points in the first dimension. Standard deviation in the second-dimensional coordinate direction Standard deviation in the third-dimensional coordinate direction :

[0100] ;

[0101] in, Indicates the first Standard deviation of the 3D coordinate direction; Indicates the coordinate dimension index. , This indicates the traversal sequence number of the remaining sample points in the standard deviation calculation; Indicates the number of remaining sample points; Indicates the first The remaining sample points at the th th Normalized values ​​of 3D coordinates; This indicates that all remaining sample points are at the th... The average value of the normalized values ​​of the dimensional coordinates.

[0102] Calculate the ratio of the largest standard deviation to the smallest standard deviation among the three standard deviations:

[0103] ;

[0104] in, Indicates the ratio of standard deviations; This represents the function that takes the maximum value. This represents the function that takes the minimum value.

[0105] Apply boundary constraints to the standard deviation ratio: when When, the kernel width parameter of the radial basis function ;when When, the kernel width parameter of the radial basis function ;when When, the kernel width parameter of the radial basis function ,in, This represents a preset lower limit value, ranging from 1.5 to 3.0. In this embodiment, based on experience, 2.0 can be used. This represents a preset upper limit value, ranging from 20 to 50. In this embodiment, based on experience, 30 can be used. The kernel width parameter represents the radial basis function.

[0106] This boundary treatment is based on the physical constraints of the dynamic response of the data center energy storage system: when the three-dimensional data is uniformly distributed (the ratio is close to 1), it indicates that the state of charge, historical load, and response speed are independent of each other. In this case, increasing the kernel width can improve the interpolation smoothness. When the data in a certain dimension is extremely dispersed (the ratio is too large), it indicates that there is strong nonlinearity or abnormal operating conditions in that dimension. Limiting the upper limit of the kernel width can prevent the fitting from diverging.

[0107] Constructing the radial basis function matrix , indicating that the dimension is The radial basis function matrix, The radial basis function matrix elements represent the number of remaining sample points:

[0108] ;

[0109] in, Represents the first element in the radial basis function matrix. The remaining sample points and the first Radial basis function values ​​between the remaining sample points; This indicates the index of the remaining sample point corresponding to the row of the matrix; This indicates the index of the remaining sample point corresponding to the column of the matrix; Indicates the first The normalized state of charge values ​​of the remaining sample points at the time of the event occurrence; Indicates the first The time integral of the absolute value of charging and discharging power within a preset time period before the event occurrence time, after normalization of the remaining sample points; Indicates the first The normalized state of charge values ​​of the remaining sample points at the time of the event occurrence; Indicates the first The time integral of the absolute value of charging and discharging power within a preset time period before the event occurrence time, after normalization of the remaining sample points; The kernel width parameter represents the radial basis functions; This represents the natural exponential function.

[0110] Solve for the weight coefficient vector:

[0111] ;

[0112] in, Represents the weight coefficient vector; This represents the transpose of the radial basis function matrix; Represents the radial basis function matrix; Represents the regularization coefficient, with a value range of 1. to In this embodiment, based on experience, the following can be adopted: ; The dimension is The identity matrix; This represents the time constant vector that indicates the rise of discharge power from its initial value to its peak value after the event occurs, after normalization. This indicates the number of remaining sample points.

[0113] The regularization coefficient is determined based on the number of remaining sample points: when the number of remaining sample points is less than 20, it is taken as follows: When the number of remaining sample points is between 20 and 50, take When the number of remaining sample points is greater than 50, take This setting is based on a trade-off between the sparsity of data samples in the data center and the ill-conditioned nature of the matrix: when there are few samples, the regularization is increased to prevent overfitting, and when there are many samples, the regularization is decreased to improve fitting accuracy.

[0114] Constructing explicit binary functions:

[0115] ;

[0116] ;

[0117] ;

[0118] in, Represents an explicit bivariate function; Indicates the index of the remaining sample points in the summation of an explicit binary function; This represents the normalized state of charge value at the moment the event occurs. This represents the time integral of the absolute value of the charging and discharging power within a preset time period prior to the normalized event occurrence time. The first element of the weight coefficient vector represents the first element of the weight coefficient vector. One element; Indicates the number of remaining sample points; Indicates the first The normalized state of charge values ​​of the remaining sample points at the time of the event occurrence; Indicates the first The time integral of the absolute value of charging and discharging power within a preset time period before the event occurrence time, after normalization of the remaining sample points; The kernel width parameter represents the radial basis functions; This represents the maximum value of the time constant for the discharge power to rise from its initial value to its peak value after the event occurs across all remaining sample points. This represents the minimum time constant for the discharge power to rise from its initial value to its peak value after the event occurs among all remaining sample points; Represents the natural exponential function; RMSE represents the root mean square error. Indicates the sample point number in the validation set; Indicates the number of sample points in the validation set; Indicates the first The predicted time constant for the discharge power to rise from its initial value to its peak value after an event occurs at each validation set sample point; Indicates the first The true value of the time constant for the discharge power to rise from its initial value to its peak value after an event occurs at each validation set sample point; MAPE represents the average relative error.

[0119] The algorithm determines whether the RMSE is less than a preset RMSE threshold and whether the MAPE is less than a preset MAPE threshold. If both are satisfied, the candidate explicit binary function is accepted as a three-dimensional mapping relationship. The preset RMSE threshold is set to 10% to 20% of the median of the time constant for the discharge power to rise from its initial value to its peak value after the event occurs. In this embodiment, 15% can be used based on experience. The preset MAPE threshold ranges from 8% to 15%. In this embodiment, 12% can be used based on experience.

[0120] This verification method is based on the diversity of operating conditions of data center energy storage systems: the hold-out method is used to verify that the three-dimensional mapping relationship can still accurately predict the response capability under unseen operating conditions, avoiding prediction failure in actual applications due to overfitting to historical data.

