Power control method of cascade energy storage system based on variational mode decomposition and K-means clustering algorithm

Abstract

The present invention provides a power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm, which decomposes the power command through a parameter-adaptive variational mode decomposition algorithm to obtain power command sub-components with obvious differences in maximum power demand, charging and discharging frequency, and total charging and discharging during scheduling. Finally, appropriate eigenvalues ​​are selected to quantify the power command sub-components and the operating characteristics of different energy storage units in the energy storage system, while optimizing the similarity metric, and completing power allocation that takes into account the life loss and power response capability of the cascade energy storage system through a clustering algorithm. The present invention improves the life loss problem of the energy storage unit while ensuring the responsiveness of the cascade energy storage system to the power demand of the power grid, thereby improving the economy and stability of the application of the cascade battery energy storage system.

Description

Technical Field

[0001] The present invention belongs to the field of frequency regulation control strategy of cascade battery energy storage systems participating in wind power grids in the direction of new energy, and specifically is a power control method of cascade energy storage systems based on variational mode decomposition (VMD) and K-means clustering algorithm. Technical Background

[0002] In recent years, the production and sales of new energy vehicles in my country have increased dramatically and maintained a sustained growth trend. With the increasing production and sales of new energy vehicles, the demand for power batteries has also increased. However, after 3-5 years of use in the car, the power battery is usually retired because the capacity and power decrease and cannot meet the driving needs. Since the power battery is generally retired when the capacity decays to about 80%, there is still a lot of room for reuse in other application scenarios. The residual economic value brought by the peak of power battery retirement and the social and environmental demand for the reuse of retired power batteries have given rise to the cascade battery utilization technology, that is, through detection, screening, reorganization and other links, the retired battery is used as an energy storage unit to build a cascade battery energy storage system, which is re-applied to household energy, photovoltaic charging stations, energy storage power stations, communication base stations and other scenarios. The cascade battery utilization technology can effectively solve the environmental pressure brought by recycling, extend the service life of power batteries, and has considerable economic and social benefits.

[0003] On the other hand, with the development of wind power generation technology in my country, wind power generation plays an increasingly important role in the construction of new power systems and has become an important pillar of my country's energy transformation and climate and environmental governance in the future. Large-scale wind power grid connection also means that more and more power generation units will be connected to the power grid through power electronic equipment, the capacity of traditional synchronous frequency regulation units will shrink, and the inertial support capacity and frequency regulation capacity of the power system will be seriously weakened. At the same time, wind power has the volatility and intermittency of power generation, which further deteriorates the stability of power system operation. Under the same frequency events, the power system with a higher proportion of wind power is subject to a shorter frequency disturbance time scale, more violent fluctuations and a larger proportion of high-frequency disturbances, and the frequency stability and power quality of the power grid operation are difficult to guarantee. Therefore, enabling wind farms to participate in the frequency support of the power system is a necessary way to alleviate the frequency regulation pressure of the system with a high proportion of wind power and improve the power grid's new energy absorption capacity.

[0004] Energy storage equipment has the ability to transfer electric energy in time and space and to flexibly dispatch it. Configuring it on the wind farm side to form a wind-storage combined system and assisting it in responding to the system frequency regulation needs is a feasible solution to the problem that new energy sites lack the ability to actively support frequency regulation due to environmental resources. Applying the technology of second-hand battery utilization to the construction of a wind-storage combined system will further improve the economic and social benefits of system operation. However, when second-hand batteries are used to build a wind-storage combined system to assist wind farm stations in frequency regulation, differences in battery health status (SOH), rated power, rated capacity, and cycle life due to factors such as different retired batches, screening, and different reorganization parameters may cause problems such as poor power system demand response capabilities, unsatisfactory frequency regulation support effects, overcharging and discharging of battery packs, and increased equipment life loss. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of overcharging and discharging of equipment, insufficient power response capability, hidden dangers in system operation safety and stability, increased life loss, inconsistent retirement time, and increased operating costs caused by unreasonable power distribution between energy storage units within the cascade battery energy storage system in the current wind-storage combined system.

[0006] The technical solution for achieving the purpose of the present invention is: a power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm, comprising:

[0007] The life loss model of the cascade battery is transformed into a practical one to obtain the quantitative data of the life loss caused by the pre-assigned power instructions on different energy storage units;

[0008] The power command received by the cascade battery energy storage system is decomposed using the variational mode decomposition algorithm to obtain the maximum power demand, charge and discharge frequency, and total charge and discharge power command sub-components during the scheduling period.

[0009] Select characteristic parameters to quantify the power instructions of the energy storage units and the operating characteristics of different energy storage units in the energy storage system into data objects in the clustering space;

[0010] The data objects in the clustering space are normalized, and a weighted coefficient matrix is ​​introduced to complete the clustering distance calculation, and the range of objects to which the power signal can be assigned is divided according to the rigid constraint conditions;

[0011] The K-means clustering algorithm is used to complete the clustering division and reorganization between power signals and energy storage units, and realize the internal power distribution of the cascade energy storage system.

[0012] Preferably, the cascade battery life loss model is transformed into a practical form, and the final form of the transformation is as follows:

[0013]

[0014] Where D A Refers to the actual discharge depth during irregular discharge, D R is the discharge depth of the lithium battery under rated conditions, γ0 and γ1 are fitting coefficients, ΔT refers to the time interval between two power dispatches of the energy storage unit, P R is the rated charge and discharge power of the battery.

