Fast nonlinear model predictive control method and device for hybrid energy storage system

Abstract

The invention discloses a fast nonlinear model prediction control method and device for a hybrid energy storage system, and the method comprises the steps: predicting the net current needed by the hybrid energy storage system at a future moment according to the net current of the hybrid energy storage system at a historical moment; according to the predicted net current and real-time sampling information of the hybrid energy storage system, a low-resolution state is searched in a state space of the hybrid energy storage system in combination with a target cost function, a first optimal path is obtained, and the target cost function is related to system efficiency, battery life and supercapacitor voltage regulation; searching a high-resolution state within a certain range around the first optimal path in the state space based on the target cost function to obtain a secondary optimal path; and determining a control input instruction of the hybrid energy storage system according to the secondary optimal path, and controlling the hybrid energy storage system according to the control input instruction.

Description

Technical Field

[0001] The present application relates to the field of control technology, and in particular to a fast nonlinear model predictive control method and device for a hybrid energy storage system. Background Art

[0002] Hybrid energy storage systems (HESSs), composed of lithium batteries and supercapacitors, combine the advantages of high power density and high energy density. They are primarily used to absorb high-frequency, large fluctuations in wave power generation, thereby coordinating the output of electricity that meets grid demand. When using a hybrid energy storage system, a specific energy management strategy (EMS) is required to determine the power distribution mechanism for the lithium batteries and supercapacitors in the HESS. Currently, the EMS commonly used in the field of wave power generation is mainly rule-based strategies represented by filtering methods. This type of EMS has low computational complexity and is easy to implement, but it relies heavily on expert experience and is a suboptimal control strategy. In contrast, model predictive control (MPC) can explicitly consider constraints and multi-objective optimization, and it is expected that more advanced EMSs will be developed based on MPC.

[0003] Existing MPC-based EMSs typically use various linearization methods to address the nonlinear characteristics of HESSs, adapting them to the linear MPC (LMPC) solution framework. This approach has the advantage of improving real-time computation speed, but also comes with a loss of control performance due to the linearization process. Summary of the Invention

[0004] To this end, this application discloses the following technical solutions:

[0005] In a first aspect, the present application provides a fast nonlinear model predictive control method for a hybrid energy storage system, comprising:

[0006] Predicting the required output net current of the hybrid energy storage system at a future time based on the net current of the hybrid energy storage system at a historical time, i.e., predicting the net current;

[0007] searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, wherein the target cost function is related to system efficiency, battery life, and supercapacitor voltage regulation;

[0008] Based on the target cost function, searching the state space for a high-resolution state within a certain range around the first optimal path to obtain a second optimal path;

[0009] A control input instruction of the hybrid energy storage system is obtained according to the quadratic optimal path, and the hybrid energy storage system is controlled according to the control input instruction.

[0010] Optionally, searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path includes:

[0011] Determining an initial state in a state space of the hybrid energy storage system according to the predicted net current and real-time sampling information of the hybrid energy storage system;

[0012] Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated according to the low-resolution states at the previous moments, and for each low-resolution state at the moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed;

[0013] An optimal low-resolution state is determined according to the target cost function value among multiple low-resolution states corresponding to the last moment in the prediction time domain, and starting from the optimal low-resolution state at the last moment, the low-resolution states corresponding to the previous moment are determined one by one to obtain the first optimal path.

[0014] Optionally, searching for a low-resolution state at a previous moment corresponding to the low-resolution state at each moment includes:

[0015] For the low-resolution state at each moment, when there are M or more low-resolution states within a search range at a moment before the low-resolution state, searching for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range;

[0016] For the low-resolution state at each moment, when the number of low-resolution states within the search range at the previous moment of the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

[0017] Optionally, searching for a corresponding low-resolution state at a previous moment from among the M low-resolution states within the search range includes:

[0018] When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current;

[0019] When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current;

[0020] When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

[0021] Optionally, obtaining a control input instruction for the hybrid energy storage system according to the quadratic optimal path includes:

[0022] determining an optimal control sequence of the hybrid energy storage system according to the quadratic optimal path;

[0023] The first item of the optimal control sequence is obtained as a control input instruction of the hybrid energy storage system.

[0024] A second aspect of the present application provides a fast nonlinear model predictive control device for a hybrid energy storage system, comprising:

[0025] an obtaining unit, configured to predict the required output net current of the hybrid energy storage system at a future moment, i.e., predict the net current, based on the net current of the hybrid energy storage system at a historical moment;

[0026] Search unit for:

[0027] searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, wherein the target cost function is related to system efficiency, battery life, and supercapacitor voltage regulation;

[0028] Based on the target cost function, searching the state space for a high-resolution state within a certain range around the first optimal path to obtain a second optimal path;

[0029] A control unit is used to obtain a control input instruction of the hybrid energy storage system according to the quadratic optimal path, and control the hybrid energy storage system according to the control input instruction.

