Power peak load regulation system based on Internet of Things

Abstract

The present invention relates to an Internet of Things-based power peak-shaving system, which aims to improve the peak-shaving capacity and efficiency of a new energy power system; the system combines advanced machine learning technology and Internet of Things infrastructure to achieve accurate monitoring and efficient management of the power grid operation status; the core components of the system include a prediction module, an Internet of Things module, a peak-shaving module and an energy storage module; the prediction module uses a machine learning algorithm to predict the new energy generation capacity and the load demand of the power grid; the Internet of Things module includes sensors and intelligent metering devices installed at key nodes of the power grid to collect power grid operation data in real time. The peak-shaving module receives data from the prediction module and the Internet of Things module, and uses a mixed integer linear programming model to determine the optimal peak-shaving strategy; finally, the energy storage module, including battery energy storage and pumped storage submodules, adjusts charging and discharging according to the instructions of the peak-shaving module to support the load and supply balance of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power peak regulation, and in particular to a power peak regulation system based on the Internet of Things. Background Art

[0002] As the global demand for sustainable energy grows, the share of renewable energy sources such as wind and solar energy in power systems is gradually increasing. However, due to the intermittent and unpredictable nature of these energy sources, power systems face major challenges in peak load regulation and load balancing. Traditional methods of power peak regulation, such as relying on fuel-driven power stations, are not only costly but also have negative impacts on the environment. Therefore, it becomes crucial to develop an efficient, economical, and environmentally friendly power peak regulation system.

[0003] In recent years, Internet of Things (IoT) technology and machine learning algorithms have shown great potential in many fields, especially in data collection, processing and predictive analysis. In the power industry, the application of these technologies provides new possibilities for achieving smarter, more efficient and reliable power grid management.

[0004] In view of these challenges and opportunities, the present invention proposes a power peak-shaving system based on the Internet of Things. Summary of the invention

[0005] The present application provides an Internet of Things-based power peak-shaving system to improve the peak-shaving capability and efficiency of new energy power systems.

[0006] The power peak-shaving system based on the Internet of Things provided in this application includes:

[0007] The prediction module is used to predict the renewable energy generation capacity and grid load demand using machine learning algorithms, and pass the prediction results to the peak load regulation module;

[0008] The Internet of Things module includes sensors and intelligent metering devices installed at the power grid measurement nodes, which are used to monitor the power grid operation data in real time; and transmit the monitored power grid operation data to the peak load module; wherein the power grid operation data includes energy flow direction, frequency, and voltage;

[0009] A peak-shaving module is used to receive the prediction results provided by the prediction module and the power grid operation data provided by the Internet of Things module; determine the peak-shaving strategy using a mixed integer linear programming model according to the prediction results and the power grid operation data; adjust the power output of the power grid and formulate a charging and discharging strategy for the energy storage module according to the peak-shaving strategy; and send charging and discharging instructions to the energy storage module according to the charging and discharging strategy;

[0010] The energy storage module includes a battery energy storage submodule and a pumped energy storage submodule; it is used to adjust the charge and discharge according to the charge and discharge instructions sent by the peak load regulation module to support the load and supply balance of the power grid.

[0011] Furthermore, the prediction module is specifically used to use a pre-trained multimodal deep learning network to predict renewable energy power generation capacity and grid load demand; wherein the multimodal deep learning network includes a convolutional neural network part, a long short-term memory network part and a data fusion and output layer; the convolutional neural network part receives satellite images, extracts spatial features using the convolutional neural network, and obtains a spatial feature representation of the satellite image for predicting solar power generation; the long short-term memory network part receives historical grid load data, meteorological data, holiday information, industrial activity data, and user behavior patterns, and obtains a time series feature representation of the input data using the long short-term memory network; the data fusion and output layer fuses the spatial feature representation and the time series feature representation, and outputs the predicted values ​​of renewable energy power generation capacity and grid load demand.

[0012] Furthermore, the data fusion and output layer includes an attention mechanism part for enhancing the prediction performance of the multimodal deep learning network; wherein the attention mechanism part is specifically used for:

[0013] Receiving the spatial feature representation and the time series feature representation;

[0014] Using a fully connected network, determining a first attention score of the spatial feature representation and a second attention score of the time series feature representation based on the spatial feature representation and the time series feature representation;

[0015] Normalizing the first attention score and the second attention score through a softmax layer to ensure that the sum of the normalized first attention score and the normalized second attention score is 1;

[0016] The spatial feature representation and the time series feature representation are weighted using the standardized first attention score and the standardized second attention score.