[0015] Preferably, the variational mode decomposition algorithm is used to decompose the power command received by the cascade battery energy storage system, and the specific method of obtaining the power command sub-components with obvious differences in maximum power demand, charge and discharge frequency, and total charge and discharge amount during scheduling is:

[0016] Create a sliding queue window with a time length of T. The window contains the power dispatch signal received by the cascade battery energy storage system when the station power dispatch instruction is issued;

[0017] In each scheduling cycle, the data in the window is refreshed, the power scheduling instruction data at the current moment is updated to enter the window, and the last data in the queue in the window is discarded to realize the sliding of the window in the time domain. Each time the sliding is completed, a VMD decomposition is performed and the sub-signal obtained in the signal decomposition result is output.

[0018] Preferably, the method for determining the sliding window length is:

[0019] The minute-level sliding window length N is selected as the initial value, and the power dispatch cycle of the cascade energy storage system in the wind farm is selected as the update step. The correlation between the sub-modes of the power signal obtained by variational mode decomposition under each window length is traversed, and the window length Nopt corresponding to the minimum sum of correlation coefficients is selected as the optimal sliding window length. Among them, the correlation coefficient selects the Pearson correlation coefficient, which is specifically defined as follows:

[0020]

[0021]

[0022] In the formula, ρ N (u i ,u k ) is the i-th sub-mode u obtained by VMD decomposition when N is the value of i and the kth submode u k Pearson correlation coefficient between Cov(u i ,u k ) is the sub-mode u i 、u k The covariance between Represent sub-mode u i 、u k The standard deviation of∑ is the sum of the Pearson correlation coefficients between all sub-modes decomposed by VMD at its N value.

[0023] Preferably, the total number of modal decompositions K during VMD decomposition is determined as follows:

[0024] The upper limit of K value meets the following requirements:

[0025] max{ω k (t)}≥max{ω T.j}

[0026] In the formula, ω k (t) represents the center frequency of each sub-mode obtained by VMD decomposition at time t; ω T.j Indicates the response frequency of each battery energy storage unit in the field;

[0027]

[0028] In the formula, H i.reo is the signal obtained by recombining all modal components decomposed by VMD when K = i; H i.ori is the original signal, Cov(H i.reo ,H i.ori ) means K=i when H i.reo With H i.ori The covariance of They respectively represent the standard deviation of the recombinant signal and the original signal when K=i.

[0029] Preferably, the data objects in the clustering space include:

[0030] The power signal side selects the maximum charge and discharge power demand Pmax and the charge and discharge frequency ω dc 、Charge and discharge quantity E during the dispatching time dc And the effective discharge capacity d of different energy storage units in response to power instructions eff.i As a quantization parameter, it is calculated as follows:

[0031] P max,i =max{P d,i ,|P c,i |}

[0032] ω dc,i =ω k,i

[0033]

[0034]

[0035] Where P d,i , P c,iThey refer to the discharge power and charging power amplitude sequences of the i-th power signal, P i (t) represents the charging and discharging power amplitude at each moment in the i-th signal, P R is the rated charge and discharge power of the battery; is the actual power of the battery during the i-th irregular charge and discharge process, D R is the discharge depth of lithium battery under rated conditions, D A Refers to the actual discharge depth during irregular discharge, P ec,i,ref It refers to the power dispatch instruction that a certain energy storage unit is expected to receive during the i-th charging and discharging process. γ0 and γ1 are both fitting coefficients. ΔT refers to the time interval between two power dispatches of the energy storage unit. ω k，i is the center frequency corresponding to the i-th submode obtained by VMD decomposition;

[0036] The energy storage unit equipment side selects the maximum charge and discharge power P of the energy storage unit ec，max , response frequency ω ec，dc , the available charge and discharge capacity E of the battery based on the current SOC state ec And the total effective discharge capacity Γ during the battery's entire life cycle R As quantization parameters, some of them are calculated as follows:

[0037]

[0038] Γ R，j =L R，j D R，j C R，j

[0039] In the formula, E ec，j is the available charge and discharge capacity of the jth energy storage unit considering the current SOC during power allocation; SOC j (t) is the SOC state of the jth energy storage unit during power distribution; SOH j is the SOH state of the jth energy storage unit during power distribution; E ec.j.n is the rated capacity of the energy storage unit.

[0040] Preferably, the specific method for normalizing the data objects in the clustering space is:

[0041]

[0042]

[0043] Where, X i , Y i are normalized parameters with dimension i on the signal side and the device side respectively; x i ,y iis the original parameter with dimension i on the signal side and the device side; y imax and imin is the maximum and minimum value of the parameter with dimension i on the equipment side; μ i is the scaling factor.

[0044] Preferably, the calculation formula of cluster distance is:

[0045] M=[m1,m2,...,m n ]

[0046]

[0047] Where M is the weighting coefficient matrix and n is the number of data.

[0048] Preferably, the range of objects to which the power signal can be allocated is divided according to the rigid constraint conditions, and when the energy storage unit meets the following conditions, it is listed as an object to which the power signal can be allocated:

[0049] P ec,max,j ≥P max,i

[0050] E ec,j ≥E dc,i

[0051] ω ec,dc,j ≥ω dc,i

[0052] Preferably, the K-means clustering algorithm is used to complete the clustering division and reorganization between the power signal and the energy storage unit, and the specific method for realizing the internal power distribution of the cascade energy storage system is:

[0053] The data objects quantized on the device side are used as the initial cluster centers for iterative solution. When the new cluster center of a certain cluster deviates from the initial cluster center by a certain distance ε during the iteration, the center of the cluster is no longer updated and the data objects of the cluster are no longer reclassified.