[0030] Optionally, the search unit searches for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, specifically for:

[0031] Determining an initial state in a state space of the hybrid energy storage system according to the predicted net current and real-time sampling information of the hybrid energy storage system;

[0032] Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated according to the low-resolution states at the previous moments, and for each low-resolution state at the moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed;

[0033] An optimal low-resolution state is determined according to the target cost function value among multiple low-resolution states corresponding to the last moment in the prediction time domain, and starting from the optimal low-resolution state at the last moment, the low-resolution states corresponding to the previous moment are determined one by one to obtain the first optimal path.

[0034] Optionally, when the search unit searches for the low-resolution state at each moment and the low-resolution state at the previous moment corresponding to the low-resolution state, the search unit is specifically configured to:

[0035] For the low-resolution state at each moment, when there are M or more low-resolution states within a search range at a moment before the low-resolution state, searching for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range;

[0036] For the low-resolution state at each moment, when the number of low-resolution states within the search range at the previous moment of the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

[0037] Optionally, when the search unit searches for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range, it is specifically configured to:

[0038] When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current;

[0039] When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current;

[0040] When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

[0041] Optionally, when the control unit obtains the control input instruction of the hybrid energy storage system according to the quadratic optimal path, it is specifically used to:

[0042] determining an optimal control sequence of the hybrid energy storage system according to the quadratic optimal path;

[0043] The first item of the optimal control sequence is obtained as a control input instruction of the hybrid energy storage system.

[0044] The beneficial effect of the present application is that when obtaining the control input instructions of the hybrid energy storage system through the forward dynamic programming method, the low-resolution state in the state space is first searched, and then the state near the first optimal path based on the first optimal path composed of the low-resolution states is searched to obtain the secondary optimal path. This can avoid the direct search of the complex state space, reduce the number of searches, and achieve the effect of reducing the amount of calculation and shortening the time spent on obtaining the control input instructions; on the other hand, this scheme combines the cost functions related to the system efficiency, battery life and supercapacitor voltage to perform state search, thereby realizing multi-objective joint optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0046] Figure 1 This invention provides a fast nonlinear model predictive control method for a hybrid energy storage system.

[0047] Figure 2 is a schematic diagram of an equivalent circuit of a hybrid energy storage system provided in an embodiment of the present application;

[0048] Figure 3 This is a schematic diagram of the principle of a fast FDP algorithm provided in an embodiment of the present application;

[0049] Figure 4 This is a flowchart of a fast FDP algorithm provided in an embodiment of the present application;

[0050] Figure 5 This is a structural diagram of a fast nonlinear model predictive control device for a hybrid energy storage system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] This embodiment provides a fast nonlinear model predictive control method for a hybrid energy storage system. Figure 1 , is a flowchart of the method, which may include the following steps.

[0053] S101, predicting the required output net current of the hybrid energy storage system at a future moment based on the net current of the hybrid energy storage system at a historical moment, that is, predicting the net current.

[0054] S102, based on the predicted net current and the real-time sampling information of the hybrid energy storage system, combined with the target cost function, searching for low-resolution states in the state space of the hybrid energy storage system to obtain the first optimal path, wherein the target cost function is related to system efficiency, battery life and supercapacitor voltage regulation.

[0055] S103 , based on the target cost function, searching the state space for a high-resolution state within a certain range around the first optimal path to obtain a second optimal path.

[0056] S104 , obtaining a control input instruction of the hybrid energy storage system according to the quadratic optimal path, and controlling the hybrid energy storage system according to the control input instruction.

[0057] The beneficial effect of the present application is that when obtaining the control input instructions of the hybrid energy storage system through the forward dynamic programming method, the low-resolution state in the state space is first searched, and then the state near the first optimal path based on the first optimal path composed of the low-resolution states is searched to obtain the secondary optimal path. This can avoid the direct search of the complex state space, reduce the number of searches, and achieve the effect of reducing the amount of calculation and shortening the time spent on obtaining the control input instructions; on the other hand, this scheme combines the cost functions related to the system efficiency, battery life and supercapacitor voltage to perform state search, thereby realizing multi-objective joint optimization.

[0058] The control method of this embodiment can be achieved by Figure 2 The controller shown in FIG. 1 is executed, and the connection relationship between the controller and the hybrid energy storage system can be used to Figure 2 Equivalent circuit representation of the hybrid energy storage system shown.

[0059] The controller may be a nonlinear model predictive control (NMPC) controller.

