Self-optimization-approaching control method applied to photovoltaic and electric vehicles

Abstract

The invention relates to a self-optimization control method applied to photovoltaic and electric vehicles. The method comprises the steps that mathematical representation models are constructed for self-optimization response constraints of the photovoltaic and electric vehicles respectively; constructing a self-optimization-approaching response model through the objective function; in the objective function, a loss function is used to describe the utility of the photovoltaic and electric vehicles; based on a primal-dual gradient projection method, proposing excitation feedback information required by a self-optimization algorithm, iteratively realizing distributed self-optimization real-time response of the electric vehicle and the photovoltaic system, solving the self-optimization algorithm, and simultaneously meeting the self-adjustment requirements of a distribution network and a load. According to the self-optimization-approaching control method applied to the photovoltaic and electric vehicles, light abandoning can be reduced at the same time, active and reactive output of the PV can be well balanced, and self-optimization-approaching response giving consideration to the requirements of the EV and the PV and the requirements of a distribution network is achieved.

Description

Technical Field

[0001] The invention and a self-optimizing control method applied to photovoltaic and electric vehicles belong to the technical field of electric power. Background Art

[0002] Disorderly charging of electric vehicles (EV) may lead to overload of the distribution system, voltage deviation and peak demand, causing more network losses. Although high-power photovoltaic power generation (PV) can reduce voltage deviation, the sharp drop in load caused by it will form a new load trough, and power plants will not be able to adjust large-scale power output in time, resulting in wind and solar power abandonment and large fluctuations in grid frequency.

[0003] Current research shows that active and reactive scheduling of electric vehicles and photovoltaics can reduce power fluctuations and overvoltages in the distribution network. For example, using photovoltaic inverters to control photovoltaic reactive power output can effectively solve the problem of voltage over-limit. Some studies have proposed a dual-scale optimization method in time and space to solve the problems of increased peak-to-valley differences and increased network losses caused by the access of electric vehicles to the grid. Some studies have comprehensively considered the electrical characteristics, production and living characteristics, and travel characteristics of electric vehicles in various regions, and used a dual-scale hierarchical scheduling method in time and space to optimize the number of charging and discharging vehicles in the region and at the node. There are also studies that group electric vehicles based on their characteristic quantities and give centralized scheduling strategies for the groups. Some studies have dynamically divided EVs in each time period into conventional vehicles with disordered charging, charging vehicles, and discharging vehicles for optimization and regulation.

[0004] The above studies distinguish the charging and load characteristics of electric vehicles in different regions and perform optimized scheduling, but they all use long time periods as iteration time limits and the optimization time is relatively long. In the rapidly and dynamically changing operating environment of distribution substations, the scheduling strategy cannot make real-time decisions.

[0005] The self-optimization algorithm advocates that the resources involved in the regulation make self-decision on power according to incentive signals such as voltage, which can solve the real-time scheduling problem of multiple types of resources connected to the distribution network. The self-optimization algorithm is widely used in virtual power plants due to its ability to track power and meet voltage constraints. At present, some studies have constructed an online distributed self-optimization optimization algorithm, which verifies the advantages of the self-optimization operation method in constraint satisfaction, optimization performance, and computational efficiency. In terms of distributed power self-optimization, some studies have achieved distributed power self-optimization output without the help of communication between distributed power sources or between distributed power sources and central controllers, but this method is only applicable to grid frequency regulation. The above studies fully utilize the advantages of self-optimization theory to support distributed power self-optimization decision-making, but do not consider in detail the impact of load characteristics such as the area where EV charging behavior belongs and the status of photovoltaic power generation on the self-optimization response.

[0006] At present, the regulation of electric vehicles and photovoltaics is mostly centralized. The adaptability is poor in diversified and time-varying environmental scenarios. The self-optimization method is suitable for real-time regulation of EVs and photovoltaics because of its advantages of supporting distributed and online optimization. However, the research on self-optimization algorithms has just started, and has not yet fully considered the power consumption characteristics and actual operating status of various loads and power grids. Summary of the invention

[0007] In order to overcome the problems in the prior art, the present invention proposes a real-time self-optimizing control method that takes into account the output characteristics of photovoltaics, the charging characteristics of EVs at different nodes and the time coupling characteristics, satisfies self-optimizing iteration, and supports real-time regulation objectives such as peak shaving and valley filling.

[0008] In order to solve the above technical problems, the technical solution proposed by the present invention is: a self-optimizing control method applied to photovoltaic and electric vehicles, comprising the following steps:

[0009] Construct self-optimizing response constraints for photovoltaic and electric vehicles respectively;

[0010] A self-optimizing response model is constructed through an objective function; a loss function is used in the objective function to describe the utility of photovoltaics and electric vehicles;

[0011] Based on the primal-dual gradient projection method, the excitation feedback information required by the self-optimizing algorithm is calculated, and the excitation feedback information is substituted into the self-optimizing response model. The self-optimizing algorithm is used to solve the optimal solution of the self-optimizing response model under the self-optimizing response constraints of photovoltaics and electric vehicles; the distribution network information and node power information are collected and input into the self-optimizing response model, and the optimal solution is combined with the distribution network information and node power information, and iterative calculation is performed to achieve distributed self-optimizing real-time response of electric vehicles and photovoltaics.

