Artificial Intelligence-Based New Energy Optimal Power Supply Strategy Method and System

Abstract

The present invention provides a method and system for an optimal power supply strategy of new energy based on artificial intelligence, which relates to the technical field of new energy power supply strategies. The method includes: constructing an electricity price prediction model, a meteorological prediction model, and a pedestrian flow-load prediction model; constructing a dynamic scheduling optimization model according to the power supply cost, the curtailment rate, and the load deviation, where the power supply cost is calculated based on the prediction result of the electricity price prediction model; the curtailment rate is determined according to the prediction result of the meteorological prediction model; the load deviation is calculated based on the prediction result of the pedestrian flow-load prediction model; and dynamically adjusting the economic, environmental, and stability strategies of the power grid according to the dynamic scheduling optimization model. The system includes a data collection module, an electricity price prediction module, a meteorological prediction module, a pedestrian flow-load prediction module, a dynamic scheduling optimization module, and a data output module. The present invention can provide an efficient and intelligent solution for the optimal power supply strategy configuration of new energy.

Description

Technical Field

[0001] The present invention relates to the technical field of renewable energy power supply strategies, and in particular to a renewable energy optimal power supply strategy method and system based on artificial intelligence. Background Art

[0002] With the rapid development of new energy technologies, their position in the power supply system is becoming increasingly important. In order to achieve efficient utilization of new energy and stable power supply, researchers have conducted in-depth exploration of the optimal power supply strategy configuration for new energy. Early research mainly focused on improving the operational safety and friendly grid-connected capabilities of new energy power stations. The relevant technologies have become relatively mature and have been applied in engineering. Subsequent research has gradually shifted to enhancing the active support grid capabilities of new energy power stations to adapt to the development trend of new energy from supplementary power sources to main power sources.

[0003] However, there are still some urgent problems to be solved in the existing technology in the configuration of the optimal power supply strategy for new energy. First, although most studies have taken into account multiple factors when configuring energy storage, in actual applications, the model is too complex, resulting in low calculation efficiency and difficulty in meeting the requirements of real-time control. Secondly, in the multi-energy complementary system, the coordination control strategy between different energy sources is not perfect, and the advantages of each energy source cannot be fully utilized, resulting in the failure to achieve the optimal overall performance of the system. In addition, the existing technology also has shortcomings in the accuracy of the prediction model and the comprehensive consideration of multiple factors, making it difficult to achieve dynamic adjustment of the economy, environmental protection and stability of the power grid. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0005] To this end, the first aspect of the present invention provides a new energy optimal power supply strategy method based on artificial intelligence.

[0006] A second aspect of the present invention provides a new energy optimal power supply strategy system based on artificial intelligence.

[0007] The present invention provides a new energy optimal power supply strategy method based on artificial intelligence, comprising:

[0008] Obtain historical basic data, including renewable energy power generation data, meteorological data, passenger flow data in the target area, and power market data;

[0009] Construct an electricity price prediction model, the input data of which includes the output data of renewable energy power generation, and the output data includes the probability distribution of electricity prices; train the electricity price prediction model based on historical basic data;

[0010] Constructing a meteorological prediction model, wherein the input data of the meteorological prediction model includes meteorological observation data, and the output data includes meteorological prediction results, wherein the meteorological prediction results at least include key meteorological events that affect the output of the power grid; training the meteorological prediction model based on historical basic data;

[0011] Construct a passenger flow-load forecasting model, the input data of which includes passenger flow data of the target area, and the output data includes future electricity demand data of the target area; train the passenger flow-load forecasting model based on historical basic data;

[0012] A dynamic scheduling optimization model is constructed based on the power supply cost, power abandonment rate and load deviation, wherein the power supply cost is calculated based on the prediction results of the electricity price prediction model; the power abandonment rate is determined based on the prediction results of the meteorological prediction model; and the load deviation is calculated based on the prediction results of the passenger flow-load prediction model;

[0013] Obtain basic data collected in real time, input the basic data collected in real time into the electricity price prediction model, meteorological prediction model and passenger flow-load prediction model to obtain prediction results, and input the prediction results into the dynamic scheduling optimization model to dynamically adjust the economy, environmental protection and stability strategies of the power grid.

[0014] The artificial intelligence-based optimal power supply strategy method for new energy according to the above technical solution of the present invention may also have the following additional technical features:

[0015] In the above technical solution, the electricity price prediction model adopts a gradient boosting regression tree model, and the expression of the electricity price prediction model is:

[0016]

[0017] in, represents the number of samples; Indicates the sample number; Indicates The actual electricity price of samples; Represents the input features, that is, the vector of input data of the electricity price prediction model; represents the integrated model predicting electricity price, ; represents the regularization coefficient; Represents the total number of regression trees; Indicates The number of leaf nodes in a regression tree; Indicates the minimum gain of leaf node splitting; Indicates The leaf node weights of a regression tree.

[0018] In the above technical solution, the input data of the weather forecast model is time series data; the weather forecast model adopts a fully connected long short-term memory network model, and its expression is:

[0019]

[0020] in, Represents the output of the weather forecast model; represents the time step of the input sequence; Indicates time Meteorological characteristics, that is, the vector composed of the input data of the meteorological forecast model; Indicates time of predicted meteorological variables; represents the attention coefficient; Represents the attention algorithm; Represents the hidden state of the long short-term memory network; Represents the attention mechanism weight.

