Active roof photovoltaic and distribution network energy bidirectional interaction control system

Abstract

The invention relates to an active roof photovoltaic and distribution network energy bidirectional interaction control system, which comprises a roof photovoltaic system, a bidirectional photovoltaic inverter, an energy storage system, a bidirectional energy storage inverter, a control center and a cloud platform, the energy storage system is in interactive connection with the power grid through the bidirectional energy storage inverter, the control center monitors the generating capacity of the photovoltaic system, the energy storage state of the energy storage system and power grid parameters in real time through a network, the bidirectional photovoltaic inverter and the bidirectional energy storage inverter interact with the control center, and the cloud platform is connected with the control center; through active bidirectional energy interaction control, collaborative optimization of a photovoltaic system and an energy storage system is realized, the electric energy management efficiency and economy are improved, reactive support and voltage regulation are provided for a power grid, and stable operation of the system and the power grid is ensured.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic power generation and power distribution. Background Art

[0002] Rooftop photovoltaic system is a typical form of distributed energy, which converts solar energy into electrical energy and transmits the electricity to the distribution network through grid-connected inverters. However, with the rapid growth of rooftop photovoltaic installations, many regions are facing challenges of grid voltage fluctuations and frequency instability.

[0003] Traditional energy management methods mostly use passive interaction: the excess power generated by rooftop photovoltaic power is directly fed back to the grid, and during peak hours, the power company dispatches the dispatchable power source to balance it. This passive mode usually lacks intelligent optimization methods, resulting in the following deficiencies in the system: the power fed back to the grid cannot be flexibly reduced during peak power consumption, which may cause grid congestion or overload; when electricity prices are low, it is impossible to actively plan charging and energy storage, missing the opportunity to participate in electricity market arbitrage.

[0004] Common energy storage management strategies include:

[0005] (1) Simple scheduled charging and discharging strategy: Some systems adopt a preset scheduled charging and discharging strategy, such as storing excess electricity during the peak period of photovoltaic power generation. However, this method does not take into account changes in the grid status and market electricity prices and lacks flexibility.

[0006] (2) Voltage regulation based on inverter reactive power support: Some PV systems provide reactive power support through inverters to regulate voltage. However, the reactive power control of such systems is mostly fixed strategy and cannot dynamically adapt to real-time load changes.

[0007] (3) Island operation mode: When a grid failure occurs, some PV systems will switch to island operation mode. However, the switching process mostly relies on passive detection, which has a slow response speed. In addition, the energy management strategy in the island mode is simple, which easily leads to resource waste.

[0008] The control method of photovoltaic energy storage integrated system currently combines photovoltaic and energy storage systems to achieve partial load power supply and energy storage control, but it is usually limited to one-way load power supply and lacks coordinated interaction with the distribution network.

[0009] Regarding photovoltaic optimization dispatch strategies in the electricity market, some schemes use price signals to guide energy storage devices to charge and discharge, but most of them are simple peak-valley electricity price responses and fail to dynamically optimize power flow in combination with real-time grid parameters.

[0010] Defects and shortcomings of existing technology:

[0011] (1) Lack of active optimization capabilities. Most existing photovoltaic systems are passively controlled and can only perform dispatch operations according to the instructions of the power company, but cannot autonomously optimize energy interaction according to the state of the power grid. When market electricity prices, weather conditions or load demand change, the charging and discharging strategy cannot be adjusted quickly, resulting in low system economy.

[0012] (2) Inflexible two-way interaction. Many photovoltaic and energy storage systems cannot achieve efficient two-way interaction with the power grid. During peak hours, photovoltaic power generation may not be fully absorbed, resulting in frequent abandonment of power. When electricity prices are low, the system cannot actively participate in charging and energy storage.

[0013] (3) The energy management system is fragmented and cannot be optimized collaboratively. The management of photovoltaic systems and energy storage systems is mostly independent, lacking an efficient collaborative optimization mechanism, and cannot achieve a balance between the overall economy of the system and the stability of the power grid. Reactive power support and grid voltage regulation functions are mostly fixed strategies, and cannot be adaptively adjusted according to the real-time status of the power grid.

[0014] (4) Lack of rapid response and intelligent control. When faced with power grid anomalies (such as voltage fluctuations and frequency anomalies), traditional systems have a slow response speed and are unable to provide timely and effective support. Since the prediction and dispatch strategies in existing solutions are relatively simple, the system cannot learn and optimize through historical data, which limits the level of intelligence. Summary of the invention

[0015] In view of the above-mentioned deficiencies in the prior art, the present invention proposes an active rooftop photovoltaic and distribution network energy two-way interactive control system, which aims to solve the problems of lack of active optimization control between distributed photovoltaic systems and energy storage systems, inflexible two-way energy interaction, and insufficient grid support.