[0121] For example, in a data center, the median time constant for the discharge power to rise from its initial value to its peak after an event is 25ms. A preset RMSE threshold of 4ms and a preset MAPE threshold of 12% are set. Seven samples are randomly selected from the remaining 30 samples as the validation set, and 23 as the training set. The validation set test yields... , The requirements are met, and the three-dimensional mapping relationship is accepted.

[0122] In step 2, constructing the instantaneous response capability evaluation function of the energy storage system based on the three-dimensional mapping relationship includes: statistically analyzing the distribution of the state of charge value at the time of occurrence of multiple voltage step drop events used in establishing the three-dimensional mapping relationship and the time integral value of the absolute value of charging and discharging power within a preset period before the time of occurrence of the event in the three-dimensional space; and calculating the number density value of voltage step drop events in each local region in the three-dimensional space.

[0123] Based on the number density value, the three-dimensional space is divided into high-density regions and low-density regions. High-density regions are those with a number density value greater than a preset density threshold, while low-density regions are those with a number density value less than or equal to the preset density threshold.

[0124] For high-density areas, a first grid step size is used for spatial grid division, and for low-density areas, a second grid step size is used for spatial grid division. The first grid step size is smaller than the second grid step size.

[0125] For each spatial grid, the first and second dimension coordinates of the center point of the spatial grid are substituted into the three-dimensional mapping relationship to calculate the pre-stored value of the time constant corresponding to the spatial grid.

[0126] When receiving the current state of charge (SOC) value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset time period prior to the current time, the spatial grid to which the SOC value and the time integral value belong are determined, the time constant pre-stored value corresponding to the spatial grid is obtained, and the time constant pre-stored value is output as the predicted value of the time constant.

[0127] The kernel density estimation method is used to calculate the number density value of voltage step drop events in each local region of three-dimensional space. The three-dimensional space is initially divided into a regular grid, with the grid size being 5% to 10% of the coordinate range of each dimension. In this embodiment, 8% can be used based on experience. The number of sample points in each grid is calculated and divided by the grid volume to obtain the number density value, expressed as:

[0128] ;

[0129] in, Represents a grid Quantity density value; Represents a grid The number of sample points within; Represents a grid The volume. This method reflects the operating frequency of the data center energy storage system under different operating conditions: high-density areas correspond to common operating conditions (such as medium state of charge, normal load changes), while low-density areas correspond to rare operating conditions (such as extremely low state of charge, large load fluctuations).

[0130] Statistical analysis was performed on the number density values ​​of all grid cells, and the median number density value was calculated. Set the preset density threshold as follows:

[0131] ;

[0132] in, Indicates the preset density threshold; The density coefficient ranges from 0.5 to 2.0. In this embodiment, 1.0 can be used based on experience. This threshold setting ensures that high-density areas contain approximately 50% of the sample points, and low-density areas contain approximately 50% of the sample points, thus achieving a reasonable allocation of computing resources.

[0133] The first grid step size is used for high-density areas. Spatial gridding is performed, and a second grid step size is used for low-density areas. Perform spatial grid generation. The first grid step size is determined based on the average nearest neighbor distance of sample points within the high-density region.

[0134] ;

[0135] in, Indicates the first grid step size; The grid coefficient ranges from 1.0 to 3.0. In this embodiment, based on experience, 2.0 can be used. It represents the average nearest neighbor distance of sample points within a high-density region.

[0136] The second grid step size is determined based on the spatial extent of the low-density region:

[0137] ;

[0138] in, Indicates the second grid step size; This is the step size ratio coefficient, which ranges from 2 to 5. In this embodiment, based on experience, 3 can be used.

[0139] This hierarchical grid strategy is based on the real-time control requirements of data centers: common operating conditions require fine-grained prediction to ensure control accuracy, while rare operating conditions allow for coarse prediction to save computation time. For example, the average nearest neighbor distance in a high-density area of ​​a data center is 0.05 (normalized coordinates). Step size ratio coefficient Then the first grid step size is 0.1, and the second grid step size is 0.3.

[0140] For each spatial grid, calculate the coordinates of the grid center point. Substituting the values ​​into the three-dimensional mapping relationship, the pre-stored time constant corresponding to the spatial grid is obtained. :

[0141] ;

[0142] in, This represents the pre-stored value of the time constant; The coordinates of the state of charge at the moment the event occurs at the center point of the spatial grid; The coordinates of the time integral of the absolute value of the charging and discharging power within a preset time period before the event occurs at the center point of the spatial grid. This represents a three-dimensional mapping relationship.

[0143] The coordinates of the center points of all spatial grids and their corresponding pre-stored time constant values ​​are stored in a lookup table. The lookup table is a two-dimensional array data structure. The first dimension index corresponds to the grid index in the direction of the state of charge value at the time of the event, and the second dimension index corresponds to the grid index in the direction of the time integral value of the absolute value of charging and discharging power within a preset time period before the event. Array elements store the pre-stored time constant values ​​of the corresponding spatial grid center points. The lookup table is pre-calculated and stored in memory during the initialization phase of the energy storage system's instantaneous response capability assessment function. During runtime, the pre-stored time constant values ​​are obtained directly by accessing the corresponding array elements through the grid index based on the input state of charge value and time integral value, eliminating the need to repeatedly calculate the radial basis function. This pre-calculation strategy transforms the online calculation of the radial basis function into a lookup table operation, reducing the execution time of the energy storage system's instantaneous response capability assessment function from milliseconds to microseconds, meeting the real-time control requirements of data centers.