[0021] In the above technical solution, the fully connected long short-term memory network model uses a forget gate, an input gate and a cell state to perform computational marking on the weather forecast, and the gating mechanism includes:

[0022]

[0023] in, represents the forget gate, which is used to control the retention ratio of historical meteorological information; Represents the Sigmoid function; Represents the gate weight matrix in the direction of the forget gate; Represents the hidden state of the previous moment; Represents the gate bias matrix in the forget gate direction; represents the input gate, which is used to control the introduction of new meteorological information; Represents the gating weight matrix in the direction of the input gate; Represents the gating weight matrix in the direction of the input gate; represents a candidate memory unit, which is used to temporarily store the candidate state of new meteorological information; tanh represents the hyperbolic tangent function; Represents the gating weight matrix in the cell state direction 0; Represents the gate bias matrix in the direction of the cell state; Represents the cell state and is used to integrate long-term meteorological trends and short-term fluctuations; Indicates the cell state at the previous moment; represents the output gate; Represents the gating weight matrix in the direction of the output gate; Represents the gate bias matrix in the output gate direction; Represents element-wise multiplication; Represents the hidden layer state.

[0024] In the above technical solution, the crowd flow-load prediction model adopts a city-level crowd flow prediction model based on big data, and its expression is:

[0025]

[0026] in, Indicates area In the period Electricity demand data; Represents the convolution kernel weight; Indicates area In the period The spatiotemporal characteristics of , i.e., the vector composed of the input data of the passenger flow-load forecasting model; Represents the photovoltaic output coupling coefficient; Indicates area In the period of photovoltaic output; Represents the index of the convolution kernel in the time dimension; Represents the index of the convolution kernel in the spatial dimension.

[0027] In the above technical solution, the dynamic dispatch optimization model is constructed according to the power supply cost, power abandonment rate and load deviation, including:

[0028]

[0029] in, represents the time step of the input sequence; represents the power supply cost at time t; represents the power abandonment rate at time t; Indicates load deviation; The weight coefficient of the power supply cost represents the economic parameter of the power supply strategy; The weight coefficient of the power abandonment rate represents the environmental protection parameter of the power supply strategy; The weight coefficient that represents the load deviation represents the stability parameter of the power supply strategy.

[0030] In the above technical solution, the power supply cost is calculated based on the electricity price result predicted by the electricity price prediction model and the energy storage cost, including:

[0031]

[0032] in, represents the wind power price at time t predicted by the electricity price prediction model; represents the wind power output at time t; represents the photovoltaic power price at time t predicted by the power price prediction model; represents the photovoltaic output at time t; represents the energy storage cost at time t; Indicates the energy storage capacity at time t;

[0033] The power abandonment rate is calculated based on the difference between wind power output and grid absorption capacity, including:

[0034]

[0035] in, represents the maximum wind power output, which is determined according to the prediction result of the meteorological prediction model; Indicates the actual wind power consumption of the power grid;

[0036] The load deviation is calculated based on the mean square error between the load demand predicted by the passenger flow-load forecasting model and the actual power supply:

[0037]

[0038] in, represents the load demand predicted by the passenger flow-load forecasting model; Indicates the actual power supply.

[0039] In the above technical solution, in the dynamic scheduling optimization model, constraints are set, and the constraints include:

[0040] Wind, solar and hydropower output ≤ installed capacity;

[0041] Energy storage charging and discharging power ≤ rated power;

[0042] Grid load ≤ transmission line capacity;

[0043] The optimization algorithm of the dynamic scheduling optimization model adopts a synchronous alternating direction multiplier method.

[0044] In the above technical solution, before inputting the basic data into the model, the basic data is also preprocessed, and the data preprocessing includes:

[0045] Perform wavelet filtering on the renewable energy power generation data to complete the denoising of the basic data;

[0046] Map meteorological data to the locations of new energy power stations and construct a spatial correlation matrix;

[0047] Standardize the data structure of heterogeneous data in the basic data and discard invalid data.

[0048] The present invention provides an artificial intelligence-based new energy optimal power supply strategy system, comprising:

[0049] Data collection module, which obtains basic data, including renewable energy power generation data, meteorological data, passenger flow data in the target area, and power market data;

[0050] The electricity price prediction module includes an electricity price prediction model, the input data of the electricity price prediction model includes the output data of new energy power generation, and the output data includes the probability distribution of electricity prices; the electricity price prediction model is trained according to historical basic data; the basic data obtained in real time is input into the electricity price prediction module to obtain the electricity price prediction result;

[0051] A meteorological forecast module includes a meteorological forecast model, the input data of the meteorological forecast model includes meteorological observation data, and the output data includes meteorological forecast results, and the meteorological forecast results at least include key meteorological events that affect the power grid output; the meteorological forecast model is trained according to historical basic data; the basic data obtained in real time is input into the meteorological forecast module to obtain the meteorological forecast results;

[0052] A passenger flow-load prediction module includes a passenger flow-load prediction model, wherein the input data of the passenger flow-load prediction model includes passenger flow data of a target area, and the output data includes future electricity demand data of the target area; the passenger flow-load prediction model is trained according to historical basic data; and the basic data obtained in real time is input into the passenger flow-load prediction module to obtain a load prediction result;

[0053] The dynamic scheduling optimization module includes a dynamic scheduling optimization model constructed according to the power supply cost, the power abandonment rate and the load deviation, wherein the power supply cost is calculated according to the prediction result of the power price prediction model; the power abandonment rate is determined according to the prediction result of the meteorological prediction model; and the load deviation is calculated according to the prediction result of the passenger flow-load prediction model;

[0054] The data output module outputs the power supply strategy formed by the dynamic scheduling optimization module according to the prediction results of the electricity price prediction module, the weather prediction module and the passenger flow-load prediction module.

