New energy large base economic optimal control method and system

Abstract

The invention relates to the technical field of new energy power generation control, and provides a new energy large base economic optimal control method and system, and the method comprises the steps: building a multi-temporal-spatial scale power prediction model, and outputting a power prediction value; constructing a time-varying multi-objective optimization model of an objective function including economic cost, equipment loss cost and power grid constraint cost and constraint conditions; adjusting weight coefficients of the economic cost, the equipment loss cost and the power grid constraint cost; and solving the time-varying multi-objective optimization model after the weight coefficient is adjusted through an improved multi-objective particle swarm algorithm to obtain an optimal control instruction set. According to the method, by constructing a collaborative architecture of a dynamic weight adjustment mechanism and a multi-target optimization model, multiple targets such as power generation efficiency, equipment life and power grid stability can be dynamically coordinated, multi-target dynamic balance is realized, the defect that multiple targets are difficult to coordinate by a traditional fixed weight coefficient is overcome, and the economical efficiency and operation benefits of a new energy large base are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of new energy power generation control, and particularly relates to a new energy large base economic optimal control method and system. BACKGROUND

[0002] With the increasing proportion of renewable energy power generation capacity in the power grid, the intermittency, uncertainty and volatility characteristics of wind power, photovoltaic and other new energy power generation put higher requirements on the flexibility, stability and safety of the power grid. Traditional power system dispatching focuses on ensuring basic load and peak demand, and it is difficult to achieve fine management and rapid response of large-scale new energy power generation. The existing new energy large base control strategy has problems such as multi-objective conflict, prediction error sensitivity, insufficient economy and high computational complexity.

[0003] The traditional method adopts fixed weight coefficients, which is difficult to coordinate the dynamic balance of power generation efficiency, equipment life, grid stability and other multiple objectives. The static optimization model has poor robustness to new energy output prediction error, resulting in frequent oscillation of control instructions. In addition, the existing method does not fully consider the time-varying influence of power fluctuation on equipment maintenance cost and market electricity price fluctuation, and the economy is insufficient. At the same time, the computational complexity of multi-objective Pareto frontier solution is high, which is difficult to meet the requirements of minute-level real-time control.

[0004] Therefore, there is an urgent need for a new type of new energy large base economic optimal control method, which can dynamically coordinate multiple objectives, improve the robustness to prediction error, enhance the economy, and reduce the computational complexity, so as to realize the efficient, economic and stable operation of the new energy large base. SUMMARY

[0005] The present application aims to solve at least one of the technical problems in the background art, and provides a new energy large base economic optimal control method and system.

[0006] To achieve the above-mentioned purpose, the present application provides a new energy large base economic optimal control method, comprising:

[0007] An LSTM-Transformer hybrid neural network architecture is adopted to establish a multi-time and space scale power prediction model, and output power prediction values;

[0008] A time-varying multi-objective optimization model including an economic cost, a device wear cost and a grid constraint cost, and a constraint condition is constructed;

[0009] The weight coefficients of the economic cost, the device wear cost and the grid constraint cost are adjusted;

[0010] The power prediction value is input into the time-varying multi-objective optimization model after the adjustment weight coefficient, and the time-varying multi-objective optimization model after the adjustment weight coefficient is solved through the improved multi-objective particle swarm optimization algorithm to obtain an optimal control instruction set.

[0011] According to one aspect of the present application, a multi-temporal and spatial scale power prediction model is established by using an LSTM-Transformer hybrid neural network architecture to output a power prediction value, which includes:

[0012] The multi-temporal and spatial scale power prediction model is established based on the LSTM-Transformer hybrid neural network architecture, wherein the number of LSTM layers is 3 layers for extracting time series features, and the depth of the Transformer encoder stack is 4 layers for capturing the spatial correlation of meteorological elements.

[0013] Data is input into the multi-temporal and spatial scale power prediction model, and the input data includes irradiance, wind speed three-dimensional field data of numerical weather prediction, photovoltaic string IV characteristic curve, and historical power data collected by the SCADA system, and the sampling frequency is 1 Hz.

[0014] The output power prediction value is a future 4-hour power prediction value and its 95% confidence interval.

[0015] According to one aspect of the present application, the target function of the economic cost is:

[0016] C_econ=∫[λ(t)·P(t)+η·(dP(t) / dt) 2 ]dt;

[0017] In the formula, λ(t) is a real-time electricity price, indicating a power market price changing with time; P(t) is a new energy system output power, indicating an actual power generation at time t; η is a power fluctuation penalty coefficient, used to quantify the negative impact of the power change rate on the system economy; and dP(t) / dt is a power change rate, reflecting the instantaneous fluctuation amplitude of the power output.