[0144] For example, a data center's high-density area is divided into a 20×15 spatial grid, and the low-density area is divided into a 10×8 spatial grid. The lookup table contains a total of 380 pre-stored values. The lookup table data structure can be represented as follows: ,in, and These are the grid indices for the direction of the state of charge value at the moment of the event and the direction of the time integral of the absolute value of the charging and discharging power within a preset period before the moment of the event, respectively. LUT stands for lookup table.

[0145] For example, the current state of charge (SOC) of a data center energy storage system is 62%, and the time integral of the absolute value of the charging and discharging power over a preset period prior to the current moment is 15 kWh. The calculated grid index is... ,pass The time constant is directly retrieved from the pre-stored value, and a single table lookup operation takes microseconds.

[0146] When receiving the current state of charge value of the energy storage system The time integral value of the absolute value of the charging and discharging power of the energy storage system within a preset time period before the current moment. At that time, the spatial grid index is determined based on the input coordinates. :

[0147] ;

[0148] ;

[0149] in, A grid index representing the direction of the state of charge at the moment the event occurs; A grid index representing the direction of the time integral of the absolute value of charging and discharging power within a preset time period prior to the occurrence of the event; This represents the current state of charge (SOC) value of the energy storage system. This represents the time integral of the absolute value of the charging and discharging power of the energy storage system within a preset time period prior to the current moment. This represents the minimum value of the state of charge at the moment the event occurs; This represents the minimum time integral value of the absolute value of the charging and discharging power within a preset time period prior to the occurrence of the event. The grid step size represents the direction of the state of charge at the moment the event occurs; The grid step size in the direction of the time integral of the absolute value of the charging and discharging power within a preset time period before the event occurs; This represents the floor function.

[0150] Retrieve the corresponding time constant pre-stored value from the lookup table based on the grid index. The pre-stored value of the time constant is output as the predicted value of the time constant. If the input coordinates exceed the three-dimensional space range, boundary extrapolation is used: the input coordinates are projected to the boundary, and the pre-stored value of the time constant of the boundary grid is used as the predicted value.

[0151] Specifically, the three-dimensional spatial range is determined based on the remaining sample points. The mesh is divided to cover the two-dimensional space formed by the first and second coordinates. to and from to The area, in which This represents the maximum value of the state of charge at the moment the event occurred among the remaining sample points. This represents the maximum time integral of the absolute value of the charging and discharging power within a preset time period prior to the event occurrence among the remaining sample points. The grid index range stored in the lookup table is: the grid index in the direction of the state of charge value at the event occurrence, from 0 to... The grid index of the time integral value of the absolute value of charging and discharging power within a preset time period before the event occurs is from 0 to... ,in The maximum value of the grid index in the direction of the state of charge at the moment the event occurs. This refers to the maximum value of the grid index in the direction of the time integral of the absolute value of charging and discharging power within a preset time period before the event occurs. The operation of projecting the input coordinates onto the boundary is as follows: if determined based on the input coordinates... If less than 0, then Corrected to 0, if Greater than Then Revised to ;like If less than 0, then Corrected to 0, if Greater than Then Revised to The time constant pre-stored value corresponding to the corrected grid index is obtained from the lookup table and used as the predicted value of the time constant.

[0152] In step 3, the voltage dynamic response equation is expressed as follows:

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] in, This represents the equivalent capacitance of the busbar. Indicates the DC bus voltage. This indicates the time from when the energy storage system begins to respond. This represents the first derivative of the DC bus voltage with respect to time. Indicates the current of the energy storage system. Indicates the load current. Indicates the peak current of the energy storage system. Represents the dynamic response factor. This represents the reduction coefficient. The predicted value representing the time constant. This indicates the rated voltage of the DC bus. Indicates the equivalent resistance of the load. Represents the natural exponential function;

[0159] The reduction coefficient Describe when the DC bus voltage Below the rated voltage of the DC bus The characteristic that the output current of the energy storage system decreases as the DC bus voltage decreases;

[0160] The peak current of the energy storage system is determined based on the preset maximum load step amplitude and the rated DC bus voltage. The equivalent resistance of the load is determined based on the load power and the rated DC bus voltage. The equivalent capacitance of the bus, the predicted value of the time constant, the peak current of the energy storage system, the equivalent resistance of the load, and the rated DC bus voltage are substituted into the voltage dynamic response equation. The rated DC bus voltage is used as the initial voltage value. The fourth-order Runge-Kutta method is used to solve the voltage dynamic response equation to obtain the numerical solution sequence of the DC bus voltage.

[0161] The minimum DC bus voltage is identified by traversing the DC bus voltage numerical solution sequence. The difference between the rated DC bus voltage and the minimum DC bus voltage is calculated as the estimated voltage drop depth.

[0162] Specifically, the preset maximum load step amplitude is determined based on data center virtual machine migration statistics. The load power step amplitude caused by all virtual machine migration events over the past 3 to 6 months is statistically analyzed, and the 95th percentile is calculated. The preset maximum load step amplitude is then set as follows:

[0163] ;

[0164] in, This indicates the preset maximum load step amplitude value; This represents the 95th percentile of the load power step amplitude; The load safety factor ranges from 1.2 to 1.5. In this embodiment, based on experience, 1.3 can be used.

[0165] This method is based on the statistical characteristics of data center load step events: 95% of virtual machine migration events have power increments of less than [a certain value]. Reserving a 30% safety margin can cover extreme situations and take into account future business growth. For example, statistics from a certain data center show... ,set up .

[0166] In step 4, the voltage dynamic response equation is solved in reverse to obtain the target value of the time constant. The state of charge value required to reach the target value of the time constant is found in reverse according to the three-dimensional mapping relationship. The difference between the retrieved state of charge value and the current state of charge value is used as the dynamic adjustment margin value. The voltage dynamic response equation is solved in reverse using the bisection method to obtain the target value of the time constant. The time constant search interval is initialized by the bisection method to the preset minimum time constant to the preset maximum time constant. In each iteration, the midpoint value of the search interval is substituted into the voltage dynamic response equation to obtain the voltage drop depth calculation value. The search interval is updated according to the relationship between the voltage drop depth calculation value and the voltage drop depth safety threshold. The iteration continues until the absolute value of the difference between the voltage drop depth calculation value and the voltage drop depth safety threshold is less than the preset error threshold.