[0055] In summary, due to the adoption of the above technical features, the beneficial effects of the present invention are:

[0056] The present invention realizes dynamic optimization and adjustment of the economy, environmental protection and stability of the new energy power grid through multi-source data fusion, advanced model construction and optimization algorithm application, effectively overcomes the problems of low computing efficiency, imperfect coordination control strategy, insufficient prediction model accuracy and so on in the prior art, and provides an efficient and intelligent solution for the configuration of the optimal power supply strategy for new energy.

[0057] Specifically, in terms of prediction accuracy, electricity price forecasting adopts the gradient boosting regression tree model, which integrates multiple factors such as new energy power generation output to improve prediction accuracy and provide a scientific basis for economic strategy adjustments; meteorological forecasting combines fully connected long short-term memory networks and attention mechanisms to accurately predict key meteorological events and enhance the foresight of grid stability strategy adjustments; passenger flow-load forecasting integrates big data and convolution operations to accurately predict electricity demand and improve the accuracy of stability strategy adjustments.

[0058] In terms of multi-factor comprehensive optimization, the dynamic scheduling optimization model focuses on power supply cost, power abandonment rate and load deviation. It combines electricity price, weather and load forecast results, and rationally arranges new energy power generation and energy storage scheduling through optimization algorithms to reduce total cost and improve economic benefits. At the same time, the power abandonment rate is included in the optimization target to reduce wind and solar power abandonment, increase the utilization rate of new energy, and achieve effective optimization of environmental protection strategies. In addition, with load deviation as one of the optimization targets, it coordinates new energy power generation and energy storage systems to match actual power supply with load demand and enhance grid stability.

[0059] In terms of real-time response and dynamic adjustment capabilities, by acquiring real-time basic data and inputting it into various prediction models, the latest prediction results can be quickly obtained, and changes in renewable energy power generation, weather and load demand can be promptly reflected, providing real-time information support for the dynamic scheduling optimization model, ensuring that the grid strategy adjustment keeps pace with the actual operating conditions; and, the real-time prediction results are input into the dynamic scheduling optimization model to achieve synchronous optimization and adjustment of the grid's economic, environmental and stability strategies, improve grid operation efficiency and reliability, and effectively respond to the challenges of intermittent and volatile renewable energy power generation.

[0060] In terms of algorithm and model optimization, historical basic data is used to train each prediction model so that the model can fully learn the data characteristics and rules, improve the prediction performance, provide high-quality prediction results for subsequent dynamic scheduling optimization, and enhance the reliability of the entire power supply strategy method; in addition, the dynamic scheduling optimization model uses the synchronous alternating direction multiplier method as the optimization algorithm to efficiently solve problems with multiple optimization objectives and constraints, ensure rapid convergence of the model, improve computing efficiency, and meet the needs of rapid decision-making in actual engineering.

[0061] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0063] Figure 1 It is a flow chart of a new energy optimal power supply strategy method based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0065] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0066] Refer to the following Figure 1 To describe the artificial intelligence-based new energy optimal power supply strategy method and system provided according to some embodiments of the present invention.

[0067] Some embodiments of the present application provide a new energy optimal power supply strategy method based on artificial intelligence.

[0068] like Figure 1 As shown, the first embodiment of the present invention proposes a new energy optimal power supply strategy method based on artificial intelligence, including the following steps S1-S7. It should be noted that the order of steps S1-S7 referred to in the present disclosure is only a schematic representation and is not a limitation of the invention. Those skilled in the art can adjust the order of steps as needed, and steps of different orders can also be executed simultaneously, for example, steps S3-S5 can be executed simultaneously.

[0069] S1. Obtain historical basic data, including new energy power generation data, meteorological data, passenger flow data in the target area, and electricity market data.

[0070] Specifically, new energy power generation data include wind power, photovoltaic, hydropower station power generation data, etc.; meteorological data include meteorological information data such as temperature, wind speed, irradiance, humidity, etc.; passenger flow data in the target area include real-time passenger flow heat maps, etc., which are used to show the distribution of urban passenger flow; electricity market data include historical electricity prices and supply and demand data, including real-time electricity prices, policy subsidies and power grid absorption capacity.

[0071] In some embodiments, the source of renewable energy power generation data may be a power grid data acquisition and monitoring control system (SCADA system); the source of meteorological data may be a weather station or satellite; the source of passenger flow data in the target area may be an urban traffic management platform (mobile phone signaling / camera); and the source of electricity market data may be an electricity trading platform.

[0072] Among them, the data transmission method of new energy power generation data, meteorological data, and passenger flow data in the target area can adopt JSON push between systems.

[0073] In a specific embodiment, the basic data collected is mainly achieved by passively receiving JSON data from a third-party system, and the data is collected and put into a database.

[0074] It should be noted that in the present disclosure, the basic data is divided into historical basic data and basic data collected in real time. The historical basic data is used for training various models, and the basic data collected in real time is used for formulating the current power supply strategy.

[0075] S2. Perform data preprocessing on the basic data, the data preprocessing includes: performing wavelet filtering on the new energy power generation data to complete the denoising of the basic data; mapping the meteorological data to the location of the new energy power station to construct a spatial correlation matrix; performing data structure standardization on the heterogeneous data in the basic data and discarding invalid data.

[0076] It is understandable that in the embodiment of the present disclosure, since the data source is not unique and the data is random, it is necessary to pre-process the data to achieve data denoising, alignment and standardization. However, step S2 is not a necessary step for the present invention and can be omitted according to actual conditions.

[0077] In a specific embodiment, in the data preprocessing stage, a Kalman filter data denoising algorithm is used to denoise the basic data. The core formula of the Kalman filter data denoising algorithm is divided into two stages: prediction and update.