[0012] The above scheme is further improved in that: the self-optimal response constraint of the photovoltaic is: assuming that the active power of the photovoltaic is Reactive power is The upper bounds of active power and reactive power are The lower bounds are Apparent capacity is The minimum power factor is The upper and lower limits of photovoltaic power are subject to constraints:

[0013]

[0014] Among them, the lower bound of photovoltaic active power is Should be 0;

[0015] The self-optimal response constraint of the electric vehicle is: let the set of all electric vehicles connected to the i-th node be Each electric vehicle in the set is indexed by k; when the kth electric vehicle is connected to node i, the SOC value of the electric vehicle at the time of connection is The expected SOC range is Start time is and departure time is Battery capacity is The upper and lower limits of charging power are The charging power constraint of the kth electric vehicle at node i at time t is: Assuming the charging efficiency is η, the SOC value at time t+1 is: The SOC value of the electric vehicle meets the following conditions when it leaves the charging station:

[0016] It is deduced that:

[0017] After equivalent transformation:

[0018] Suppose that the dual multiplier vector of the upper and lower power constraints of all nodes connected to electric vehicles is denoted as and Then we have:

[0019]

[0020]

[0021] The above scheme is further improved in that: the specific steps of constructing the self-optimizing response model through the objective function are to establish the self-optimizing response model with the goal of minimizing the amount of abandoned photovoltaic power, minimizing the amount of photovoltaic reactive power response, no offset of the charging power of electric vehicles, and minimizing the node voltage deviation; the objective function of the self-optimizing response model is:

[0022]

[0023] Among them, γ||V t -V nom || 2 Refers to the numerical sum of the squares of the voltage amplitude V of all nodes at time t and the offset voltage rated value Vnom; V is an N-dimensional column vector V = [V1; V2; ...; VN], ||.|| is the 2-norm, and γ is the weight coefficient of the voltage fluctuation value; and They represent the utility loss functions of photovoltaic and electric vehicles respectively, which are specifically expressed as:

[0024]

[0025] in, and are the utility loss coefficients of photovoltaic active power, photovoltaic reactive power and electric vehicle respectively.

[0026] The above scheme is further improved in that: the specific steps of solving the optimal solution of the self-optimizing response model under the self-optimizing response constraint conditions of photovoltaic and electric vehicles are, based on the basic primal-dual gradient projection method, changing the multiple iteration strategy to a single iteration strategy, that is, after completing the update of the primal variable and the dual variable once, directly enter the model solution at the next moment, and at the same time, the incentive value generated by the change of the operating environment acts on the real-time self-optimizing response of photovoltaic and electric vehicles;

[0027] The single iteration formula and excitation value solved by the model are as follows:

[0028]

[0029] α t =R(u L,t -u U,t )+(λ L,t -λ U,t )(2RP t +M T )-2γR(V t -V nom );

[0030] β t =X(u L,t -u U,t )+(λ L,t -λ U,t )(2RQ t +N T )-2γX(V t -V nom );

[0031] in, Indicates z i,t In the set φ t The projection on t and β t is an N-dimensional column vector, which represents the active and reactive incentive feedback values ​​of the power flow changes of N nodes at time t; the node injected active power P t and reactive power Q t It is an N×1 dimensional vector, which is equal to the net load obtained by subtracting the node base load from the node’s injected distributed generation power. is the total loss of the distribution network; and is the coefficient obtained from the branch impedance of the distribution network, and is the dual multiplier vector for all nodes connected to the EV power upper and lower limits, u U,tand u L,t is the dual multiplier corresponding to the voltage upper and lower limit constraints, λ U,t and λ L,t is the dual multiplier corresponding to the upper and lower power constraints of the boundary nodes, and σ is the step size of the iterative update.

[0032] In order to solve the above technical problems, the present invention also proposes a technical solution: a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any step of the self-optimizing control method for photovoltaic and electric vehicles described in any one of claims 1 to 4 is implemented.

[0033] In order to solve the above technical problems, the present invention also proposes a technical solution: a self-optimizing control device applied to photovoltaics and electric vehicles, used to implement the self-optimizing control method for photovoltaics and electric vehicles described in any one of claims 1 to 4, including: a photovoltaic power generation component and an electric vehicle charging component; the output of the photovoltaic power generation component is adjusted by an inverter, and the inverter is used to adjust the active and reactive output of the photovoltaic power generation component, and the upper and lower bounds of the active and reactive power can be controlled; the electric vehicle charging component is used to collect the charging power of the electric vehicle, the time when the electric vehicle is connected and the time when it leaves.

[0034] In order to solve the above technical problems, the present invention also proposes a technical solution: a self-optimizing control system applied to photovoltaic and electric vehicles, used to implement the self-optimizing control method for photovoltaic and electric vehicles described in any one of claims 1 to 4, comprising: an incentive feedback information module, a photovoltaic self-optimizing response constraint module, an electric vehicle self-optimizing response constraint module and a self-optimizing response model module;

[0035] Among them, the excitation feedback information module obtains excitation feedback information based on the primal-dual gradient projection method and outputs it to the self-optimizing response model module; the photovoltaic self-optimizing response constraint module and the electric vehicle self-optimizing response constraint module are connected to the self-optimizing response model module; the self-optimizing response model module uses the self-optimizing algorithm to calculate the optimal solution according to the input of the excitation feedback information module, the photovoltaic self-optimizing response constraint module and the electric vehicle self-optimizing response constraint module. The self-optimizing response model module is also connected to the distribution network and the node to obtain the distribution network information and the node power information, and iteratively calculates in combination with the optimal solution, so as to realize the distributed self-optimizing real-time response of electric vehicles and photovoltaics.

[0036] The above scheme is further improved in that: in the photovoltaic self-optimization response constraint module, the active power of photovoltaic is set to Reactive power is The upper bounds of active power and reactive power are The lower bounds are Apparent capacity is The minimum power factor is The upper and lower limits of the photovoltaic power output by the photovoltaic self-optimizing response constraint module are subject to the following constraints:

[0037]

[0038] Among them, the lower bound of photovoltaic active power is Should be 0;

[0039] In the self-optimizing response constraint module of the electric vehicle, the set of all electric vehicles connected to the i-th node is Each electric vehicle in the set is indexed by k; when the kth electric vehicle is connected to node i, the SOC value of the electric vehicle at the time of connection is The expected SOC range is Start time is and departure time is Battery capacity is The upper and lower limits of charging power are The charging power constraint of the kth electric vehicle at node i at time t is: Assuming the charging efficiency is η, the SOC value at time t+1 is: The SOC value of the electric vehicle meets the following conditions when it leaves the charging station:

[0040] It is deduced that:

[0041] After equivalent transformation:

[0042] Suppose that the dual multiplier vector of the upper and lower power constraints of all nodes connected to electric vehicles is denoted as and Then we have:

[0043]

[0044] The above scheme is further improved in that: the self-optimizing response model module has an objective function module; the objective function module is set to minimize the amount of abandoned photovoltaic power, minimize the amount of photovoltaic reactive power response, ensure that the charging power of the electric vehicle is not offset, and minimize the node voltage deviation. The objective function provided by the objective function module is:

[0045]

[0046] Among them, γ||V t -V nom || 2Refers to the numerical sum of the squares of the voltage amplitude V of all nodes at time t and the offset voltage rated value Vnom; V is an N-dimensional column vector V = [V1; V2; ...; VN], ||.|| is the 2-norm, and γ is the weight coefficient of the voltage fluctuation value; and They represent the utility loss functions of photovoltaic and electric vehicles respectively, which are specifically expressed as:

[0047]

[0048] in, and are the utility loss coefficients of photovoltaic active power, photovoltaic reactive power and electric vehicle respectively.