[0167] Based on the three-dimensional mapping relationship, the contour lines of the target value whose time constant is equal to the time constant are identified in the two-dimensional space formed by the time integral of the absolute value of charging and discharging power during a preset period before the current time of the energy storage system.

[0168] The target state point is the point on the contour line corresponding to the minimum Euclidean distance from the current state point, which is calculated by the time integral of the current state value of the energy storage system and the absolute value of the charging and discharging power during a preset period before the current state.

[0169] Extract the coordinates of the state of charge (SOC) value of the target state point as the target SOC value, and calculate the difference between the target SOC value and the current SOC value of the energy storage system as the dynamic adjustment margin value.

[0170] Specifically, the contour lines for identifying time constants equal to the target value of the time constant include: using the spatial grid established in step 2 when constructing the instantaneous response capability evaluation function of the energy storage system, traversing each spatial grid; for each spatial grid, substituting the first-dimensional coordinate value and the second-dimensional coordinate value of the center point of the spatial grid into the explicit binary function to calculate the corresponding calculated value of the time constant; identifying several spatial grids with the smallest absolute value of the difference between the calculated value of the time constant and the target value of the time constant, and connecting the center points of the identified spatial grids according to their spatial positions to form the contour lines.

[0171] Specifically, the preset minimum time constant and preset maximum time constant define the search interval boundaries of the time constant when solving the voltage dynamic response equation in reverse using the bisection method.

[0172] The preset minimum time constant is determined based on the physical limits of the energy storage system converter response speed:

[0173] ;

[0174] in, This indicates the preset minimum time constant; This indicates the output filter inductance value of the energy storage system converter; This represents the equivalent resistance value of the active damping.

[0175] This time constant reflects the fastest rise rate of the current response in the energy storage system and is constrained by the time constant of the converter hardware circuit. Typical values ​​range from 5ms to 20ms. For example, if the converter output filter inductance of a data center energy storage system is 2mH and the active damping equivalent resistance is 0.2Ω, then the preset minimum time constant is 10ms.

[0176] The preset maximum time constant is determined based on the response speed of the energy storage system under the worst operating conditions:

[0177] ;

[0178] in, Indicates the preset maximum time constant; This represents the nominal time constant of the energy storage system, and its value is a typical value of the time constant obtained by testing the energy storage system under rated operating conditions. This represents the amplification factor of the time constant caused by the low-charge state, with a value ranging from 0.2 to 0.5; This represents the time constant amplification factor caused by low temperature, with a value ranging from 0.3 to 0.8; This represents the time constant increase factor caused by battery aging, with a value ranging from 0.2 to 0.6.

[0179] This time constant reflects the slowest response speed of an energy storage system under the combined effects of multiple adverse factors such as low state of charge, low temperature environment, and battery aging. For example, a data center energy storage system has a nominal time constant of 25ms. Considering the combined effects of state of charge, temperature, and aging, a preset maximum time constant is set to... .

[0180] The preset error threshold represents the maximum allowable deviation between the calculated voltage drop depth and the safe voltage drop depth threshold at the termination of the bisection method iteration. This threshold is determined based on the accuracy of the DC bus voltage measurement.

[0181] ;

[0182] in, Indicates the preset error threshold; The value is the error coefficient, ranging from 1 to 3. In this embodiment, based on experience, 2 can be used. This indicates the accuracy of DC bus voltage measurement.

[0183] The accuracy of DC bus voltage measurement is determined by the accuracy class of the voltage sensor, typically ranging from 0.1% to 0.5% of the rated voltage. For example, if the rated voltage of a data center's DC bus is 500V and the voltage sensor accuracy is 0.2%, then the DC bus voltage measurement accuracy is 1V, with a preset error threshold of 2V.

[0184] This threshold setting ensures that the accuracy of the bisection method matches the accuracy of the measurement system: a preset error threshold that is too small leads to too many iterations without meaningful accuracy improvement, while a preset error threshold that is too large results in insufficient solution accuracy, affecting the control effect. Setting the preset error threshold to twice the measurement accuracy allows convergence within 5 to 8 iterations, meeting real-time control requirements.

[0185] The safe threshold for voltage dip depth is determined based on the voltage tolerance of data center IT equipment. When the voltage dip depth exceeds 15% of the rated voltage, IT equipment may experience server restarts or data loss.

[0186] The voltage drop depth safety threshold is set as follows:

[0187] ;

[0188] in, Indicates the safe threshold for voltage drop depth; The voltage safety factor ranges from 0.08 to 0.12. In this embodiment, based on experience, 0.10 can be used. This indicates the rated voltage of the DC bus.

[0189] This coefficient is set based on data center power supply reliability requirements: controlling the voltage drop depth within 10% of the rated voltage ensures that 99.99% of IT equipment is unaffected, with a 5% margin to account for equipment differences and measurement errors. For example, if the rated voltage of a data center's DC bus is 500V, the safe threshold for voltage drop depth is set at 50V, meaning that the DC bus voltage must not be lower than 450V after a voltage drop.

[0190] Furthermore, the voltage drop depth safety threshold can be set according to the data center service level classification: Tier IV data centers adopt... Tier III data centers adopt Tier II data centers adopt .

