Energy management method of renewable energy system and energy storage system based on FFO-DADRCNN

Abstract

An FFO-DADRCNN-based energy management method for a renewable energy system and an energy storage system belongs to the technical field of renewable energy systems and energy storage systems, and comprises the steps of constructing an energy storage system ESS cost model including ESS capital cost and ESS operation cost; constructing a wind generating set model, a solar generating set model, a micro-grid load model and a frequency response load model; and setting voltage and current input factors of a wind turbine generator and a photovoltaic system, and carrying out optimal distribution of micro-grid energy by adopting an FFO algorithm. According to the method, the neural network architecture DADRCNN is adopted to predict the future energy demand load of the electric vehicle, various innovative technologies are combined, the operation management level of the micro-grid is improved, the RESs are allowed to operate independently of an EMS reference power supply, the system achieves higher flexibility and autonomy in the aspect of power generation, and therefore efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of renewable energy systems and energy storage systems, and in particular relates to a hybrid technology for reliability management of energy storage systems containing a high proportion of new energy. Background Art

[0002] Hybrid ESSs for microgrids have attracted much attention recently. Microgrids include both controllable and uncontrollable loads, as well as ESSs and DERs. DERs can be successfully included in the distribution network using MGs. MGs are a type of distribution network that can be disconnected from the larger grid. The stability of the current grid is not affected by the connection of MGs. An independent control system is required that can autonomously adjust the operating state of the microgrid based on signals from the grid and loads. As uncontrolled loads dominate, it becomes challenging to accurately estimate power consumption. Traditional power systems have limited ability to effectively dispatch generation resources. The intermittent and variability of renewable energy sources leads to mismatches in energy supply and demand. These mismatches can adversely affect the stability and reliability of the grid. Technical challenges such as compatibility issues, grid integration complexity, and the need for complex control and management strategies can also hinder performance.

[0003] Energy management in smart microgrids involves multiple disciplines, including electrical engineering, computer science, and environmental science. This research provides interdisciplinary insights that can promote innovation and integration of various technical and scientific principles. Smart microgrids are complex systems that require complex management techniques. This research helps to deepen the understanding of these systems, especially in terms of their dynamic behavior, interdependencies, and the impact of various factors on their performance. In summary, large power plants produce most of the electrical energy used in traditional power systems, and the use of time-varying droop controllers for distributed generators introduces complexity. Scenario-based analysis may encounter scalability issues when dealing with large systems. In addition, the computational complexity of scenario-based analysis methods increases significantly with the number of scenarios considered, which is particularly noteworthy for large microgrid systems.

[0004] In the existing technology, although the energy management strategy of smart microgrids introduces a new energy management system (EMS) that can adapt to the changing frequency, it often focuses too much on distributed generation (DGs) and demand response programs, ignoring other key aspects of energy management. In addition, the scalability of genetic algorithms is limited when dealing with large-scale optimization problems, resulting in a significant increase in computational complexity as the size of the optimization problem increases. The mixed integer linear programming (MILP) model may not be able to fully capture the nonlinear dynamics or complexity of the microgrid system due to linear constraints. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN, which combines multiple innovative technologies to improve the operation and management level of the microgrid. By allowing RESs to operate independently of the EMS reference power supply, the system achieves greater flexibility and autonomy in power generation, thereby improving efficiency.

[0006] An energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN is characterized by comprising the following steps:

[0007] Step 1: Build an energy storage system ESS cost model, including ESS capital cost and ESS operating cost;

[0008] Step 2: Construct wind turbine model, solar turbine model, microgrid load model and frequency response load model;

[0009] Step 3: Set the voltage and current input factors of the wind turbine and photovoltaic system, and use the FFO algorithm to optimally allocate the energy of the microgrid;

[0010] Step 4: Use the neural network architecture DADRCNN to predict the future energy demand load of electric vehicles;

[0011] Step 5: Combine steps 1 to 4 to perform hybrid management of the energy management system EMS of the renewable energy system and energy storage system of the smart microgrid.