[0060] like Figure 2 As shown in Figure 1, a hybrid energy storage system (HESS) can include two modules: a supercapacitor (SC) and a lithium battery (BAT). The SC is connected to the DC bus through a supercapacitor DC converter (DC-DC converter). The equivalent circuit of the SC can be an ideal capacitor and a resistor (R sc)、Inductance(L SC ) series circuit, where the voltage across the ideal capacitor is denoted as u c The BAT is connected to the DC bus through a lithium battery DC converter (DC-DC converter). The equivalent circuit of the BAT can be regarded as an ideal voltage source and a resistor (R b )、Inductance(L b ) in series, the voltage provided by the ideal voltage source can be recorded as u oc . Figure 2 In the figure, d represents the width of the pulse signal output by the pulse width modulation.

[0061] SC is used to stabilize the DC bus voltage u dc , BAT is responsible for tracking the optimal current increment Δi calculated by the controller b * The control method of this embodiment can be applied to the wave power generation scenario. In this scenario, the wave power generation device and the grid-side system can be regarded as equivalent current sources, wherein the wave power generation device can be regarded as an equivalent current source i wave , the grid side system can be regarded as an equivalent current source i grid .i grid with i wave The difference between net Represents the net output current required by the HESS.

[0062] The controller can be connected to the prediction module, which can be connected to the storage module. In step S101, the storage module can store historical moments, that is, store the net current output by the HESS at multiple recent moments. The prediction module can obtain the net current at these historical moments from the storage module and determine the predicted net current i that the hybrid energy storage system needs to output at a future moment based on the net current at the historical moments. netp The working principle of the prediction module can be found in the related prior art and will not be described in detail in this embodiment.

[0063] The controller can determine the optimal battery current increment based on the predicted net current and real-time sampling information, and control the current of the lithium battery in the HESS based on the optimal battery current increment, thereby distributing the net current between the lithium battery and the supercapacitor.

[0064] like Figure 2 As shown, the real-time sampling information obtained may include u at the current moment sc 、u b 、i sc and i b The control input command that needs to be determined can be the optimal current increment Δi of BAT b * .u sc and i scRepresent the voltage and current of the supercapacitor, u b and i b The voltage and current of the lithium battery respectively.

[0065] In this embodiment, the process of determining the optimal current increment based on the real-time sampling information and the predicted net current can be regarded as a process of solving the optimal control problem (OCP) based on the real-time sampling information and the predicted net current. The optimal control problem can be expressed as: on the basis of satisfying the following formulas (1) to (3), Q in formula (4) is EMS (k) is minimized. The meaning of the parameters in formula (4) can be found in formulas (5) to (7).

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073] In the above formula, Q loss (k) represents the cost function used to reduce system energy loss at time k, Q b (k) represents the cost function for extending the battery life at time k, Q p (k) represents the cost function for managing the supercapacitor voltage at time k. w1 to w4 represent the preset weight factors, and N represents the prediction time domain of NMPC. Its specific value can be pre-set as needed and is not limited. EMS (k) represents the sum of the above cost functions, that is, the total cost function. For any possible state at time k+i in the prediction time domain N, the target cost function O(k+i|k) can be defined as the following formula (8):

[0074]

[0075] P sc Indicates the supercapacitor discharge power, P b Indicates the discharge power of lithium battery.

[0076] u screfrepresents the reference voltage of the supercapacitor, k+i|k represents the corresponding variable value at time k+i predicted based on the information at time k, for example, x(k+i|k) represents the variable value of the x variable at time k+i predicted based on the information at time k, η1 represents the efficiency of the DC-DC converter connected to the supercapacitor, and η2 represents the efficiency of the DC-DC converter connected to the lithium battery.

[0077] x represents the state vector of NMPC, that is, x=[u sc ,u b ,i sc ,i b ] T , u represents the control variable of NMPC, that is, the current increment Δi of the lithium battery b ,Ω u (k) represents the permissible control set of the control variable u at time k, w represents the disturbance of NMPC, i.e., the predicted value of the net output current of HESS i netp , that is, w=i netp f() represents the nonlinear state space equation of HESS, x max and x min represent the upper and lower limits of the state vector respectively.

[0078] Q EMS (k) refers to the total cost function in the prediction time domain (1 to N) at the real time k, and its function value is related to the system efficiency, battery life and supercapacitor voltage regulation. Each time k+i (i∈[1, N]) in the prediction time domain can be called a virtual time. The purpose of the method of this embodiment is to determine a value such that Q EMS (k) Minimizing the optimal current increment (i.e., the control input instruction of step S104). To achieve this goal, when executing the method of this embodiment, the target cost function value of the possible states at each virtual moment in the prediction time domain can be calculated to determine the optimal current increment.