[0191] The bisection iterative process is as follows: Initialize the search interval as... Each iteration takes the midpoint of the interval. Substituting the values ​​into the voltage dynamic response equation, we can obtain the calculated voltage sag depth. ,like Then update the search range as follows ,like Then update the search range as follows Repeat the iteration until ,in, This represents the time constant value corresponding to the midpoint of the search interval; This represents the calculated voltage drop depth. This indicates the preset minimum time constant; Indicates the preset maximum time constant; Indicates the safe threshold for voltage drop depth; This indicates the preset error threshold.

[0192] The final result This is the target value of the time constant.

[0193] In step 5, when the distance between the current state of charge (SOC) value of the energy storage system and the SOC boundary is less than the dynamic adjustment margin value, the distributed power source and load regulation are invoked to move the SOC value to the region that meets the requirements of the dynamic adjustment margin value. This includes calculating the amount of SOC adjustment required to move the current SOC value of the energy storage system to a position that is equal to the dynamic adjustment margin value from the SOC boundary.

[0194] Calculate the charging and discharging power adjustment of the energy storage system based on the state of charge adjustment amount and the energy storage system capacity.

[0195] Determine the sign of the energy storage system's charge and discharge power adjustment amount. When the energy storage system's charge and discharge power adjustment amount is positive, it is determined that the power flowing into the energy storage system needs to be increased. When the energy storage system's charge and discharge power adjustment amount is negative, it is determined that the power flowing into the energy storage system needs to be reduced.

[0196] When it is necessary to increase the power flowing into the energy storage system, the distributed power source is called to increase the output power, and the load regulation is called to reduce the data center load power.

[0197] When it is necessary to reduce the power flowing into the energy storage system, distributed power sources are called to reduce output power, and load regulation is called to increase the load power of the data center.

[0198] The state-of-charge boundary of an energy storage system includes the upper boundary of the state of charge. and boundary under charged state The upper boundary of the state of charge is determined based on the charging cutoff threshold of the energy storage system's battery management system, with a typical value of 90% to 95%. In this embodiment, 90% can be used based on experience. The lower boundary of the state of charge is determined based on the discharging cutoff threshold of the energy storage system's battery management system, with a typical value of 10% to 20%. In this embodiment, 20% can be used based on experience.

[0199] When the distance between the current state of charge (SOC) value of the energy storage system and the SOC boundary is less than the dynamic adjustment margin value, the judgment logic is as follows:

[0200] If the current state of charge value of the energy storage system Calculate the distance to the upper boundary of the charged state. ,judge Whether it holds true; if the current state of charge value of the energy storage system is... Calculate the distance to the boundary in the charged state. ,judge Whether it is true or not, among which, This represents the current state of charge (SOC) value of the energy storage system. Indicates the upper boundary of the charged state; Indicates the boundary under a charged state; This indicates the distance between the current state of charge and the boundary of the state of charge; This indicates the dynamic adjustment margin value.

[0201] This judgment logic clarifies the bidirectional adjustment space requirements of data center energy storage systems: when the state of charge is close to the upper boundary, a discharge margin needs to be reserved to cope with sudden load increases; when the state of charge is close to the lower boundary, a charging margin needs to be reserved to cope with sudden load drops. The nearest boundary is selected for judgment with 50% as the dividing point.

[0202] Specifically, the steps of increasing output power by calling upon distributed power and reducing data center load power by calling upon load regulation include: determining the relationship between the absolute value of the energy storage system's charge / discharge power adjustment and the amount of output power that the distributed power can increase; when the absolute value of the energy storage system's charge / discharge power adjustment is less than or equal to the amount of output power that the distributed power can increase, the distributed power is called to increase output power equal to the energy storage system's charge / discharge power adjustment; when the absolute value of the energy storage system's charge / discharge power adjustment is greater than the amount of output power that the distributed power can increase, the distributed power is called to increase output power equal to the amount of output power that the distributed power can increase, the remaining power demand is calculated as the absolute value of the energy storage system's charge / discharge power adjustment minus the amount of output power that the distributed power can increase, and the load regulation is called to reduce the data center load power equal to the remaining power demand. When it is necessary to reduce the power flowing into the energy storage system, the same allocation principle is used for both reducing output power by calling upon distributed power and increasing data center load power by calling upon load regulation.

[0203] In step 6, during the adjustment process, the actual value of the time constant is monitored, the deviation between the actual value of the time constant and the predicted value of the time constant is calculated, and the three-dimensional mapping relationship is updated based on the deviation using the recursive least squares method. This includes detecting the energy storage system's charging and discharging power response process triggered by the call to distributed power sources and load regulation, and extracting the measured value of the time constant reflecting the energy storage system's response speed from the charging and discharging power response process as the actual value of the time constant.

[0204] The state of charge (SOC) value at the corresponding moment of the energy storage system response process and the time integral value of the absolute value of the charging and discharging power within a preset period before the corresponding moment are obtained. The SOC value and the time integral value are input into a three-dimensional mapping relationship to calculate the predicted value of the time constant.

[0205] The difference between the actual value and the predicted value of the time constant is calculated as the prediction bias;

[0206] The cumulative number of times the energy storage system's charging and discharging power response process is triggered within a preset statistical time window is counted. The operating intensity of the energy storage system is determined based on the cumulative number of times. The forgetting factor of the recursive least squares method is adjusted based on the operating intensity. The higher the operating intensity, the smaller the forgetting factor.

[0207] The radial basis function value vector is calculated by inputting the state of charge value and time integral value into the radial basis function in the three-dimensional mapping relationship. The weight coefficient vector of the radial basis function in the three-dimensional mapping relationship is updated by the recursive least squares method based on the radial basis function value vector, prediction bias and forgetting factor, so that the three-dimensional mapping relationship can adapt to the changes in the characteristics of the energy storage system.