[0018] The device loss cost function is:

[0019]

[0020] In the formula, α cycle is a battery cycle aging coefficient, reflecting the influence weight of battery cycle aging on the cost; DOD(t) is a discharge depth of the battery at time t, reflecting the discharge degree of the battery at the time; β inv is an inverter harmonic loss coefficient, used to measure the influence degree of inverter harmonic loss on the cost; and THD(t) is a total harmonic distortion at time t, representing the harmonic content of the current or voltage.

[0021] The power grid constraint cost function is:

[0022]

[0023] In the formula, μ is a frequency deviation penalty coefficient, used to measure the influence degree of frequency deviation on cost; Δf(t) is the frequency deviation at t moment, that is, the difference between actual frequency and rated frequency, reflecting the fluctuation of power grid frequency; ν is a voltage out-of-limit penalty coefficient, embodying the influence weight of voltage out-of-limit on cost; V n (t) is the voltage of the nth node at t moment; V lim,n is the voltage limit value of the nth node;

[0024] The time-varying multi-objective optimization model is constructed by combining the economic cost objective function, the equipment wear cost objective function and the power grid constraint cost objective function:

[0025] minf=α(t)·C_econ+β(t)·C_wear+γ(t)·C_grid;

[0026] In the formula, α(t), β(t) and γ(t) are weight coefficients of economic cost, equipment wear cost and power grid constraint cost at t moment respectively, satisfying the following constraints:

[0027] α(t)+β(t)+γ(t)=1.

[0028] According to one aspect of the present application, the constraint conditions of the time-varying multi-objective optimization model include:

[0029] Power balance constraint:

[0030] P pv (t)+P wind (t)+P bess (t)=P grid (t)

[0031] In the formula, P pv (t) is photovoltaic power generation power; P wind (t) is wind power generation power; P bess (t) is energy storage charge and discharge power; P grid (t) is power grid load power;

[0032] Photovoltaic and wind power constraint:

[0033]

[0034] In the formula, is the maximum power generation power of photovoltaic power prediction at t moment; is the maximum power generation power of wind power prediction at t moment;

[0035] Wind turbine climbing constraint:

[0036] P winddown ≤P wind (t)-P wind (t-1)≤P windup ;

[0037] P windup , P winddown are the up and down climbing rates of the wind turbine;

[0038] Voltage and frequency constraint:

[0039] V n,min ≤V n (t)≤V n,max , f min ≤f(t)≤f max ;

[0040] V n,min , V n,max are the minimum and maximum voltages allowed at the nth node; f(t) is the frequency of the system at time t; f min , f max are the maximum frequencies allowed by the system.

[0041] According to an aspect of the present application, the weight coefficient α(t) of the economic cost is:

[0042] α(t)=α base +k1·σ pv (t)+k2·|dλ / dt|;

[0043] α base is the base value of the economic weight, indicating the default weight when there is no external disturbance; k1 is the photovoltaic output fluctuation sensitive coefficient, adjusting the influence intensity of photovoltaic power fluctuation on the economic weight; σ pv (t) is the instantaneous standard deviation of photovoltaic output, reflecting the fluctuation amplitude of photovoltaic power in the time window; k2 is the price change rate gain coefficient, quantifying the adjustment of the price fluctuation on the economic weight; dλ / dt is the real-time price change rate, representing the dynamic change rate of the power market price;

[0044] The weight coefficient β(t) of the equipment loss cost is:

[0045] β(t)=β base ·e -τ·SOC(t) ;

[0046] β baseis the initial value of the device life weight, corresponding to the default weight when the battery health is best; τ is the battery health decay factor, controlling the influence of state of charge on the weight decay speed; SOC(t) is the real-time state of charge of the battery, with a value range of 0≤SOC(t)≤100%, reflecting the current remaining capacity of the battery;

[0047] The weight coefficient γ(t) of the power grid constraint cost is:

[0048] γ(t)=γ max -k3·Δf(t-Δt);

[0049] In the formula, γ max is the upper limit value of the power grid stability weight, corresponding to the maximum weight when there is no frequency deviation; k3 is a frequency deviation compensation coefficient, adjusting the suppression intensity of frequency fluctuation on the power grid stability weight; Δf(t-Δt) is the power grid frequency deviation in a delay time window, and the calculation formula is:

[0050] Δf(t-Δt)=f(t-Δt)-f nomal ;

[0051] In the formula, f nomal is the rated frequency; and Δt is the response delay time of the control system.