[0079] The prediction time domain is a preset parameter that defines the time range covered by the state space. For example, if the prediction time domain is set to 10, the state space can include 10 consecutive moments after the current moment (excluding the current moment).

[0080] The interval between each moment is equal in length and can be set as needed without restriction.

[0081] In this embodiment, in order to solve the above OCP problem and obtain appropriate control input instructions, a forward dynamic programming (FDP) algorithm may be used for the solution. Steps S102 and S103 are equivalent to the process of solving the above problem based on the FDP algorithm.

[0082] When solving the above OCP problem, the method of this embodiment can select the lithium battery current i b As a state variable for dynamic programming. Figure 3 Each circle represents a state. The low-resolution state in step S102 can be represented by Figure 3 The orange circle shown indicates the high resolution state of step S103, which can be Figure 3 The black circles shown are indicated.

[0083] In this embodiment, the low-resolution state can be defined as:

[0084] In the value range of the state variable [i b1 ,i bM ], the variable values ​​of multiple state variables are determined within the range according to a larger division interval Y1, and the states corresponding to the variable values ​​divided according to the interval of Y1 can be regarded as low-resolution states.

[0085] by Figure 3 For example, the upper limit of the value range i bM The corresponding state is used as a low-resolution state, and the upper limit i bM -Y1 corresponds to another low-resolution state, and the upper limit i bM The state corresponding to -2*Y1 is another low-resolution state.

[0086] Correspondingly, the high-resolution state can be defined as dividing the variable value of at least one state variable between the variable values ​​corresponding to two adjacent low-resolution states according to a smaller dividing interval Y2. The states corresponding to the variable values ​​divided according to the interval Y2 can be regarded as high-resolution states.

[0087] by Figure 3 For example, you can use i bM to i bM -The interval of Y1 is divided by Y2, and the state corresponding to the divided variable value is regarded as i bM to i bM -High resolution state between Y1.

[0088] Y2 is smaller than Y1. The values ​​of Y1 and Y2 can be set as needed. As an example, in this embodiment, Y1 can be equal to the maximum value of |u(k+i-1|k)|, that is, the maximum value of the increment of the lithium battery at time k+i-1, and Y2 can be equal to one half or one third of Y1.

[0089] In some embodiments, the high-resolution state divided by the smaller interval Y2 may include the low-resolution state divided by the larger interval.

[0090] In step S102, the first best path can be searched and obtained as follows:

[0091] 1. Determine the initial state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system;

[0092] 2. Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated based on the low-resolution state at each previous moment. Moreover, for each low-resolution state at a moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed;

[0093] 3. Determine an optimal low-resolution state among the multiple low-resolution states corresponding to the last moment of the prediction time domain according to the target cost function value, and starting from the optimal low-resolution state at the last moment, determine the low-resolution states corresponding to the previous moment one by one to obtain the first optimal path.

[0094] In step 1, the state at time k can be determined based on the real-time information obtained, that is, the initial state x(k|k)=[u sc (k|k),u b (k|k), i sc (k|k), i b (k|k)].

[0095] In step 2, starting from the initial state, according to the above formulas (1) to (7), the target cost function value of the low-resolution state at the subsequent moment can be calculated one by one using the state at the previous moment. That is, the target cost function value O(k|k) of the initial state at moment k can be calculated first, and then the target cost function value O(k+1|k) of the low-resolution state at moment k+1 can be calculated, and the target cost function value O(k+2|k) of the low-resolution state at moment k+2 can be calculated, and so on, until the target cost function value of each low-resolution state in the state space is calculated.

[0096] Taking the k+i moment as an example, for a low-resolution state at a moment, the method of searching for the low-resolution state at the previous moment corresponding to the low-resolution state can be:

[0097] Search each low-resolution state within the search range of the previous moment (i.e., moment k+i-1) to determine a low-resolution state at moment k+i-1 corresponding to the low-resolution state at moment k+i, and then determine the optimal control variable that makes the following formula (9) valid based on this corresponding state.

[0098]

[0099] Among them, Q(k+i|k) represents the single-step cost function of transferring x(k+i-1|k) to x(k+i|k) using u(k+i-1|k), V(k+i|k) represents the minimum cost from x(k|k) to x(k+i-1|k), and u * (k+i-1|k) represents the optimal control variable at time k+i-1, that is, Δi b * (k+i-1|k). Combined Figure 3 , by making any i at time k+i b The forward recursion of the state can determine the optimal control variable at time k+i-1. N represents the total number of time periods included in the prediction horizon, that is, the prediction horizon includes N time periods after the current time (time k).