[0208] Specifically, the measured time constant reflecting the response speed of the energy storage system is extracted from the charge and discharge power response process, and an exponential fitting method is used. The time series of the energy storage system's charge and discharge power from the start of the response to reaching the steady-state value is collected, and the power value at each sampling moment is recorded. The measured power response curve is then fitted to a first-order exponential model.

[0209] ;

[0210] in, Indicates the elapsed time since the start of the response. The charging and discharging power of the energy storage system at that time; This indicates the time from when the energy storage system begins to respond; This represents the steady-state power value; This represents the measured value of the time constant; This represents the natural exponential function.

[0211] The measured time constant and steady-state power are solved using the nonlinear least squares method, and the objective function is minimized.

[0212] ;

[0213] in, This represents the minimize operator; This represents the measured value of the time constant; This represents the steady-state power value; Indicates the power sampling point number during the response process; Indicates the total number of sampling points; Indicates the first Measurement values ​​of the charging and discharging power of the energy storage system at each sampling point; Indicates the first The relative time from the start of the response corresponds to each sampling point; This represents the natural exponential function.

[0214] For example, during a certain adjustment, the charging and discharging power of the energy storage system was increased from 50kW to 180kW. With a sampling interval of 50ms, 100 data points were collected, and the measured value of the time constant was obtained by fitting, which was 22.8ms, and this value was taken as the actual value of the time constant.

[0215] The preset statistical time window is determined based on the data center load change cycle and is set to 0.5 to 2 times the main load change cycle. In this embodiment, based on experience, 1 times can be used. For data centers with a predominantly daily cycle, the preset statistical time window is set to 12 to 48 hours. In this embodiment, based on experience, 24 hours can be used. Let the number of hours in the preset statistical time window be... (Unit: hours)

[0216] The forgetting factor of the recursive least squares method is adjusted based on the cumulative number of times the energy storage system's charging and discharging power response process is triggered within a preset statistical time window. A higher cumulative number of times indicates that the energy storage system's characteristics change more rapidly, and a smaller forgetting factor allows the model to adapt quickly; a lower cumulative number of times indicates that the energy storage system's characteristics change more slowly, and a larger forgetting factor maintains model stability.

[0217] The forgetting factor in the recursive least squares method is determined based on the cumulative number of times: when the cumulative number of times the energy storage system's charging and discharging power response process is triggered within a preset statistical time window is less than... When using the recursive least squares method, the forgetting factor is between 0.98 and 0.99. In this embodiment, 0.99 can be used based on experience.

[0218] When the cumulative number of times the energy storage system's charging and discharging power response process is triggered within the preset statistical time window is greater than or equal to and less than When using the recursive least squares method, the forgetting factor is between 0.95 and 0.97. In this embodiment, based on experience, 0.96 can be used.

[0219] When the cumulative number of times the energy storage system's charging and discharging power response process is triggered within the preset statistical time window is greater than or equal to When using the recursive least squares method, the forgetting factor is typically between 0.90 and 0.93; in this embodiment, based on experience, 0.92 can be used. This indicates the number of hours in the preset statistical time window.

[0220] For example, the preset statistical time window is 24 hours, that is... When the cumulative number of times is less than 48, the forgetting factor of the recursive least squares method is 0.99; when the cumulative number of times is between 48 and 96, the forgetting factor of the recursive least squares method is 0.96; and when the cumulative number of times is greater than or equal to 96, the forgetting factor of the recursive least squares method is 0.92.

[0221] The weight coefficient vector of the radial basis function in the three-dimensional mapping relationship is updated using the recursive least squares method. The recursive formula is as follows:

[0222] ;

[0223] ;

[0224] ;

[0225] in, This represents the updated weight coefficient vector; This represents the weight coefficient vector before the update; Represents the gain vector; This represents the updated covariance matrix; This represents the covariance matrix before the update. Represents the radial basis function value vector; This represents the forgetting factor in recursive least squares; Indicates prediction bias; This represents the transpose of the radial basis function value vector.

[0226] The radial basis function value vector is calculated by inputting the current state of charge and time integral values ​​into the radial basis function in the three-dimensional mapping relationship. The first radial basis function value vector is... The elements are:

[0227] ;

[0228] in, The first value of the radial basis function value vector represents the... One element; This represents the normalized state of charge value at the corresponding moment in the energy storage system's response process; This represents the time integral value of the normalized absolute value of the charging and discharging power within a preset time period before the corresponding moment of the energy storage system's response process. Indicates the first The normalized state of charge values ​​of the remaining sample points at the time of the event occurrence; Indicates the first The time integral of the absolute value of charging and discharging power within a preset time period before the event occurrence time, after normalization of the remaining sample points; The kernel width parameter represents the radial basis functions; This represents the natural exponential function.

[0229] Prediction bias is calculated as the difference between the normalized measured value and the normalized predicted value of the time constant:

[0230] ;

[0231] in, This represents the measured value of the normalized time constant. This represents the predicted value of the time constant after normalization.

[0232] The initial covariance matrix is ​​set to ,in The initialization coefficients range from 100 to 1000. In this embodiment, based on experience, 500 can be used. The dimension is The identity matrix; Indicates the number of remaining sample points; Let represent the initial covariance matrix.

[0233] Example 2: This example also provides an electronic device applicable to a data center DC power supply load-storage coordinated control method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the data center DC power supply load-storage coordinated control method proposed in the above examples.

[0234] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements a data center DC power supply load-storage coordinated control method as proposed in the above embodiments.