[0052] According to one aspect of the present application, the improved multi-objective particle swarm algorithm is used to calculate the time-varying multi-objective optimization model, which comprises:

[0053] Pareto solution set management: an adaptive K-means clustering algorithm is adopted, and the non-inferior solution set is divided into K clusters according to the Euclidean distance in the target space, and the value of K is dynamically adjusted to be 10%-15% of the population size;

[0054] A hybrid convergence enhancement mechanism is adopted, and the inertia weight is adjusted to be a nonlinear parameter:

[0055] ω=[0.9-0.4·(t / T) 2 ]·(1+δ·sin(πt / 2T));

[0056] In the formula, δ=0.15 is a quantum disturbance factor;

[0057] The individual learning factor c1 adopts a U-shaped curve adjustment strategy, and is adjusted to be:

[0058]

[0059] In the formula, c 1max , c 1min are the maximum value and the minimum value of c1.

[0060] To achieve the above object, the application further provides a new energy large base economic optimal control system, comprising:

[0061] A multi-time-and-space scale power prediction model construction module adopts an LSTM-Transformer hybrid neural network architecture to establish a multi-time-and-space scale power prediction model and output a power prediction value;

[0062] A time-varying multi-objective optimization model construction module constructs a time-varying multi-objective optimization model including an objective function of economic cost, equipment wear cost and grid constraint cost and constraint conditions;

[0063] A weight coefficient adjustment module adjusts weight coefficients of the economic cost, the equipment wear cost and the grid constraint cost;

[0064] An optimal control instruction set acquisition module inputs the power prediction value into the time-varying multi-objective optimization model after the weight coefficients are adjusted, solves the time-varying multi-objective optimization model after the weight coefficients are adjusted through an improved multi-objective particle swarm optimization algorithm, and obtains an optimal control instruction set.

[0065] To achieve the above object, the application further provides an electronic device, comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the computer program is executed by the processor to implement the new energy large base economic optimal control method.

[0066] To achieve the above object, the application further provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the computer program is executed by a processor to implement the new energy large base economic optimal control method.

[0067] According to the scheme of the application, the dynamic weight adjustment mechanism and the multi-objective optimization model are cooperatively constructed, the multiple targets such as power generation efficiency, equipment life and grid stability are dynamically coordinated, the dynamic balance of the multiple targets is achieved, the shortcoming of the traditional fixed weight coefficient that is difficult to coordinate the multiple targets is overcome, and the economy and operation benefit of the new energy large base are improved.

[0068] The dynamic optimization model is introduced in the application, the robustness to the new energy output prediction error is improved, the frequent oscillation of the control instruction is avoided, and the uncertainty and volatility of the new energy output are effectively coped with;

[0069] The application considers the time-varying influence of power fluctuation on equipment maintenance cost and market electricity price fluctuation, and improves the economy and operation benefit of the new energy large base;

[0070] The improved multi-objective particle swarm optimization algorithm is designed, and the following is introduced: (1) Pareto solution set management: an adaptive K-means clustering algorithm is adopted, the non-inferior solution set is divided into K clusters according to the Euclidean distance of the target space, and the value of K is dynamically adjusted to be 10%-15% of the population size; (2) a hybrid convergence enhancement mechanism is adopted, and the inertia weight is adjusted to be a nonlinear parameter; (3) the individual learning factor adopts a U-shaped curve adjustment strategy, the solution time is controlled within 30 seconds / cycle, the time requirement of minute-level real-time control is met, and the calculation complexity is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 A flow chart schematically representing a new energy large base economic optimal control method according to an embodiment of the application. DETAILED DESCRIPTION

[0072] The content of the present application will now be discussed with reference to exemplary embodiments. It should be understood that the discussed embodiments are only to enable those of ordinary skill in the art to better understand and thus implement the content of the present application, and are not intended to imply any limitation on the scope of the present application.

[0073] As used herein, the term "comprising" and variations thereof are to be construed as meaning "including, but not limited to". The term "based on" is to be construed as "based at least in part on". The terms "one embodiment" and "an embodiment" are to be construed as "at least one embodiment".

[0074] Figure 1 A flow chart schematically representing a new energy large base economic optimal control method according to an embodiment of the application. As shown in Figure 1 In the present embodiment, the new energy large base economic optimal control method comprises:

[0075] An LSTM-Transformer hybrid neural network architecture is adopted to establish a multi-time and space scale power prediction model, and output power prediction values;

[0076] A time-varying multi-objective optimization model including an economic cost, a device loss cost and a power grid constraint cost, a target function and a constraint condition is constructed;

