A substation-energy storage station-photovoltaic station coordinated power control method

Abstract

The present invention discloses a substation-energy storage station-photovoltaic station coordinated power control method, which belongs to the field of electric power technology. By integrating heterogeneous data from substations, energy storage stations, photovoltaic stations and meteorological platforms, a dynamic prediction of photovoltaic power and a smooth trajectory of bus power are constructed. Combined with rolling optimization control and millisecond-level fluctuation suppression, the method realizes smooth grid power output coordinated by the three stations, and solves the technical problems of power coordination, prediction feedforward and rapid disturbance compensation of substation-energy storage station-photovoltaic station. The present invention improves the accuracy of photovoltaic prediction, outputs pre-regulation instructions in advance, enhances the system feedforward control capability, improves the bus power tracking capability, and controls the energy storage power fluctuation and the SOC safety margin, realizing local rapid response and closed-loop dynamic compensation, significantly improving the system's anti-disturbance capability and stability.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to a transformer substation-energy storage station-photovoltaic station coordinated power control method. Background Art

[0002] With the large-scale integration of renewable energy, intermittent power sources like photovoltaic power generation are placing higher demands on the grid's power stability and regulation capabilities. To improve the ability to absorb renewable energy, technologies such as photovoltaic-storage coordinated control or combined source-load-storage regulation are often used to mitigate PV output fluctuations through energy storage systems.

[0003] However, existing technologies are mostly centered on a single station, only realizing simple tracking and control of the energy storage system and local photovoltaics. They are unable to coordinate the power flow between substations, photovoltaic stations and energy storage stations at the system level, and it is difficult to meet the power regulation requirements of high precision and fast dynamic response on the bus side.

[0004] Furthermore, existing control strategies typically employ fixed strategies or static optimization methods, lacking the ability to respond in real time to forecast uncertainties and rapid disturbances. In scenarios where PV power experiences significant fluctuations and frequent jumps, control systems are prone to response lags, amplified bus power fluctuations, and even voltage / frequency violations. Furthermore, while some current systems incorporate forecasting algorithms, they fail to fully integrate environmental information and weather factors, resulting in significant deficiencies in forecast accuracy and dynamic adaptability. Summary of the Invention

[0005] The purpose of the present invention is to provide a transformer substation-energy storage station-photovoltaic station coordinated power control method, which solves the technical problems of transformer substation-energy storage station-photovoltaic station power coordination, predictive feedforward and rapid disturbance compensation.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A transformer substation-energy storage station-photovoltaic station coordinated power control method includes the following steps:

[0008] Step 1: Collect and preprocess heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms to construct a unified time-series raw data set. The heterogeneous data includes bus voltage, active power, dispatch instructions, current SOC status, predicted SOC status, energy storage charge and discharge capacity, battery state of health (SOH), current photovoltaic power generation, local environmental data, and weather forecasts. Feature extraction is performed on the raw data set to obtain features reflecting bus status, energy storage capacity limits, and photovoltaic power generation trends.

[0009] Step 2: Based on historical data and local environmental data, a short-term PV power forecast model is constructed to generate a primary forecast curve for PV power generation over the next 10 minutes. A weather factor vector extracted from weather forecasts is introduced to build a multi-model fusion mechanism to obtain a PV power dynamic trend forecast and generate confidence intervals. Based on the PV power dynamic trend forecast, a target bus active power curve for the next 10 minutes is constructed, incorporating slope limits and fluctuation range annotations to form a smooth desired power trajectory. Energy storage pre-regulation instructions are then output in advance based on the predicted fluctuation trend.

[0010] Step 3: With the goal of stable bus power output, a collaborative optimization control model is constructed. The inputs of the collaborative optimization control model include the PV power dynamic trend prediction results, the target bus active power curve, the current SOC state, the predicted SOC state, the energy storage charge and discharge capacity, the battery health state (SOH), the real-time load power, the bus voltage and frequency, and the dispatch instruction.

[0011] Set constraints for the collaborative optimization control model, including PV output constraints, energy storage power constraints, SOC boundary constraints, voltage and frequency constraints, and power change rate constraints; output a minute-by-minute control instruction set by performing a rolling solution on the collaborative optimization control model;

[0012] After the control instruction set is generated, a millisecond-level fluctuation suppression mechanism is embedded in the energy storage PCS. When a rapid disturbance occurs in the bus voltage or frequency, the millisecond-level fluctuation suppression mechanism triggers the droop control logic, quickly corrects the power offset, and performs dynamic response compensation.