[0078] During the prediction phase:

[0079]

[0080] in, express State prediction value at the moment (such as wind power output prediction value); express The predicted value of the state at the moment; Represents the state transfer matrix, which is used to describe system dynamics (such as the inertia characteristics of the power system); express The forecast error covariance matrix at time quantifies the forecast uncertainty; express The forecast error covariance matrix at time instant; Represents the process noise covariance, which is used to reflect the system model error (such as the prediction deviation caused by sudden change in wind speed); Represents the control input matrix at time k, mapping the external control quantity to the state space, such as the dynamic influence matrix of light and temperature on power output in photovoltaic projects; Represents the control input vector, which is used to modify the state prediction model to improve the estimation accuracy (such as light intensity and temperature in environmental variables).

[0081] In the update phase (fusing observations):

[0082]

[0083] in, represents the Kalman gain, which is used to dynamically balance the weights of model predictions and observed data; Represents the observation matrix, which maps the state to the observation space (such as the conversion of voltage to current); represents the observation noise covariance, which characterizes the sensor error (such as current transformer noise); Represents actual observed values ​​(such as real-time current or voltage data of the power grid); represents the updated error covariance matrix; Represents the updated optimal estimate of the state (such as wind power output value).

[0084] Among the above parameters, According to the dynamic characteristics of the power system, such as the wind speed-power conversion relationship in the wind power output model, specifically, when predicting wind power output, It can be set as the linearization coefficient matrix of wind speed and power generation; Through historical data statistics or adaptive algorithm dynamic adjustment, it reflects the uncertainty of sudden change of wind speed and equipment aging, such as in typhoon weather. It needs to be increased to cope with the dramatic fluctuations in wind speed; According to the sensor accuracy calibration, such as the current transformer error range (±0.5%), for example, in grid current monitoring, Set as sensor error variance; Usually it is an identity matrix (directly observed state quantity) or a linear mapping matrix (such as voltage → power conversion). For example, when predicting the power of a photovoltaic power station, It can represent the irradiance-power conversion factor.

[0085] In a specific embodiment, the actual effect When it is equal to the linearized model based on the wind speed-power curve, (reflects wind speed fluctuations), (sensor noise variance), the root mean square error (RMSE) of the power data after denoising is reduced by 60%, and the wind curtailment rate is reduced from 8% to 2%.

[0086] After denoising the input data, the system will align and standardize the data structure of the reasonable data and discard invalid data.

[0087] S3. Construct an electricity price prediction model. The input data of the electricity price prediction model include new energy power generation output data, installed capacity and policy control coefficients, etc. The output data include the probability distribution of electricity prices in different time periods in the next 24 hours (such as the confidence interval of the difference between peak and valley electricity prices); train the electricity price prediction model based on historical basic data. For example, the prediction results of the electricity price prediction model show that when the peak wind power output is predicted at night, the electricity price drops to 0.2 yuan / kWh.

[0088] In some embodiments, the electricity price prediction model adopts a gradient boosted regression tree model (GBRT) to minimize the prediction error and prevent overfitting. The expression of the electricity price prediction model is:

[0089]

[0090] in, represents the number of samples; Indicates the sample number; Indicates The actual electricity price of each sample (unit: yuan / kWh); Represents the input features, that is, the vector of input data of the electricity price prediction model; represents the integrated model predicting electricity price, ,In a specific embodiment, the learning rate is 0.1, which controls the contribution of a single tree and prevents overfitting; represents the regularization coefficient (value range: 0.1~1.0); represents the total number of regression trees (typical value: 300 trees); Indicates The number of leaf nodes in a regression tree (generated automatically by the model, typical value: 8~16); Indicates the minimum gain for splitting a leaf node (value range: 0.1~0.3); Indicates The leaf node weights of a regression tree are automatically learned by the model.

[0091] For example, a city's power market needs to predict electricity prices 24 hours in advance to optimize energy storage charging and discharging strategies. Based on the trained electricity price prediction model proposed in this disclosure, by inputting historical electricity prices, wind speed, irradiance, industrial load, holiday marks, etc., the electricity price curve for the next 24 hours can be output with an error of ≤3.5%.

[0092] S4. Construct a meteorological prediction model. The input data of the meteorological prediction model includes meteorological observation data. The meteorological observation data can be a time series of meteorological observation data (temperature, irradiance, wind speed) from multiple sites. The output data includes meteorological forecast results for the next 72 hours. The meteorological forecast results at least include key meteorological events that affect the output of the power grid, such as the impact of strong wind weather on the sudden increase in wind power output. Train the meteorological prediction model based on historical basic data.

[0093] In a specific embodiment, the prediction result of the meteorological prediction model shows that in the predicted typhoon weather, the wind speed will suddenly increase from 5m / s to 15m / s. After calculation or query, it can be found that the wind power output will increase by 60%.

[0094] In some embodiments, the input data of the weather forecast model is time series data; the weather forecast model adopts a fully connected long short-term memory network model (FC-LSTM), which mainly uses LSTM and attention mechanism to predict future meteorological variables (such as wind speed, irradiance) and capture key meteorological events, and its expression is:

[0095]

[0096] in, Represents the output of the weather forecast model; Indicates the time step of the input sequence, for example, predicting the next 24 hours based on 72 hours of historical data; Indicates time Meteorological characteristics, that is, the vector composed of the input data of the meteorological forecast model; Indicates time of predicted meteorological variables; represents the attention coefficient; represents the attention algorithm, , The feature vector representing the input content x, b is the bias term, and the softmax function is used to convert the input feature vector x into a probability distribution so that the sum of the probabilities between different feature vectors is 1; Represents the hidden state of the long short-term memory network; Represents the attention mechanism weight.