[0077] Adjust the weight coefficients of economic cost, equipment wear cost and grid constraint cost; specifically, consider the weight coefficient of economic cost of photovoltaic output fluctuation sensitivity coefficient, photovoltaic output instantaneous standard deviation, price change rate gain coefficient and real-time price change rate; consider the weight coefficient of equipment wear cost of battery health degree attenuation factor and real-time state of charge; consider the weight coefficient of grid stability weight upper limit value, frequency deviation compensation coefficient and grid frequency deviation in delay time window to adjust the weight coefficient of grid constraint cost. The time-varying multi-objective optimization model constructed above is calculated to achieve the purpose of maximizing economic benefit of each energy output plan;

[0078] The power prediction value is input into the time-varying multi-objective optimization model after adjusting the weight coefficient, and the time-varying multi-objective optimization model after adjusting the weight coefficient is solved by an improved multi-objective particle swarm optimization algorithm to obtain an optimal control instruction set (i.e. the output plan curve of each energy in the new energy base).

[0079] Further, according to an embodiment of the present application, a LSTM-Transformer hybrid neural network architecture is used to establish a multi-time and space scale power prediction model to output a power prediction value, which includes:

[0080] The LSTM-Transformer hybrid neural network architecture is used to establish a multi-time and space scale power prediction model, wherein the number of LSTM layers is 3 layers for extracting time sequence features, and the depth of the Transformer encoder stack is 4 layers for capturing the spatial correlation of meteorological elements;

[0081] The data is input into the multi-time and space scale power prediction model, and the input data includes irradiance, wind speed three-dimensional field data of numerical weather prediction (NWP), photovoltaic string IV characteristic curve and historical power data collected by the SCADA system, and the sampling frequency is 1Hz;

[0082] In the inference stage, the Monte Carlo Dropout (with a reservation probability of 0.8) is enabled, 50 random forward propagations are performed, and future 4-hour power prediction values and their 95% confidence intervals are generated.

[0083] Further, according to an embodiment of the present application, the target function of economic cost is:

[0084] C_econ=∫[λ(t)·P(t)+η·(dP(t) / dt) 2 ]dt;

[0085] In the formula: λ(t) is the real-time electricity price (unit: yuan / kWh), which represents the power market price changing with time; P(t) is the new energy system output power (unit: kW), which represents the actual power generation at time t; η is the power fluctuation penalty coefficient (unit: yuan / (kW / s)), which is used to quantify the negative impact of power change rate on system economy; dP(t) / dt is the power change rate (unit: kW / s), which reflects the instantaneous fluctuation amplitude of power output; 2

[0086] Among them, the real-time electricity price is obtained through the power market API interface, updated every 5 minutes, and synchronized with the power prediction value; the penalty coefficient is dynamically adjusted according to the grid dispatching instruction.

[0087] By accumulating the battery cycle aging loss and inverter harmonic loss at each time, the equipment loss cost of the whole time period is obtained, and the equipment loss cost function is:

[0088]

[0089] In the formula: α cycle is the battery cycle aging coefficient, which reflects the influence weight of battery cycle aging on cost; DOD(t) is the discharge depth of the battery at t, which reflects the discharge degree of the battery at that time; β inv is the inverter harmonic loss coefficient, which is used to measure the influence degree of inverter harmonic loss on cost; THD(t) is the total harmonic distortion at t, which represents the harmonic content of current or voltage;

[0090] Among them, the battery cycle aging coefficient is calculated based on the Arrhenius equation, which is in exponential relationship with the discharge depth (DOD), and the real-time state of charge (SOC) is fed back by the BMS system; the total harmonic distortion (THD) is collected in real time through the grid-connected point power quality monitoring device, and the loss coefficient is proportional to the square of THD.

[0091] By accumulating the frequency deviation cost and the node voltage out-of-limit cost at each time, the grid constraint cost of the whole time period is obtained, and the grid constraint cost function is:

[0092]

[0093] In the formula: μ is the frequency deviation penalty coefficient, which is used to measure the influence degree of frequency deviation on cost; Δf(t) is the frequency deviation at t, that is, the difference between the actual frequency and the rated frequency, which reflects the fluctuation of grid frequency; ν is the voltage out-of-limit penalty coefficient, which reflects the influence weight of voltage out-of-limit on cost; V n (t) is the voltage of the nth node at t; V lim,n is the voltage limit value of the nth node; ​

[0094] Wherein, the frequency deviation is measured by the PMU device at a sampling rate of 50Hz; the node voltage overrun cost is calculated by segmenting the overrun amplitude.

[0095] In combination with the economic cost target function, the equipment wear cost target function and the grid constraint cost target function, a time-varying multi-objective optimization model is constructed:

[0096] minf=α(t)·C_econ+β(t)·C_wear+γ(t)·C_grid;

[0097] In the formula, α(t), β(t) and γ(t) are weight coefficients of the economic cost, the equipment wear cost and the grid constraint cost at t moment respectively, and satisfy the following constraints:

[0098] α(t)+β(t)+γ(t)=1.