[0049] The above scheme is further improved in that: the module for solving the self-optimizing response model, based on the basic primal-dual gradient projection method, changes the multiple iteration strategy to a single iteration strategy, that is, after completing the update of the primal variable and the dual variable once, directly enters the model solution at the next moment, and at the same time, the incentive value generated by the change of the operating environment acts on the real-time self-optimizing response of photovoltaic and electric vehicles;

[0050] The single iteration formula and excitation value solved by the model are as follows:

[0051]

[0052]

[0053] α t =R(u L,t -u U,t )+(λ L,t -λ U,t )(2RP t +M T )-2γR(V t -V nom );

[0054] β t =X(u L,t -u U,t )+(λ L,t -λ U,t )(2RQ t +N T )-2γX(V t -V nom );

[0055] in, Indicates z i,t In the set φ t The projection on t and β tis an N-dimensional column vector, which represents the active and reactive incentive feedback values ​​of the power flow changes of N nodes at time t; the node injected active power P t and reactive power Q t It is an N×1 dimensional vector, which is equal to the net load obtained by subtracting the node base load from the node’s injected distributed generation power. is the total loss of the distribution network; and is the coefficient obtained from the branch impedance of the distribution network, and is the dual multiplier vector for all nodes connected to the EV power upper and lower limits, u U,t and u L,t is the dual multiplier corresponding to the voltage upper and lower limit constraints, λ U,t and λ L,t is the dual multiplier corresponding to the upper and lower power constraints of the boundary nodes, and σ is the step size of the iterative update. The above scheme is further improved in that: the mathematical characterization model for photovoltaics is: assuming that the active power of photovoltaics is Reactive power is The upper bounds of active power and reactive power are The lower bounds are Apparent capacity is The minimum power factor is The upper and lower limits of photovoltaic power are subject to constraints:

[0056]

[0057] Among them, the lower bound of photovoltaic active power is Should be 0;

[0058] The mathematical representation model for electric vehicles is: let the set of all electric vehicles connected to the i-th node be Each electric vehicle in the set is indexed by k; when the kth electric vehicle is connected to node i, the SOC value of the electric vehicle at the time of connection is The expected SOC range is Start time is and departure time is Battery capacity is The upper and lower limits of charging power are The charging power constraint of the kth electric vehicle at node i at time t is: Assuming the charging efficiency is η, the SOC value at time t+1 is: The SOC value of the electric vehicle meets the following conditions when it leaves the charging station:

[0059] It is deduced that:

[0060] After equivalent transformation:

[0061] Suppose that the dual multiplier vector of the upper and lower power constraints of all nodes connected to electric vehicles is denoted as and Then we have:

[0062]

[0063]

[0064] The self-optimizing control method applied to photovoltaic and electric vehicles provided by the present invention describes the utility of photovoltaic and electric vehicles through the self-optimizing response constraints and objective functions of photovoltaic and electric vehicles; it can simultaneously reduce abandoned light and balance the active and reactive outputs of PV; on the premise of meeting the SOC requirements of each EV when it is off-grid, it can also maximize the cooperation with the distribution network to achieve peak shaving and valley filling, reduce node voltage fluctuations, and reduce abandoned light, thus achieving a self-optimizing response that takes into account the needs of EV, PV itself and the needs of the distribution network. . BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The present invention will be further described below in conjunction with the accompanying drawings.

[0066] Figure 1 The present invention is a flowchart of a self-optimizing control method applied to photovoltaic and electric vehicles according to a preferred embodiment of the present invention.

[0067] Figure 2 This is an 11-node photovoltaic output curve of a preferred embodiment of the present invention.

[0068] Figure 3 This is a curve of EV output values ​​at each node of a preferred embodiment of the present invention.

[0069] Figure 4 This is a SOC value curve of EV at each node according to a preferred embodiment of the present invention.

[0070] Figure 5 A boundary node power curve of a preferred embodiment of the present invention.

[0071] Figure 6 This is a voltage value curve of disordered charging nodes in a photovoltaic distribution network according to a preferred embodiment of the present invention.

[0072] Figure 7 This is a curve of node voltage values ​​at each moment after self-optimizing operation according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0073] Embodiment: The self-optimizing control method applied to photovoltaic and electric vehicles in this embodiment considers the charging characteristics and constraints of electric vehicles in different partitions and the upper limit constraints of active and reactive power change of photovoltaic in real time, and constructs a self-optimizing scheduling model that can effectively track the power regulation target of the distribution network, reduce the node voltage deviation, and reduce the utility loss of photovoltaic and electric vehicles. Based on the original dual gradient projection method, combined with the incentive feedback information provided by the self-optimizing algorithm calculation, the distributed self-optimizing real-time response is iteratively realized. This method can simultaneously reduce the abandonment of light and balance the active and reactive output of photovoltaics. On the premise of meeting the SOC requirements of each electric vehicle when it is off the grid, it can also maximize the cooperation with the distribution network to achieve peak shaving and valley filling, reduce node voltage fluctuations, and reduce abandonment of light, so as to achieve a self-optimizing response that takes into account the needs of electric vehicles, photovoltaics themselves, and the distribution network. At the same time, compared with the traditional centralized control algorithm that is limited to long-term scale control, the real-time solution time is greatly reduced, so that it can follow the real-time output of photovoltaics, adjust the output of electric vehicles in time, and respond to the dynamic changes of the distribution network operation environment.