[0100] Specifically, under the premise of determining the low-resolution state at time k+i, for each low-resolution state within the search range at time k+i-1, the Q(k+i|k)+V(k+i|k) of the state at time k+i-1 to the state at time k+i can be calculated. For different states at time k+i-1, the value of the formula is different, so in step 2, the value of the formula corresponding to each state at time k+i-1 in the search range can be obtained, and the state with the smallest value is determined as the state at time k+i-1 corresponding to the low-resolution state at time k+i, and the lithium battery current increment when the corresponding state at time k+i-1 is converted to the low-resolution state at time k+i is the optimal control variable u at time k+i-1. * (k+i-1|k).

[0101] Among them, the search range at time k+i-1 is determined according to the low-resolution state at time k+i. Specifically, if the absolute value of the lithium battery current increment of a state A at time k+i exceeds the preset upper limit of the lithium battery current increment compared with a state B at time k+i-1, it can be considered that the state B at time k+i-1 cannot be converted to the state A at time k+i. If it does not exceed the upper limit of the lithium battery current increment, it can be considered that the state B at time k+i-1 can be converted to the state A at time k+i. For a specific low-resolution state A at time k+i, the search range at time k+i-1 can be the set of all low-resolution states at time k+i-1 that can be converted to state A. For example, if there are 10 low-resolution states at time k+i-1 that can be converted to state A, then when searching for the corresponding low-resolution state in the previous moment for state A, the search range is these 10 low-resolution states.

[0102] In step 3, the target cost function value of each low-resolution state in the last moment of the prediction time domain can be traversed, and the low-resolution state with the smallest target cost function value can be determined as the optimal low-resolution state. Then, starting from the optimal low-resolution state at the last moment, the low-resolution state corresponding to the previous moment can be determined one by one to obtain the first optimal path consisting of the optimal low-resolution state and the corresponding low-resolution state in the previous moment.

[0103] For example, assuming that the last moment is k+10, after determining the optimal low-resolution state at k+10, the search method of the aforementioned step 2 can be used to obtain the state corresponding to the optimal low-resolution state at k+9, which is recorded as the optimal state at k+9. Then, the search method is used to obtain the optimal state corresponding to k+9 at k+8, which is recorded as the optimal state at k+8. This process is repeated until the initial state is found. The path formed by the initial state and the optimal states from k+1 to k+10 is the first optimal path mentioned above.

[0104] Optionally, to further improve search efficiency, an early stopping strategy can be introduced into the above search process. Specifically, if the number of low-resolution states within the search range exceeds M, the search can be limited to a maximum of M states each time the state at time k+i-1 is searched. M can be a preset parameter whose value can be set as needed and is not limited. For example, if the target range contains 10 states (including high-resolution and low-resolution states), M can be set to 5 or 6.

[0105] That is, in step 2, for each low-resolution state at a moment, searching for the low-resolution state at the previous moment corresponding to the low-resolution state may be performed in the following manner:

[0106] For each low-resolution state at a moment, if there are M or more low-resolution states within the search range at the moment before the low-resolution state, search the M low-resolution states within the search range for the corresponding low-resolution state at the moment before.

[0107] For each low-resolution state at a moment, when the number of low-resolution states within the search range at a moment before the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

[0108] When the early stopping strategy is introduced and the number of low-resolution states included in the search range is greater than M, the search can be performed randomly for M states within the search range, or at equal intervals for M states within the search range. Alternatively, M states can be continuously searched within the searchable range in a specific order based on the real-time sampling information obtained.

[0109] For example, the corresponding low-resolution state at the previous moment can be searched from the M low-resolution states within the search range as follows:

[0110] When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current;

[0111] When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current;

[0112] When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

[0113] For example, suppose that it is necessary to search for a state corresponding to the low-resolution state at time k+i at time k+i-1, where the value of the state variable of the low-resolution state at time k+i is i b5 , assuming that the search range at time k+i-1 contains all low-resolution states, the corresponding maximum and minimum values ​​of the state variables are i b6 and i b4 , that is, the interval of the state variable corresponding to the search range is [i b4 ,i b6 ].

[0114] When the real-time sampling information meets the first search condition, you can first search for i b6 The corresponding state, obtain the V+Q of the state, and then search i in turn b6 -Y1 corresponding state, i b6 -2*Y1 corresponding state, and so on, until the search is for i b6 -(M-1)*Y1 corresponding state,

[0115] When the real-time sampling information meets the second search condition, you can first search for i b4 The corresponding state, obtain the V+Q of the state, and then search i in turn b4 +Y1 corresponding state, i b4 +2*Y1 corresponding state, and so on, until the search is for i b4 + (M-1) * Y1 corresponding state.

[0116] When the real-time sampling information meets the third search condition, you can first search for i b5 The corresponding state, obtain the V+Q of the state, and then search i in turn b5 +Y1 corresponding state, i b5+2*Y1 corresponding state, and so on, until the search is for i b5 + (M-1) / 2 * Y1 corresponding state, and then search i in turn b5 -Y1 corresponding state, i b5 -2*Y1 corresponding state, and so on, until the search is for i b5 -(M-1) / 2*Y1 corresponding state.