[0054] The error sum of squares SSE is selected as the clustering effectiveness evaluation index for iterative solution. The error sum of squares of the clustering result is expressed as:

[0055]

[0056] In the formula, C i represents the i-th cluster center; K represents the total number of cluster centers in the clustering result; if object x belongs to C i The Euclidean distance between the two is calculated, and the clustering of all objects to their cluster centers is summed up. When the SSE is the smallest, the sample difference in the same cluster is minimized, and the clustering division and reorganization between the power signal and the energy storage unit are completed.

[0057] Compared with the prior art, the present invention has the following significant advantages:

[0058] The present invention realizes the preliminary decomposition of the power command of the energy storage system through the parameter adaptive variational mode decomposition algorithm. Compared with the commonly used signal decomposition algorithms such as empirical mode decomposition and wavelet packet decomposition, this method reduces the dependence on the basis function, improves the mode aliasing phenomenon, has better robustness, and ensures that the power command of the energy storage unit obtained by decomposition has obvious distinguishable differences in a certain number of features, which lays a good foundation for the subsequent power distribution between energy storage units.

[0059] The present invention selects the maximum power amplitude, charge and discharge frequency, charge and discharge capacity during the scheduling period, and the maximum charge and discharge power of the energy storage unit and supercapacitor, the response frequency, and the charge and discharge capacity under the current SOC state. These operating characteristic parameters quantify the energy storage unit power instruction and the operating characteristics of different energy storage units in the energy storage system into data objects, creating conditions for clustering division;

[0060] The present invention improves the similarity metric standard, optimizes the power control process, and uses a clustering algorithm to complete the internal power distribution of the cascade energy storage system that takes into account life loss and power responsiveness. While ensuring the power demand responsiveness of the energy storage system, it effectively improves the operating state of the energy storage equipment, enhances the system operation stability, reduces equipment operation loss, and improves the economy and stability of the application of the cascade battery energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is the structure diagram of the simulation verification system;

[0062] Figure 2 It is the input wind speed of the wind turbine units in the wind farm within 12 hours;

[0063] Figure 3 It is the comparison of system frequency stability under different power control strategies;

[0064] Figure 4 It is the power decomposition effect of the cascade energy storage system;

[0065] Figure 5 It is the comparison of SOC status of energy storage unit under different control strategies;

[0066] Figure 6 It is a comparison of energy storage unit life loss under different control strategies;

[0067] Table 1 is the specific parameter configuration in the simulation verification. DETAILED DESCRIPTION

[0068] A power control method for a cascade energy storage system based on variational mode decomposition (VMD) and K-means clustering algorithm. The cascade battery energy storage system is composed of multiple lithium iron phosphate battery module energy storage units, which are installed on the wind farm side and controlled by the station dispatch. The real-time power command received by the cascade battery energy storage system is obtained by the difference between the power dispatch demand of the power grid and the maximum power that can be increased (reduced) by the wind turbine in the station at the current moment. By selecting characteristic values ​​that take into account the operating characteristics of the energy storage device and the characteristics of the power command signal, the energy storage device and the energy storage unit power command obtained by parameter adaptive VMD decomposition are quantified into data objects in the clustering space. The clustering problem is solved by the K-means algorithm to achieve the power control of the cascade energy storage system while ensuring the power response capability of the energy storage system and improving the operating status of the equipment, thereby improving the system operation stability, reducing the equipment operation loss, and improving the economy and stability of the cascade energy storage system application. The method includes the following steps:

[0069] The first step is to transform the cascade battery life loss model into a practical one, so that before the internal power distribution of the cascade energy storage system is completed, the impact of the current distribution method on the operating life of different energy storage units can be quantified.

[0070] Based on the battery life cumulative loss model, the influence of battery discharge depth and discharge speed is considered, and the power command is converted into the effective discharge capacity of the energy storage unit under rated conditions, which is convenient for comparison with the total effective discharge capacity in its entire life cycle, providing a reference for the rationality of power allocation. The specific method is as follows:

[0071] The expression for determining the total effective discharge capacity of a battery under rated conditions is as follows:

[0072] Γ R =L R D R C R (1)

[0073] In the formula, Γ R L is the total effective discharge capacity of the lithium battery in its entire life cycle; R D is the cycle life of lithium battery under rated conditions; R C is the discharge depth of lithium battery under rated conditions; R is the capacity of the lithium battery under rated conditions;

[0074] The main factors that affect the battery cycle life are the depth of discharge and the charge and discharge rate. The deeper the discharge depth, the lower the battery cycle life. The relationship between the discharge depth and the actual cycle life of the energy storage unit can be fitted into the form shown in the following formula:

[0075]

[0076] Where, L refers to the actual life of the battery; DA Refers to the actual discharge depth during a certain irregular discharge process; γ0 and γ1 are both fitting coefficients.

[0077] It is approximately believed that the life loss caused by the charge and discharge rate of the battery can be expressed by the charge and discharge power. The specific expression is as shown in the formula:

[0078]

[0079] Where P R is the rated charge and discharge power of the battery; P i ES is the actual power of the battery during the i-th irregular charge and discharge process. k0 is the influence coefficient.