[0099] Further, according to an embodiment of the present application, the constraint conditions of the time-varying multi-objective optimization model include:

[0100] Power balance constraint:

[0101] P pv (t)+P wind (t)+P bess (t)=P grid (t)

[0102] In the formula, P pv (t) is photovoltaic power; P wind (t) is wind power; P bess (t) is energy storage charging and discharging power; P grid (t) is grid load power.

[0103] Photovoltaic and wind power constraint:

[0104]

[0105] In the formula: is the maximum power generation at t moment of photovoltaic power prediction; is the maximum power generation at t moment of wind power prediction;

[0106] Wind turbine climbing constraint:

[0107] P winddown ≤P wind (t)-P wind (t-1)≤P windup ;

[0108] In the formula: P windup , P winddownThe up and down ramp rates of the wind turbine;

[0109] Voltage and frequency constraints:

[0110] V n,min ≤V n (t)≤V n,max , f min ≤f(t)≤f max ;

[0111] Wherein: V n,min , V n,max are the minimum and maximum voltages allowed for the nth node; f(t) is the frequency of the system at time t; f min , f max are the maximum frequencies allowed by the system.

[0112] Further, according to an embodiment of the present application, the weight coefficient a(t) of the economic cost is:

[0113] a(t)=a base +k1·σ pv (t)+k2·|dλ / dt|;

[0114] Wherein: a ase b is the base value of the economic weight, indicating the default weight when there is no external disturbance; k1 is the photovoltaic output fluctuation sensitivity coefficient (unit: 1 / kW), adjusting the influence intensity of photovoltaic power fluctuation on the economic weight; σ pv (t) is the instantaneous standard deviation of photovoltaic output (unit: kW), reflecting the fluctuation amplitude of photovoltaic power in the time window; k2 is the price change rate gain coefficient (unit: 1 / (yuan / kWh·s), quantifying the adjustment of the price fluctuation on the economic weight; dλ / dt is the real-time price change rate (unit: yuan / (kWh·s)), representing the dynamic change rate of the power market price;

[0115] Specifically, when the photovoltaic output fluctuation standard deviation exceeds 50 kW or the price change rate is greater than 0.1 yuan / (kWh·s), the weight coefficient is automatically increased by 20%-40%, giving priority to the economic target;

[0116] The weight coefficient b(t) of the equipment loss cost is:

[0117] b(t)=b base ·e -τ·SOC(t) ;

[0118] Wherein: b baseis the initial value of the device life weight, corresponding to the default weight when the battery health is best; τ is the battery health decay factor (unit: 1 / %) to control the influence of state of charge (SOC) on the decay speed of the weight; SOC(t) is the real-time state of charge of the battery (unit: %), the value range is 0≤SOC(t)≤100%, reflecting the current remaining capacity of the battery;

[0119] Specifically, when the state of charge (SOC) is lower than 20%, the battery health decay factor triggers a nonlinear growth mechanism, and the weight coefficient is reduced to 30%-50% of the initial value, avoiding deep discharge loss;

[0120] The weight coefficient γ(t) of the grid constraint cost is:

[0121] γ(t)=γ max -k3·Δf(t-Δt);

[0122] In the formula, γ max is the upper limit value of the grid stability weight, corresponding to the maximum weight when there is no frequency deviation; k3 is the frequency deviation compensation coefficient (unit: 1 / Hz), adjusting the suppression intensity of the frequency fluctuation on the grid stability weight; Δf(t-Δt) is the grid frequency deviation in the delay time window (unit: Hz), and the calculation formula is:

[0123] Δf(t-Δt)=f(t-Δt)-f nomal ;

[0124] In the formula, f nomal is the rated frequency (such as 50Hz or 60Hz); and Δt is the response delay time of the control system (unit: s);

[0125] Specifically, the frequency deviation time window is set to 10 seconds, and if the average frequency deviation in the continuous 3 windows exceeds ±0.2Hz, the grid constraint weight coefficient is increased to 1.5 times of the upper limit value, and the safety constraint is forced to be met.