[0097] In the above technical solution, the fully connected long short-term memory network model uses a forget gate, an input gate and a cell state to perform computational marking on the weather forecast, and the gating mechanism includes:

[0098]

[0099] in, Represents the forget gate, which is used to control the retention ratio of historical meteorological information (for example, the memory of the high-pressure system before a typhoon); Represents the Sigmoid function, with an output range of [0,1]; Represents the gate weight matrix in the direction of the forget gate; Represents the hidden state of the previous moment (short-term memory); Represents the gate bias matrix in the forget gate direction; represents the input gate, which is used to control the introduction of new meteorological information (e.g., the impact of sudden rainfall); Represents the gating weight matrix in the direction of the input gate; Represents the gating weight matrix in the direction of the input gate; represents a candidate memory unit, which is used to temporarily store the candidate state of new meteorological information; tanh represents the hyperbolic tangent function, which is used to compress the value into the range of [−1,1]; Represents the gating weight matrix in the direction of the cell state; Represents the gate bias matrix in the direction of the cell state; Represents cell state (long-term memory) and is used to synthesize long-term meteorological trends (such as monsoon cycles) and short-term fluctuations (such as temperature drops); Indicates the cell state at the previous moment; represents the output gate; Represents the gating weight matrix in the direction of the output gate; Represents the gate bias matrix in the output gate direction; Represents element-wise multiplication; Represents the hidden layer state.

[0100] In a specific embodiment, a wind farm needs to predict the wind speed in the next 24 hours to adjust the wind turbine output plan. The input parameters are historical wind speed, temperature, and humidity data of multiple sites. For example, it is predicted that the wind speed will increase from 5m / s to 15m / s during a typhoon, and the wind power output will be increased by 60%. The wind speed forecast for the next 24 hours is output, with an error of ≤9%.

[0101] S5. Construct a passenger flow-load prediction model. The input data of the passenger flow-load prediction model includes the passenger flow data of the target area, such as urban passenger flow heat map, holiday markers, traffic hub flow, etc. The output data includes the future electricity demand data of the target area, that is, the time-segment load curve of the electricity demand of each target area in the next 24 hours; train the passenger flow-load prediction model based on historical basic data.

[0102] In a specific embodiment, the prediction result of the passenger flow-load prediction model shows that the passenger flow in a certain commercial area increases sharply during the evening peak, and the power load increases from 10MW to 15MW.

[0103] In some embodiments, the crowd flow-load forecasting model adopts a city-level crowd flow forecasting model (FCCF) based on big data, captures the spatial distribution (such as commercial area aggregation) and time periodicity (such as morning and evening peaks) of crowd flow through spatiotemporal convolution, and combines photovoltaic output to correct load forecast. According to the spatiotemporal characteristics of crowd flow and the power generation of renewable energy, it can provide more practical reference information for formulating power supply strategies, thereby helping to optimize power grid scheduling and improve energy utilization efficiency; its expression is:

[0104]

[0105] in, Indicates area In the period Electricity demand data; Represents the convolution kernel weight; Indicates area In the period The spatiotemporal characteristics of , i.e., the vector composed of the input data of the passenger flow-load forecasting model; Represents the photovoltaic output coupling coefficient; Indicates area In the period of photovoltaic output; Represents the index of the convolution kernel in the time dimension; Represents the index of the convolution kernel in the spatial dimension.

[0106] Based on the above model, it is predicted that the evening peak load in the commercial area will increase from 10MW to 15MW, with an error of ≤4.2%.

[0107] S6. Construct a dynamic scheduling optimization model based on the power supply cost, power abandonment rate and load deviation. The power supply cost is calculated based on the prediction results of the electricity price prediction model; the power abandonment rate is determined based on the prediction results of the meteorological prediction model; and the load deviation is calculated based on the prediction results of the passenger flow-load prediction model.

[0108] Specifically, based on the comprehensive application of the core model established in the above steps S3 to S5, a basic data set about the optimal power generation and power supply can be derived. Subsequently, by introducing a dynamic scheduling optimization model for precise calculation, this series of basic data can be further converted into a specific and efficient power supply strategy. The dynamic scheduling optimization model adopted in the present disclosure can analyze the operating status of the power system in real time, including key factors such as load demand, power generation equipment performance, and transmission line capacity, and can also dynamically adjust and optimize the power generation plan based on these real-time data.

[0109] In some embodiments, the step of constructing a dynamic scheduling optimization model based on power supply cost, power abandonment rate and load deviation includes:

[0110]

[0111] in, represents the time step of the input sequence; represents the power supply cost at time t; represents the power abandonment rate at time t; Indicates load deviation; The weight coefficient of the power supply cost represents the economic parameter of the power supply strategy; The weight coefficient of the power abandonment rate represents the environmental protection parameter of the power supply strategy; The weight coefficient of load deviation represents the stability parameter of the power supply strategy. , to achieve normalization constraints.

[0112] Specifically, the power supply cost is calculated based on the power price result predicted by the power price prediction model and the energy storage cost, including:

[0113]

[0114] in, represents the wind power price at time t predicted by the electricity price prediction model; represents the wind power output at time t; represents the photovoltaic power price at time t predicted by the power price prediction model; represents the photovoltaic output at time t; represents the energy storage cost at time t; Indicates the energy storage capacity at time t;

[0115] The power abandonment rate is calculated based on the difference between wind power output and grid absorption capacity, including:

[0116]

[0117] in, represents the maximum wind power output, which is determined according to the prediction result of the meteorological prediction model; Indicates the actual wind power consumption of the power grid;

[0118] The load deviation is calculated based on the mean square error between the load demand predicted by the passenger flow-load forecasting model and the actual power supply:

[0119]

[0120] in, represents the load demand predicted by the passenger flow-load forecasting model; Indicates the actual power supply.