[0013] Preferably, when executing step 1, the following steps are specifically included:

[0014] When executing step 1, the specific steps include:

[0015] Step 1-1: Collect heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms. Specifically:

[0016] On the substation side, bus voltage, active power and dispatch instructions are obtained through the SCADA system;

[0017] On the energy storage station side, the energy storage SOC status, current power, maximum power capability, and battery health status SOH are collected through the BMS / EMS interface;

[0018] On the PV station side, the current PV power generation is obtained through the inverter interface, and the local environmental transmitter is connected to obtain local environmental data;

[0019] On the meteorological platform side, access the third-party weather forecast platform through API to obtain the short-term weather forecast in the future;

[0020] Step 1-2: Format the heterogeneous data, including cleaning units, timestamps, and missing values; align the time axis based on the bus power sampling period; perform noise filtering and interpolation to form a continuous time series that can be used for modeling;

[0021] Steps 1-3: Integrate the formatted data to obtain the original data set. The variable fields of the original data set include: bus voltage, active power, SOC data, energy storage capacity, SOH data, photovoltaic power generation, local environmental data, and weather forecast factors;

[0022] Each data in the original dataset is aligned by timestamp;

[0023] Steps 1-4: Extract features from the original dataset to obtain three types of business features and a structured feature set, including:

[0024] Bus status characteristics, used to indicate bus voltage fluctuation amplitude and active power change trend;

[0025] Energy storage capacity boundary characteristics, used to represent SOC intervals, maximum or minimum power response capabilities, and health status estimation;

[0026] Photovoltaic power generation trend characteristics are used to represent the power generation power increase / decrease rate, current irradiation trend, and future weather factor vector.

[0027] Preferably, feature extraction is performed on each indicator in the original data set, specifically:

[0028] The busbar status feature extraction uses a sliding window method to calculate the statistics of voltage and active power:

[0029] The active power change rate can be expressed as:

[0030] ;

[0031] in, is the active power at time t, represents the sampling time interval, Indicates the rate of change of active power;

[0032] The voltage fluctuation amplitude is calculated as:

[0033] ;

[0034] Where T represents the sliding window width, is a time variable, Indicates the voltage fluctuation amplitude, express The bus voltage at ;

[0035] Energy storage capacity boundary feature extraction:

[0036] The current SOC range can be determined to be high, medium, or low based on the SOC value provided by the BMS interface (BMS system) on the energy storage station side; the maximum power capability is provided by the BMS interface;

[0037] The state of health (SOH) can affect the maximum charge and discharge capacity. The modified energy storage boundary capacity is defined as:

[0038] ;

[0039] Among them, SOH represents the battery health status, that is, SOH data, Indicates the maximum power capability of the energy storage system after correction; is the rated maximum power of the energy storage system;

[0040] The energy storage system is deployed at the energy storage station side;

[0041] Photovoltaic power generation trend feature extraction:

[0042] ;

[0043] in, is the rate of change of photovoltaic power generation at time t, which represents the power change per unit time; It represents the actual power generation of the photovoltaic system at time t.

[0044] Preferably, when executing step 2, the following steps are specifically included:

[0045] Step 2-1: Using historical operating data and local environmental data, the LSTM model is used to build a short-term photovoltaic power prediction model to quickly predict the primary prediction curve of photovoltaic power generation in the next 10 minutes, that is, the primary prediction result;

[0046] Step 2-2: Use a multi-model fusion weighted algorithm to fuse the primary forecast results and the weather factor vector to obtain a secondary forecast result, which is used to represent the dynamic trend forecast result of photovoltaic power in the next 10 minutes, namely the photovoltaic dynamic forecast;

[0047] The multi-model fusion weighted algorithm is shown in the following formula:

[0048] ;

[0049] in, represents the secondary prediction value after fusion, represents the primary prediction result, It represents the prediction result of the lightweight neural network model after introducing weather factors; represents the weighting coefficient, represents the time step, ;

[0050] Based on the secondary prediction results, a confidence interval is generated to reflect the upper and lower limits of the prediction value;

[0051] Step 2-3: Based on the PV forecast results, construct the target bus active power curve for the next 10 minutes;

[0052] If there is a fluctuating time period on the target bus active power curve, i.e., a fluctuating section, upper and lower power buffer zones and markers are added to reserve adjustment space for the energy storage system.

[0053] Setting tolerance margin on the target bus active power curve to cope with forecast errors or sudden weather disturbances;

[0054] The final output target bus active power curve, fluctuation section identification and dynamic adjustment margin marking;

[0055] Step 2-4: When power fluctuations are predicted in the future, energy storage pre-regulation instructions are generated and output in advance and sent to the energy storage system;

[0056] The generated energy storage pre-regulation instructions are as follows:

[0057] If an upward fluctuation is predicted, a pre-regulation instruction for energy storage to enter the "pre-charge" state is generated in advance and sent to the energy storage station;

[0058] If a downward fluctuation is predicted, a pre-regulation instruction for energy storage to enter the "pre-discharge" state is generated in advance and sent to the energy storage station.

[0059] Preferably, a lightweight neural network model, namely the LightGBM model, is used to model and introduce the weather factor vector and predict it. The model input includes the weather factor vector x obtained from the weather forecast. weather , specifically:

[0060] x weather =[I solar ,C cloud ,P rain ,v wind ,T air ];

[0061] Among them, I solar Indicates the predicted solar radiation intensity, C cloud represents the cloud cover factor, P rain represents the probability of rainfall, v wind Indicates wind speed, T air Indicates the temperature;

[0062] The output of the LightGBM model is :

[0063] ;

[0064] That is the LightGBM model;

[0065] The confidence intervals are as follows:

[0066] ;

[0067] in, Indicates the width of the upper and lower confidence bands; this interval is used to determine the subsequent energy storage pre-adjustment space;

[0068] When executing steps 2-3, specifically including introducing a slope limiter based on the photovoltaic prediction results to limit the jump of the target bus active power curve;

[0069] According to the secondary prediction results , construct the target bus active power sequence for the next 10 minutes ;

[0070] The slope limiter is used to constrain the change rate to prevent the target bus power from jumping too quickly, as follows:

[0071] ;

[0072] Indicates the maximum allowed power change slope, which is the preset value;

[0073] When the power first-order difference of the prediction curve exceeds the threshold, it is marked as a fluctuation segment , that is, the fluctuation segment identifier:

[0074] ;

[0075] Reserve upper and lower power buffer zones for the fluctuation range:

[0076] ;

[0077] ;

[0078] in, is the power buffer margin;

[0079] In order to cope with prediction errors and sudden disturbances, a small amount of tolerance redundancy is also added in the non-fluctuation section:

[0080] ;

[0081] in, represents the tolerance redundancy, Much smaller than , Indicates the adjusted reference power value.