[0017] Furthermore, the data fusion and output layer includes a dynamic weight adjustment mechanism, which is specifically used to:

[0018] Encoding state information of the power grid to obtain a state code, wherein the state information includes season information or special event information;

[0019] Using the added input layer in the multimodal deep learning network, the receiving state encoding;

[0020] Calculate the weight of feature fusion based on the received state code;

[0021] According to the calculated feature fusion weight, the spatial feature representation and the time series feature representation are fused.

[0022] Furthermore, the objective function Objective of the mixed integer linear programming model adopts the following formula 1:

[0023] Objective=w1·C total +w2·U renew -w3·E carbon +w4·R system Formula 1

[0024] Among them, C total is the economic cost function; U renew is the renewable energy utilization function; E carbon is the carbon emission function; R system is the system reliability function; w1, w2, w3, w4 are weight coefficients;

[0025] Economic cost function C total Use the following formula 2:

[0026] ∑ t (C gen (t)·P gen (t)-C sell (t)·F grid (t)+C store (S bat (t),S pump (t))) Formula 2

[0027] Among them, C gen (t) is the electricity generation cost at time t; C sell (t) is the electricity price at time t; C store is the storage cost function; P gen (t) is the predicted total power generation at time t; F grid (t) is the energy flowing from the grid to the storage facility at time t; S bat (t) and S pump (t) represent the states of battery energy storage and pumped storage at time t, respectively;

[0028] Renewable energy utilization function U renew Using the following formula 3:

[0029]

[0030] Among them, P renew (t) represents the renewable energy power generation at time t;

[0031] Carbon emission function Ecarbon Using the following formula 4:

[0032] E carbon =∑ t E factor ·(P gen (t)-P renew (t)) Formula 4

[0033] Among them, E factor is the carbon emission factor, which represents the carbon emission per unit of non-renewable energy generation;

[0034] System reliability function R system Using the following formula 5:

[0035] R system =∑ t (W5R storage (t)+w6·R match (t)+w7·R renewable (t)) Formula 5

[0036] Among them, R storage (t) is the energy storage responsiveness score, which can be evaluated based on the difference between the remaining capacity of the energy storage facility and the predicted load; R match (t) is the load matching score, which can be evaluated by comparing the difference between the dispatchable generation capacity and the forecast load; R renewable (t) is the renewable energy utilization efficiency score, which can be evaluated based on the ratio of renewable energy output to total power generation; w5, w6, and w7 are weight coefficients.

[0037] Furthermore, the constraints of the mixed integer linear programming model include an energy balance constraint, and the energy balance constraint is expressed using the following formula:

[0038] P gen (t) = P dem (t)+F grid (t)

[0039] Among them, P dem (t) is the grid demand at time t.

[0040] Furthermore, the constraints of the mixed integer linear programming model include energy storage device constraints, and the energy storage device constraints are expressed using the following formula:

[0041] S min ≤S bat (t)

[0042] S pump (t)≤S max

[0043] Among them, S min and S max They respectively refer to the minimum and maximum storage capacity limits of energy storage facilities in operation.

[0044] Furthermore, the constraints of the mixed integer linear programming model include power generation resource allocation optimization constraints, and the power generation resource allocation optimization constraints are expressed using the following formula:

[0045] ΔP gen (t)≤Rate change ·P gen (t-1)

[0046] Where ΔP gen (t) is the change in power generation between time t and t-1; Rate change is the maximum rate of change allowed in power generation.

[0047] Furthermore, the constraints of the mixed integer linear programming model include an environmental impact minimization constraint, and the environmental impact minimization constraint is expressed using the following formula:

[0048] E impact (t) = EnvImpactFactor·(P gen (t)-P renew (t))

[0049] Among them, E impact (t) is the environmental impact at time t; EnvImpactFactor is the environmental impact factor used to quantify the environmental impact of each unit of non-renewable energy generation.

[0050] Furthermore, the objective function of the mixed integer linear programming model further includes a network loss minimization factor C loss , which is expressed using the following formula:

[0051] C loss =∑ t LossFactor·(P gen (t)-P dem (t)) 2

[0052] Among them, LossFactor is the network loss factor, which is used to quantify the energy loss in the power grid.

[0053] The beneficial effects of this application mainly include: (1) By predicting the new energy power generation capacity and grid load demand through machine learning algorithms, this system can more effectively utilize renewable energy and reduce dependence on traditional fossil fuels, thereby improving energy utilization efficiency. (2) Real-time monitoring of grid operation data (including energy flow, frequency, voltage, etc.) enables the system to quickly respond to changes in grid status, adjust the peak-shaving strategy in a timely manner, and enhance the stability and reliability of the grid. (3) The peak-shaving strategy determined by the mixed integer linear programming model can more accurately adjust the power output, while intelligently formulating the charging and discharging strategy of the energy storage module to optimize the energy distribution and storage management of the entire grid. (4) The system helps to reduce the operating cost of the power system by optimizing the use of new energy and reducing energy waste. At the same time, more efficient energy utilization and the priority use of renewable energy reduce the negative impact on the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a schematic diagram of an Internet of Things-based power peak-shaving system provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0055] Many specific details are described in the following description to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present application, so the present application is not limited by the specific implementation disclosed below.