[0012] The ESS operating cost in step 1 is generated by degradation and aging. The aging cost AC t The formula is:

[0013] AC t =η t *IC ESS

[0014]

[0015] Where: IC ESS is the ESS installation cost, η t is the aging rate of ESS in a single cycle, and the aging rate is related to the number of cycles. about, and is the average SOC and SOC changes within a cycle The wind turbine model described in step 2 uses a two-state upper and lower Markov model to evaluate the reliability of the wind turbine. The availability probability of WTG in the model is equal to 1-q WTS and p WTS.rated The rated capacity is unavailable with a probability equal to q WTS;

[0016]

[0017] The reliability of solar generators is modeled using a two-state Markov model:

[0018]

[0019] Where: MTTR WTS MTTF WTS MTTR PVS MTTF PVS are the mean time between failures and mean repair time for WT and PV respectively.

[0020] The microgrid load model uses the normal distribution model with uncertainty probability distribution function:

[0021]

[0022] Where P L represents the load demand of MG, μ L and σ L are the mean and standard deviation, respectively;

[0023] Step 3 The method of the FFO algorithm for optimal energy allocation in the microgrid is: using an automatic parameter adjustment method, using fuzzy logic FL to dynamically set the parameter value of FFO; the overall size of FFO is determined using heuristic Dimensionality determination, by using heuristics to determine the SL (survival list) size;

[0024] Automatically and dynamically control the value of each parameter, the distance δ of each solution from the best tested solution so far, and a function Ψ for the improvement of the objective function value of each potential solution relative to the previous iteration;

[0025] The distance δ between two foxes is calculated as:

[0026]

[0027] Where i and j are the i-th and j-th flying foxes respectively, |·| is the corresponding distance metric, is the position of the i-th and j-th flying fox;

[0028] Consider the current and previous positions of fox i in the last two time steps as:

[0029]

[0030] In the formula, δ max is the distance between the worst solution detected and the best solution detected;

[0031] δ is used to calculate the distance between the fox and the coldest point, and the function Ψ characterizes how the solution improves based on the objective function values ​​of the current and previous iterations.

[0032] Step 4 The neural network architecture DADRCNN performs future energy demand load prediction for electric vehicles, including the following steps:

[0033] DCU (Dilation Convolutional Unit): Dilated convolution is an expansion of traditional convolution; the receptive field of the second-layer dilated convolution gradually becomes entangled with the receptive field of the previous layer, and the equation is:

[0034]

[0035] In the formula, R k and R k-1 Represent the receptive fields of the convolution kernels at the K layer and the K1 layer respectively; where f K Indicates the convolution kernel size; l k represents the expansion component, S n Indicates the convolution size;

[0036] DRM (Deep Residual Module): In order to extract more accurate data features, the Res network structure is introduced. The feature data is transferred from the front layer to the back layer through the shallow network and the residual function mapping layer. After the residual jump connection is established, the residual connection F(E)+E0 is obtained.

[0037] F(C)=ω(W2(Rel(ω(W1C+b1))))+b2)

[0038] C'=ω(W3E+b3)

[0039] Where W1, W2, W3 represent the weights of a single extended convolutional layer, and b1, b2, b3 represent the nonlinear activation function, the offset of the batch layer, and the normalization;

[0040] The CARM that considers the variable values ​​of several landslide evaluation factors is introduced. The CARM formula is as follows:

[0041]

[0042] Where A(·) represents average pooling, FE represents a fully interconnected layer, ρ represents an sigmoid curve, Z represents matrix multiplication, and Y represents matrix addition.

[0043] Through the above design scheme, the present invention can bring the following beneficial effects: an energy management method for renewable energy systems and energy storage systems based on FFO-DADRCNN, which combines the Flying Fox Optimization (FFO) algorithm with the Deep Attention Dilated Residual Convolutional Neural Network (DADRCNN), not only realizes the function improvement of the algorithm itself, but also introduces a unique combination method to significantly improve the operation and management level of the microgrid. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0045] Figure 1 The present invention is a schematic diagram of an energy management method EMS configuration of a renewable energy system and an energy storage system based on FFO-DADRCNN.