[0016] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: an active rooftop photovoltaic and distribution network energy bidirectional interactive control system, including a rooftop photovoltaic system, a bidirectional photovoltaic inverter, an energy storage system, a bidirectional energy storage inverter, a control center and a cloud platform, the photovoltaic system and the energy storage system are connected through a bidirectional inverter, the energy storage system and the power grid are interactively connected through a bidirectional energy storage inverter, the control center monitors the power generation of the photovoltaic system, the energy storage status of the energy storage system and the power grid parameters in real time through the network, the bidirectional photovoltaic inverter and the bidirectional energy storage inverter interact with the control center, and the cloud platform is connected to the control center;

[0017] Monitor the status data of the photovoltaic system, power grid and energy storage system in real time, and update the forecast data; formulate a two-way energy interaction plan based on the real-time monitoring data and forecast results, optimize the scheduling and dynamically adjust the power output of photovoltaic power generation and energy storage equipment; after each scheduling cycle, record the actual execution results, compare them with the forecast plan, and perform feedback learning and parameter optimization based on the error.

[0018] The status data of the photovoltaic system, power grid and energy storage system include: the power demand Pload(t) of the user-side load, the current energy state E(t) and charging and discharging power Pstorage(t) of the energy storage device, the voltage V(t), frequency f(t) and current power interaction status Pgrid(t) of the distribution network, and the real-time electricity prices Cbuy(t) and Csell(t) obtained from the cloud platform.

[0019] The method for predicting data is: using a hybrid prediction model of ARIMA and LSTM,

[0020] The ARIMA model handles the linear portion of the time series:

[0021] The linear trend of the time series is identified through the ARIMA model, and the linear part forecast value of the time series is obtained:

[0022]

[0023] Where: y t : represents the actual data at time t;

[0024] ARIMA: Autoregressive Integrated Moving Average model, used to identify and predict linear trends in time series;

[0025] represents the linear prediction value obtained by the ARIMA model at time t;

[0026] The LSTM model handles the nonlinear part of the time series:

[0027] Subtract the linear prediction result of ARIMA from the original time series to get the residual series, which contains the nonlinear part of the data:

[0028]

[0029] Where: r(t): residual sequence, which represents the linear prediction result of subtracting ARIMA from the original time series y(t) The difference after

[0030] y(t): the observed value or true value of the original time series;

[0031] The linear part results predicted by the ARIMA model;

[0032] Use the LSTM model to predict the residual sequence and obtain the nonlinear part prediction value:

[0033]

[0034] in: Tong represents the nonlinear prediction value at time t;

[0035] LSTM: Long short-term memory network, used to capture nonlinear relationships and long-term dependency characteristics in time series;

[0036] r(1),r(2)...,r(t): the values ​​of the residual sequence at each moment, used as the input of the LSTM model;

[0037] Combined prediction results:

[0038] The prediction results of the ARIMA model and the LSTM model are weighted and combined to obtain the final prediction value:

[0039]

[0040] in: The final hybrid model prediction result includes the superposition of linear and nonlinear parts.

[0041] The linear part of the ARMIA model prediction;

[0042] The nonlinear part of the LSTM model prediction;

[0043] Hybrid prediction model:

[0044] Input monitored time series data y(t),

[0045] ARIMA linear prediction: Use the ARIMA model to predict the linear trend and get the linear part prediction value

[0046] Calculate the residual sequence: Calculate the residual

[0047]

[0048] Where: r(t): residual sequence, representing the difference between the original time series y(t) and the ARIMA linear prediction value The residual sequence r(t) represents the part of the time series that cannot be explained by the linear model, that is, the nonlinear part;

[0049] LSTM nonlinear prediction: Use the LSTM model to predict the residual sequence r(t) and obtain the nonlinear part prediction value

[0050] Combined prediction results: Add the prediction results of the linear part and the nonlinear part to get the final prediction value

[0051] The method for optimizing scheduling and dynamically adjusting the power output of photovoltaic power generation and energy storage equipment is:

[0052] Scheduling objectives and optimization functions:

[0053]

[0054] Where: Z: optimization objective function value;

[0055] T: total number of time periods in the optimization time period;

[0056] P buy (t): power purchased from the grid at time t;

[0057] C buy (t): the price of electricity purchased from the grid at time t;

[0058] P sell (t): power sold to the grid at time t;

[0059] C sell (t): the price of electricity sold to the grid at time t;

[0060] C storage : Energy storage cost, representing the operation or maintenance cost of the energy storage system;

[0061] Constraints:

[0062] Power balance constraint: In each time period t, the following power balance equation is satisfied:

[0063] P PV (t)+P buy (t) = P load (t)+P storage (t)+P sell (t)

[0064] Where: P PV (t): output power of the photovoltaic system at time t;

[0065] P load (t): load demand power of the system at time t;

[0066] P storage (t): the charge and discharge power of the energy storage device at time t, a positive value indicates discharge, and a negative value indicates charge;

[0067] SOC limits of energy storage devices:

[0068] E min ≤E(t)≤E max

[0069] Where: Emin is the minimum energy storage charge and discharge capacity, E(t) is the energy storage charge and discharge capacity at time t, E max The maximum charge and discharge capacity for energy storage;

[0070] Charging and discharging control of energy storage system:

[0071] During off-peak electricity price periods, active charging from the grid to energy storage devices:

[0072] P storage (t)=-P charge

[0073] Where: P storage (t) is the energy storage power, -P charge is the charging power;

[0074] During peak electricity price periods, priority is given to discharging to meet local loads, and the remaining electricity is sold:

[0075] P storage (t) = P discharge

[0076] Where: P storage (t) is the energy storage power, -P charge Charging power

[0077] Energy storage power limit:

[0078]

[0079] Where: P storage max : Maximum charge and discharge power limit of energy storage equipment;

[0080] Reactive power regulation:

[0081] The system adjusts the reactive power Q(t) in real time through the photovoltaic inverter and energy storage inverter to ensure the stability of the grid voltage:

[0082] Q(t)=k·(V ref -V(t)

[0083] Where: Q(t): reactive power output or absorbed at time t;

[0084] k: reactive power regulation coefficient;

[0085] V ref : The target voltage value of the system;

[0086] V(t): the actual voltage value measured at time t;

[0087] If the voltage is too low, the system actively outputs reactive power compensation:

[0088] Q(t)=Q inject

[0089] Where: Q inject : Refers to the amount of reactive power output by the system when the voltage is too low;

[0090] If the voltage is too high, the system absorbs reactive power:

[0091] Q(t)=Q absorb

[0092] Where: Q absorb : refers to the amount of reactive power absorbed by the system when the voltage is too high;

[0093] Based on weather forecast data and historical load data, forecast photovoltaic power generation and load demand in the next 24 hours

[0094] The model predictive control algorithm is used for dispatch optimization to dynamically adjust the power flow according to the latest market electricity prices and grid status:

[0095]

[0096] Where: T: total number of time periods;

[0097] P grid (t): actual power interacting with the grid at time t;

[0098] P opt (t): The optimized target power calculated at time t, which represents the power flow under ideal conditions.

[0099] The feedback learning and parameter optimization method is:

[0100] Recorded actual data on PV generation, load demand, energy storage status, and grid interaction;

[0101] Perform error calculation:

[0102] Error(t)=|P actual (t) = P predicted (t)|

[0103] Where: Error(t): prediction error at time t, representing the absolute difference between actual power and predicted power;

[0104] P actual (t): actual power value at time t, representing the real power data of photovoltaic power generation, load demand, energy storage status and grid interaction during system operation;

[0105] P predicted(t): the predicted power value at time t, calculated by the system's prediction model, indicating the power value expected in the scheduling plan;

[0106] Feedback learning and parameter optimization:

[0107] Error adjustment and compensation: The system analyzes the sources of forecast errors and adjusts the parameters of the forecast model;

[0108] Strategy optimization: The system adjusts the control strategy according to the actual operation results;

[0109] Self-learning: Use machine learning algorithms to optimize scheduling strategies.

[0110] The active two-way interactive control system for energy between rooftop photovoltaic and distribution network of the present invention realizes the coordinated optimization of photovoltaic system and energy storage system through active two-way interactive energy control, improves the efficiency and economy of power management, and provides reactive power support and voltage regulation for the power grid to ensure the stable operation of the system and the power grid. This technical solution will make up for the defects in the traditional distributed energy management system and provide an innovative solution for the efficient and coordinated development of smart grid and distributed energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 It is a schematic diagram of the active rooftop photovoltaic and distribution network energy two-way interactive control system of the present invention. DETAILED DESCRIPTION

[0112] The active rooftop photovoltaic and distribution network energy two-way interactive control system of the present invention is as follows Figure 1 As shown, it includes the following main equipment and control systems:

[0113] 1. Rooftop photovoltaic system (PV system): rooftop photovoltaic modules, photovoltaic inverters (supporting bidirectional active and reactive power control);

[0114] 2. Energy storage system (ESS): lithium battery pack or supercapacitor, bidirectional energy storage inverter (supporting charge and discharge control);

[0115] 3. Grid connection interface: distribution network connection point (PCC, Point of Common Coupling), transformers and feeders in the distribution network.

[0116] 4. Control center and communication system: Local controller: realizes coordinated control of photovoltaic and energy storage; Cloud platform: supports real-time status monitoring of the power grid and acquisition of market electricity prices; Communication interface: supports LoRa, 485, Ethernet and other communication methods to connect with the platform.

[0117] The photovoltaic system and the energy storage system are connected through a bidirectional inverter to support bidirectional energy flow. The energy storage system is connected to the grid at the distribution connection point (PCC) and realizes bidirectional interaction with the grid through a bidirectional charging and discharging inverter. The control center monitors the photovoltaic power generation, energy storage status and grid parameters in real time through Ethernet or wireless network. The photovoltaic inverter and energy storage inverter in the system interact with the control center through communication protocols (such as Modbus) to achieve real-time control.