[0080] During the operation of the second-life battery system, it is difficult to ensure that each charge and discharge process of the energy storage unit is carried out under rated conditions. Therefore, in order to intuitively quantify the impact of the power command on the operating life of each unit, the power command is converted into the effective discharge capacity of the energy storage unit under rated conditions, which is convenient for comparison with the total effective discharge capacity during its entire life cycle, providing a reference for the rationality of power allocation. The conversion method is as follows:

[0081] d effi =ε i μ i d a.i (4)

[0082] Where, d eff.i The actual power of the lithium battery in the i-th charge and discharge process is converted to the effective power under rated conditions; ε i is the discharge depth influence coefficient of the lithium battery in the i-th charge and discharge process; μ i is the charge and discharge rate influence coefficient of the lithium battery in the i-th charge and discharge process; d a，i is the actual capacity of the lithium battery during the i-th charge and discharge process.

[0083] The calculation method of the discharge depth influence coefficient is as follows:

[0084]

[0085] Where, L i is the actual cycle life corresponding to the i-th charge and discharge process of the battery. The calculation method has been given in formula (2).

[0086] The calculation method of the discharge rate influence coefficient is as follows:

[0087]

[0088] In the formula, I R is the rated charge and discharge current of the battery; is the actual current of the battery during the i-th charge and discharge process.

[0089] Combining equations (1) and (4)-(6) can obtain the effective discharge capacity in the i-th charge and discharge process taking into account the effects of discharge depth and discharge rate:

[0090]

[0091] where d a，i It can also be converted into the following form:

[0092] d a，i =P ec，i，ref ΔT (8)

[0093] Where P ec，i，ref Refers to the power dispatch instruction that a certain energy storage unit is expected to receive during the i-th charging and discharging process. The calculation assumes that the power instruction can be fully responded to, that is, P i ES =P ec，i，ref ; ΔT refers to the time interval between two power dispatches of the energy storage unit.

[0094] Formula (7) can be transformed into the following form:

[0095]

[0096] Formula (9) can be used to obtain the quantitative data on the life loss caused by the pre-allocated power instructions on different energy storage units. This data will provide support for subsequent more reasonable power allocation decisions.

[0097] In the second step, the power command received by the cascade battery energy storage system is decomposed using the variational modal decomposition (VMD) algorithm to obtain power command sub-components with significant differences in maximum power demand, charge and discharge frequency, and total charge and discharge during scheduling. In order to realize the real-time application of the VMD algorithm, it is combined with the first in, first out (FIFO) queue, and the adaptive adjustment of important parameters that determine the signal decomposition effect is achieved through the adaptive window length method and the adaptive total modal decomposition method.

[0098] The real-time application of the parameter adaptive variational mode decomposition algorithm includes the following three parts. First, the first-in-first-out queue is combined with the VMD algorithm. A sliding queue window with a time length of T is created. The window contains the power dispatch signal received by the cascade battery energy storage system when the station power dispatch instruction is issued. The data in the window is refreshed in each dispatch cycle, the power dispatch instruction data at the current moment is updated to enter the window, and the last data in the queue in the window is discarded to realize the sliding of the window in the time domain. Each time the sliding is completed, a VMD decomposition is performed and the sub-signal obtained in the signal decomposition result is output.

[0099] The second part is the adaptive selection of sliding window length. The sliding window length N of the minute level is selected as the initial value, and the power dispatch cycle of the cascade energy storage system in the wind farm is selected as the update step. The correlation between the sub-modes of the power signal obtained by variational mode decomposition under each window length is traversed, and the window length N corresponding to the minimum sum of correlation coefficients is selected. opt As the optimal sliding window length. Among them, the correlation coefficient selects the Pearson correlation coefficient, which is specifically defined as follows:

[0100]

[0101]

[0102] In the formula, ρ i，n (u i ,u k ) is the sub-mode u obtained by VMD decomposition when N = i i 、u k Pearson correlation coefficient between Cov(u i ,u k ) is the sub-mode u i 、u k The covariance between . Represent sub-mode u i 、u k The standard deviation of ∑，i is the sum of the Pearson correlation coefficients between all sub-modalities obtained by VMD decomposition when N=i.

[0103] The third part is the adaptive selection of the total number of modal decompositions K. When the maximum center frequency of the power signal vibration component obtained based on the current K value exceeds the maximum response frequency of the cascade battery energy storage unit in the wind farm, the energy storage device can no longer effectively respond to the high-frequency vibration component generated by continuing to increase the K value. At this time, the K value has reached the upper limit and no longer needs to be increased. This condition can be expressed by the following formula:

[0104] max{ω k (t)}≥max{ω T.j}

[0105] In the formula, ω k (t) represents the center frequency of each sub-mode obtained by VMD decomposition at time t; ω T.j Indicates the response frequency of each battery energy storage unit in the field.

[0106] On the other hand, consider using the Person correlation coefficient to quantify the correlation between all modal components decomposed under different K values ​​and the original signal after reorganization. The larger the correlation coefficient, the closer the decomposition is to complete, and it is meaningless to continue to increase the K value. When the correlation coefficient reaches the threshold δ, it means that the maximum K value that can be selected for signal decomposition at the current moment has been obtained. The construction of the Person correlation coefficient is shown in the following formula:

[0107]

[0108] In the formula, H i.reo is the signal obtained by recombining all modal components decomposed by VMD when K = i; H i.ori is the original signal. Cov(H i.reo , H i.ori ) represents the covariance of the above two when K=i; They respectively represent the standard deviation of the recombinant signal and the original signal when K=i.