[0046] Figure 2 The present invention is a schematic flow chart of an energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN.

[0047] Figure 3 The present invention is a FFO algorithm flow chart of an energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN.

[0048] Figure 4 This is a typical power curve diagram of a wind turbine generator in an energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN of the present invention. DETAILED DESCRIPTION

[0049] An energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN mainly includes:

[0050] (1) Hybrid technology innovation in energy management system (EMS): The role of FFO algorithm in optimization problems in energy management system is introduced; the role of deep attention dilated residual convolutional neural network (DADRCNN) in load forecasting modeling is explained, and how to combine it with FFO algorithm.

[0051] (2) Application of FFO-DADRCNN method: This paper explains how to improve the stability, efficiency and load forecasting accuracy of the power grid by combining the FFO algorithm and DADRCNN. It is clear that this method aims to improve the stability and efficiency of the power grid, reduce the cost of the microgrid system, and improve the load forecasting accuracy.

[0052] (3) Components of microgrid system: The main components of the microgrid system are listed, and how to use FFO technology to most effectively regulate the microgrid system and estimate loads is explained.

[0053] Specifically, the present invention is an energy management system (EMS) method based on renewable energy systems (RES) and energy storage systems (ESS). Through real-time optimization scheduling and load forecasting, the FFO-DADRCNN method can better balance supply and demand and reduce grid instability caused by the intermittency and variability of renewable energy; the FFO algorithm provides an effective search mechanism for finding the optimal solution under complex grid operation constraints, thereby improving the overall operation efficiency of the grid; through accurate load forecasting and optimized scheduling, it can reduce dependence on expensive peak power and reduce the operating cost of the energy storage system; the DADRCNN model uses deep learning technology to more accurately predict load demand, thereby improving the scheduling accuracy of the grid. Through the combination of optimization algorithms and deep learning models, the economy of the grid can be improved while meeting technical requirements.

[0054] Figure 1 The configuration of the proposed EMS is depicted. In order to reduce the number of conversion stages, the present invention proposes an AC / DC microgrid. The AC bus regulates voltage, frequency and power balance and is connected to the main grid. In contrast, DERs are connected to the DC bus and require local controller management due to power generation fluctuations. Unlike AC loads connected to the AC bus, DC loads are linked to the DC bus. In order to achieve power flow regulation within the MG using an optimization approach, the EMS delivers a planned reference power to the controller. When the MG has surplus power, it reaches its maximum limit of power transfer to the main utility grid. In response, it operates in a grid-interactive manner to balance the system power.

[0055] The proposed FFO technique is used to regulate the MG system and load forecasting is performed using DADRCNN. The flowchart of the proposed system is shown in Figure 2 After proposing the configuration of the EMS, we delve into other modeling details.

[0056] ESS units are modeled based on their characteristics, such as power and energy capacity, location, charge and discharge cycles, and implementation requirements. Their main limitations are the maximum and minimum state of charge (SOC), charge and discharge efficiency, and ESS charge and discharge power limits. The two main parameters of ESSs are rated capacity and rated power. The rated power will determine the energy charge and discharge rate of the energy storage system. On the other hand, the rated capacity is the amount of energy that the storage system can store. A high-capacity ESS is capable of storing high power but does not have efficient charge and discharge capabilities; on the contrary, a small-capacity ESS has a suitable charge and discharge power. The state of charge (SOC) of an ESS is expressed as a percentage of the available energy at time t, expressed as formula (1):

[0057]

[0058] En is the upper limit of ESS capacity.