[0118] Actively optimize power flow through two-way interaction between local photovoltaic system, energy storage system and power grid. The system predicts future status and dynamically adjusts the energy interaction between photovoltaic and energy storage based on real-time data (such as load, photovoltaic output, market electricity price, and energy storage status). The main functions of the system are as follows:

[0119] Reduce the amount of PV power fed into the grid during peak hours to avoid congestion and give priority to supplying power to local loads;

[0120] Charging energy storage from the grid during off-peak periods to prepare for discharge during peak periods;

[0121] Dynamically adjust reactive power output according to voltage and frequency fluctuations to ensure grid stability;

[0122] Participate in peak-valley arbitrage based on market electricity price signals to maximize profits.

[0123] 1. The control method is as follows:

[0124] 1. Power flow control of photovoltaic system and energy storage system:

[0125] In each time period t, the following power balance equation is satisfied:

[0126] P PV (t)+P buy (t) = P load (t)+P storge (t)+P sell t)

[0127] Where: P PV (t): output power of the photovoltaic system; P buy (t): power purchased from the grid;

[0128] P load (t): local load power; P storage (t): charging and discharging power of the energy storage system (positive for discharging, negative for charging);

[0129] P sell(t) P sell (t): Power sold to the grid.

[0130] 2. Charge and discharge control of energy storage system:

[0131] During the off-peak electricity price period (nighttime), the energy storage device is actively charged from the grid:

[0132] P storage (t)=-P charge

[0133] Where: P storage (t) is the energy storage power, -P charge is the charging power.

[0134] During peak electricity price periods, priority is given to discharging to meet local loads, and the remaining electricity is sold:

[0135] P storage (t) = P discharge

[0136] Where: P discharge is the discharge power.

[0137] Power Constraints:

[0138] E min ≤E(t)≤E max

[0139] Where: E min is the minimum charge and discharge capacity of energy storage, E(t) is the charge and discharge capacity of energy storage, E max Maximum charge and discharge capacity for energy storage

[0140] 3. Reactive power regulation:

[0141] The system adjusts the reactive power Q(t) in real time through the photovoltaic inverter and energy storage inverter to ensure the stability of the grid voltage:

[0142] Q(t)=k·(V ref -V(t)

[0143] Where: Q(t): reactive power output or absorbed at time t, regulating the voltage level of the system;

[0144] k: Reactive power regulation coefficient, used to control the regulation strength of reactive power;

[0145] V ref : Reference voltage, indicating the target voltage value of the system;

[0146] V(t): the actual voltage value measured at time t;

[0147] Through this formula, the system calculates the required reactive power based on the deviation of the current voltage from the reference voltage to stabilize the voltage.

[0148] If the voltage is too low, the system will actively output reactive power compensation to increase the voltage:

[0149] Q(t)=Q inject

[0150] Where: Q inject : Refers to the amount of reactive power output by the system when the voltage is too low, used to compensate for the low voltage condition.

[0151] If the voltage is too high, the system absorbs reactive power to reduce the voltage:

[0152] Q(t)=Q absorb

[0153] Where: Q absorb : Refers to the amount of reactive power absorbed by the system when the voltage is too high, used to reduce the voltage to a normal range.

[0154] 4. Model Predictive Control (MPC) and Rolling Optimization:

[0155] (1) Predictive control:

[0156] The system predicts photovoltaic power generation and load demand in the next 24 hours based on meteorological forecast data and historical load data.

[0157]

[0158] in: The predicted photovoltaic power generation power represents the predicted output power of the photovoltaic system at the future time t;

[0159] The predicted load demand power represents the predicted system load demand at the future time t.

[0160] These predicted power values ​​are based on weather forecast data and historical load data and are typically used to forecast PV power generation and load demand for the next 24 hours.

[0161] (2) Scrolling optimization:

[0162] The dispatch optimization is performed every 15 minutes to dynamically adjust the power flow according to the latest market electricity prices and grid status:

[0163]

[0164] This formula is used to minimize the grid interaction power P in each time period t. grid (t) and target power P opt (t), thereby achieving dynamic optimization of power flow.

[0165] Where: T: total number of time periods;

[0166] P grid (t): actual power interacting with the grid at time t;

[0167] P opt (t): The optimized target power calculated at time t, which represents the power flow under ideal conditions.

[0168] 2. Working process:

[0169] 1. Initialization phase

[0170] The system obtains market electricity prices and weather forecast data from the cloud platform and generates an initial 24-hour power dispatch plan.

[0171] 2. Real-time monitoring and prediction stage

[0172] The system monitors PV output, load power, grid voltage and energy storage status in real time, and updates forecast data.