[0109] The third step is to select characteristic parameters and quantify the power instructions of the energy storage units and the operating characteristics of different energy storage units in the energy storage system into data objects in the clustering space.

[0110] The power signal side selects the maximum charge and discharge power requirement P max 、Charge and discharge frequencyω dc 、Charge and discharge quantity E during the dispatching time dc And the effective discharge capacity d of different energy storage units in response to power instructions eff.i As a quantization parameter, it is calculated as follows:

[0111] P max，i =max{P d，i ,|P c，i |}

[0112] ω dc，i =ω k，i

[0113]

[0114]

[0115] Where P d，i , P c，iThey refer to the discharge power and charging power amplitude sequences of the i-th power signal respectively; P i (t) represents the charging and discharging power amplitude at each moment in the i-th signal;

[0116] The energy storage unit equipment side selects the maximum charge and discharge power P of the energy storage unit ec，max , response frequency ω ec，dc , the available charge and discharge capacity E of the battery based on the current SOC state ec And the total effective discharge capacity Γ during the battery's entire life cycle R As quantization parameters, some parameters are calculated as follows:

[0117]

[0118] Γ R，j =L R，j D R，j C R，j

[0119] In the formula, E ec，j is the available charge and discharge capacity of the j-th energy storage unit considering the current SOC during power allocation; SOC(t) is the SOC state of the j-th energy storage unit during power allocation; SOH j is the SOH state of the jth energy storage unit during power distribution; E ec.j.n is the rated capacity of the energy storage unit.

[0120] The fourth step is to normalize the data objects in the clustering space, introduce a weighted coefficient matrix, complete the clustering distance calculation, and divide the range of energy storage units to be allocated for the power signal. This step improves the similarity metric of the clustering algorithm and optimizes the power control process.

[0121] Improve the similarity metric of clustering algorithms. The signal side and the device side have the same dimension for the same type of feature values, but due to the short time interval between two power allocations, there will be a large difference in the value of some parameters on both sides. Therefore, in order to ensure the rationality of the clustering results when the clustering data set has a large number of dimensions and large differences in dimensions and values, a scaling factor is introduced to normalize the data, and its form is as follows:

[0122]

[0123]

[0124] Where, X i , Y i are normalized parameters with dimension i on the signal side and the device side respectively; x i ,y i is the original parameter with dimension i on the signal side and the device side; y imaxand imin is the maximum and minimum value of the parameter with dimension i on the equipment side; μ i is the scaling factor, responsible for converting the device-side parameters to the same order of magnitude as the signal-side parameters;

[0125] At the same time, in order to avoid the influence of some parameters on the clustering results being too small and to facilitate the field-side control system to consider the priority of multiple factors when deciding power allocation, the weighted coefficient matrix M is introduced when calculating the clustering distance. Finally, the calculation of the clustering distance is shown as follows:

[0126] M=[m1,m2,...,m n ]

[0127]

[0128] Power control process optimization. In order to ensure the safe and stable operation of the cascade energy storage system, it is necessary to define rigid constraints before the "screening and reorganization" of the power signal, select the parameter with the most serious limit-crossing consequence among the quantitative parameters as the judgment factor, and divide the range of energy storage units to be allocated for a certain power signal. The determination form of the allocation of energy storage units is shown in the following formula:

[0129] P ec，max，j ≥P max，i

[0130] E ec，j ≥E dc，i

[0131] ω ec，dc，j ≥ω dc，i

[0132] When the energy storage unit meets the above conditions, it is listed as an assignable object of the power signal, and further consideration is given to realizing power allocation through a clustering algorithm.

[0133] The fifth step is to use the K-means clustering algorithm to complete the clustering division and reorganization between the power signal and the energy storage unit to achieve the internal power distribution of the cascade energy storage system.

[0134] The basic process of K-means algorithm is:

[0135] STEP 1: Generate sample data and select K objects as initial cluster centers;

[0136] STEP 2: Determine the similarity metric (distance measure) and standardize the data objects;

[0137] STEP 3: Calculate the distance from each sample data to the cluster center and divide the clusters;

[0138] STEP 4: Return to STEP 3 until the clustering stop condition (iteration number, SSE drop threshold) is met; STEP 5: Output the clustering results;

[0139] In order to ensure that the "screening and recombination" results of the power signal are more in line with the equipment operating parameters, restrictions on the iterative update of the cluster center are added, as described below:

[0140] The data object quantified on the device side is used as the initial cluster center, and when the new cluster center of a cluster deviates from the initial cluster center by a certain distance ε during the iteration process, the center of the cluster is no longer updated and the data object attribution of the cluster is no longer reclassified. The judgment condition can be described as:

[0141]

[0142] In the formula, d(C o , C n ) is the Euclidean distance between the center of the new cluster generated by a certain iteration process and the center of the initial cluster; C o is the initial cluster center of a certain cluster; C k It is the new cluster center generated by a certain type of cluster during the iteration process.

[0143] The sum of squared errors (SSE) is selected as the clustering effectiveness evaluation index for iterative solution. The sum of squared errors of the clustering results can be expressed as:

[0144]

[0145] In the formula, C i represents the i-th cluster center; K represents the total number of cluster centers in the clustering result; if object x belongs to C i If there are clusters of classes, the Euclidean distance between them is calculated, and the clustering of all objects to their cluster centers is summed up. When SSE is minimum, the difference between samples in the same cluster is minimized, and the algorithm solution is completed.

[0146] Example

[0147] The following simulation example in Matlab / Simulink environment illustrates the specific implementation of this method and verifies the feasibility of the method.