[0059] SOC is the state of charge of the ESS, which changes during the charging and discharging process. The state of charge of the battery increases during charging and decreases during discharging. Sometimes the energy of the ESS is expressed as the depth of discharge (DOD) instead of SOC, which is defined as the energy (charge) removed in a given time. It determines the total amount of electricity that the battery can store in a certain state, expressed as formula (2).

[0060] DoD(t)=1-SOC(t) (2)

[0061] ESS can operate in three different modes: charging, discharging, and idle. The SOC of ESS cells during battery discharging and charging is defined as Equations (3) and (4), respectively. In charging mode, the SOC of the battery depends on the battery charging efficiency (η ch ), battery charging power (P ch ) and battery self-discharge rate (ζ).

[0062]

[0063] η disc is the ESS discharge efficiency, k ch , k disc It is the charge and discharge flag.

[0064] The installation of ESS brings huge costs to the microgrid. Therefore, an accurate and practical ESS cost model will enhance the modeling of system operation from the perspective of economy and safety. In order to arrange the operation of ESS, it is crucial to obtain an accurate ESS cost model. ESS cost consists of ESS capital cost and ESS operating cost. Batteries do not use any fuel, so their operating cost is mainly due to their degradation and aging costs.

[0065] The aging cost of ESS is related to its degradation level. The degradation of ESS depends on the cycle and calendar aging of the battery. Cycle aging comes from C-rate, temperature, DOD usage and number of cycles, while calendar aging depends on SOC, temperature and time. The ESS aging cost depends on the lifetime degradation ratio and the installed cost (IC), as shown in equation (5):

[0066] AC t =η t *IC ESS (5)

[0067] Where: IC ESS is the ESS installation cost, η tis the aging rate of ESS in a single cycle. The aging rate of the battery per cycle can be obtained by the following formula and used to calculate the aging cost. The aging cost will be added to the cost minimization problem.

[0068]

[0069] Aging rate and cycle number about, and is the average SOC and SOC changes within a cycle function.

[0070] The power generation of a wind turbine depends on the wind speed. Considering the uncertainty of wind speed prediction, the wind force needs to be modeled. Previous studies have shown that wind speed can be successfully modeled using the Weibull distribution:

[0071]

[0072] Where: γ and k are the scale parameter and shape parameter of the Weibull distribution function respectively; ν is the wind speed. When the cut-off wind speed is reached, the turbine starts to generate electricity. As the wind speed increases, the output of the turbine also increases. When the wind speed is too high, the turbine equipment will be automatically dismantled to protect the turbine. Therefore, the output power of the wind turbine can generally be expressed by a piecewise function, as shown in equation (8). The output power of the wind turbine expressed in terms of wind speed depends on the wind turbine characteristic curve, which is given by the following equation:

[0073]

[0074] Where p is the output power, A, B, C are the shape parameters of WT, N is the number of wind turbines in the wind farm, P r is the rated power, SW t is the wind speed, V r is the rated power wind speed, V ci 、V co The typical power curve of a wind turbine is as follows: Figure 4 shown.

[0075] The reliability of wind turbines is studied using a two-state up-down Markov model as a probability model. In this model, the probability of availability of WTG is equal to 1-q WTS and p WTS.rated rated capacity, otherwise the probability of unavailability is equal to q WTS .q WTS is defined as follows:

[0076]

[0077] The generation of solar power generation system depends on solar irradiance, solar cell characteristics and ambient temperature. By comparing with the existing literature modeling, the results show that the Beta probability density function can be well used to simulate solar irradiance. Therefore, the present invention uses beta distribution to represent the uncertainty of solar irradiance as follows:

[0078]

[0079] Where: c and υ are the scale parameter and shape parameter of the beta distribution function respectively. When the solar irradiance is known, the power generation of the photovoltaic cell is obtained by equation (11). pv ) and predicted solar radiation (G t ) and a certain radiation point (R c )related.

[0080]

[0081] Where: η c , K c are photovoltaic conversion efficiency and utilization rate respectively.