[0173] 3. Dynamic scheduling and execution phase

[0174] Rolling optimization is performed every 15 minutes to adjust the power interaction between photovoltaics and energy storage according to the real-time status of the power grid and load.

[0175] 4. Feedback learning and optimization stage

[0176] The system records the effect of each dispatch and optimizes future dispatch strategies through a feedback mechanism.

[0177] Detailed working process:

[0178] In order to achieve efficient coordinated optimization control between photovoltaic power generation, energy storage system and power grid, this system uses real-time monitoring and prediction, intelligent scheduling and execution, feedback learning and optimization to continuously execute in three stages to ensure the maximum economy and stability of the system. The following will describe each stage in detail, including the monitoring and prediction methods used, scheduling execution basis and formula, feedback learning and optimization strategy.

[0179] 1. Real-time monitoring and prediction stage:

[0180] During this stage, the system obtains real-time data through sensors, communication interfaces and data platforms, and uses historical data and external data to make predictions and formulate scheduling plans for the future period in advance.

[0181] 1.1 Monitoring content and methods

[0182] Photovoltaic power generation monitoring: real-time collection of the output voltage, current and power of the photovoltaic array, and transmission of data through the inverter.

[0183] P PV(t) = V (t) × I (t)]

[0184] Load power monitoring: Real-time monitoring of the power demand Pload(t) of the user-side load in order to respond to load fluctuations in a timely manner;

[0185] Energy storage status monitoring: obtain the current energy status E(t) and charging and discharging power Pstorage(t) of the energy storage device;

[0186] Grid status monitoring: Collect the voltage V(t), frequency f(t) and current power interaction status Pgrid(t) of the distribution network;

[0187] Market electricity price monitoring: obtain real-time electricity prices Cbuy(t) and Csell(t) from the cloud platform.

[0188] 1.2 Prediction methods and formulas

[0189] In order to improve the prediction accuracy, this solution adopts a hybrid prediction model of ARIMA and LSTM to combine the advantages of both. The ARIMA model is good at processing linear trends in time series, while LSTM is more suitable for capturing nonlinear relationships and long-term dependencies. Through this combination, the system can learn both the linear and nonlinear parts of the data at the same time, achieving more accurate predictions of photovoltaic power generation, load demand and market electricity prices.

[0190] (1) Working principle of hybrid prediction model:

[0191] The ARIMA model handles the linear portion of the time series:

[0192] The linear trend of the time series is identified through the ARIMA model, and the linear part prediction value of the time series is obtained.

[0193]

[0194] Where: y t : Observed values ​​or real data of time series, representing actual data at time t (such as photovoltaic power generation, load demand or market electricity price)

[0195] ARIMA: Autoregressive Integrated Moving Average model, used to identify and predict linear trends in time series. The ARIMA model consists of an autoregressive (AR) part, a differencing (I) part, and a moving average (MA) part.

[0196] The linear part of the time series predicted by the ARIMA model represents the linear prediction value obtained by the ARIMA model at time t.

[0197] (2) The LSTM model processes the nonlinear part of the time series:

[0198] Subtract the linear prediction result of ARIMA from the original time series to get the residual series. The residual series contains the nonlinear part of the data.

[0199]

[0200] Where: r(t): residual sequence, which represents the linear prediction result of subtracting ARIMA from the original time series y(t) The residual series retains the nonlinear characteristics of the time series that cannot be explained by the linear model;

[0201] y(t): the observed value or true value of the original time series;

[0202] The linear part results predicted by the ARIMA model;

[0203] Use the LSTM model to predict the residual sequence and obtain the nonlinear part prediction value:

[0204]

[0205] in: The nonlinear part of the time series predicted by the LSTM model represents the nonlinear prediction value at time t;

[0206] LSTM: Long short-term memory network, used to capture nonlinear relationships and long-term dependency characteristics in time series. LSTM is particularly suitable for processing time series data with nonlinear characteristics.

[0207] r(1),r(2)…,r(t): The values ​​of the residual sequence at each moment, used as the input of the LSTM model to extract nonlinear features.

[0208] (3) Combined prediction results:

[0209] The prediction results of the ARIMA model and the LSTM model are weighted and combined to obtain the final prediction value.

[0210]

[0211] in: The final hybrid model prediction result includes the superposition of linear and nonlinear parts.

[0212] The linear part of the ARMIA model prediction;

[0213] The nonlinear part of the LSTM model prediction;

[0214] The final predicted value It is a combination of linear and nonlinear parts, which can more accurately reflect the complex characteristics of time series.

[0215] The algorithm flow of the hybrid prediction model:

[0216] Input: time series data y(t) (such as photovoltaic power generation, load demand or market electricity price).