[0148] The simulation system structure is as follows Figure 1 As shown in the figure, the capacity of the thermal power synchronous unit in the model is 200MW; the capacity of the wind turbine unit on the wind farm side is 60MW, the rated wind speed is 10m / s, and 10% of the reserve capacity is reserved; the total capacity of the cascade energy storage system is 60MW, which is composed of energy storage units composed of 4 battery modules. The specific settings of other parameters are shown in Table 1. The input wind speed of the wind turbine unit in the wind farm station is as follows Figure 2 Finally, the energy storage unit power average distribution method is set as a control group to verify the effectiveness of the strategy proposed in this invention.

[0149] Table 1 Specific parameter configuration in simulation verification

[0150]

[0151] The implementation plan is as follows:

[0152] In order to verify the validity of the technology patent under simulation conditions, a cascade battery energy storage system model containing multiple battery module energy storage units was constructed. Battery modules with different capacities, maximum powers, and response frequencies were selected to construct the cascade battery energy storage system model. The model includes the energy storage equipment response time constant, maximum power limit, rated capacity, SOC state and other operating characteristics as well as output control strategies.

[0153] At the same time, a system model of wind-storage combined system participating in grid-related calls is constructed to simulate application scenarios. A wind-storage combined system is formed by configuring a cascade battery energy storage system on the wind farm side, which responds to the grid power call demand after being connected to the grid. The power demand of the energy storage system is obtained as P hess =P all -P wind , where P all is the dispatch value sent by the power dispatch center to the wind farm station, P wind is the maximum power that the wind turbine can generate or reduce at the current moment, P hess It is a power instruction issued to the energy storage system.

[0154] The following is a specific implementation plan:

[0155] The first step is to determine the total effective discharge capacity of each energy storage unit according to the simulation conditions.

[0156] The total effective discharge capacity of the energy storage unit under rated conditions is expressed as follows:

[0157] Γ R =L R D R C R

[0158] In the formula, Γ R L is the total effective discharge capacity of the lithium battery in its entire life cycle; R D is the cycle life of lithium battery under rated conditions; R C is the discharge depth of lithium battery under rated conditions; R It is the capacity of lithium battery under rated conditions.

[0159] The second step is to complete the preliminary decomposition of the energy storage system power command through the parameter adaptive VMD algorithm

[0160] The parameter-adaptive VMD algorithm is used to decompose the power instructions sent to the energy storage system into a limited number of energy storage unit power instructions. The power instructions of energy storage units at different unit levels have obvious differences in some characteristics, which lays the foundation for the subsequent distribution of power instructions between energy storage units and supercapacitors.

[0161] The third step is to calculate the various operating parameters on the power signal side and the device side, and quantify the energy storage unit power instructions and the operating characteristics of different energy storage units in the energy storage system into data objects in the clustering space.

[0162] The power signal side selects the maximum charge and discharge power requirement P max 、Charge and discharge frequencyω dc 、Charge and discharge quantity E during the dispatching time dc And the effective discharge capacity d of different energy storage units in response to power instructions eff.i As a quantization parameter, it is calculated as follows:

[0163] P max，i =max{P d，i ,|P c，i |}

[0164] ω dc，i =ω k，i

[0165]

[0166]

[0167] Where P d，i , P c，i They refer to the discharge power and charging power amplitude sequences of the i-th power signal respectively; P i (t) represents the charging and discharging power amplitude at each moment in the i-th signal;

[0168] The energy storage unit equipment side selects the maximum charge and discharge power P of the energy storage unit ec，max , response frequency ω ec，dc , the available charge and discharge capacity E of the battery based on the current SOC state ec And the total effective discharge capacity Γ during the battery's entire life cycle R As quantization parameters, some parameters are calculated as follows:

[0169]

[0170] Γ R，j =L R，j D R，j C R，j

[0171] In the formula, E ec，j is the available charge and discharge capacity of the j-th energy storage unit considering the current SOC during power allocation; SOC(t) is the SOC state of the j-th energy storage unit during power allocation; SOHj is the SoH state of the j-th energy storage unit during power allocation; E ec.j.n is the rated capacity of the energy storage unit.

[0172] The fourth step is to divide the range of objects to which the power signal can be allocated according to the rigid constraints.

[0173] The determination form of the allocable energy storage unit division is as follows:

[0174] P ec，max，j ≥P max，i

[0175] E ec，j ≥E dc，i

[0176] ω ec，dc，j ≥ω dc，i

[0177] When the energy storage unit meets the above conditions, it is listed as an assignable object of the power signal, and further consideration is given to realizing power allocation through a clustering algorithm.

[0178] The fifth step is to normalize the data objects in the clustering space.

[0179] A scaling factor is introduced to normalize the data, and its form is as follows:

[0180]

[0181]

[0182] Where, X i , Y i Normalized parameter with dimension i on the signal side and the device side; x i ,y i is the original parameter with dimension i on the signal side and the device side; y imax and imin is the maximum and minimum value of the parameter with dimension i on the equipment side; μ i is the scaling factor, responsible for converting the device-side parameters to the same order of magnitude as the signal-side parameters;

[0183] The weighted coefficient matrix M is introduced when calculating the cluster distance. Finally, the calculation of the cluster distance is as follows:

[0184] M=[m1,m2,..,m n ]

[0185]

[0186] In the sixth step, the K-means clustering algorithm is used to calculate the distance between the data object and the cluster center through the Euclidean distance. The optimal solution is determined with the minimum sum of squared errors as the objective function, and the cluster center deviation distance is used as an additional termination condition to achieve clustering division between power signals and energy storage units, that is, to realize power distribution control of the cascade battery energy storage system.