[0235] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for coordinating control of DC power supply load and storage in a data center proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0236] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0237] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for coordinated control of DC power supply and energy storage in a data center, characterized in that: include, The system collects DC bus voltage sequences, energy storage system charge / discharge power sequences, and state of charge sequences. It identifies voltage step drop events in the voltage sequences and extracts the state of charge value at the time of the event, the time integral of the absolute value of charge / discharge power within a preset period before the event, and the time constant for the discharge power to rise from the initial value to the peak value after the event for each voltage step drop event. It then establishes a three-dimensional mapping relationship between the state of charge value at the time of the event, the time integral of the absolute value of charge / discharge power within a preset period before the event, and the time constant for the discharge power to rise from the initial value to the peak value after the event. An instantaneous response capability evaluation function for an energy storage system is constructed based on a three-dimensional mapping relationship. The instantaneous response capability evaluation function is then input with the current state of charge value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset time period prior to the current time to obtain the predicted value of the time constant. Substitute the predicted value of the time constant, the equivalent capacitance of the bus, and the preset maximum load step amplitude into the voltage dynamic response equation to obtain the estimated value of the voltage drop depth. When the estimated voltage drop depth exceeds the safe threshold for voltage drop depth, the voltage dynamic response equation is solved in reverse to obtain the target value of the time constant. The state of charge value required to reach the target value of the time constant is looked up in reverse according to the three-dimensional mapping relationship. The difference between the state of charge value obtained in reverse and the state of charge value at the current moment is used as the dynamic adjustment margin value. When the distance between the current state of charge (SOC) value of the energy storage system and the SOC boundary is less than the dynamic adjustment margin value, the distributed power source and load regulation are invoked to move the SOC value to the region that meets the requirements of the dynamic adjustment margin value. During the adjustment process, the actual value of the time constant is monitored, the deviation between the actual value of the time constant and the predicted value of the time constant is calculated, and the three-dimensional mapping relationship is updated based on the deviation using the recursive least squares method.

2. The data center DC power supply load-storage coordinated control method as described in claim 1, characterized in that: The identification of voltage step drop events in the voltage sequence includes performing a first-order time difference operation on the DC bus voltage sequence to obtain a voltage change rate sequence. Identify the moment when the absolute value of the voltage change rate in the voltage change rate sequence exceeds a preset change rate threshold, determine that the voltage change direction at that moment is negative, and take that moment as a candidate voltage drop moment; Extract the power reference value of the energy storage system's charging and discharging power sequence before the candidate voltage drop moment, extract the power peak value of the energy storage system's charging and discharging power sequence within a preset verification time window after the candidate voltage drop moment, and determine whether the power peak value shows a step increase relative to the power reference value. If a step increase occurs, the candidate voltage drop moment is confirmed as the voltage step drop event moment. Based on the timestamp of the voltage step drop event, the state of charge value corresponding to the timestamp is found in the state of charge sequence as the state of charge value at the time of the voltage step drop event.

3. The data center DC power supply load-storage coordinated control method as described in claim 2, characterized in that: The establishment of a three-dimensional mapping relationship between the state of charge value at the moment of the event, the time integral value of the absolute value of the charging and discharging power within a preset period before the moment of the event, and the time constant for the discharge power to rise from the initial value to the peak value after the event includes establishing a three-dimensional spatial coordinate system with the state of charge value at the moment of the event as the first coordinate, the time integral value of the absolute value of the charging and discharging power within a preset period before the moment of the event as the second coordinate, and the time constant for the discharge power to rise from the initial value to the peak value after the event as the third coordinate. The state of charge value at the time of occurrence of multiple voltage step drop events, the time integral value of the absolute value of charging and discharging power within a preset period before the time of occurrence of the event, and the time constant of the discharge power rising from the initial value to the peak value after the event are used as the coordinate values ​​of sample points in the three-dimensional spatial coordinate system. Calculate the Euclidean distance between any two sample points in the three-dimensional spatial coordinate system, identify sample point pairs whose Euclidean distance is less than a preset distance threshold, remove one sample point from the sample point pair, and obtain the remaining sample points; Calculate the standard deviation of the remaining sample points in the first-dimensional coordinate direction, the standard deviation in the second-dimensional coordinate direction, and the standard deviation in the third-dimensional coordinate direction. Calculate the ratio of the maximum standard deviation to the minimum standard deviation among the three standard deviations, and use the ratio as the kernel width parameter of the radial basis function. The remaining sample points are fitted with radial basis function surfaces using kernel width parameters to obtain an explicit binary function of the time constant from the initial value to the peak value of the discharge power after the event. The independent variables of the explicit binary function are the state of charge value at the time of the event and the time integral value of the absolute value of the charge and discharge power within a preset period before the event. The explicit binary function is used as a three-dimensional mapping relationship.

4. The data center DC power supply load-storage coordinated control method as described in claim 3, characterized in that: The instantaneous response capability evaluation function of the energy storage system constructed based on the three-dimensional mapping relationship includes: statistically analyzing the distribution of the state of charge value at the time of occurrence of multiple voltage step drop events used in the establishment of the three-dimensional mapping relationship and the time integral value of the absolute value of charging and discharging power within a preset period before the time of occurrence of the event in the three-dimensional space; and calculating the number density value of voltage step drop events in each local region in the three-dimensional space. Based on the number density value, the three-dimensional space is divided into high-density regions and low-density regions. High-density regions are those with a number density value greater than a preset density threshold, while low-density regions are those with a number density value less than or equal to the preset density threshold. For high-density areas, a first grid step size is used for spatial grid division, and for low-density areas, a second grid step size is used for spatial grid division. The first grid step size is smaller than the second grid step size. For each spatial grid, the first and second coordinate values ​​of the center point of the spatial grid are substituted into the three-dimensional mapping relationship to calculate the pre-stored value of the time constant corresponding to the spatial grid. When receiving the current state of charge value of the energy storage system and the time integral value of the absolute value of the charging and discharging power within a preset time period before the current time, the spatial grid to which the state of charge value and the time integral value belong are determined, the time constant pre-stored value corresponding to the spatial grid is obtained, and the time constant pre-stored value is output as the predicted value of the time constant.