[0074] The photovoltaic output in this embodiment is adjusted by the inverter, and the constraint determination method that meets the charging target of the electric vehicle user when off-grid is to transform the long-term constraint into a real-time constraint by constructing the maximum and minimum values ​​of the average power that meets the charging demand, thereby satisfying the real-time online solution of the model. According to the time-varying and uncertainty of the photovoltaic output, the photovoltaic inverter can adjust the active and reactive output of the photovoltaic, and the upper and lower bounds of the active and reactive power can be controlled, and it is related to the apparent capacity and the minimum power factor. At the same time, the complex power of the photovoltaic output should not be greater than the rated capacity of the inverter, and the upper and lower limits of the reactive output at each moment are constructed to change with the upper limit of the active output. For electric vehicles, the SOC constraint related to the charging power is a long-term constraint, and the SOC value at time t will affect the charging power range at time t+1, thereby generating a time coupling problem. In order to meet the long-term SOC demand through real-time constraints, this embodiment sets the upper and lower limits of power related to SOC, which are determined by the ratio of the upper and lower limits of the battery charging demand when the electric vehicle is off the grid and the charging time, respectively, so as to obtain the equivalent real-time constraints of the long-term constraints of the EV charging power, and meet the SOC demand when the EV is off the grid. The real-time charging power of each EV only needs to meet the upper and lower limits of the power average value to achieve the charging demand under the long-term scale, and the constraint allows a certain degree of constraint violation during the charging time period, which is conducive to timely tracking the real-time changing distribution network demand. The self-optimizing response model in this embodiment takes into account the goals of minimum abandoned photovoltaic power, minimum PV reactive response, no deviation of EV charging power, and minimum node voltage deviation, combined with electric vehicle travel constraints, battery load state constraints, charging and discharging time constraints, and photovoltaic active and reactive processing constraints. When participating in the self-optimizing response, photovoltaics are required to minimize the amount of abandoned photovoltaics and reactive response; EVs should prevent overcharging or excessive deviation from the rated charging power. To this end, a utility loss function is constructed to punish the abandonment of photovoltaics and the degree of deviation of charging power. In view of the phenomenon that the self-optimizing response quantity deviates from the expected value of photovoltaic and EV users themselves, a utility loss function is constructed for evaluation. Therefore, this embodiment takes the utility loss of photovoltaic and EV at time t and the minimum voltage deviation of the distribution network node as the optimization goal. For the distribution network, basic power flow constraints should be followed, including boundary node power constraints and node voltage amplitude constraints that meet the needs of peak shaving and valley filling. In order to quickly and efficiently solve the objective function of this embodiment, linear power flow equations are used to map these constraints. At the same time, considering the demand for peak shaving and valley filling of the distribution network, the power of the boundary nodes should have a certain power upper limit at peak times.

[0075] The method for solving the photovoltaic and electric vehicle self-optimizing scheduling model in this embodiment is based on the gradient projection method solution algorithm. In order to solve the multi-constraint model established in this embodiment, and to ensure that the model has a global optimal solution, a regularized Lagrangian function is constructed to transform the original problem into a convex problem, and the constrained problem is transformed into an unconstrained problem that can approach the optimal solution by relaxing the constraints. The traditional original dual gradient projection method iteration is a common distributed solution algorithm, which requires sufficient time for iterative optimization and cannot track changes in the distribution network operating environment in a timely manner, making it difficult to apply in practice. Therefore, this embodiment uses a single iteration instead of multiple iterations for optimization, that is, when the dual variable is iterated once, its updated value is directly used to update the original variable, and the original variable update value is used as the scheduling value of photovoltaic and EV at the next moment. This embodiment no longer pursues the optimal value of the objective function, but obtains the optimal value through simple iterative information, thereby achieving real-time optimization. This embodiment generates an excitation signal at time t+1 based on the distribution network node voltage and boundary node power information collected at time t, and transmits it to each node via the DMS. At the same time, the photovoltaic power plants and EVs connected to different nodes generate the output at time t+1 based on the output at time t. The solution is repeated over time until the real-time solution is completed every minute, and finally the real-time self-optimizing scheduling of electric vehicles in the distribution network is realized.

[0076] The self-optimizing control method applied to photovoltaic and electric vehicles in this embodiment specifically includes the following steps:

[0077] Step 1: Construct mathematical characterization models for the self-optimizing response constraints of photovoltaics and electric vehicles respectively.

[0078] Since photovoltaic power generation has lower cost and larger installed capacity than wind power generation, this embodiment focuses on the self-optimizing response model of photovoltaic power generation and electric vehicles in the distribution network. In order to study the impact of the self-optimizing response power of electric vehicles (EV) and photovoltaic power generation (PV) on the total load and node voltage of the distribution network, the set of all nodes in the distribution network except the boundary node (node ​​0) is Γ: = {1, 2, ..., N}, where N represents the total number of distribution network nodes, indexed by i, and ф t It is the feasible set of outputs of all PVs and EVs in the distribution network.

[0079] Photovoltaic output is time-varying and uncertain. Photovoltaic inverters can adjust the active power of photovoltaics. and reactive Output, and controllable upper limit of active and reactive power and the lower bound Apparent Capacity and minimum power factor Related [9]. The complex power of photovoltaic output should not be greater than the rated capacity of the inverter, and the upper and lower limits of reactive output at each moment change with the upper limit of active output. Considering that photovoltaic power only generates reactive power during the daytime, and the upper and lower limits of power are subject to the constraints:

[0080]

[0081] Among them, the lower limit of photovoltaic active output Should be 0.