[0187] The cluster center deviation judgment condition can be described as:

[0188]

[0189] In the formula, d(C o , C n ) is the Euclidean distance between the center of the new cluster generated by a certain iteration process and the center of the initial cluster; C o is the initial cluster of a certain cluster; C k It is the new cluster center generated by a certain type of cluster during the iteration process.

[0190] The sum of squared errors (SSE) is selected as the clustering effectiveness evaluation index for iterative solution. The sum of squared errors of the clustering results can be expressed as:

[0191]

[0192] In the formula, C i represents the i-th cluster center; K represents the total number of cluster centers in the clustering result; if object x belongs to C i If there are clusters of classes, the Euclidean distance between them is calculated, and the clustering of all objects to their cluster centers is summed up. When SSE is minimum, the difference between samples in the same cluster is minimized, and the algorithm solution is completed.

[0193] The power control of the cascade energy storage system designed by the present invention should first ensure the stability of the system frequency and be able to respond to the grid dispatching instructions in a timely manner. Therefore, the frequency stability of the system operation under different strategies is first compared, and the comparison results are as follows: Figure 3 As shown in the figure, it can be seen that both the proposed strategy and the control strategy can ensure that the system frequency fluctuation is within a reasonable range and can ensure the stable operation of the system. Afterwards, the parameter adaptive VMD algorithm is verified to decompose the power signal of the cascade energy storage system. The decomposition results are shown in Figure 4 As shown in the figure, the power signal received by the energy storage system is decomposed into energy storage unit power signals with obvious differences in maximum amplitude, charge and discharge capacity, and charge and discharge frequency, which lays the foundation for subsequent power distribution. Comparing the SOC state of the energy storage system when it is running after the power distribution is completed, the results are as follows: Figure 5As shown. It can be seen that after applying the strategy proposed by the present invention, the SOC state of battery modules 3 and 4 with weaker working parameters has been significantly improved, and overcharge and discharge phenomena no longer occur. In addition, the allocated power instructions are also adapted to their operating characteristics, and energy storage devices with different operating characteristics can be used more reasonably. Finally, the cumulative life loss of battery modules 3 and 4 is compared, and the results are as follows Figure 6 As shown. Since the influence of life loss is taken into account in power distribution, the life loss rate of battery modules 3 and 4 is significantly reduced after adopting the present invention. In summary, it can be explained that the application of the cascade energy storage system power control method proposed by the present invention can achieve energy storage system power control that ensures the power response capability of the energy storage system while improving the equipment operating status, improve system operation stability, reduce equipment operation loss, and improve the economy of hybrid energy storage system application.

[0194] In view of the above problems, the present invention adopts a cascade energy storage system power control method based on variational mode decomposition and K-means clustering algorithm. The VMD is used to realize the initial decomposition of the energy storage system power command to obtain the power commands of the energy storage units to be allocated with obvious differences. Then, appropriate eigenvalues ​​are selected to quantify the power signal and the energy storage equipment into sample points to be clustered and cluster centers in the clustering space, respectively. By solving the clustering problem, the power control of the energy storage system considering multiple factors is realized. While ensuring the power response capability of the system, the operating state of the energy storage equipment is effectively improved, the system operation stability is enhanced, the equipment operation loss is reduced, and the economy of the application of the hybrid energy storage system is improved.

[0195] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm, characterized in that: include: The life loss model of the cascade battery is transformed into a practical one to obtain the quantitative data of the life loss caused by the pre-assigned power command on different energy storage units; The power command received by the cascade battery energy storage system is decomposed using the variational mode decomposition algorithm to obtain the maximum power demand, charge and discharge frequency, and total charge and discharge power command sub-components during the scheduling period. Select characteristic parameters to quantify the power instructions of the energy storage units and the operating characteristics of different energy storage units in the energy storage system into data objects in the clustering space; The data objects in the clustering space are normalized, and the weighted coefficient matrix is ​​introduced to complete the clustering distance calculation. The power signal can be assigned to the object range according to the rigid constraint conditions. The specific method for normalizing the data objects in the clustering space is as follows: In the formula, X i , Y i are normalized parameters with dimension i on the signal side and the device side respectively; x i ,y i The dimensions for the signal side and the device side are i The original parameters of y imax and imin is the maximum and minimum value of the parameter with dimension i on the equipment side; μ i is the scaling factor; The calculation formula of cluster distance is: M=[m1,m2,...,m n ] In the formula, M is the weight coefficient matrix, and n is the number of data; The K-means clustering algorithm is used to complete the clustering division and reorganization between power signals and energy storage units, and realize the internal power distribution of the cascade energy storage system.

2. The power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The practical transformation of the cascade battery life loss model is shown below: Where D A Refers to the actual discharge depth during irregular discharge, D R is the discharge depth of the lithium battery under rated conditions, γ0 and γ1 are fitting coefficients, ΔT refers to the time interval between two power dispatches of the energy storage unit, P R is the rated charge and discharge power of the battery.