[0082] Preferably, when executing step 3, the following steps are specifically included:

[0083] Step 3-1: Obtain PV dynamic forecast, bus target power curve, current SOC status, predicted SOC status, energy storage charging and discharging capacity, battery health status (SOH), current real-time load active power, current bus voltage, current bus frequency, and superior dispatch instructions to construct a control optimization input set; the control optimization input set is updated every minute.

[0084] Step 3-2: Set multidimensional constraints, including:

[0085] Photovoltaic output constraints are used to control the output power of the photovoltaic inverter within the predicted upper and lower bounds;

[0086] Energy storage power constraint, used to control the energy storage output power not to exceed the energy storage charging and discharging capacity;

[0087] SOC boundary constraint is used to control the energy storage SOC to always operate within the set healthy range;

[0088] Voltage-frequency constraint, used to control the bus voltage to be stable within the allowable deviation range and the frequency to be kept within the preset fluctuation band;

[0089] Power change rate constraint is used to introduce the bus power change rate as a limiting item to avoid shock or disturbance caused by command jump;

[0090] Step 3-3: Construct a collaborative optimization control model, as shown in the following formula:

[0091] ;

[0092] in, Indicates the actual bus output power, represents the predicted bus output power, represents the variance of energy storage charging and discharging power, represents the first-order derivative of bus power, are weight factors;

[0093] Step 3-4: The collaborative optimization control model is solved on a rolling basis on a minute-by-minute basis, and the control instructions for the current time slice are output. This control instruction set is generated and sent to the energy storage station and PV station to drive the energy storage system and PV inverter to operate in coordination, ensuring stable bus power output.

[0094] Step 3-5: After the control instruction set is issued, the millisecond-level fluctuation suppression mechanism is synchronously activated. The millisecond-level fluctuation suppression mechanism runs inside the local controller of the energy storage PCS;

[0095] When a rapid disturbance in bus voltage or frequency is detected, the energy storage PCS automatically performs real-time power compensation based on the droop control mechanism. The specific steps are as follows:

[0096] Step 3-5-1: Frequency droop compensation, if the frequency offset is , then the instantaneous output change of energy storage is:

[0097] ;

[0098] in, Indicates the frequency response coefficient set locally by the energy storage PCS. Indicates the instantaneous output change of energy storage;

[0099] Step 3-5-2: Voltage sag compensation, if the voltage offset is , then the change in energy storage output is:

[0100] ;

[0101] in, Indicates the voltage response coefficient set locally by the energy storage PCS. Indicates the change in energy storage output;

[0102] Step 3-5-3: The final calculation formula for the actual output power of energy storage is as follows:

[0103] ;

[0104] in, Indicates the actual output power of the energy storage.

[0105] The substation-energy storage station-photovoltaic station collaborative power control method described in the present invention solves the technical problems of power coordination, predictive feedforward and rapid disturbance compensation of the substation-energy storage station-photovoltaic station. The present invention adopts multi-dimensional data of substation, photovoltaic, energy storage and meteorological platforms to construct a unified time series and structured feature set to provide reliable input for subsequent modeling and control. It combines the LSTM deep model with meteorological factors to improve the accuracy of photovoltaic prediction, output pre-regulation instructions in advance, enhance the system's feedforward control capability, introduce a collaborative optimization model with multi-dimensional constraints, improve the bus power tracking capability, and control the energy storage power fluctuation and SOC safety boundary. It embeds a millisecond-level fluctuation suppression mechanism to achieve local rapid response and closed-loop dynamic compensation, significantly improving the system's anti-disturbance capability and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] Figure 1 It is the main flow chart of the present invention;

[0107] Figure 2 is a flow chart of step 1 of the present invention;

[0108] Figure 3 is a flow chart of step 2 of the present invention;

[0109] Figure 4 is a flow chart of step 3 of the present invention;

[0110] Figure 5 is a flow chart of steps 3-5 of the present invention. DETAILED DESCRIPTION

[0111] Depend on Figure 1-Figure 5 A transformer substation-energy storage station-photovoltaic station coordinated power control method is shown, comprising the following steps:

[0112] Step 1: Collect and preprocess heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms to construct a unified time-series raw data set. The heterogeneous data includes bus voltage, active power, dispatch instructions, current SOC status, predicted SOC status, energy storage charge and discharge capacity, battery state of health (SOH), current photovoltaic power generation, local environmental data, and weather forecasts. Feature extraction is performed on the raw data set to obtain features reflecting bus status, energy storage capacity limits, and photovoltaic power generation trends.