[0121] In some embodiments, in the dynamic scheduling optimization model, constraints are set, and the constraints include:

[0122] Wind, solar and hydropower output ≤ installed capacity;

[0123] Energy storage charging and discharging power ≤ rated power;

[0124] Grid load ≤ transmission line capacity;

[0125] The optimization algorithm of the dynamic scheduling optimization model adopts a synchronous alternating direction multiplier method and combines it with a reinforcement learning dynamic adjustment strategy.

[0126] S7. Obtain basic data collected in real time, input the basic data collected in real time into the electricity price prediction model, the weather prediction model and the passenger flow-load prediction model to obtain the prediction results, and input the prediction results into the dynamic scheduling optimization model to dynamically adjust the economy, environmental protection and stability strategies of the power grid.

[0127] pass , , Different value combinations of can set the power supply strategy mode of the power grid. For example, in the economic model, , , ; In Eco mode, , , .

[0128] You can also set one of the parameters according to different weather conditions to adjust the power supply strategy. For example, in typhoon weather, wind power is given priority. , and called on backup hydropower and energy storage equipment to prioritize power supply to commercial areas, reducing the wind curtailment rate from 15% to 1%, saving 18% in costs.

[0129] In some embodiments, the above-mentioned models are continuously adjusted in a real-time feedback and correction manner to dynamically adjust the power supply strategy. For example, new energy power generation data, meteorological data, passenger flow data in the target area, and electricity market data are received and updated every 30 minutes, and model parameters (tree structure of GBRT, convolution kernel weight of FCCF) are updated through online learning to dynamically adjust the power supply strategy according to the latest prediction results.

[0130] In other embodiments, the present invention provides an artificial intelligence-based new energy optimal power supply strategy system, including a data collection module, an electricity price prediction module, a weather prediction module, a passenger flow-load prediction module, a dynamic scheduling optimization module and a data output module.

[0131] The data collection module obtains basic data, including new energy power generation data, meteorological data, passenger flow data in the target area, and electricity market data.

[0132] The electricity price prediction module includes an electricity price prediction model. The input data of the electricity price prediction model includes the output data of new energy power generation, and the output data includes the probability distribution of electricity prices. The electricity price prediction model is trained based on historical basic data. The basic data obtained in real time is input into the electricity price prediction module to obtain the electricity price prediction results.

[0133] The meteorological forecast module includes a meteorological forecast model, the input data of the meteorological forecast model includes meteorological observation data, and the output data includes meteorological forecast results, which at least include key meteorological events that affect the output of the power grid; the meteorological forecast model is trained based on historical basic data; and the basic data obtained in real time is input into the meteorological forecast module to obtain meteorological forecast results.

[0134] The passenger flow-load prediction module includes a passenger flow-load prediction model. The input data of the passenger flow-load prediction model includes the passenger flow data of the target area, and the output data includes the future electricity demand data of the target area. The passenger flow-load prediction model is trained based on historical basic data. The basic data obtained in real time is input into the passenger flow-load prediction module to obtain the load prediction result.

[0135] The dynamic scheduling optimization module includes a dynamic scheduling optimization model constructed according to the power supply cost, power abandonment rate and load deviation. The power supply cost is calculated according to the prediction results of the electricity price prediction model; the power abandonment rate is determined according to the prediction results of the meteorological prediction model; and the load deviation is calculated according to the prediction results of the passenger flow-load prediction model.

[0136] The data output module outputs the power supply strategy formed by the dynamic scheduling optimization module according to the prediction results of the electricity price prediction module, the weather prediction module and the passenger flow-load prediction module.

[0137] In a specific embodiment, the data output module includes a visualization interface to display the prediction results, scheduling strategies and operating indicators. The contents of the prediction results display include: electricity price curve prediction, a chart showing the trend of future electricity price changes, including peak and valley electricity prices, electricity price fluctuation range, etc.; wind speed change prediction, real-time wind speed prediction value and trend chart, including wind speed size, wind direction, etc.; crowd flow heat map prediction, predicting the heat map of future crowd flow distribution, showing crowded areas, flow direction, etc.; irradiance prediction, predicting the chart of future solar irradiance, providing reference for photovoltaic power generation; temperature and humidity prediction, predicting the trend of future temperature and humidity changes, and providing environmental data for wind power, photovoltaic and power grid operation; meteorological disaster warning, predicting possible disaster weather such as typhoons and rainstorms based on meteorological data, and providing warning information.

[0138] The dispatch strategy display includes: energy storage charging and discharging plan, which displays the charging and discharging plan of energy storage equipment, including charging time, discharging time, charging and discharging power, etc.; grid load distribution, which displays the distribution of grid load in real time, including the load size of each region, load change trend, etc.; new energy power generation dispatching, which formulates a dispatch plan for new energy power generation equipment based on the forecast results of new energy power generation to ensure stable power supply; demand response strategy, which formulates a demand response strategy based on power market data and user electricity consumption behavior to adjust user electricity load; grid fault response plan, which formulates a fault response plan based on the operating status of the grid, including fault location, isolation, and recovery.

[0139] The operation indicators displayed include: cost, which displays the total cost of renewable energy power generation and grid operation; power abandonment rate, which displays the proportion of power abandonment caused by the inability of renewable energy power generation to be connected to the grid; load deviation, which displays the deviation between the actual load and the predicted load; energy efficiency indicators, which display the energy efficiency indicators of grid operation, such as transmission loss and distribution loss; carbon emissions, which calculates carbon emissions based on renewable energy power generation and grid operation; user satisfaction, which evaluates user satisfaction based on indicators such as user electricity usage behavior and number of power outages.