[0082] EV has good controllable characteristics and is a typical self-optimizing response resource. Assume that the charging pile is equipped with an intelligent chip with independent information processing and computing capabilities, responsible for formulating charging strategies for connected EVs and communicating with the upper distribution management system (DMS). Suppose the set of all EVs connected to the i-th node is The total is Each EV in the set is indexed by k. When the kth EV connects to node i, the charging pile collects the SOC value of the EV when it connects Expected SOC range Start time and departure time Battery capacity and charging power upper and lower limits Assume that the upper and lower limits of charging power of EVs connected to the same node are the same. In this embodiment, the charging power is set to a negative value and the discharging power is set to a positive value. The charging power constraint of the kth EV at node i at time t is:

[0083]

[0084] The EV participating in the self-optimizing operation supports real-time online power adjustment, and its SOC value also changes with time. If the charging efficiency is η, the SOC value at time t+1 is:

[0085] In order to meet the charging needs of EV, the SOC value when leaving the charging station should meet the following conditions:

[0086] Combining the above formula, we can deduce:

[0087] After equivalent transformation, it is expressed as:

[0088]

[0089] Therefore, the SOC constraint of EV related to charging power is a long-term constraint, and the SOC value at time t will affect the charging power range at time t+1, thus causing a time coupling problem. In order to meet the long-term SOC demand through real-time constraints, this embodiment sets the upper and lower limits of power related to SOC, namely and The constraint is thus equivalent to the following formula, thereby obtaining the equivalent real-time constraint of the long-term constraint of EV charging power to meet the SOC requirement when EV is off-grid. In addition, the dual multiplier vector of the upper and lower limit constraints of EV power access for all nodes is denoted as and

[0090]

[0091] From the above analysis, it can be seen that the real-time charging power of each EV only needs to meet the above constraints to achieve the charging demand over a long time scale, and this constraint allows a certain degree of constraint violation during the charging time period, which is conducive to timely tracking of real-time changing distribution network requirements.

[0092] The self-optimizing dispatch of flexible resources in the distribution network is required to meet the actual operation needs of the distribution network. This section will establish a self-optimizing dispatch model based on the characteristics of the distribution network, PV and EV, with the goals of minimizing the amount of abandoned photovoltaic power, minimizing the amount of PV reactive response, maintaining the EV charging power, and minimizing the node voltage deviation.

[0093] When PV participates in the self-optimizing response, it is required to minimize the amount of abandoned light and reactive response; EV should prevent overcharging or excessive deviation from the rated charging power. To this end, a utility loss function is constructed to punish the abandonment of light and the degree of charging power deviation. In view of the phenomenon that the self-optimizing response deviates from the expected value of PV and EV users themselves, a utility loss function is constructed to evaluate it. Therefore, this embodiment takes the minimum utility loss of PV and EV and the voltage deviation of the distribution network node at time t as the optimization goal, and constructs the self-optimizing objective function as follows:

[0094]

[0095] Among them, γ||V t -V nom || 2 It refers to the square sum of the voltage amplitude V of all nodes at time t and the voltage rating Vnom. Where V is an N-dimensional column vector V = [V1; V2; ...; VN], ||.|| is the 2-norm, and γ is the weight coefficient of the voltage fluctuation value. and They represent the utility loss functions of PV and EV respectively. Specifically, they are:

[0096]

[0097]

[0098] in, and are the utility loss coefficients of PV active power, reactive power and EV respectively.

[0099] For distribution networks, basic power flow constraints should be followed, including the power of boundary nodes to meet the needs of peak shaving and valley filling. Constraints and node voltage amplitude constraints. In order to quickly and efficiently solve the objective function of this embodiment, the linear power flow equation is used to map these constraints:

[0100] V t ≈RP t +XQ t +V nom ,

[0101]

[0102] Among them, the node injects active power P t and reactive power Q t It is an N×1 dimensional vector, which is equal to the net load obtained by subtracting the node base load from the node’s injected distributed generation power. is the total loss of the distribution network. ,, and. is the coefficient obtained from the branch resistance of the distribution network, is the branch reactance, is an N-dimensional unit vector, is the relationship coefficient between boundary node power and reactive power, Indicates the calculation deviation when calculating the boundary node power through linear power flow.

[0103] The node voltage is constrained to: V L ≤V t ≤V U ;

[0104] V L and V U are the N-dimensional voltage amplitude upper and lower limit matrices respectively. Meanwhile, the dual multipliers corresponding to the voltage upper and lower limit constraints are denoted as u U,t and u L,t In addition, considering the demand for peak load shaving and valley filling in the distribution network, the power of the boundary nodes should have a certain upper limit during peak hours. The specific constraints are as follows:

[0105]

[0106] Among them, τ t Whether the power of the boundary node participates in the power tracking during the peak shaving and valley filling phase, τ t =1 means tracking, τ t=0 means no tracking; is the target power value to be tracked; ε represents the maximum error ratio when the target power value is met; at the same time, the dual multiplier corresponding to the upper and lower power constraints of the boundary nodes is λ U,t and λ L,t .

[0107] Solution of self-optimizing scheduling model based on gradient projection method:

[0108] In order to realize the distributed real-time online self-optimizing response of PV and EV, this embodiment adopts the primal-dual gradient projection algorithm considering the change of the operating environment to solve the self-optimizing model.

[0109] (1) Convex relaxation of the scheduling model

[0110] In order to solve the multi-constraint model established in this embodiment and to ensure that the model has a global optimal solution, the regularized Lagrangian function constructed in this embodiment transforms the original problem into a convex problem, and transforms the constrained problem into an unconstrained problem that can approach the optimal solution by relaxing the constraints. The objective function is transformed into the regularized Lagrangian function:

[0111]

[0112] Among them, z t and ω t denote the set of dual variables corresponding to the original variables and constraints respectively, g t ≤0 indicates a compact form of the constraint. When , the dual function D of the above formula can be ensured t (ω t ) is strongly concave. Because f t The corresponding utility loss function is a convex function, and the regularized Lagrangian function is a strictly feasible convex problem, then its strong duality property holds.

[0113] (2) Establish the original dual variable iteration formula

[0114] Based on the above analysis, the regularized Lagrangian function The original dual gradient projection method iteration relationship is:

[0115]

[0116] in, Indicates z i,t In the set φ t The projection on , σ is the step size of the iterative update, n represents the number of iterations, and in this embodiment, the EVs connected to the same node are considered to have the same charging power and are recorded as the same original variable. Generally, the solution obtained by iteration is the optimal solution, recorded as z i,t,*, and the iteration strategy of the original variable is recorded as Λ t (z t,n ), there exists 0<δ<1, so that the following equation holds:

[0117] ‖Λ t (z t,n+1 )-Λ t (z t,n )‖≤δ‖z t,n+1 -z t,n ‖.