[0113] When executing step 1, the specific steps include:

[0114] Step 1-1: Collect heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms. Specifically:

[0115] On the substation side, bus voltage, active power and dispatch instructions are obtained through the SCADA system;

[0116] On the energy storage station side, the energy storage SOC status, current power, maximum power capability, and battery health status SOH are collected through the BMS / EMS interface;

[0117] On the PV station side, the current PV power generation is obtained through the inverter interface, and the local environmental transmitter is connected to obtain local environmental data. The local environmental data is collected by environmental sensors on the PV station side, such as temperature and humidity transmitters, and light intensity transmitters.

[0118] On the meteorological platform side, access the third-party weather forecast platform through API to obtain the short-term weather forecast in the future;

[0119] In this embodiment, the control center platform and the collaborative control platform are deployed on the substation side. The collaborative control platform communicates with the control center platform through the Internet and adopts IEC104 protocol or MQTT to realize data interaction, such as the issuance of control instructions, configuration of constraint parameters, and feedback of status.

[0120] The collaborative control platform communicates with terminal systems or devices such as inverters, environmental sensors, energy storage station PCS systems, energy storage BMS systems, smart meters, and synchronized phasor measurement devices (PMUs) on the photovoltaic station side or energy storage station side through the Internet (TCP), serial port (Modbus), or wireless network.

[0121] The collaborative control platform is responsible for real-time collection of heterogeneous data, photovoltaic prediction and target construction, collaborative optimization and regulation, control instruction generation, issuance, feedback and other process operations.

[0122] Step 1-2: Format the heterogeneous data, including cleaning units, timestamps, and missing values; align the time axis based on the bus power sampling period; perform noise filtering and interpolation to form a continuous time series that can be used for modeling;

[0123] Steps 1-3: Integrate the formatted data to obtain the original data set. The variable fields of the original data set include: bus voltage, active power, SOC data, energy storage capacity, SOH data, photovoltaic power generation, local environmental data, and weather forecast factors;

[0124] Each data in the original dataset is aligned by timestamp;

[0125] Steps 1-4: Extract features from the original dataset to obtain three types of business features and a structured feature set, including:

[0126] Bus status characteristics, used to indicate bus voltage fluctuation amplitude and active power change trend;

[0127] Energy storage capacity boundary characteristics, used to represent SOC intervals, maximum or minimum power response capabilities, and health status estimation;

[0128] Photovoltaic power generation trend characteristics are used to represent the power generation power increase / decrease rate, current irradiation trend, and future weather factor vector.

[0129] In this embodiment, statistical analysis and time series feature engineering methods can be used to extract features from various indicators in the original data set, specifically:

[0130] The busbar status feature extraction uses a sliding window method to calculate the statistics of voltage and active power:

[0131] The active power change rate can be expressed as:

[0132] ;

[0133] in, is the active power at time t, represents the sampling time interval, Indicates the active power change rate.

[0134] The voltage fluctuation amplitude is calculated as:

[0135] ;

[0136] Where T represents the sliding window width, is a time variable, Indicates the voltage fluctuation amplitude, express The bus voltage at .

[0137] Energy storage capacity boundary feature extraction:

[0138] The current SOC range can be determined to be high, medium, or low based on the SOC value provided by the BMS interface (BMS system) on the energy storage station side; the maximum power capability is provided by the BMS interface;

[0139] The state of health (SOH) can affect the maximum charge and discharge capacity. The modified energy storage boundary capacity is defined as:

[0140] ;

[0141] Among them, SOH represents the battery health status, that is, SOH data, Indicates the maximum power capability of the energy storage system after correction; is the rated maximum power of the energy storage system;

[0142] The energy storage system is deployed at the energy storage station side.

[0143] Photovoltaic power generation trend feature extraction:

[0144] ;

[0145] in, is the rate of change of photovoltaic power generation at time t, which represents the power change per unit time; represents the actual power generation of the photovoltaic system at time t;

[0146] The irradiance variation trend is calculated by the sliding average of local sensor data, and the weather factor vector is obtained from the weather forecast.

[0147] Step 2: Based on historical data and local environmental data, a short-term PV power forecast model is constructed to generate a primary forecast curve for PV power generation over the next 10 minutes. A weather factor vector extracted from weather forecasts is introduced to build a multi-model fusion mechanism to obtain a PV power dynamic trend forecast and generate confidence intervals. Based on the PV power dynamic trend forecast, a target bus active power curve for the next 10 minutes is constructed, incorporating slope limits and fluctuation range annotations to form a smooth desired power trajectory. Energy storage pre-regulation instructions are then output in advance based on the predicted fluctuation trend.

[0148] When executing step 2, the specific steps include:

[0149] Step 2-1: Using historical operating data and local environmental data, the LSTM model is used to build a short-term photovoltaic power prediction model to quickly predict the primary prediction curve of photovoltaic power generation in the next 10 minutes, that is, the primary prediction result;

[0150] The LSTM model is an existing technology and will not be described in detail.

[0151] The input data of the LSTM model include historical photovoltaic power, light intensity, and local temperature. Light intensity and local temperature are both local environmental data.