[0082] For wind power generation, the reliability of solar power generation is modeled using a two-state Markov model:

[0083]

[0084] The load consumption of a microgrid is affected by the type of its users and depends on the day of the week and the month of the year. The probabilistic behavior of the microgrid load is modeled using a normal distribution function as follows:

[0085]

[0086] Where: μ and σ are the mean and standard deviation respectively.

[0087] The probability distribution function of load uncertainty is modeled using normal distribution, which is in the form of:

[0088]

[0089] Where P L represents the load demand of MG, μ L and σ L are the mean and standard deviation respectively.

[0090] Below is the DC bus voltage modeling, which is a key determinant of energy storage capacity in a microgrid. The DC bus voltage determines the capacity of the battery and SC. If the ESS losses are reduced, the DC bus voltage can be easily achieved. Below is the differential form of the bus voltage:

[0091]

[0092] In the formula, E bus (t) is the bus energy loss at time t, P sc (t), P b (t), P L (t) are the power of renewable energy, battery and electrical load respectively.

[0093] Factors that affect the DC bus voltage are:

[0094]

[0095] In the formula, C bus is the total DC bus capacitance.

[0096] Total capacitance of the DC bus:

[0097]

[0098] In the formula: c1, c2, c3, c4, c5 are the branch capacitances of the DC bus respectively.

[0099] By using the SC current voltage saturation function, the limitation is minimized. The following describes the minimum and maximum SC currents:

[0100]

[0101] The maximum SC voltage is The measured SC voltage is defined as represents the minimum SC voltage; Δv represents the voltage change; Indicates the maximum charging SC current; Indicates the minimum charging SC current; Indicates the extreme charge SC voltage.

[0102] The performance of the battery is affected by the discharge and charging characteristics of the battery at different speeds and the charging current. This section will introduce the basic EEC of lithium-ion batteries. The battery terminal voltage under constant current and constant time charging is:

[0103]

[0104] For the discharge scenario, the battery terminal voltage is:

[0105]

[0106] Where: Q is the nominal capacity of the battery, I c ,I d are the charge and discharge currents respectively, R1 and R2 are the internal resistances of the battery, V0 is the open circuit voltage, C is the adjustment factor, and c1 is the capacitance.

[0107] Following our discussion of batteries, we will move on to the modeling of frequency response loads, which plays a key role in grid stability and management.

[0108] Frequency response load modeling, natural reaction loads and frequency control loads can be divided into two categories of loads, which contribute to MG frequency management; DC-DC boost converter modeling It is well known that the voltage produced by the fuel cell stack varies greatly, and it differs greatly under full load and no-load conditions. To address this variability, the fuel cell requires a DC to DC converter to adjust the voltage output according to the specific working environment; Pricing model, to determine the price of electricity within a 24-hour period, there are various rates available, including real-time pricing, critical peak pricing, time of use (ToU) pricing, and peak pricing (PP). Time-of-use electricity pricing is intended to be used for electricity energy pricing in most electrical appliance scheduling systems. In addition, the ToU pricing structure divides the total length pricing into different labs and assigns a fixed price to each slab; User comfort refers to the degree to which users feel comfortable and satisfied when interacting with a product, service, or environment. It includes various factors that contribute to a positive user experience, such as latency, thermal, visual, and air quality comfort.

[0109] The detailed description of the proposed technique is as follows. FFO is a population-based approach inspired by the endurance program of flying foxes during heat waves, and its goal is to achieve the optimal distribution of energy within the microgrid. The FFO algorithm flow chart is shown in Figure 3 As shown, the following is the definition of the FFO step:

[0110] Step 1: Initialization

[0111] Set the voltage, current and other input factors of wind turbines and photovoltaic systems.

[0112] Step 2: Random Generation

[0113] The input parameters are randomly generated after initialization.

[0114]

[0115] Among them, d is the dimension of the expected variable and n is the initialization parameter.