[0118] The above iterative formula is a common distributed solution algorithm, which requires sufficient time for iterative optimization and cannot track the changes in the distribution network operating environment in a timely manner, making it difficult to apply in practice. Therefore, this embodiment uses a single iteration instead of multiple iterations for optimization, that is, after the dual variable is iterated once, its updated value is directly used to update the original variable, and the updated value of the original variable is used as the scheduling value of PV and EV at the next moment:

[0119]

[0120] As can be seen from the above formula, this embodiment no longer seeks to obtain the optimal value of the objective function, but obtains the optimal value through simple iterative information, thereby achieving real-time optimization. At the same time, the simplified iterative formula of the dual variable is obtained as follows:

[0121]

[0122] Among them, g t The voltage value and boundary node power value at time t in are obtained by measurement. In this embodiment, the linear power flow equation is used instead. Then, there is an error between the measured value and the calculated value, and the error is bounded in theory. The simplified iterative formula of the original variable is:

[0123] The iteration strategy of the above formula is denoted as Due to the error caused by the linear power flow equation, And the upper limit of error d g With g t The upper limit of error is a linear relationship.

[0124] The general formula for power iteration of PV and EV is:

[0125]

[0126] α t =R(u L,t -u U,t )+(λ L,t -λ U,t )(2RPt +M T )-2γR(V t -V nom )

[0127] β t =X(u L,t -u U,t )+(λ L,t -λ U,t )(2RQ t +N T )-2γX(V t -V nom )

[0128] Among them, α t and β t is an N-dimensional column vector, which represents the active and reactive incentive feedback values ​​of the power flow changes of N nodes at time t. It can be seen that the incentive feedback value of each node is the same and affects the power output of PV and EV at time t+1.

[0129] Although the algorithm used in this embodiment optimizes the solution rate, it causes errors in the target value, so the accuracy of the algorithm needs to be explained.

[0130] Assume first that f t for YesL f -Lipschitz continuous. Then assume that in the real-time solution strategy of this embodiment, the theoretical optimal solution z at time t and t+1 in the regularized Lagrangian function is t,* 、z t+1,* , in a small time interval, there exists d x ≥0, so that ‖z t+1,* -z t,* ‖≤d x The difference between the original dual gradient projection strategy based on a single iteration and the original dual gradient projection strategy based on multiple iterations is:

[0131]

[0132] It can be seen from the above formula that the self-optimization strategy adopted in this embodiment pursues the optimization of the real-time objective function, and the difference between the objective function value and the optimal value tends to a fixed value over a long period of time and will not increase infinitely with time.

[0133] (3) Self-optimization algorithm solution framework

[0134] The self-optimization algorithm solution framework of this embodiment is as follows Figure 1As shown in the figure. First, the EV travel behavior is simulated by Monte Carlo, and then the objective function of minimizing the PV, EV utility loss and distribution network voltage deviation is constructed. The total load power tracking is used as the constraint of the model, and then the model is solved by the original dual gradient projection method. According to the distribution network node voltage and boundary node power information collected at time t, the excitation signal α at time t+1 is generated. t and β t , transmitted to each node through DMS. At the same time, the PV and EV connected to different nodes generate the output at time t+1 according to the output at time t. The solution is repeated over time until the real-time solution of each minute is completed, and finally the real-time self-optimizing scheduling of electric vehicles in the distribution network is realized.

[0135] Specific experimental examples and analysis:

[0136] (1) Experimental data and parameter settings

[0137] This embodiment uses the IEEE33 node system for simulation. The change of the basic load of the distribution network over time is shown in Figure 5 . Assume that there are 200 EVs participating in self-optimizing scheduling in one day, with 62.5% in residential areas, 25% in office areas, and 12.5% ​​in commercial areas. The charging start time of EVs connected to nodes 27, 15, 28, and 16 in the residential area follows the N(17.6, 3.41) distribution, and the off-grid time follows the N(8.92, 3.24) distribution; the charging time of EVs connected to nodes 31 and 30 in the office area follows the N(8.7, 0.8) distribution, and the off-grid time follows the N(18, 2); the charging time of EVs connected to node 21 in the commercial area follows the generalized extreme value distribution

[13] , the initial SOC is 0.35, the expected SOC value when leaving is in the range of 0.8-0.95, the charging efficiency is 0.95, the battery capacity is 60kWh, the upper limit of the charging power of EV connected to 21 nodes is 10kW, and the upper limit of the charging power of EV connected to other nodes is 7kW. The peak power of a single PV is 300kW, the complex power is 300kVA, and the 4 PVs are connected to nodes 11, 17, 24 and 32 respectively. The simulation time interval of this embodiment is 1min, and there are 1440 time periods in a day.

[0138] (2) Self-optimizing scheduling results

[0139] The strategy proposed in this embodiment is used for simulation, and the output values ​​of PV and EV are obtained as follows: Figure 2 and Figure 3 As shown. Figure 2 It can be seen that the photovoltaic output is relatively large during 9-16h. In order to meet the demand of peak shaving and valley filling, a certain degree of light abandonment will occur. During 0-7h and after 19h, PV will not have active output due to the influence of sunlight. During 7-19h, PV should reduce reactive output as much as possible.

[0140] Depend on Figure 3 It can be seen that when EV participates in the self-optimization response, it will choose to charge during the low electricity consumption period of 0-7h, reduce the charging power during the peak electricity consumption period of 8-11 and 17-20, and increase the charging power after 20h of the low electricity consumption period, so as to meet its own charging needs when leaving the power grid. For the short-term charging demand (charging time is 2h) connected to the 21 nodes, when encountering the peak electricity consumption period, the self-optimization strategy can also reduce the charging power as much as possible and play a regulatory role. Figure 4 It can be seen that the SOC of all EVs can reach above 0.8 when they are finally off the grid, which shows that the self-optimizing scheduling strategy proposed in this embodiment can meet the charging needs of each EV. This embodiment uses the self-optimizing algorithm to perform scheduling every minute for 1 day, and the total simulation time is about 3.93 seconds, and the simulation time per unit time is 0.003 seconds.