[0117] The first search condition may be that the supercapacitor voltage in the real-time sampling information is less than the low voltage threshold u L The second search condition can be that the supercapacitor voltage in the real-time sampling information is greater than the high voltage threshold u H The third search condition may be that the supercapacitor voltage in the real-time sampling information is greater than or equal to a low voltage threshold and less than or equal to a high voltage threshold.

[0118] In summary, when the supercapacitor voltage is less than the low voltage threshold u L When i b , so the state space is searched from top to bottom until the number of searches increases to M. When the supercapacitor voltage is greater than the high voltage threshold u H In other cases, the state space is searched from the center to both sides.

[0119] It should be noted that the above search order can also be applied without introducing the early stopping strategy, that is, when the first search condition is met, each state in the search range is traversed in descending order starting from the upper limit of the corresponding state variable; when the second search condition is met, each state in the search range is traversed in ascending order starting from the lower limit; when the third search condition is met, each state in the search range is traversed from the best state at the next moment to both sides.

[0120] After obtaining the first best path, the target cost function can be combined with the method of obtaining the first best path described above to traverse multiple high-resolution states adjacent to the first best path at each moment to obtain the second best path.

[0121] Specifically, the method for obtaining the quadratic optimal path may be:

[0122] Starting from the initial state, the target cost function value of the high-resolution state adjacent to the first optimal path at the next moment is calculated one by one according to the state at the previous moment;

[0123] For each moment, the high-resolution state adjacent to the first best path at that moment may include: the state where the first best path is located at that moment is the center, and the two adjacent low-resolution states on both sides of the center are used as boundaries. All high-resolution states contained in this range are considered to be the high-resolution states adjacent to the first best path at that moment.

[0124] For example, assume that the value of the state variable of the optimal state corresponding to the first optimal path at time k+i is i b5 , with this as the center, the state variables of the two adjacent low-resolution states on both sides of the center are i b4 =i b5 -Y1,i b6 =i b5 +Y1, then all state variables at time k+i are in [i b4 ,i b6 ] interval can be used as the high-resolution state adjacent to the first optimal path at time k+i.

[0125] After calculating each high-resolution state adjacent to the first optimal path in the state space, for each high-resolution state at a moment, search for the high-resolution state at the previous moment corresponding to the high-resolution state until every high-resolution state adjacent to the first optimal path in the state space is traversed.

[0126] The above-mentioned search method can refer to the search method for obtaining the first best path in the above-mentioned embodiment, and will not be described in detail.

[0127] When searching for a high-resolution state in the above-mentioned search manner, the search range may include a high-resolution state that is adjacent to the first optimal path at a previous moment and can be converted to a next moment.

[0128] Optionally, when searching for the quadratic optimal path, if the number of high-resolution states within the search range exceeds M, the aforementioned early stopping strategy and related search order may be introduced. For details, please refer to the previous article and will not be repeated here.

[0129] After the search is complete, the optimal high-resolution state with the minimum target cost function value can be determined from all high-resolution states adjacent to the initial optimal path at the last moment in the prediction time domain. Then, starting from the last moment, the high-resolution states corresponding to the previous moments can be determined one by one. In the above example, assuming that the last moment is time k+10, after determining the optimal high-resolution state adjacent to the initial optimal path at time k+10, the optimal high-resolution state at time k+9 corresponding to the optimal high-resolution state at time k+10 can be obtained in sequence, and the optimal high-resolution state at time k+8 corresponding to the optimal high-resolution state at time k+9 can be obtained in sequence, until returning to the initial state. The path formed by the initial state and the optimal high-resolution states from time k+1 to time k+10 is the aforementioned secondary optimal path.

[0130] In summary, the method of this embodiment uses the FDP method to solve the OCP problem. The FDP is repeatedly used twice in the state space. When the FDP is used for the first time, only the low-resolution states in the state space are traversed to obtain the first optimal path. When the FDP is used for the second time, only the high-resolution states near the first optimal path are searched to obtain the second optimal path. The advantage of this is that it can avoid the direct search of the complex state space, thereby improving the search efficiency. The search process of the above steps S102 and S103 can be used Figure 4 Flowchart representation of .

[0131] In step S104 , an optimal control sequence of the hybrid energy storage system may be determined according to the quadratic optimal path, and the first item of the optimal control sequence may be obtained as a control input instruction of the hybrid energy storage system.