3. The power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The variational mode decomposition algorithm is used to decompose the power command received by the cascade battery energy storage system to obtain the power command sub-components with obvious differences in maximum power demand, charging and discharging frequency, and total charging and discharging during scheduling. The specific method is as follows: Create a sliding queue window with a time length of T. The window contains the power dispatch signal received by the cascade battery energy storage system when the station power dispatch instruction is issued; In each scheduling cycle, the data in the window is refreshed, the power scheduling instruction data at the current moment is updated to enter the window, and the last data in the queue in the window is discarded to realize the sliding of the window in the time domain. Each time the sliding is completed, a VMD decomposition is performed and the sub-signal obtained in the signal decomposition result is output.

4. The power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The method to determine the sliding window length is: The minute-level sliding window length N is selected as the initial value, and the power dispatch cycle of the cascade energy storage system in the wind farm is selected as the update step. The correlation between the sub-modes of the power signal obtained by variational mode decomposition under each window length is traversed, and the window length N corresponding to the minimum sum of correlation coefficients is selected. opt As the optimal sliding window length, the correlation coefficient selects the Pearson correlation coefficient, which is specifically defined as follows: In the formula, ρ i,n (u i ,u k ) is the i-th sub-mode u obtained by VMD decomposition under the current N value i With the kth submode u k Pearson correlation coefficient between Cov(u i ,u k ) is the sub-mode u i 、u k The covariance between Represent sub-mode u i 、u k The standard deviation of ∑ is the sum of the Pearson correlation coefficients between all sub-modalities decomposed by VMD under the N value.

5. The power control method for cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The total number of modal decompositions K during VMD decomposition is determined as follows: The upper limit of K value meets the following requirements: max{ω k (t)}≥max{ω T.j } In the formula, ω k (t) represents the center frequency of each sub-mode obtained by VMD decomposition at time t; ω T.j Indicates the response frequency of each battery energy storage unit in the field; In the formula, H i.reo is the signal obtained by recombining all modal components decomposed by VMD when K = i; H i.ori is the original signal, Cov(H i.reo ,H i.ori ) means K=i when H i.reo With H i.ori The covariance of They respectively represent the standard deviation of the recombinant signal and the original signal when K=i.

6. The power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The data objects in the clustering space include: The power signal side selects the maximum charge and discharge power requirement P max 、Charge and discharge frequencyω dc 、Charge and discharge quantity E during the dispatching time dc And the effective discharge capacity d of different energy storage units in response to power instructions eff.i As a quantization parameter, it is calculated as follows: P max,i =max{P d,i ,|P c,i |} oh dc,i =ω k,i Where P d,i , P c,i They refer to the discharge power and charging power amplitude sequences of the i-th power signal, P i (t) represents the charging and discharging power amplitude at each moment in the i-th signal, P R is the rated charge and discharge power of the battery; P i ES is the actual power of the battery during the i-th irregular charge and discharge process, D R is the discharge depth of lithium battery under rated conditions, D A Refers to the actual discharge depth during irregular discharge, P ec,i,ref It refers to the power dispatch instruction that a certain energy storage unit is expected to receive during the i-th charging and discharging process. γ0 and γ1 are both fitting coefficients. ΔT refers to the time interval between two power dispatches of the energy storage unit. ω k,i is the center frequency corresponding to the i-th submode obtained by VMD decomposition; The energy storage unit equipment side selects the maximum charge and discharge power P of the energy storage unit ec,max , response frequency ω ec,dc , the available charge and discharge capacity E of the battery based on the current SOC state ec And the total effective discharge capacity Γ during the battery's entire life cycle R As quantization parameters, some of them are calculated as follows: C R,j =L R,j D R,j C R,j In the formula, E ec,j is the available charge and discharge capacity of the jth energy storage unit considering the current SOC during power allocation; SOC j (t) is the SOC state of the jth energy storage unit during power distribution; SOH j is the SOH state of the jth energy storage unit during power distribution; E ec.j.n is the rated capacity of the energy storage unit.

7. The power control method for a cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The range of objects that can be allocated to the power signal is divided according to the rigid constraints. When the energy storage unit meets the following conditions, it is listed as an object that can be allocated to the power signal: P ec,max,j ≥P max,i AND ec,j ≥E dc,i oh ec,dc,j ≥ω dc,i Where P ec,max,j is the maximum charge and discharge power of the jth energy storage unit, E ec,j is the available charge and discharge capacity of the jth energy storage unit considering the current SOC during power allocation, P max,i is the maximum charge and discharge power requirement of the i-th power signal, E dc,i is the charge and discharge quantity within the scheduling time of the i-th power signal, ω ec,dc,j is the most responsive frequency of the jth energy storage unit, ω dc,i is the charging and discharging frequency of the i-th power signal.

8. The power control method for cascade energy storage system based on variational mode decomposition and K-means clustering algorithm according to claim 1, characterized in that: The K-means clustering algorithm is used to complete the clustering division and reorganization between power signals and energy storage units. The specific method to achieve internal power distribution of the cascade energy storage system is as follows: The data objects quantized on the device side are used as the initial cluster centers for iterative solution. When the new cluster center of a certain cluster deviates from the initial cluster center by a certain distance ε during the iteration, the center of the cluster is no longer updated and the data objects of the cluster are no longer reclassified. The error sum of squares SSE is selected as the clustering effectiveness evaluation index for iterative solution. The error sum of squares of the clustering results is expressed as: In the formula, C i represents the i-th cluster center; K represents the total number of cluster centers in the clustering result; if object x belongs to C i The Euclidean distance between the two is calculated, and the clustering of all objects to their cluster centers is summed up. When the SSE is the smallest, the sample difference in the same cluster is minimized, and the clustering division and reorganization between the power signal and the energy storage unit are completed.

Images