[0217] (1) ARIMA linear prediction: Use the ARIMA model to predict the linear trend and obtain the linear part prediction value

[0218] (2) Calculate the residual sequence: Calculate the residual:

[0219]

[0220] Where: r(t): residual sequence, representing the difference between the original time series y(t) and the ARIMA linear prediction value The residual sequence r(t) represents the part of the time series that cannot be explained by the linear model, that is, the nonlinear part.

[0221] (3) LSTM nonlinear prediction: Use the LSTM model to predict the residual sequence r(t) and obtain the nonlinear part prediction value

[0222] (4) Combined prediction results: Add the prediction results of the linear part and the nonlinear part to get the final prediction value

[0223] ARIMA captures the linear trend in the time series, and LSTM processes the nonlinear residual, making the model have higher prediction accuracy. This solution is suitable for complex time series predictions such as photovoltaic power generation, load demand, and market electricity prices. The forecast data is updated every 15 minutes, and the forecast results are dynamically adjusted through the hybrid model to provide support for the scheduling and execution of the system.

[0224] 2. Scheduling and Execution Phase

[0225] At this stage, the system formulates a two-way energy interaction plan based on real-time monitoring data and prediction results, uses the model predictive control (MPC) algorithm for optimal scheduling, and dynamically adjusts the power output of photovoltaic power generation and energy storage equipment.

[0226] 2.1 Scheduling Objectives and Optimization Functions

[0227] The optimization goal of the system is to maximize economic benefits and ensure grid stability. The specific objective function is as follows:

[0228]

[0229] Where: Z: optimization objective function value

[0230] T: The total number of time periods in the optimization time period

[0231] P buy (t): Power purchased from the grid at time t

[0232] C buy (t): The price of electricity purchased from the power grid at time t

[0233] P sell (t): Power sold to the grid at time t

[0234] C sell (t): The price of electricity sold to the grid at time t

[0235] C storage : Energy storage cost, representing the operation or maintenance cost of the energy storage system

[0236] Constraints:

[0237] 1. Power balance constraints:

[0238] P PV (t)+P buy (t) = P load (t)+P storage (t)+P sell (t)

[0239] Where: P PV (t): Output power of the photovoltaic system at time t

[0240] P load (t): load power demand of the system at time t

[0241] P storage (t): The charge and discharge power of the energy storage device at time t. Positive values ​​indicate discharge and negative values ​​indicate charge.

[0242] 2. SOC (State of Charge) restrictions for energy storage devices:

[0243] E min ≤E(t)≤E max

[0244] Where: E min is the minimum energy storage charge and discharge capacity, E(t) is the energy storage charge and discharge capacity at time t, E max The maximum charge and discharge capacity of energy storage.

[0245] 3. Energy storage power limit:

[0246]

[0247] Where: P storage max :The maximum charging and discharging power limit of the energy storage device ensures that the energy storage device operates within a safe range.

[0248] 2.2 Model Predictive Control (MPC) Algorithm

[0249] MPC ensures that the system makes the best decision in each time period through rolling optimization.

[0250] 1 Establish a prediction model: Based on the predicted photovoltaic power generation and load demand, calculate the power balance state in the future.

[0251] 2 Solve the optimization problem: In each scheduling cycle, the photovoltaic output, energy storage charging and discharging, and grid interaction power at each moment are determined by solving the optimization function.

[0252] 3. Rolling optimization execution: Update forecast data every 15 minutes and adjust the plan based on the latest data.

[0253] 3. Feedback learning and optimization stage

[0254] After each scheduling cycle, the system will record the actual execution results, compare them with the predicted plan, and perform feedback learning and parameter optimization based on the error.

[0255] 3.1 Data recording and analysis

[0256] Recording content: Record actual data of photovoltaic power generation, load demand, energy storage status and grid interaction.

[0257] Error calculation:

[0258] Error(t)=|P actual (t) = P predicted (t)|

[0259] Where: Error(t): prediction error at time t, representing the absolute difference between actual power and predicted power.

[0260] P actual (t): The actual power value at time t, representing the real power data of photovoltaic power generation, load demand, energy storage status and grid interaction during system operation

[0261] P predicted (t): The predicted power value at time t, calculated by the system's prediction model, which represents the power value expected in the scheduling plan.

[0262] 3.2 Feedback Learning and Parameter Optimization

[0263] 1. Error adjustment and compensation:

[0264] The system analyzes the sources of forecast errors and adjusts the parameters of the forecast model, such as improving the accuracy of weather forecasts or improving load forecast models.

[0265] 2. Strategy Optimization:

[0266] The system adjusts the control strategy based on the actual operating results, such as reducing the upper limit of photovoltaic power fed into the grid during peak load periods, or optimizing the charging and discharging thresholds of energy storage.

[0267] 3. Self-learning function:

[0268] Using machine learning algorithms, such as reinforcement learning, we continuously optimize the scheduling strategy so that the system can adapt to changes in different seasons and electricity usage habits.