[0116] Step 3: Fitness Calculation

[0117] The degree of response is chosen based on the fairness function as follows:

[0118] Fitness=Min(F) (23)

[0119] Step 4: Search the space

[0120] During or after the morning heatwave, the flying foxes (iterations) will seek cooler trees (better solutions) to relax in order to protect each other. Most of the time, individuals trying to plant trees with moderate heat will be overwhelmed by additional flying foxes. As a result, they tend to die (discard).

[0121] Step 5: Exploration Phase

[0122] When flying foxes follow a route or seek out a neighboring tree, if the habitat tree does not provide the ideal minimum temperature for the flying fox, it can be inferred that they will move in front of another tree. The following equation is used to explain how this relationship is formed:

[0123]

[0124] for, represents the jth component of FF(i), cool j is the lowest temperature position. When x is repeated t, it is a stable value. The position of FF with the lowest temperature in the tree is passed to rand~∪(0,1) before it cools down. were organized, in which position the flying fox appeared in the most striking posture, which is the best explanation for the possibility of the largest reserved area, and the two flying foxes were considered to be close to isolation.

[0125] Step 6: Development Phase

[0126] A flying fox searches for the nearest area in the cold air To help prevent suffocation. The following calculations provide more details about the event:

[0127]

[0128] Here, rand ~ ∪ (0, 1), rnd j represents a subjective integer between 0 and 1, and in are two random associations, they are in the correct position of the current population, pa indicates possible stability. Finally, k is guaranteed to be in {1, 2, ....., m} by random selection, from Started by The original and new results are chosen to ensure that there is no duplication. The proposed solution is assumed to have been evaluated. Otherwise, it falls back to the most recent position.

[0129] Step 7: Death and Replacement of the Foxbat

[0130] The death of flying foxes is caused by many factors. For example, if they are looking for the optimal tree, they may find themselves in a hot, remote place. Therefore, the following formula can be used as a reference for the newly constructed flying fox position to obtain a random integer n∈[2,NL]:

[0131]

[0132] In the formula, represents the kth RL on the FF in the entire t repetitions. The following formula is to maximize the selection of labeling and sufficient area.

[0133]

[0134] Here, nc represents the number of FFs and is related to the objective function of the best explanation.

[0135] Step 8: Termination

[0136] Verify the parking criteria; if the optimal result is obtained, the procedure is complete; if not, go to step 3.

[0137] In order to make the FFO algorithm independent of any initial parameter settings, an automatic parameter tuning approach is adopted, where the parameter setting values ​​of the FFO are dynamically determined through fuzzy logic (FL). Initially, the overall size of the FFO is determined by utilizing the heuristic The SL size is determined by the number of dimensions. The SL size is determined by utilizing heuristics. Therefore, the size of the SL is set according to the overall size.

[0138] To automatically and dynamically control the value of each parameter, we consider the distance δ of each solution (fox) from the most tested solution (coldest spot) so far, and a function Ψ that determines the improvement of the objective function value of each potential solution relative to the previous iteration.

[0139] The distance δ between two foxes is calculated as:

[0140]

[0141] where i and j are the i-th and j-th flying foxes respectively, and |·| is the corresponding distance metric.

[0142] Consider the current and previous positions of fox i in the last two time steps as:

[0143]

[0144] where δ max is the distance between the worst detected solution and the best detected solution.

[0145] The main idea of ​​using δ is to calculate the distance between the fox and the coldest spot, while Ψ is to characterize how the solution has improved based on the objective function value of the current and previous iterations.

[0146] The FFO technique proposed above is used to regulate the MG system, and then DADRCNN is used for load forecasting. The flow chart is as follows Figure 2 As shown. DADRCNN is a neural network architecture that combines several advanced techniques to restore the performance of image recognition and processing tasks. The CNN architecture is called DADR-CNN. It combines dilated convolution and attention methods to improve the performance of various computer vision tasks. This technique alleviates the gradient vanishing problem and makes it possible to train deeper networks. The deep residual module can sample the data and extract more precise features. Among the many complex and diverse features of landslide assessment, the channel attention residual module can focus on important aspects. The DADRCNN method is used to predict the future energy demand of electric vehicles considering the dynamic charging mode of electric vehicles.