[0152] Historical operating data includes historical photovoltaic output sequences, and local environmental data includes light intensity and temperature. The output of the LSTM model is as follows:

[0153] ;

[0154] The update frequency of the LSTM model is 1 minute sliding update. Indicates the time step, the initial default is 1 minute.

[0155] Step 2-2: Use a multi-model fusion weighted algorithm to fuse the primary forecast results and the weather factor vector to obtain a secondary forecast result, which is used to represent the dynamic trend forecast result of photovoltaic power in the next 10 minutes, namely the photovoltaic dynamic forecast;

[0156] The multi-model fusion weighted algorithm is shown in the following formula:

[0157] ;

[0158] in, represents the secondary prediction value after fusion, represents the primary prediction result, Indicates the forecast result in the weather forecast; represents the weighting coefficient, represents the time step, ;

[0159] In this embodiment, a lightweight neural network model, namely the LightGBM model, is used to model and introduce the weather factor vector and predict the weather factor vector. The model input includes the weather factor vector x obtained by the weather forecast. weather , specifically:

[0160] x weather =[I solar ,C cloud ,P rain ,v wind ,T air ];

[0161] Among them, I solar Indicates the predicted solar radiation intensity, C cloud represents the cloud cover factor, P rain represents the probability of rainfall, v wind Indicates wind speed, T air Indicates the temperature;

[0162] The output of the LightGBM model is :

[0163] ;

[0164] This is the LightGBM model.

[0165] The LightGBM model is an existing technology and will not be described in detail.

[0166] Based on the secondary prediction results, a confidence interval is generated to reflect the upper and lower limits of the prediction value;

[0167] The confidence intervals are as follows:

[0168] ;

[0169] in, Indicates the width of the upper and lower confidence bands; this interval is used to determine the subsequent energy storage pre-adjustment space.

[0170] Step 2-3: Based on the PV forecast results, construct the target bus active power curve for the next 10 minutes, including:

[0171] Based on the photovoltaic prediction results, a slope limiter is introduced to limit the jump of the target bus active power curve;

[0172] According to the secondary prediction results , construct the target bus active power sequence for the next 10 minutes .

[0173] The slope limiter is used to constrain the change rate to prevent the target bus power from jumping too quickly, as follows:

[0174] ;

[0175] Indicates the maximum allowed power change slope, which is a preset value.

[0176] If there is a fluctuating time period on the target bus active power curve, i.e., a fluctuating section, upper and lower power buffer zones and markers are added to reserve adjustment space for the energy storage system.

[0177] When the power first-order difference of the prediction curve exceeds the threshold, it is marked as a fluctuation segment , that is, the fluctuation segment identifier:

[0178] ;

[0179] Reserve upper and lower power buffer zones for the fluctuation range:

[0180] ;

[0181] ;

[0182] in, is the power buffer margin.

[0183] Setting tolerance margin on the target bus active power curve to cope with forecast errors or sudden weather disturbances;

[0184] In order to cope with prediction errors and sudden disturbances, this embodiment also adds a small amount of tolerance redundancy in the non-fluctuation segment:

[0185] ;

[0186] in, represents the tolerance redundancy, Much smaller than , Indicates the adjusted reference power value.

[0187] The final output target bus active power curve, fluctuation section identification and dynamic adjustment margin marking;

[0188] Step 2-4: When power fluctuations are predicted in the future, energy storage pre-regulation instructions are generated and output in advance and sent to the energy storage system;

[0189] The generated energy storage pre-regulation instructions are as follows:

[0190] If an upward fluctuation is predicted, a pre-regulation instruction for energy storage to enter the "pre-charge" state is generated in advance and sent to the energy storage station;

[0191] If a downward fluctuation is predicted, a pre-regulation instruction for energy storage to enter the "pre-discharge" state is generated in advance and sent to the energy storage station.

[0192] In this embodiment, for all fluctuation segments Execute trend judgment and generate energy storage pre-regulation status in advance:

[0193] If the power forecast shows an upward trend (discharge reduction or charging is required), a "pre-charge" instruction is generated:

[0194] ;

[0195] If the power forecast shows a downward trend (reduced power generation), a "pre-discharge" instruction is generated:

[0196] ;

[0197] in, Indicates the control instructions for dynamic adjustment margin and energy storage, They represent the pre-charge set value and pre-discharge set value respectively.

[0198] Step 3: With the goal of stable bus power output, a collaborative optimization control model is constructed. The inputs of the collaborative optimization control model include the PV power dynamic trend prediction results, the target bus active power curve, the current SOC state, the predicted SOC state, the energy storage charge and discharge capacity, the battery health state (SOH), the real-time load power, the bus voltage and frequency, and the dispatch instruction.

[0199] Set constraints for the collaborative optimization control model, including PV output constraints, energy storage power constraints, SOC boundary constraints, voltage and frequency constraints, and power change rate constraints; output a minute-by-minute control instruction set by performing a rolling solution on the collaborative optimization control model;

[0200] After the control instruction set is generated, a millisecond-level fluctuation suppression mechanism is embedded in the energy storage PCS. When a rapid disturbance occurs in the bus voltage or frequency, the millisecond-level fluctuation suppression mechanism triggers the droop control logic, quickly corrects the power offset, and performs dynamic response compensation.