[0132] Specifically, based on the obtained quadratic optimal path, the optimal control variable when the optimal high-resolution state at each moment is converted to the optimal high-resolution state at the next moment can be obtained, that is, the optimal control variable u when the initial state at time k is converted to the optimal high-resolution state at time k+1 can be determined. * (k|k), the optimal control variable u when the optimal high-resolution state at time k+1 is converted to the optimal high-resolution state at time k+2 * (k+1|k), the optimal control variable u when the optimal high-resolution state at time k+2 is converted to the optimal high-resolution state at time k+3 * (k+2|k), and so on, until the optimal control variable u is obtained when the optimal high-resolution state at time k+N-1 is converted to the optimal high-resolution state at time k+N. * (k+N-1|k), these optimal control variables constitute the optimal control sequence: U*(k)=[u * (k|k),u *(k+1|k), ...u * (k+N-1|k)]. Where k+N is the last moment in the prediction time domain.

[0133] When controlling the hybrid energy storage system, the first control variable in the optimal control sequence can be obtained to control the hybrid energy storage system. Since the current increment of the lithium battery is used as the control variable in this embodiment, in S104, the first control variable u in the optimal control sequence can be used to control the hybrid energy storage system. * (k|k) controls the lithium battery current of the hybrid energy storage system. For example, at time k+1, the lithium battery current is adjusted to i b (k)+u * (k|k), i b (k) represents the discharge current i of the lithium battery at time k b ,i b (k) can be obtained from real-time sampling information.

[0134] Through the above control method, this embodiment can take into account multiple competing control objectives, and maximize the utilization of the supercapacitor while ensuring that the supercapacitor voltage meets the constraints, and can achieve a rapid solution to the nonlinear programming problem corresponding to OCP in NMPC.

[0135] The present application also provides a fast nonlinear model predictive control device for a hybrid energy storage system. Figure 5 , the device may include the following units.

[0136] An obtaining unit 501 is configured to predict the required output net current of the hybrid energy storage system at a future time based on the net current of the hybrid energy storage system at a historical time, i.e., predict the net current;

[0137] The search unit 502 is configured to:

[0138] Based on the predicted net current and real-time sampling information of the hybrid energy storage system, combined with the target cost function, a low-resolution state is searched in the state space of the hybrid energy storage system to obtain the first optimal path. The target cost function is related to system efficiency, battery life and supercapacitor voltage regulation.

[0139] Based on the target cost function, the high-resolution state within a certain range around the first optimal path is searched in the state space to obtain the second optimal path;

[0140] The control unit 503 is configured to obtain a control input instruction of the hybrid energy storage system according to the quadratic optimal path, and control the hybrid energy storage system according to the control input instruction.

[0141] Optionally, the search unit 502 searches for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with the target cost function to obtain the first optimal path, specifically for:

[0142] determining an initial state in a state space of the hybrid energy storage system according to the predicted net current and real-time sampling information of the hybrid energy storage system;

[0143] Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated based on the low-resolution states at each previous moment. Moreover, for each low-resolution state at a moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed;

[0144] An optimal low-resolution state is determined according to the target cost function value among multiple low-resolution states corresponding to the last moment in the prediction time domain. Starting from the optimal low-resolution state at the last moment, the low-resolution states corresponding to the previous moment are determined one by one to obtain the first optimal path.

[0145] Optionally, when the search unit 502 searches for the low-resolution state at each moment and the low-resolution state at the previous moment corresponding to the low-resolution state, it is specifically configured to:

[0146] For each low-resolution state at a moment, if there are M or more low-resolution states within the search range at the moment before the low-resolution state, search the M low-resolution states within the search range for the corresponding low-resolution state at the moment before.

[0147] For each low-resolution state at a moment, when the number of low-resolution states within the search range at a moment before the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

[0148] Optionally, when searching for the corresponding low-resolution state at the previous moment from among the M low-resolution states within the search range, the search unit 502 is specifically configured to:

[0149] When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current;

[0150] When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current;

[0151] When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

[0152] Optionally, when the control unit 503 obtains the control input instruction of the hybrid energy storage system according to the quadratic optimal path, it is specifically used to:

[0153] Determine the optimal control sequence of the hybrid energy storage system based on the quadratic optimal path;

[0154] The first item of the optimal control sequence is obtained as the control input instruction of the hybrid energy storage system.

[0155] The working principle of the fast nonlinear model predictive control device for a hybrid energy storage system provided in this embodiment can be found in the relevant steps of the fast nonlinear model predictive control method for a hybrid energy storage system provided in this embodiment, and will not be repeated here.

[0156] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other.