[0141] (3) Comparative analysis of peak shaving and valley filling and voltage regulation effects

[0142] Figure 5 The power comparison curve of the distribution network boundary node under the disordered charging and the self-optimizing response strategy of this embodiment is shown. As can be seen from the figure, photovoltaic output can reduce the load peak, but there is no photovoltaic output during the load peak period of 10-15. At this time, the self-optimizing scheduling strategy is needed to reduce the load peak and meet the target load curve requirements. Similarly, the self-optimizing strategy can also be used to increase the load value during the valley period, such as optimizing the load valley period caused by a large amount of PV output at noon.

[0143] The objective function values ​​of photovoltaic and EV in the same area after self-optimization response are shown in the following table. The charging power deviation of EV is the smallest. Since photovoltaic has a large reactive output, the total utility loss function value is large. The voltage deviation is the sum of 1440 time points, so the value is the largest. The total voltage deviation is 23.0998, which is 72% lower than the 82.08 of disordered charging. For EV, the utility loss function in the commercial area is the smallest. This is because the charging time of EV in the commercial area is the shortest, which is closer to the rated power charging. From the perspective of node voltage deviation, the commercial area has the smallest deviation and the residential area has the largest deviation. This is because the number of nodes in the residential area is the largest, and the commercial area is close to the 0 node at the transformer end, which is close to the voltage per unit value.

[0144]

[0145]

[0146] Figure 6 and Figure 7The node voltage values ​​after disordered charging and self-optimizing operation are shown respectively. As can be seen from the figure, disordered charging of EV will lead to voltage over-limit, and it mainly over-limits at peak load. The self-optimizing operation strategy proposed in this embodiment enables EV to reduce charging power at peak load and increase charging power at valley load. At the same time, PV releases active and reactive power at peak load, effectively reducing the amount of abandoned light, preventing voltage over-limit, greatly reducing voltage deviation value, and the minimum node voltage deviation is reduced by about 50%.

[0147] The present invention is not limited to the above embodiments, and any technical solutions formed by equivalent replacement shall fall within the protection scope required by the present invention.

Claims

1. A self-optimizing control method for photovoltaic and electric vehicles, characterized in that: The steps include: Construct self-optimizing response constraints for photovoltaic and electric vehicles respectively; A self-optimizing response model is constructed through an objective function; a loss function is used in the objective function to describe the utility of photovoltaics and electric vehicles; Based on the primal-dual gradient projection method, the excitation feedback information required by the self-optimizing algorithm is calculated, and the excitation feedback information is substituted into the self-optimizing response model. The self-optimizing algorithm is used to solve the optimal solution of the self-optimizing response model under the self-optimizing response constraints of photovoltaics and electric vehicles; the distribution network information and node power information are collected and input into the self-optimizing response model, and the optimal solution is combined with the distribution network information and node power information, and iterative calculation is performed to achieve distributed self-optimizing real-time response of electric vehicles and photovoltaics.

2. The self-optimizing control method for photovoltaic and electric vehicles according to claim 1, characterized in that: The self-optimal response constraint of the photovoltaic is: let the active power of the photovoltaic be Reactive power is The upper bounds of active power and reactive power are The lower bounds are Apparent capacity is The minimum power factor is The upper and lower limits of photovoltaic power are subject to constraints: Among them, the lower bound of photovoltaic active power is Should be 0; The self-optimal response constraint of the electric vehicle is: let the set of all electric vehicles connected to the i-th node be Each electric vehicle in the set is indexed by k; when the kth electric vehicle is connected to node i, the SOC value of the electric vehicle at the time of connection is The expected SOC range is Start time is and departure time is Battery capacity is The upper and lower limits of charging power are The charging power constraint of the kth electric vehicle at node i at time t is: Assuming the charging efficiency is η, the SOC value at time t+1 is: The SOC value of the electric vehicle meets the following conditions when it leaves the charging station: It is deduced that: After equivalent transformation: Suppose that the dual multiplier vector of the upper and lower power constraints of all nodes connected to electric vehicles is denoted as and Then we have:

3. The self-optimizing control method for photovoltaic and electric vehicles according to claim 2 is characterized in that: The specific steps of constructing the self-optimizing response model through the objective function are: setting the goals of minimizing the amount of abandoned photovoltaic power, minimizing the amount of photovoltaic reactive power response, not offsetting the charging power of electric vehicles, and minimizing the node voltage deviation, and establishing the self-optimizing response model; the objective function of the self-optimizing response model is: Among them, γ||V t -V nom || 2 Refers to the numerical sum of the squares of the voltage amplitude V of all nodes at time t and the offset voltage rated value Vnom; V is an N-dimensional column vector V = [V1; V2; ...; VN], ||.|| is the 2-norm, and γ is the weight coefficient of the voltage fluctuation value; and They represent the utility loss functions of photovoltaic and electric vehicles respectively, which are specifically expressed as: in, and are the utility loss coefficients of photovoltaic active power, photovoltaic reactive power and electric vehicle respectively.