[0126] Further, according to an embodiment of the present application, the improved multi-objective particle swarm algorithm is used to calculate the time-varying multi-objective optimization model constructed above, and the method comprises:

[0127] Pareto solution set management: an adaptive K-means clustering algorithm is used to divide the non-inferior solution set into K clusters according to the Euclidean distance in the target space, and the value of K is dynamically adjusted to be 10%-15% of the population size;

[0128] A hybrid convergence enhancement mechanism is used, and the inertia weight is adjusted to be a nonlinear parameter:

[0129] ω=[0.9-0.4·(t / T) 2(1+delta*sin(pi*t / 2T));

[0130] delta=0.15 is quantum disturbance factor;

[0131] The individual learning factor c1 adopts a U-shaped curve adjustment strategy and is adjusted as follows:

[0132]

[0133] In the formula, c 1max , c 1min are the maximum and minimum values of c1.

[0134] The improved multi-objective particle swarm optimization algorithm is used to solve the time-varying multi-objective optimization model constructed by the application to obtain the economic optimal energy output results of the new energy large base, and the specific steps are as follows:

[0135] Step one: initialize the particle swarm, randomly initialize the position and speed of the particle, each particle represents a set of decision variables, including the charge and discharge power of the energy storage system, the power distribution of the photovoltaic power station and the wind power station, etc. The search space is dynamically contracted according to the upper and lower limits of the predicted power.

[0136] Step two: calculate the fitness value, according to the objective function of the time-varying multi-objective optimization model, calculate the fitness value of each particle. The lower the fitness value, the better the decision scheme corresponding to the particle.

[0137] Step three: Pareto solution set management, adopt adaptive K-means clustering algorithm, divide the non-inferior solution set into K clusters according to the Euclidean distance of the target space, and dynamically adjust the K value to 10%-15% of the population size. Through clustering, similar non-inferior solutions can be gathered together, which is convenient for subsequent selection and updating.

[0138] Step four: update the speed and position of the particle according to the update formula, so that the particle moves to a better solution direction. The inertia weight is adjusted as a nonlinear parameter and dynamically changes according to the iteration number. In the early stage (<50 generations), the inertia weight is kept at 0.9 to enhance global search, and in the later stage (>150 generations), it is reduced to 0.4 to improve the local optimization ability, and in the middle stage, it is smoothly transitioned according to the Sigmoid function; the individual learning factor adopts a U-shaped curve adjustment strategy, and the value is smaller in the early and late stages of iteration, and the value is larger in the middle stage of iteration. The initial value of the individual learning factor is 2.5, which changes in a U-shaped manner with the iteration number: linearly decreases to 1.0 in the first 1 / 3 stage, and rises to 2.0 in the last 1 / 3 stage, balancing the exploration and development ability.

[0139] Step five: judge the termination condition, if the maximum iteration number is reached or other termination conditions are met, stop iteration and output the optimal solution.

[0140] According to the above scheme of the present application, the present application can dynamically coordinate multiple targets such as power generation efficiency, equipment life, and power grid stability by constructing a dynamic weight adjustment mechanism and a collaborative architecture of a multi-objective optimization model, realize dynamic balance of multiple targets, overcome the shortcomings of traditional fixed weight coefficients that are difficult to coordinate multiple targets, and improve the economy and operation benefit of a new energy large base.

[0141] The present application introduces a dynamic optimization model, improves the robustness of new energy output prediction error, avoids frequent oscillation of control instructions, and effectively deals with the uncertainty and volatility of new energy output.

[0142] The present application considers the time-varying influence of power fluctuation on equipment maintenance cost and market electricity price fluctuation, improves the economy and operation benefit of a new energy large base.

[0143] The present application designs an improved multi-objective particle swarm optimization algorithm, introduces (1) Pareto solution set management: an adaptive K-means clustering algorithm is used to divide the non-inferior solution set into K clusters according to the Euclidean distance of the target space, and the value of K is dynamically adjusted to 10%-15% of the population size; (2) a hybrid convergence enhancement mechanism is used, and the inertia weight is adjusted to a nonlinear parameter; (3) the individual learning factor is adjusted using a U-shaped curve strategy, and the solution time is controlled within 30 seconds / period, meeting the time requirements of minute-level real-time control and reducing the computational complexity.

[0144] Further, in order to achieve the above purpose, the present application also provides a new energy large base economic optimal control system, comprising:

[0145] A multi-time and space scale power prediction model construction module adopts an LSTM-Transformer hybrid neural network architecture to establish a multi-time and space scale power prediction model and output a power prediction value.

[0146] A time-varying multi-objective optimization model construction module constructs a time-varying multi-objective optimization model including an economy cost, an equipment wear cost, and a power grid constraint cost, and a constraint condition.

[0147] A weight coefficient adjustment module adjusts the weight coefficients of the economy cost, the equipment wear cost, and the power grid constraint cost.

[0148] An optimal control instruction set acquisition module solves the time-varying multi-objective optimization model after adjusting the weight coefficients by using an improved multi-objective particle swarm optimization algorithm to obtain an optimal control instruction set.