[0269] 3.3 System Optimization Example

[0270] 1. Discover and optimize energy storage usage efficiency: If it is found that the energy storage equipment is not discharging enough during peak hours, the system will adjust the minimum limit of the energy storage SOC and increase the discharge power.

[0271] 2. Dynamically adjust reactive power support strategy: If the grid voltage fluctuates frequently, the system will increase the reactive power compensation and optimize the reactive power output curve of the inverter.

[0272] The present invention actively optimizes power flow according to electricity price, photovoltaic power generation, load demand and grid status, reduces grid feed during peak hours, and actively charges and stores energy during valley hours to maximize economic benefits. Reduce photovoltaic abandonment: When power generation exceeds demand, store excess power in the energy storage system to improve energy utilization. Participate in peak-valley electricity price arbitrage: Charge when electricity prices are low and discharge during peak hours to maximize revenue. Reduce grid load pressure: Optimize power flow by shaving peaks and filling valleys to avoid grid overload during peak hours. Use the ARIMA-LSTM hybrid prediction model to capture the linear trend and nonlinear characteristics of the time series at the same time, and achieve high-precision prediction of photovoltaic power generation, load demand and market electricity prices. Improve prediction accuracy: Compared with a single model, the hybrid model is more adaptable to complex data and improves the accuracy of the scheduling plan. Rolling prediction and real-time optimization: The model supports updating the prediction results every 15 minutes to provide the latest data support for scheduling decisions. Enhance system flexibility: Plan charging and discharging strategies in advance to effectively respond to weather changes and load fluctuations. The energy storage system and the power grid realize two-way energy interaction, and automatically adjust charging and discharging according to strategies during peak and trough periods of electricity consumption. Improve energy storage utilization: Through reasonable planning of charging and discharging, efficient use of energy storage resources can be achieved. Reduce electricity purchase costs: Charge when electricity prices are low to avoid high-priced electricity purchases during peak periods. Support island operation: Automatically switch to island mode when the power grid fails to ensure uninterrupted power supply to critical loads. Based on the voltage and frequency status of the grid, the system dynamically adjusts the reactive output of the photovoltaic inverter and the energy storage system to provide reactive support for the grid. Improve grid voltage stability: Dynamically adjust reactive output, reduce voltage fluctuations, and improve power quality. Optimize grid power factor: Reduce reactive losses and improve transmission efficiency. Enhance the anti-interference ability of the grid: Actively provide support when the grid frequency or voltage is abnormal to ensure stable operation of the system.

Claims

1. An active two-way interactive control system for energy between rooftop photovoltaic and distribution network, characterized by: It includes a rooftop photovoltaic system, a bidirectional photovoltaic inverter, an energy storage system, a bidirectional energy storage inverter, a control center and a cloud platform. The photovoltaic system and the energy storage system are connected through a bidirectional inverter, and the energy storage system and the power grid are interactively connected through a bidirectional energy storage inverter. The control center monitors the photovoltaic system power generation, the energy storage status of the energy storage system and the power grid parameters in real time through the network. The bidirectional photovoltaic inverter and the bidirectional energy storage inverter interact with the control center, and the cloud platform is connected to the control center. Monitor the status data of the photovoltaic system, power grid and energy storage system in real time, and update the forecast data; formulate a two-way energy interaction plan based on the real-time monitoring data and forecast results, optimize the scheduling and dynamically adjust the power output of photovoltaic power generation and energy storage equipment; after each scheduling cycle, record the actual execution results, compare them with the forecast plan, and perform feedback learning and parameter optimization based on the error.

2. According to claim 1, an active rooftop photovoltaic and distribution network energy two-way interactive control system is characterized by: The status data of the photovoltaic system, power grid and energy storage system include: the power demand Pload(t) of the user-side load, the current energy state E(t) and charging and discharging power Pstorage(t) of the energy storage device, the voltage V(t), frequency f(t) and current power interaction status Pgrid(t) of the distribution network, and the real-time electricity prices Cbuy(t) and Csell(t) obtained from the cloud platform.