[0201] When executing step 3, the specific steps include:

[0202] Step 3-1: Obtain PV dynamic forecast, bus target power curve, current SOC status, predicted SOC status, energy storage charging and discharging capacity, battery health status (SOH), current real-time load active power, current bus voltage, current bus frequency, and superior dispatch instructions to construct a control optimization input set; the control optimization input set is updated every minute.

[0203] In this embodiment, the real-time load active power can be acquired by the smart meter, and the current bus voltage and the current bus frequency can be acquired by the synchronized phasor measurement device PMU.

[0204] In this embodiment, the acquired data is classified as follows:

[0205] Photovoltaic system: instantaneous output power , light intensity , component temperature ;

[0206] Energy storage system: charging and discharging power , SOC status , SOH status ;

[0207] Busbar electrical quantity: voltage ,frequency , bus active power ;

[0208] Dispatch information: instructions from superiors ;

[0209] Forecast results: Photovoltaic dynamic forecast , bus target power curve , predict SOC curve .

[0210] Use a unified time base for timestamp mapping and construct a standard time series index X:

[0211] ;

[0212] in, Indicates the time, with a value from 0 to n.

[0213] The aligned data vector corresponding to each moment is expressed as:

[0214] .

[0215] Step 3-2: Set multidimensional constraints, including:

[0216] Photovoltaic output constraints are used to control the output power of the photovoltaic inverter within the predicted upper and lower bounds;

[0217] ;

[0218] in, They represent the lower and upper limits of the PV power prediction, respectively.

[0219] Energy storage power constraint, used to control the energy storage output power not to exceed the energy storage charging and discharging capacity;

[0220] ;

[0221] in, They represent the minimum charging and discharging power and the maximum charging and discharging power of the energy storage system respectively.

[0222] SOC boundary constraint is used to control the energy storage SOC to always operate within the set healthy range;

[0223] ;

[0224] in, They represent the minimum allowable SOC and maximum allowable SOC of the energy storage system respectively.

[0225] Voltage-frequency constraint, used to control the bus voltage to be stable within the allowable deviation range and the frequency to be kept within the preset fluctuation band;

[0226] ;

[0227] Respectively represent the allowable deviation range of voltage and frequency;

[0228] Power change rate constraint is used to introduce the bus power change rate as a limiting item to avoid shock or disturbance caused by command jump;

[0229] ;

[0230] Indicates the maximum allowable rate of change of bus power.

[0231] Step 3-3: Construct a collaborative optimization control model, as shown in the following formula:

[0232] ;

[0233] in, Indicates the actual bus output power, represents the predicted bus output power, represents the variance of energy storage charging and discharging power, represents the first-order derivative of bus power, are weight factors;

[0234] Step 3-4: The collaborative optimization control model is solved on a rolling basis on a minute-by-minute basis, and the control instructions for the current time slice are output. This control instruction set is generated and sent to the energy storage station and PV station to drive the energy storage system and PV inverter to operate in coordination, ensuring stable bus power output.

[0235] The control values ​​in the control instruction set include:

[0236] Expected bus power value at the current time point , used as the target of bus power regulation;

[0237] Reference charge / discharge power of energy storage system , used to drive the energy storage system to participate in regulation;

[0238] Output limit of photovoltaic inverter , used to prevent inverter overload caused by prediction deviation;

[0239] Step 3-5: After the control instruction set is issued, the millisecond-level fluctuation suppression mechanism is synchronously activated. The millisecond-level fluctuation suppression mechanism runs inside the local controller of the energy storage PCS;

[0240] When a rapid disturbance in bus voltage or frequency is detected, the energy storage PCS automatically performs real-time power compensation based on the droop control mechanism. The specific steps are as follows:

[0241] Step 3-5-1: Frequency droop compensation, if the frequency offset is , then the instantaneous output change of energy storage is:

[0242] ;

[0243] in, Indicates the frequency response coefficient set locally by the energy storage PCS. Indicates the instantaneous output change of energy storage;

[0244] Step 3-5-2: Voltage sag compensation, if the voltage offset is , then the change in energy storage output is:

[0245] ;

[0246] in, Indicates the voltage response coefficient set locally by the energy storage PCS. Indicates the change in energy storage output;

[0247] Step 3-5-3: The final calculation formula for the actual output power of energy storage is as follows:

[0248] ;

[0249] in, Indicates the actual output power of the energy storage.

[0250] The substation-energy storage station-photovoltaic station collaborative power control method described in the present invention solves the technical problems of power coordination, predictive feedforward and rapid disturbance compensation of the substation-energy storage station-photovoltaic station. The present invention adopts multi-dimensional data of substation, photovoltaic, energy storage and meteorological platforms to construct a unified time series and structured feature set to provide reliable input for subsequent modeling and control. It combines the LSTM deep model with weather factor fusion to improve the accuracy of photovoltaic prediction, output pre-regulation instructions in advance, enhance the system's feedforward control capability, introduce a collaborative optimization model with multi-dimensional constraints, improve the bus power tracking capability, and control the energy storage power fluctuation and SOC safety boundary. It embeds a millisecond-level fluctuation suppression mechanism to achieve local rapid response and closed-loop dynamic compensation, significantly improving the system's anti-disturbance capability and stability.