[0149] The above-mentioned new energy large base economic optimal control system according to the present application can realize the above-mentioned new energy large base economic optimal control method, and the specific process steps are as described above, which will not be repeated here.

[0150] Further, to achieve the above object, the present application provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, and the computer program, when executed by the processor, implements the new energy large base economic optimal control method.

[0151] Further, to achieve the above object, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the new energy large base economic optimal control method.

[0152] Those skilled in the art can understand that the modules and algorithm steps described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0153] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the above-described device and equipment can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0154] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. In actual implementation, another division mode can be used. For example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed modules can be indirect coupling or communication connection through some interface, device or module, and can be electrical, mechanical or other forms.

[0155] The modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, i.e. they can be located in one place or distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0156] In addition, each functional module in the embodiments of the present application can be integrated into a processing module, or each module can exist physically independently, or two or more modules can be integrated into one module.

[0157] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the energy-saving signal transmission / reception method of the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk, and various media that can store program codes.

[0158] The above description is merely preferred embodiments of the present application and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the application involved in the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features can be replaced with the technical features disclosed in the present application (but not limited to) having similar functions to form technical solutions.

[0159] It should be understood that the sequence of the steps in the summary and embodiments of the present application does not absolutely mean the order of execution, and the execution order of the processes should be determined according to their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

Claims

1. The economic optimal control method for a large new energy base is characterized by: include: Adopting the LSTM-Transformer hybrid neural network architecture, a multi-temporal and spatial-scale power prediction model is established to output power prediction values; Construct a time-varying multi-objective optimization model that includes objective functions and constraints including economic cost, equipment loss cost, and grid constraint cost; Adjust the weight coefficients of economic cost, equipment loss cost and grid constraint cost; The power prediction value is input into the time-varying multi-objective optimization model after adjusting the weight coefficient. The time-varying multi-objective optimization model after adjusting the weight coefficient is solved by the improved multi-objective particle swarm algorithm to obtain the optimal control instruction set.

2. The economic optimal control method for a large new energy base according to claim 1 is characterized in that: Using the LSTM-Transformer hybrid neural network architecture, a multi-time and space-scale power prediction model is established to output power prediction values, including: A multi-spatiotemporal power prediction model was established based on a hybrid LSTM-Transformer neural network architecture. Three LSTM layers were used to extract time series features, and a four-layer Transformer encoder stack was used to capture the spatial correlation of meteorological elements. Input data into the multi-temporal-scale power prediction model. The input data includes irradiance and three-dimensional wind speed field data from numerical weather forecasts, PV string IV characteristic curves, and historical power data collected by the SCADA system. The sampling frequency is 1 Hz. The output power forecast value is the power forecast value for the next 4 hours and its 95% confidence interval.

3. The economic optimal control method for a large new energy base according to claim 1 is characterized in that: The objective function of the economic cost is: C_econ=∫[λ(t)·P(t)+η·(dP(t) / dt) 2 ]dt; Where: λ(t) is the real-time electricity price, which represents the time-varying electricity market price; P(t) is the output power of the renewable energy system, which represents the actual generated power at time t; η is the power fluctuation penalty coefficient, which is used to quantify the negative impact of the power change rate on the system economy; dP(t) / dt is the power change rate, which reflects the instantaneous fluctuation amplitude of power output; The equipment loss cost function is: Where: α cycle is the battery cycle aging coefficient, which reflects the weight of the impact of battery cycle aging on cost; DOD(t) is the depth of discharge of the battery at time t, which reflects the degree of discharge of the battery at that moment; β inv is the harmonic loss coefficient of the inverter, which is used to measure the impact of the inverter harmonic loss on the cost; THD(t) is the total harmonic distortion at time t, representing the harmonic content of current or voltage; The grid constraint cost function is: Where: μ is the frequency deviation penalty coefficient, which is used to measure the impact of frequency deviation on cost; Δf(t) is the frequency deviation at time t, that is, the difference between the actual frequency and the rated frequency, reflecting the fluctuation of the grid frequency; ν is the voltage over-limit penalty coefficient, which reflects the impact weight of voltage over-limit on cost; V n (t) is the voltage of the nth node at time t; V lim,n is the voltage limit of the nth node; Combining the economic cost objective function, equipment loss cost objective function and grid constraint cost objective function, a time-varying multi-objective optimization model is constructed: minf=α(t)·C_econ+β(t)·C_wear+γ(t)·C_grid; Where: α(t), β(t) and γ(t) are the weight coefficients of economic cost, equipment loss cost and grid constraint cost at time t, respectively, satisfying the following constraints: α(t)+β(t)+γ(t)=1.