3. According to claim 1, an active rooftop photovoltaic and distribution network energy two-way interactive control system is characterized by: The data prediction method is: using a hybrid prediction model of ARIMA and LSTM, The ARIMA model handles the linear part of the time series: The linear trend of the time series is identified through the ARIMA model, and the linear part forecast value of the time series is obtained: Where: y t : represents the actual data at time t; ARIMA: Autoregressive Integrated Moving Average model, used to identify and predict linear trends in time series; represents the linear prediction value obtained by the ARIMA model at time t; The LSTM model handles the nonlinear part of the time series: Subtract the linear prediction result of ARIMA from the original time series to get the residual series, which contains the nonlinear part of the data: Where: r(t): residual sequence, which represents the linear prediction result of subtracting ARIMA from the original time series y(t) The difference after y(t): the observed value or true value of the original time series; The linear part results predicted by the ARIMA model; Use the LSTM model to predict the residual sequence and obtain the nonlinear part prediction value: in: Tong represents the nonlinear prediction value at time t; LSTM: Long short-term memory network, used to capture nonlinear relationships and long-term dependency characteristics in time series; r(1),r(2)...,r(t): the values ​​of the residual sequence at each moment, used as the input of the LSTM model; Combined prediction results: The prediction results of the ARIMA model and the LSTM model are weighted and combined to obtain the final prediction value: in: The final hybrid model prediction result includes the superposition of linear and nonlinear parts. The linear part of the ARMIA model prediction; The nonlinear part of the LSTM model prediction; Hybrid prediction model: Input monitored time series data y(t), ARIMA linear prediction: Use the ARIMA model to predict the linear trend and get the linear part prediction value Calculate the residual sequence: Calculate the residual Where: r(t): residual sequence, representing the difference between the original time series y(t) and the ARIMA linear prediction value The residual sequence r(t) represents the part of the time series that cannot be explained by the linear model, that is, the nonlinear part; LSTM nonlinear prediction: Use the LSTM model to predict the residual sequence r(t) and obtain the nonlinear part prediction value Combined prediction results: Add the prediction results of the linear part and the nonlinear part to get the final prediction value 4. According to claim 1, an active rooftop photovoltaic and distribution network energy two-way interactive control system is characterized by: The method for optimizing scheduling and dynamically adjusting the power output of photovoltaic power generation and energy storage equipment is: Scheduling objectives and optimization functions: Where: Z: optimization objective function value; T: total number of time periods in the optimization time period; P buy (t): power purchased from the grid at time t; C buy (t): the price of electricity purchased from the grid at time t; P sell (t): power sold to the grid at time t; C sell (t): the price of electricity sold to the grid at time t; C storage : Energy storage cost, representing the operation or maintenance cost of the energy storage system; Constraints: Power balance constraint: In each time period t, the following power balance equation is satisfied: P PV (t)+P buy (t)=P load (t)+P storage (t)+P sell (t) Where: P PV (t): output power of the photovoltaic system at time t; P load (t): load demand power of the system at time t; P storage (t): the charge and discharge power of the energy storage device at time t, a positive value indicates discharge, and a negative value indicates charge; SOC limit of energy storage equipment: E min ≤E(t)≤E max Where: E min is the minimum energy storage charge and discharge capacity, E(t) is the energy storage charge and discharge capacity at time t, E max The maximum charge and discharge capacity for energy storage; Charging and discharging control of energy storage system: During off-peak electricity price periods, active charging from the grid to energy storage devices: P storage (t)=-P charge Where: P storage (t) is the energy storage power, -P charge is the charging power; During peak electricity price periods, priority is given to discharging to meet local loads, and the remaining electricity is sold: P storage (t)=P discharge Where: P storage (t) is the energy storage power, -P charge Charging power Energy storage power limit: Where: P storage max : Maximum charge and discharge power limit of energy storage equipment; Reactive power regulation: The system adjusts the reactive power Q(t) in real time through the photovoltaic inverter and energy storage inverter to ensure the stability of the grid voltage: Q(t)=k·(V ref -V(t)) Where: Q(t): reactive power output or absorbed at time t; k: reactive power regulation coefficient; V ref : The target voltage value of the system; V(t): the actual voltage value measured at time t; If the voltage is too low, the system actively outputs reactive power compensation: Q(t)=Q inject Where: Q inject : Refers to the amount of reactive power output by the system when the voltage is too low; If the voltage is too high, the system absorbs reactive power: Q(t)=Q absorb Where: Q absorb : refers to the amount of reactive power absorbed by the system when the voltage is too high; Based on weather forecast data and historical load data, forecast photovoltaic power generation and load demand in the next 24 hours The model predictive control algorithm is used for dispatch optimization to dynamically adjust the power flow according to the latest market electricity prices and grid status: Where: T: total number of time periods; P grid (t): actual power interacting with the grid at time t; P opt (t): The optimized target power calculated at time t, which represents the power flow under ideal conditions.

5. According to claim 1, an active rooftop photovoltaic and distribution network energy two-way interactive control system is characterized by: The feedback learning and parameter optimization method is: Recorded actual data on PV generation, load demand, energy storage status, and grid interaction; Perform error calculation: Error(t)=|P actual (t)=P predicted (t) Where: Error(t): prediction error at time t, representing the absolute difference between actual power and predicted power; P actual (t): actual power value at time t, representing the real power data of photovoltaic power generation, load demand, energy storage status and grid interaction during system operation; P predicted (t): the predicted power value at time t, calculated by the system's prediction model, indicating the power value expected in the scheduling plan; Feedback learning and parameter optimization: Error adjustment and compensation: The system analyzes the sources of forecast errors and adjusts the parameters of the forecast model; Strategy optimization: The system adjusts the control strategy according to the actual operation results; Self-learning: Use machine learning algorithms to optimize scheduling strategies.

Images