[0147] DCU: Dilated convolution is an extension of the traditional convolution. The receptive field of the second dilated convolution layer gradually becomes entangled with the receptive field of the previous layer. The equation is:

[0148]

[0149] In the formula, R k and R k-1 Represent the receptive fields of the convolution kernels at the K layer and the K1 layer respectively. K Indicates the convolution kernel size; l k represents the expansion component, S n Represents the convolution size.

[0150] DRM: In order to extract more accurate data features, the ResNet structure is introduced. The main purpose of this structure is to transfer feature data from the front layer to the back layer through a shallow network and a residual function mapping layer. After the residual jump connection is established, the residual connection F(E)+E0 is obtained.

[0151] F(C)=ω(W2(Rel(ω(W1C+b1))))+b2) (32)

[0152] C'=ω(W3E+b3) (33)

[0153] Among them, W1, W2, W3 represent the weights of a single dilated convolutional layer, while b1, b2, b3 represent the non-linear activation function and the offset and normalization of the batch layer are well known.

[0154] CARM: CARM is introduced, which takes into account the variable values ​​of several landslide assessment factors. The CARM formula is as follows:

[0155]

[0156] Where A(·) represents average pooling, FE represents a fully interconnected layer, ρ represents an sigmoid curve, Z represents matrix multiplication, and Y represents matrix addition.

[0157] A hybrid EMS technology based on renewable energy system and energy storage system based on smart microgrid is proposed below. The primary and secondary control levels of EMS are discussed in detail. The primary and secondary control levels are part of EMS. In EMS, the microgrid controller (MGCC) handles the most important secondary control level in a centralized manner. Each control level will be described in depth along with the corresponding mathematical formulas.

[0158] The droop controlled DGs have characteristics used by the main controller. These characteristics are automatically used for frequency drift and distribution. Frequency drift is caused by the unpredictability of RES. Changes in MG load are also the cause of frequency deviation.

[0159]

[0160] Where: P g,h DG_ref is the active power reference, P s,g,h DG is the actual active power, is the power regulation cost factor, F 0 is the base cost.

[0161] The primary control level ensures that the voltage produced by all controlled distributed generators has the same steady-state angular frequency. Therefore, the frequency of the MG can be calculated as follows:

[0162]

[0163] An important component of the EMS is the Management Center located on the second level. The MGCC is actually responsible for ensuring that the IMGs operate as optimally as possible, within all technical and economic constraints. The local MGCC sets the strain level of the frequency response loads. The DGs with droop control and their active power references are established. In addition, it sets their droop gains.

[0164] A stochastic optimization function is used to model the MGCC objective in the manner described below.

[0165]

[0166] In the formula, α is the fixed operating cost, u is a binary variable representing the operating status of DG's, β is the first stage operating cost, and ρ DRP is the cost of demand response regulation, PDG_ref is the active power benchmark, δ is the startup / shutdown cost, u ON and u OFF Respectively represent the start and shutdown state of DG, ΔP DG represents the power deviation between DG and its reference, λ PV is the operating cost of PV, λ WT is the operating cost of WT, π is the scenario probability, ΔP FR is the power deviation related to the frequency response load, VLL is the loss load value, and LSH is the involuntary load reduction.

[0167] In developing the proposed EMS's second layer, constraints, power balance, controllable DGs, frequency security, droop constraints and renewable resources need to be considered.

Claims

1. An energy management method for a renewable energy system and an energy storage system based on FFO-DADRCNN, characterized by: The following steps are included: Step 1: Build an energy storage system ESS cost model, including ESS capital cost and ESS operating cost; Step 2: Construct wind turbine model, solar turbine model, microgrid load model and frequency response load model; Step 3: Set the voltage and current input factors of the wind turbine and photovoltaic system, and use the FFO algorithm to optimally allocate the energy of the microgrid; Step 4: Use the neural network architecture DADRCNN to predict the future energy demand load of electric vehicles; Step 5: Combine steps 1 to 4 to perform hybrid management of the energy management system EMS of the renewable energy system and energy storage system of the smart microgrid.