[0140] In addition, the data output module can also output the dispatching strategy to the power grid SCADA system and energy storage management platform to realize the automatic execution of the power supply strategy.

[0141] In a specific embodiment, the visualization interface includes: a strategy configurator page, a real-time strategy deduction page, a historical strategy comparison page, a strategy effect dashboard page, and an AI strategy recommendation page.

[0142] The Weight Adjustment panel in the Strategy Configurator page contains the economics ( )、Environmental protection( ),stability( ) Three-target slider adjustment, real-time linkage strategy results. Preset mode switching (economic priority / environmental protection priority / extreme weather plan), support for custom strategy template saving; constraint settings in the strategy configurator page include wind and solar output upper limit (linked FC-LSTM prediction value), energy storage charging and discharging threshold, and grid stability tolerance.

[0143] The real-time strategy deduction page uses a dynamic Gantt chart to display the power output plan (wind power / photovoltaic / hydropower / energy storage) for the next 24 hours, and supports dragging to adjust the output ratio during the period. Through the cost-benefit heat map: the cost, power abandonment rate, and load deviation risk level of the strategy execution are marked by hourly dimension.

[0144] The historical strategy comparison page compares the parameter combinations of different strategies through parallel coordinate graphs And result indicators (total cost, power abandonment rate, stability score). You can also perform strategy retrospective analysis by clicking on any historical strategy to load the meteorological data, load curve and actual execution effect at that time.

[0145] The strategy effect dashboard page displays the current strategy execution progress (such as energy storage SOC status, actual power abandonment rate) and abnormal warnings (such as sudden drop in wind and solar power output triggering strategy adaptive correction). This page comes with an Excel information export function.

[0146] The AI ​​strategy recommendation page recommends strategy adjustments based on real-time data (such as "it is recommended to increase the weight of energy storage discharge when a typhoon passes").

[0147] In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0148] Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A new energy optimal power supply strategy method based on artificial intelligence, characterized in that: include: Obtain historical basic data, including renewable energy power generation data, meteorological data, passenger flow data in the target area, and power market data; Construct an electricity price prediction model, the input data of which includes the output data of renewable energy power generation, and the output data includes the probability distribution of electricity prices; train the electricity price prediction model based on historical basic data; Constructing a meteorological prediction model, wherein the input data of the meteorological prediction model includes meteorological observation data, and the output data includes meteorological prediction results, wherein the meteorological prediction results at least include key meteorological events that affect the output of the power grid; training the meteorological prediction model based on historical basic data; Construct a passenger flow-load forecasting model, the input data of which includes passenger flow data of the target area, and the output data includes future electricity demand data of the target area; train the passenger flow-load forecasting model based on historical basic data; A dynamic scheduling optimization model is constructed based on the power supply cost, power abandonment rate and load deviation, wherein the power supply cost is calculated based on the prediction results of the electricity price prediction model; the power abandonment rate is determined based on the prediction results of the meteorological prediction model; and the load deviation is calculated based on the prediction results of the passenger flow-load prediction model; Obtain basic data collected in real time, input the basic data collected in real time into the electricity price prediction model, meteorological prediction model and passenger flow-load prediction model to obtain prediction results, and input the prediction results into the dynamic dispatch optimization model to dynamically adjust the economic, environmental protection and stability strategies of the power grid; The dynamic dispatch optimization model is constructed according to the power supply cost, power abandonment rate and load deviation, including: in, represents the time step of the input sequence; represents the power supply cost at time t; represents the power abandonment rate at time t; Indicates load deviation; The weight coefficient of the power supply cost represents the economic parameter of the power supply strategy; The weight coefficient of the power abandonment rate represents the environmental protection parameter of the power supply strategy; The weight coefficient representing the load deviation represents the stability parameter of the power supply strategy; The power supply cost is calculated based on the power price result predicted by the power price prediction model and the energy storage cost, including: in, represents the wind power price at time t predicted by the electricity price prediction model; represents the wind power output at time t; represents the photovoltaic power price at time t predicted by the power price prediction model; represents the photovoltaic output at time t; represents the energy storage cost at time t; Indicates the energy storage capacity at time t; The power abandonment rate is calculated based on the difference between wind power output and grid absorption capacity, including: in, represents the maximum wind power output, which is determined according to the prediction result of the meteorological prediction model; Indicates the actual wind power consumption of the power grid; The load deviation is calculated based on the mean square error between the load demand predicted by the passenger flow-load forecasting model and the actual power supply: in, represents the load demand predicted by the passenger flow-load forecasting model; Indicates the actual power supply.

2. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 1 is characterized in that: The electricity price prediction model adopts a gradient boosting regression tree model, and the expression of the electricity price prediction model is: in, represents the number of samples; Indicates the sample number; Indicates The actual electricity price of samples; Represents the input features, that is, the vector of input data of the electricity price prediction model; represents the integrated model predicting electricity price, ; represents the regularization coefficient; Represents the total number of regression trees; Indicates The number of leaf nodes in a regression tree; Indicates the minimum gain of leaf node splitting; Indicates The leaf node weights of a regression tree.

3. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 1 is characterized in that: The input data of the weather forecast model is time series data; the weather forecast model adopts a fully connected long short-term memory network model, and its expression is: in, Represents the output of the weather forecast model; represents the time step of the input sequence; Indicates time Meteorological characteristics, that is, the vector composed of the input data of the meteorological forecast model; Indicates time of predicted meteorological variables; represents the attention coefficient; Represents the attention algorithm; Represents the hidden state of the long short-term memory network; Represents the attention mechanism weight.

4. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 3 is characterized in that: The fully connected long short-term memory network model uses forget gate, input gate and cell state to perform calculation mark on weather forecast, and the gating mechanism includes: in, represents the forget gate, which is used to control the retention ratio of historical meteorological information; Represents the Sigmoid function; Represents the gate weight matrix in the direction of the forget gate; Represents the hidden state of the previous moment; Represents the gate bias matrix in the forget gate direction; represents the input gate, which is used to control the introduction of new meteorological information; Represents the gating weight matrix in the direction of the input gate; Represents the gating weight matrix in the direction of the input gate; represents a candidate memory unit, which is used to temporarily store the candidate state of new meteorological information; tanh represents the hyperbolic tangent function; Represents the gating weight matrix in the direction of the cell state; Represents the gate bias matrix in the direction of the cell state; Represents the cell state and is used to integrate long-term meteorological trends and short-term fluctuations; Indicates the cell state at the previous moment; represents the output gate; Represents the gating weight matrix in the direction of the output gate; Represents the gate bias matrix in the output gate direction; Represents element-wise multiplication; Represents the hidden layer state.

5. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 1 is characterized in that: The crowd flow-load prediction model adopts a city-level crowd flow prediction model based on big data, and its expression is: in, Indicates area In the period Electricity demand data; Represents the convolution kernel weight; Indicates area In the period The spatiotemporal characteristics of , i.e., the vector of input data of the passenger flow-load forecasting model; Represents the photovoltaic output coupling coefficient; Indicates area In the period of photovoltaic output; Represents the index of the convolution kernel in the time dimension; Represents the index of the convolution kernel in the spatial dimension.

6. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 1 is characterized in that: In the dynamic scheduling optimization model, constraints are set, and the constraints include: Wind, solar and hydropower output ≤ installed capacity; Energy storage charging and discharging power ≤ rated power; Grid load ≤ transmission line capacity; The optimization algorithm of the dynamic scheduling optimization model adopts a synchronous alternating direction multiplier method.

7. The optimal power supply strategy method for new energy based on artificial intelligence according to claim 1 is characterized in that: Before inputting the basic data into the model, the basic data is preprocessed, and the data preprocessing includes: Perform wavelet filtering on the renewable energy power generation data to complete the denoising of the basic data; Map meteorological data to the locations of new energy power stations and construct a spatial correlation matrix; Standardize the data structure of heterogeneous data in the basic data and discard invalid data.

8. An optimal power supply strategy system for new energy based on artificial intelligence, characterized in that: include: Data collection module, which obtains basic data, including renewable energy power generation data, meteorological data, passenger flow data in the target area, and power market data; The electricity price prediction module includes an electricity price prediction model, the input data of the electricity price prediction model includes the output data of new energy power generation, and the output data includes the probability distribution of electricity prices; the electricity price prediction model is trained according to historical basic data; the basic data obtained in real time is input into the electricity price prediction module to obtain the electricity price prediction result; A meteorological forecast module includes a meteorological forecast model, the input data of the meteorological forecast model includes meteorological observation data, and the output data includes meteorological forecast results, and the meteorological forecast results at least include key meteorological events that affect the power grid output; the meteorological forecast model is trained according to historical basic data; the basic data obtained in real time is input into the meteorological forecast module to obtain the meteorological forecast results; A passenger flow-load prediction module includes a passenger flow-load prediction model, wherein the input data of the passenger flow-load prediction model includes passenger flow data of a target area, and the output data includes future electricity demand data of the target area; the passenger flow-load prediction model is trained according to historical basic data; and the basic data obtained in real time is input into the passenger flow-load prediction module to obtain a load prediction result; The dynamic scheduling optimization module includes a dynamic scheduling optimization model constructed according to the power supply cost, the power abandonment rate and the load deviation, wherein the power supply cost is calculated according to the prediction result of the power price prediction model; the power abandonment rate is determined according to the prediction result of the meteorological prediction model; and the load deviation is calculated according to the prediction result of the passenger flow-load prediction model; The data output module outputs the power supply strategy formed by the dynamic scheduling optimization module according to the prediction results of the electricity price prediction module, the weather prediction module and the passenger flow-load prediction module; The dynamic dispatch optimization model is constructed according to the power supply cost, power abandonment rate and load deviation, including: in, represents the time step of the input sequence; represents the power supply cost at time t; represents the power abandonment rate at time t; Indicates load deviation; The weight coefficient of the power supply cost represents the economic parameter of the power supply strategy; The weight coefficient of the power abandonment rate represents the environmental protection parameter of the power supply strategy; The weight coefficient representing the load deviation represents the stability parameter of the power supply strategy; The power supply cost is calculated based on the power price result predicted by the power price prediction model and the energy storage cost, including: in, represents the wind power price at time t predicted by the electricity price prediction model; represents the wind power output at time t; represents the photovoltaic power price at time t predicted by the power price prediction model; represents the photovoltaic output at time t; represents the energy storage cost at time t; Indicates the energy storage capacity at time t; The power abandonment rate is calculated based on the difference between wind power output and grid absorption capacity, including: in, represents the maximum wind power output, which is determined according to the prediction result of the meteorological prediction model; Indicates the actual wind power consumption of the power grid; The load deviation is calculated based on the mean square error between the load demand predicted by the passenger flow-load forecasting model and the actual power supply: in, represents the load demand predicted by the passenger flow-load forecasting model; Indicates the actual power supply.

Images