5. The data center DC power supply load-storage coordinated control method as described in claim 4, characterized in that: The voltage dynamic response equation is expressed as follows: in, This represents the equivalent capacitance of the busbar. Indicates the DC bus voltage. This indicates the time from when the energy storage system begins to respond. This represents the first derivative of the DC bus voltage with respect to time. Indicates the current of the energy storage system. Indicates the load current. Indicates the peak current of the energy storage system. Represents the dynamic response factor. This represents the reduction coefficient. This represents the predicted value of the time constant. This indicates the rated voltage of the DC bus. Indicates the equivalent resistance of the load. Represents the natural exponential function; The peak current of the energy storage system is determined based on the preset maximum load step amplitude and the rated DC bus voltage. The equivalent resistance of the load is determined based on the load power and the rated DC bus voltage. The equivalent capacitance of the bus, the predicted value of the time constant, the peak current of the energy storage system, the equivalent resistance of the load, and the rated DC bus voltage are substituted into the voltage dynamic response equation. The rated DC bus voltage is used as the initial voltage value. The fourth-order Runge-Kutta method is used to solve the voltage dynamic response equation to obtain the numerical solution sequence of the DC bus voltage. The minimum DC bus voltage is identified by traversing the DC bus voltage numerical solution sequence. The difference between the rated DC bus voltage and the minimum DC bus voltage is calculated as the estimated voltage drop depth.

6. The data center DC power supply load-storage coordinated control method as described in claim 5, characterized in that: The target value of the time constant is obtained by inversely solving the voltage dynamic response equation. The state of charge value required to reach the target value of the time constant is retrieved according to the three-dimensional mapping relationship. The difference between the retrieved state of charge value and the current state of charge value is used as the dynamic adjustment margin value. The target value of the time constant is obtained by inversely solving the voltage dynamic response equation using the bisection method. The time constant search interval is initialized by the bisection method to a preset minimum time constant to a preset maximum time constant. In each iteration, the midpoint value of the search interval is substituted into the voltage dynamic response equation to obtain the voltage drop depth calculation value. The search interval is updated according to the relationship between the voltage drop depth calculation value and the voltage drop depth safety threshold. The iteration continues until the absolute value of the difference between the voltage drop depth calculation value and the voltage drop depth safety threshold is less than a preset error threshold. Based on the three-dimensional mapping relationship, the contour lines in the two-dimensional space formed by the time integral values ​​of the absolute values ​​of charging and discharging power during a preset period before the current time of the energy storage system are identified so that the time constant is equal to the target value of the time constant. The target state point is the point on the contour line corresponding to the minimum Euclidean distance from the current state point, which is calculated by the time integral of the current state value of the energy storage system and the absolute value of the charging and discharging power during a preset period before the current state. Extract the coordinates of the state of charge (SOC) value of the target state point as the target SOC value, and calculate the difference between the target SOC value and the current SOC value of the energy storage system as the dynamic adjustment margin value.

7. The data center DC power supply load-storage coordinated control method as described in claim 6, characterized in that: The step of calling distributed power sources and load regulation to move the state of charge (SCC) value to a region that meets the requirements of the dynamic regulation margin when the distance between the current SCC value and the SCC boundary of the energy storage system is less than the dynamic regulation margin value includes calculating the amount of SCC value regulation required to move the current SCC value of the energy storage system to a position that is equal to the dynamic regulation margin value from the SCC boundary. Calculate the charging and discharging power adjustment of the energy storage system based on the state of charge adjustment amount and the energy storage system capacity. Determine the sign of the energy storage system's charge and discharge power adjustment amount. When the energy storage system's charge and discharge power adjustment amount is positive, it is determined that the power flowing into the energy storage system needs to be increased. When the energy storage system's charge and discharge power adjustment amount is negative, it is determined that the power flowing into the energy storage system needs to be reduced. When it is necessary to increase the power flowing into the energy storage system, the distributed power source is called to increase the output power, and the load regulation is called to reduce the data center load power. When it is necessary to reduce the power flowing into the energy storage system, distributed power sources are called to reduce output power, and load regulation is called to increase the load power of the data center.

8. The data center DC power supply load-storage coordinated control method as described in claim 7, characterized in that: During the adjustment process, monitoring the actual value of the time constant, calculating the deviation between the actual value of the time constant and the predicted value of the time constant, and updating the three-dimensional mapping relationship based on the deviation using the recursive least squares method includes detecting the energy storage system charging and discharging power response process triggered by calling distributed power sources and load regulation, and extracting the measured value of the time constant reflecting the response speed of the energy storage system from the charging and discharging power response process as the actual value of the time constant. The state of charge value at the corresponding moment of the energy storage system response process and the time integral value of the absolute value of the charging and discharging power within a preset period before the corresponding moment are obtained. The state of charge value and the time integral value are input into a three-dimensional mapping relationship to calculate the predicted value of the time constant. The difference between the actual value of the time constant and the predicted value of the time constant is calculated as the prediction deviation; The cumulative number of times the energy storage system's charging and discharging power response process is triggered within a preset statistical time window is counted. The operating intensity of the energy storage system is determined based on the cumulative number of times. The forgetting factor of the recursive least squares method is adjusted based on the operating intensity. The higher the operating intensity, the smaller the forgetting factor. The radial basis function value vector is calculated by inputting the state of charge value and time integral value into the radial basis function in the three-dimensional mapping relationship. The weight coefficient vector of the radial basis function in the three-dimensional mapping relationship is updated by the recursive least squares method based on the radial basis function value vector, prediction bias and forgetting factor, so that the three-dimensional mapping relationship can adapt to the changes in the characteristics of the energy storage system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the data center DC power supply load-storage coordinated control method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the data center DC power supply load-storage coordinated control method according to any one of claims 1 to 8.

Images