4. The economic optimal control method for a large new energy base according to claim 1 is characterized in that: The constraints of the time-varying multi-objective optimization model include: Power balance constraints: P pv (t)+P wind (t)+P bess (t)=P grid (t) Where: P pv (t) is the photovoltaic power generation power; P wind (t) is the wind power generation power; P bess (t) is the energy storage charging and discharging power; P grid (t) is the grid load power; Photovoltaic and wind power power constraints: Where: is the maximum power generation power at time t predicted by photovoltaic power; is the maximum power generated by wind power at time t; Wind turbine climbing constraints: P winddown ≤P wind (t)-P wind (t-1)≤P windup ; Where: P windup 、P winddown is the up and down ramp rate of the wind turbine; Voltage and frequency constraints: V n,min ≤V n (t)≤V n,max ,f min ≤f(t)≤f max ; Where: V n,min 、V n,max is the minimum and maximum voltage allowed at the nth node; f(t) is the frequency of the system at time t; f min 、f max The maximum frequency allowed by the system.

5. The economic optimal control method for a large new energy base according to claim 3 is characterized in that: The weight coefficient α(t) of the economic cost is: α(t)=α base +k1·s pv (t)+k2·|dλ / dt|; Where: αb ase is the basic value of the economic weight, indicating the default weight when there is no external disturbance; k1 is the photovoltaic output fluctuation sensitivity coefficient, which adjusts the intensity of the impact of photovoltaic power fluctuation on the economic weight; σ pv (t) is the instantaneous standard deviation of photovoltaic output, reflecting the fluctuation amplitude of photovoltaic power generation within the time window; k2 is the electricity price change rate gain coefficient, which quantifies the adjustment of electricity price fluctuations on economic weight; dλ / dt is the real-time electricity price change rate, which represents the dynamic change rate of electricity market price; The weight coefficient β(t) of the equipment loss cost is: β(t)=β base ·e -τ·SOC(t) ; Where: β base is the initial value of the device life weight, corresponding to the default weight when the battery is in optimal health; τ is the battery health decay factor, which controls the impact of the state of charge on the weight decay rate; SOC(t) is the real-time state of charge of the battery, with a value range of 0≤SOC(t)≤100%, reflecting the current remaining capacity of the battery; The weight coefficient γ(t) of the grid constraint cost is: γ(t)=γ max -k3·Δf(t-Δt); Where: γ max is the upper limit of the grid stability weight, corresponding to the maximum weight when there is no frequency deviation; k3 is the frequency deviation compensation coefficient, which adjusts the suppression intensity of frequency fluctuation on the grid stability weight; Δf(t-Δt) is the grid frequency deviation within the delay time window, and the calculation formula is: Δf(t-Δt)=f(t-Δt)-f nomal ; Where: f nomal is the rated frequency; Δt is the response delay time of the control system.

6. The economic optimal control method for a large new energy base according to claim 1 is characterized in that: The improved multi-objective particle swarm optimization algorithm is used to calculate the time-varying multi-objective optimization model, which includes: Pareto solution set management: Adaptive K-means clustering algorithm is used to divide the non-inferior solution set into K clusters based on the Euclidean distance in the target space. The K value is dynamically adjusted to 10%-15% of the population size. A hybrid convergence enhancement mechanism is used, and the inertia weight is adjusted to a nonlinear parameter: ω=[0.9-0.4·(t / T) 2 ]·(1+δ·sin(πt / 2T)); Where: δ = 0.15 is the quantum perturbation factor; The individual learning factor c1 adopts a U-shaped curve adjustment strategy and is adjusted to: Where: c 1max 、c 1min are the maximum and minimum values ​​of c1.

7. The economic optimal control system of a large new energy base is characterized by: include: The multi-temporal and spatial scale power prediction model construction module uses the LSTM-Transformer hybrid neural network architecture to establish a multi-temporal and spatial scale power prediction model and output power prediction values; A time-varying multi-objective optimization model construction module, which constructs a time-varying multi-objective optimization model including objective functions and constraints including economic cost, equipment loss cost, and grid constraint cost; A weight coefficient adjustment module adjusts the weight coefficients of economic cost, equipment loss cost and grid constraint cost; The optimal control instruction set acquisition module inputs the power prediction value into the time-varying multi-objective optimization model after adjusting the weight coefficient, and solves the time-varying multi-objective optimization model after adjusting the weight coefficient through the improved multi-objective particle swarm algorithm to obtain the optimal control instruction set.

8. An electronic device, characterized in that The method comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the method for optimal economic control of a large new energy base as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the economic optimal control method for a large new energy base according to any one of claims 1 to 6 is implemented.

Images