Claims

1. A method for coordinated power control of a substation, energy storage station, and photovoltaic station, characterized by: The steps include: Step 1: Collect and preprocess heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms to construct a unified time-series raw data set. The heterogeneous data includes bus voltage, active power, dispatch instructions, current SOC status, predicted SOC status, energy storage charge and discharge capacity, battery state of health (SOH), current photovoltaic power generation, local environmental data, and weather forecasts. Feature extraction is performed on the raw data set to obtain features reflecting bus status, energy storage capacity limits, and photovoltaic power generation trends. Step 2: Based on historical data and local environmental data, a short-term PV power forecast model is constructed to generate a primary forecast curve for PV power generation over the next 10 minutes. A weather factor vector extracted from weather forecasts is introduced to build a multi-model fusion mechanism to obtain a PV power dynamic trend forecast and generate confidence intervals. Based on the PV power dynamic trend forecast, a target bus active power curve for the next 10 minutes is constructed, integrating slope limits and fluctuation range identifiers to form a smooth desired power trajectory. Energy storage pre-regulation instructions are then output in advance based on the predicted fluctuation trend. When executing step 2, the specific steps include: Step 2-1: Using historical operating data and local environmental data, the LSTM model is used to build a short-term photovoltaic power prediction model to quickly predict the primary prediction curve of photovoltaic power generation in the next 10 minutes, that is, the primary prediction result; Step 2-2: Use a multi-model fusion weighted algorithm to fuse the primary prediction results and the weather factor vector to obtain a secondary prediction result, which is used to represent the dynamic trend prediction result of photovoltaic power in the next 10 minutes, that is, the photovoltaic dynamic prediction result; The multi-model fusion weighted algorithm is shown in the following formula: ； in, represents the secondary prediction value after fusion, represents the primary prediction result, It represents the prediction result of the lightweight neural network model after introducing weather factors; represents the weighting coefficient, represents the time step, ; Based on the secondary prediction results, a confidence interval is generated to reflect the upper and lower limits of the prediction value; Step 2-3: Construct the target bus active power curve for the next 10 minutes based on the PV forecast results; If there is a fluctuating time period on the target bus active power curve, i.e., a fluctuating section, upper and lower power buffer zones and markers are added to reserve adjustment space for the energy storage system. Setting tolerance margin on the target bus active power curve to cope with forecast errors or sudden weather disturbances; The final output target bus active power curve, fluctuation section identification and dynamic adjustment margin marking; Step 2-4: When power fluctuations are predicted in the future, energy storage pre-regulation instructions are generated and output in advance and sent to the energy storage system; The generated energy storage pre-regulation instructions are as follows: If an upward fluctuation is predicted, a pre-regulation instruction for energy storage to enter the "pre-charging" state is generated in advance and sent to the energy storage station; If a downward fluctuation is predicted, a pre-discharge energy storage adjustment instruction is generated in advance and sent to the energy storage station; Step 3: With the goal of stable bus power output, a collaborative optimization control model is constructed. The inputs of the collaborative optimization control model include the PV power dynamic trend prediction results, the target bus active power curve, the current SOC state, the predicted SOC state, the energy storage charge and discharge capacity, the battery health state (SOH), the real-time load power, the bus voltage and frequency, and the dispatch instruction. Set constraints for the collaborative optimization control model, including PV output constraints, energy storage power constraints, SOC boundary constraints, voltage and frequency constraints, and power change rate constraints; output a minute-by-minute control instruction set by performing a rolling solution on the collaborative optimization control model; After the control instruction set is generated, a millisecond-level fluctuation suppression mechanism is embedded in the energy storage PCS. When a rapid disturbance occurs in the bus voltage or frequency, the millisecond-level fluctuation suppression mechanism triggers the droop control logic, quickly corrects the power offset, and performs dynamic response compensation.

2. The transformer substation-energy storage station-photovoltaic station coordinated power control method according to claim 1, characterized in that: When executing step 1, the specific steps include: Step 1-1: Collect heterogeneous data from substations, energy storage stations, photovoltaic stations, and meteorological platforms. Specifically: On the substation side, bus voltage, active power and dispatching instructions are obtained through the SCADA system; On the energy storage station side, the energy storage SOC status, current power, maximum power capability, and battery health status SOH are collected through the BMS / EMS interface; On the PV station side, the current PV power generation is obtained through the inverter interface, and the local environmental transmitter is connected to obtain local environmental data; On the meteorological platform side, access the third-party weather forecast platform through API to obtain the short-term weather forecast in the future; Step 1-2: Format the heterogeneous data, including cleaning units, timestamps, and missing values; align the time axis based on the bus power sampling period; perform noise filtering and interpolation to form a continuous time series for modeling; Steps 1-3: Integrate the formatted data to obtain the original data set. The variable fields of the original data set include: bus voltage, active power, SOC data, energy storage capacity, SOH data, photovoltaic power generation, local environmental data, and weather forecast factors; Each data in the original dataset is aligned by timestamp; Steps 1-4: Extract features from the original dataset to obtain three types of business features and a structured feature set, including: Bus status characteristics, used to indicate bus voltage fluctuation amplitude and active power change trend; Energy storage capacity boundary characteristics, used to represent SOC intervals, maximum or minimum power response capabilities, and health status estimation; Photovoltaic power generation trend characteristics are used to represent the power generation power increase / decrease rate, current irradiation trend, and future weather factor vector.