2. The energy management method of a renewable energy system and an energy storage system based on FFO-DADRCNN according to claim 1 is characterized by: step The operating cost of the ESS is caused by degradation and aging. The aging cost formula is: AC t =η t *IC ESS Where, IC ESS is the ESS installation cost, η t is the aging rate of ESS in a single cycle, and the aging rate is related to the number of cycles. about, and is the average SOC and SOC changes within a cycle function.

3. The energy management method of a renewable energy system and an energy storage system based on FFO-DADRCNN according to claim 1 is characterized in that: The wind turbine model described in step 2 uses a two-state up-down Markov model to evaluate the reliability of the wind turbine. The availability probability of the WTG in the model is equal to 1-q WTS and p WTS.rated The rated capacity is unavailable with a probability equal to q WTS ; Where: MTTR WTS MTTF WTS MTTR PVS MTTF PVS are the mean time between failures and mean time to repair for WT and PV respectively; The reliability of solar generators is modeled using a two-state Markov model: The microgrid load model uses the normal distribution model with uncertainty probability distribution function: Where P L represents the load demand of MG, μ L and σ L are the mean and standard deviation respectively.

4. The energy management method of a renewable energy system and an energy storage system based on FFO-DADRCNN according to claim 1 is characterized in that: Step 3 The method of the FFO algorithm for optimal energy allocation in the microgrid is: using an automatic parameter adjustment method, using fuzzy logic FL to dynamically set the parameter value of FFO; the overall size of FFO is determined using heuristic Dimensionality determination, by using heuristics to determine the size of the SL survival list; Automatically and dynamically control the value of each parameter, the distance δ of each solution from the best tested solution so far, and a function Ψ for the improvement of the objective function value of each potential solution relative to the previous iteration; The distance δ between two foxes is calculated as: Where i and j are the i-th and j-th flying foxes respectively, |·| is the corresponding distance metric, is the position of the i-th and j-th flying fox; Consider the current and previous positions of fox i in the last two time steps as: In the formula, δ max is the distance between the worst solution detected and the best solution detected; δ is used to calculate the distance between the fox and the coldest spot, and the function Ψ characterizes how the solution improves based on the objective function values ​​of the current and previous iterations.

5. The energy management method of a renewable energy system and an energy storage system based on FFO-DADRCNN according to claim 1 is characterized in that: Step 4: The neural network architecture DADRCNN for predicting future energy demand load of electric vehicles includes the following modules: DCU extended convolution unit: extended convolution is extended on the basis of traditional convolution; the receptive field of the second layer of dilated convolution is gradually entangled with the receptive field of the previous layer, and the equation is: In the formula, R k and R k-1 Represent the receptive fields of the convolution kernels at the K layer and the K1 layer respectively; where f K Indicates the convolution kernel size; l k represents the expansion component, S n Indicates the convolution size; DRM deep residual module: In order to extract more accurate data features, the Res network structure is introduced to transfer feature data from the front layer to the back layer through the shallow network and residual function mapping layer; after establishing the residual jump connection, the residual connection F(E)+E0 is obtained; F(C)=ω(W2(Rel(ω(W1C+b1)))+b2) C'=ω(W3E+b3) Where W1, W2, W3 represent the weights of a single extended convolutional layer, and b1, b2, b3 represent the nonlinear activation function, the offset of the batch layer, and the normalization; The CARM that considers the variable values ​​of several landslide evaluation factors is introduced. The CARM formula is as follows: Where A(·) represents average pooling, FE represents a fully interconnected layer, ρ represents an sigmoid curve, Z represents matrix multiplication, and Y represents matrix addition.

Images