[0157] For the convenience of description, the above systems or devices are described as being divided into various modules or units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0158] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0159] Finally, it should be noted that, in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0160] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A fast nonlinear model predictive control method for a hybrid energy storage system, characterized in that: include: Predicting the required output net current of the hybrid energy storage system at a future time based on the net current of the hybrid energy storage system at a historical time, i.e., predicting the net current; searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, wherein the target cost function is related to system efficiency, battery life, and supercapacitor voltage regulation; Based on the target cost function, searching the state space for a high-resolution state within a certain range around the first optimal path to obtain a second optimal path; A control input instruction of the hybrid energy storage system is obtained according to the quadratic optimal path, and the hybrid energy storage system is controlled according to the control input instruction.

2. The method according to claim 1, characterized in that The step of searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path includes: Determining an initial state in a state space of the hybrid energy storage system according to the predicted net current and real-time sampling information of the hybrid energy storage system; Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated according to the low-resolution states at the previous moments, and for each low-resolution state at the moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed; An optimal low-resolution state is determined according to the target cost function value among multiple low-resolution states corresponding to the last moment in the prediction time domain, and starting from the optimal low-resolution state at the last moment, the low-resolution states corresponding to the previous moment are determined one by one to obtain the first optimal path.

3. The method according to claim 2, characterized in that The step of searching for a low-resolution state at a previous moment corresponding to the low-resolution state at each moment includes: For the low-resolution state at each moment, when there are M or more low-resolution states within a search range at a moment before the low-resolution state, searching for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range; For the low-resolution state at each moment, when the number of low-resolution states within the search range at the previous moment of the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

4. The method according to claim 3, characterized in that The searching for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range includes: When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current; When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current; When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

5. The method according to claim 1, wherein Obtaining a control input instruction for the hybrid energy storage system according to the quadratic optimal path includes: determining an optimal control sequence of the hybrid energy storage system according to the quadratic optimal path; The first item of the optimal control sequence is obtained as a control input instruction of the hybrid energy storage system.

6. A fast nonlinear model predictive control device for a hybrid energy storage system, characterized in that: include: an obtaining unit, configured to predict the required output net current of the hybrid energy storage system at a future moment, i.e., predict the net current, based on the net current of the hybrid energy storage system at a historical moment; Search unit for: searching for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, wherein the target cost function is related to system efficiency, battery life, and supercapacitor voltage regulation; Based on the target cost function, searching the state space for a high-resolution state within a certain range around the first optimal path to obtain a second optimal path; A control unit is used to obtain a control input instruction of the hybrid energy storage system according to the quadratic optimal path, and control the hybrid energy storage system according to the control input instruction.

7. The device according to claim 6, characterized in that The search unit searches for a low-resolution state in the state space of the hybrid energy storage system based on the predicted net current and the real-time sampling information of the hybrid energy storage system in combination with a target cost function to obtain a first optimal path, specifically for: Determining an initial state in a state space of the hybrid energy storage system according to the predicted net current and real-time sampling information of the hybrid energy storage system; Starting from the initial state, the target cost function value of the low-resolution state at the next moment is calculated according to the low-resolution states at the previous moments, and for each low-resolution state at the moment, the low-resolution state at the previous moment corresponding to the low-resolution state is searched until every low-resolution state in the state space is traversed; An optimal low-resolution state is determined according to the target cost function value among multiple low-resolution states corresponding to the last moment in the prediction time domain, and starting from the optimal low-resolution state at the last moment, the low-resolution states corresponding to the previous moment are determined one by one to obtain the first optimal path.

8. The device according to claim 7, characterized in that When the search unit searches for the low-resolution state at each moment and the low-resolution state at the previous moment corresponding to the low-resolution state, the search unit is specifically configured to: For the low-resolution state at each moment, when there are M or more low-resolution states within a search range at a moment before the low-resolution state, searching for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range; For the low-resolution state at each moment, when the number of low-resolution states within the search range at the previous moment of the low-resolution state is less than or equal to M, the corresponding low-resolution state at the previous moment is searched within the search range.

9. The device according to claim 8, characterized in that When the search unit searches for the corresponding low-resolution state at the previous moment from the M low-resolution states within the search range, it is specifically configured to: When the real-time sampling information satisfies the first search condition, starting from the upper limit of the search range, searching for M low-resolution states in descending order of lithium battery current; When the real-time sampling information satisfies the second search condition, starting from the lower limit of the search range, searching for M low-resolution states in increasing order of lithium battery current; When the real-time sampling information satisfies the third search condition, starting from the middle of the search range, M low-resolution states are searched in increasing and decreasing order of lithium battery current.

10. The device according to claim 6, characterized in that When the control unit obtains the control input instruction of the hybrid energy storage system according to the quadratic optimal path, it is specifically used to: determining an optimal control sequence of the hybrid energy storage system according to the quadratic optimal path; The first item of the optimal control sequence is obtained as a control input instruction of the hybrid energy storage system.

Images