4. The self-optimizing control method for photovoltaic and electric vehicles according to claim 3 is characterized in that: The specific steps of solving the optimal solution of the self-optimizing response model under the self-optimizing response constraint conditions of photovoltaic and electric vehicles are: on the basis of the basic primal-dual gradient projection method, the multiple iteration strategy is changed to a single iteration strategy, that is, after completing the update of the primal variable and the dual variable once, directly enter the model solution at the next moment, and at the same time, the incentive value generated by the change of the operating environment acts on the real-time self-optimizing response of photovoltaic and electric vehicles; The single iteration formula and excitation value solved by the model are as follows: α t =R(u L,t -u U,t )+(λ L,t -λ U,t )(2RP t +M T )-2γR(V t -V nom ); b t =X(u L,t -u U,t )+(λ L,t -l U,t )(2RQ t +N T )-2γX(V t -V nom ); in, Indicates z i,t In the set φ t The projection on t and β t is an N-dimensional column vector, which represents the active and reactive incentive feedback values ​​of the power flow changes of N nodes at time t; the node injected active power P t and reactive power Q t It is an N×1 dimensional vector, which is equal to the net load obtained by subtracting the node base load from the node’s injected distributed generation power. is the total loss of the distribution network; and is the coefficient obtained from the branch impedance of the distribution network, and is the dual multiplier vector for all nodes connected to the EV power upper and lower limits, u U,t and u L,t is the dual multiplier corresponding to the voltage upper and lower limit constraints, λ U,t and λ L,t is the dual multiplier corresponding to the upper and lower power constraints of the boundary nodes, and σ is the step size of the iterative update.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, any step of the self-optimizing control method for photovoltaic and electric vehicles described in any one of claims 1 to 4 is implemented.

6. A self-optimizing control device for photovoltaic and electric vehicles, used to implement the self-optimizing control method for photovoltaic and electric vehicles as described in any one of claims 1 to 4, characterized in that: include: Photovoltaic power generation components and electric vehicle charging components; the output of the photovoltaic power generation components is adjusted by an inverter, and the inverter is used to adjust the active and reactive output of the photovoltaic power generation components, and the upper and lower limits of the active and reactive power can be controlled; the electric vehicle charging component is used to collect the charging power of the electric vehicle, the time when the electric vehicle is connected and the time when it leaves.

7. A self-optimizing control system for photovoltaic and electric vehicles, used to implement the self-optimizing control method for photovoltaic and electric vehicles as described in any one of claims 1 to 4, characterized in that: include: Incentive feedback information module, photovoltaic self-optimizing response constraint module, electric vehicle self-optimizing response constraint module and self-optimizing response model module; Among them, the excitation feedback information module obtains excitation feedback information based on the primal-dual gradient projection method and outputs it to the self-optimizing response model module; the photovoltaic self-optimizing response constraint module and the electric vehicle self-optimizing response constraint module are connected to the self-optimizing response model module; the self-optimizing response model module uses the self-optimizing algorithm to calculate the optimal solution according to the input of the excitation feedback information module, the photovoltaic self-optimizing response constraint module and the electric vehicle self-optimizing response constraint module. The self-optimizing response model module is also connected to the distribution network and the node to obtain the distribution network information and the node power information, and iteratively calculates in combination with the optimal solution, so as to realize the distributed self-optimizing real-time response of electric vehicles and photovoltaics.

8. The self-optimizing control system for photovoltaic and electric vehicles according to claim 7 is characterized in that: In the photovoltaic self-optimizing response constraint module, the active power of photovoltaic is assumed to be Reactive power is The upper bounds of active power and reactive power are The lower bounds are Apparent capacity is The minimum power factor is The upper and lower limits of the photovoltaic power output by the photovoltaic self-optimizing response constraint module are subject to the following constraints: Among them, the lower bound of photovoltaic active power is Should be 0; In the self-optimizing response constraint module of the electric vehicle, the set of all electric vehicles connected to the i-th node is Each electric vehicle in the set is indexed by k; when the kth electric vehicle is connected to node i, the SOC value of the electric vehicle at the time of connection is The expected SOC range is Start time is and departure time is Battery capacity is The upper and lower limits of charging power are The charging power constraint of the kth electric vehicle at node i at time t is: Assuming the charging efficiency is η, the SOC value at time t+1 is: The SOC value of the electric vehicle meets the following conditions when it leaves the charging station: It is deduced that: After equivalent transformation: Suppose that the dual multiplier vector of the upper and lower power constraints of all nodes connected to electric vehicles is denoted as and Then we have:

9. The self-optimizing control system for photovoltaic and electric vehicles according to claim 8, characterized in that: The self-optimizing response model module has an objective function module; the objective function module is set to minimize the amount of abandoned photovoltaic power, minimize the amount of photovoltaic reactive power response, ensure that the charging power of the electric vehicle is not offset, and minimize the node voltage deviation. The objective function provided by the objective function module is: Among them, γ||V t -V nom || 2 Refers to the numerical sum of the squares of the voltage amplitude V of all nodes at time t and the offset voltage rated value Vnom; V is an N-dimensional column vector V = [V1; V2; ...; VN], ||.|| is the 2-norm, and γ is the weight coefficient of the voltage fluctuation value; and They represent the utility loss functions of photovoltaic and electric vehicles respectively, which are specifically expressed as: in, and are the utility loss coefficients of photovoltaic active power, photovoltaic reactive power and electric vehicle respectively.

10. The self-optimizing control system for photovoltaic and electric vehicles according to claim 9, characterized in that: The self-optimizing response model solving module, based on the basic primal-dual gradient projection method, changes the multiple iteration strategy into a single iteration strategy, that is, after completing the update of the primal variable and the dual variable once, directly enters the model solving at the next moment, and at the same time, the incentive value generated by the change of the operating environment acts on the real-time self-optimizing response of photovoltaic and electric vehicles; The single iteration formula and excitation value solved by the model are as follows: α t =R(u L,t -u U,t )+(λ L,t -λ U,t )(2RP t +M T )-2γR(V t -V nom ); b t =X(u L,t -u U,t )+(λ L,t -l U,t )(2RQ t +N T )-2γX(V t -V nom ); in, Indicates z i,t In the set φ t The projection on t and β t is an N-dimensional column vector, which represents the active and reactive incentive feedback values ​​of the power flow changes of N nodes at time t; the node injected active power P t and reactive power Q t It is an N×1 dimensional vector, which is equal to the net load obtained by subtracting the node base load from the node’s injected distributed generation power. is the total loss of the distribution network; and is the coefficient obtained from the branch impedance of the distribution network, and is the dual multiplier vector for all nodes connected to the EV power upper and lower limits, u U,t and u L,t is the dual multiplier corresponding to the voltage upper and lower limit constraints, λ U,t and λ L,t is the dual multiplier corresponding to the upper and lower power constraints of the boundary nodes, and σ is the step size of the iterative update.

Images