3. The transformer substation-energy storage station-photovoltaic station coordinated power control method according to claim 2, characterized in that: Perform feature extraction on the original data set, specifically: The busbar status feature extraction uses a sliding window method to calculate the statistics of voltage and active power: The active power change rate is expressed as: ； in, is the active power at time t, represents the sampling time interval, Indicates the rate of change of active power; The voltage fluctuation amplitude is calculated as: ； Where T represents the sliding window width, is a time variable, Indicates the voltage fluctuation amplitude, express The bus voltage at ; Energy storage capacity boundary feature extraction: The current SOC range is determined to be high, medium, or low based on the SOC value provided by the BMS interface on the energy storage station side; the maximum power capability is provided by the BMS interface; The revised energy storage capacity boundary is defined as: ； Among them, SOH represents the battery health status, that is, SOH data, Indicates the maximum power capability of the energy storage system after correction; is the rated maximum power of the energy storage system; The energy storage system is deployed at the energy storage station; Photovoltaic power generation trend feature extraction: ； in, is the rate of change of photovoltaic power generation at time t, which represents the power change per unit time; It represents the actual power generation of the photovoltaic system at time t.

4. The transformer substation-energy storage station-photovoltaic station coordinated power control method according to claim 3, characterized in that: A lightweight neural network model, namely the LightGBM model, is used for modeling to introduce weather factor vectors and make predictions. The model input includes the weather factor vector x obtained from the weather forecast. weather , specifically: x weather =[I solar ,C cloud ,P rain ,v wind ,T air ]; Among them, I solar Indicates the predicted solar radiation intensity, C cloud represents the cloud cover factor, P rain represents the probability of rainfall, v wind Indicates wind speed, T air Indicates the temperature; The output of the LightGBM model is : ； That is the LightGBM model; The confidence intervals are as follows: ; in, Indicates the width of the upper and lower confidence bands; this interval is used to determine the subsequent energy storage pre-adjustment space; When executing steps 2-3, specifically including introducing a slope limiter based on the photovoltaic prediction results to limit the jump of the target bus active power curve; According to the secondary prediction results , construct the target bus active power sequence for the next 10 minutes ; The slope limiter is used to constrain the change rate to prevent the target bus power from jumping too quickly, as follows: ； Indicates the maximum allowed power change slope, which is the preset value; When the power first-order difference of the prediction curve exceeds the threshold, it is marked as a fluctuation segment , that is, the fluctuation segment identifier: ； Reserve upper and lower power buffer zones for the fluctuation range: ； ； in, is the power buffer margin; In order to cope with forecast errors and sudden weather disturbances, a small amount of tolerance redundancy is also added in the non-fluctuating section: ； in, represents the tolerance redundancy, Indicates the adjusted reference power value.

5. The transformer substation-energy storage station-photovoltaic station coordinated power control method according to claim 1, characterized in that: When executing step 3, the specific steps include: Step 3-1: Obtain the PV dynamic prediction results, target bus active power curve, current SOC status, predicted SOC status, energy storage charging and discharging capabilities, battery health status (SOH), current real-time load active power, current bus voltage, current bus frequency, and superior dispatch instructions to construct a control optimization input set; the control optimization input set is updated every minute. Step 3-2: Set multidimensional constraints, including: Photovoltaic output constraints are used to control the output power of the photovoltaic inverter within the predicted upper and lower bounds; Energy storage power constraint, used to control the energy storage output power not to exceed the energy storage charging and discharging capacity; SOC boundary constraint is used to control the energy storage SOC to always operate within the set healthy range; Voltage-frequency constraint, used to control the bus voltage to be stable within the allowable deviation range and the frequency to be kept within the preset fluctuation band; Power change rate constraint is used to introduce the bus power change rate as a limiting item to avoid shock or disturbance caused by command jump; Step 3-3: Construct a collaborative optimization control model, as shown in the following formula: ； in, Indicates the actual bus output power, represents the predicted bus output power, represents the variance of energy storage charging and discharging power, represents the first-order derivative of the actual bus output power, are all weight factors; Step 3-4: The collaborative optimization control model is solved on a rolling basis on a minute-by-minute basis, and the control instructions for the current time slice are output. This control instruction set is generated and sent to the energy storage station and PV station to drive the energy storage system and PV inverter to operate in coordination, ensuring stable bus power output. Step 3-5: After the control instruction set is issued, the millisecond-level fluctuation suppression mechanism is synchronously activated. The millisecond-level fluctuation suppression mechanism runs inside the local controller of the energy storage PCS; When a rapid disturbance in bus voltage or frequency is detected, the energy storage PCS automatically performs real-time power compensation based on the droop control mechanism. The specific steps are as follows: Step 3-5-1: Frequency droop compensation, if the frequency offset is , then the instantaneous output change of energy storage is: ； in, Indicates the frequency response coefficient set locally by the energy storage PCS. Indicates the instantaneous output change of energy storage; Step 3-5-2: Voltage sag compensation, if the voltage offset is , then the change in energy storage output is: ； in, Indicates the voltage response coefficient set locally by the energy storage PCS. Indicates the change in energy storage output; Step 3-5-3: The final calculation formula for the actual output power of energy storage is as follows: ； in, Indicates the actual output power of energy storage, is the reference charge / discharge power of the energy storage system.

Images