[0056] The first embodiment of the present application provides a power peak load regulation system based on the Internet of Things. Figure 1 , which is a schematic diagram of the first embodiment of the present application. Figure 1 The first embodiment of the present application provides a power peak-shaving system based on the Internet of Things for detailed description.

[0057] The system includes a prediction module 101, an Internet of Things module 102, a peak-shaving module 103 and an energy storage module 104.

[0058] The prediction module 101 is used to predict the renewable energy generation capacity and grid load demand using a machine learning algorithm, and pass the prediction results to the peak regulation module.

[0059] In the power peak-shaving system based on the Internet of Things provided in this embodiment, the prediction module 101 plays a core role. The module uses advanced machine learning algorithms, such as neural networks, to predict the ability of renewable energy to generate electricity and the load demand of the power grid. Predicting the power generation capacity of renewable energy means that the module can estimate the power generation of renewable energy such as wind power and solar energy in a specific time period based on factors such as weather conditions, seasonal changes, historical data and current trends. This prediction is crucial because the production capacity of renewable energy is usually affected by environmental factors and therefore fluctuates.

[0060] Meanwhile, the forecasting module 101 also focuses on forecasting the load demand of the power grid. This involves analyzing and estimating the power demand of the power grid at a specific time, taking into account factors such as user behavior patterns, industrial activities, weather conditions, holidays and historical load data. By accurately forecasting the load demand of the power grid, the module helps the peak load system to balance power supply and demand more effectively, especially during peak hours or when renewable energy generation is reduced.

[0061] The prediction module 101 is not just a passive data processing unit, it also has the ability to self-learn and adapt. Over time, by continuously receiving and analyzing new data, this module is able to continuously optimize its prediction model, thereby improving the accuracy of the prediction. This means that as more and more data is analyzed, the prediction results will become more accurate, making the entire power peak-shaving system more efficient and reliable.

[0062] In addition, another key function of the prediction module 101 is to pass its prediction results to the peak-shaving module 103. This step ensures that the peak-shaving strategy is based on the latest and most accurate information. In this way, the prediction module 101 directly affects the overall operating efficiency of the power system, helping the peak-shaving module to more intelligently decide when to increase or decrease the supply of electricity, and how to best utilize the energy storage facilities in the energy storage module 104, such as the battery energy storage submodule and the pumped storage submodule.

[0063] This embodiment provides a multi-modal deep learning network to predict the renewable energy generation capacity and grid load demand. The following is a detailed construction scheme of this multi-modal deep learning network:

[0064] (I) Network input includes:

[0065] 1. Time series data: historical grid load data, including hourly or daily electricity consumption records.

[0066] 2. Spatial data: Satellite imagery, especially for solar power generation areas.

[0067] 3. Meteorological data: temperature, wind speed, solar irradiance, precipitation, etc. These factors directly affect the performance of renewable energy power generation.

[0068] 4. Other data: such as holiday information, industrial activity data, user behavior patterns, etc., which can help the model predict the load demand of the power grid more accurately.

[0069] These other data (such as holiday information, industrial activity data, user behavior patterns, etc.) are converted into time series data so that they can be effectively processed by the LSTM layer together with the power grid load data and meteorological data. Here are some examples of how to convert these data into time series data:

[0070] Holiday Information:

[0071] Conversion method: Specific dates can be marked as holidays or non-holidays. For example, a binary sequence corresponding to the timestamp of power grid load data can be created, where holidays are marked as 1 and non-holidays are marked as 0.

[0072] Example: If National Day is a holiday, then the data points on that day will be labeled as 1, and other days will be labeled as 0.

[0073] Industrial activity data:

[0074] Transformation method: Arrange the intensity or output of industrial activities in chronological order. This can be a continuous sequence of values ​​reflecting the level of industrial activity at different points in time.

[0075] Example: If a plant operates 24 / 7 on weekdays and is closed on weekends, data points for weekdays will show higher activity levels and significantly lower activity levels on weekends.

[0076] User behavior patterns:

[0077] Conversion method: The user's electricity usage pattern is encoded in a time series and can be based on historical electricity usage data.

[0078] For example, if household electricity consumption peaks at night, then this part of the time series will show higher electricity consumption. Accordingly, the values ​​of these data points will be higher in the evening.

[0079] By converting this data into a time series format, it can be combined with historical grid load data and meteorological data to provide richer and more contextual input to the LSTM layer. This comprehensive time series data helps improve the model's ability to predict grid load demand, especially during special events (such as holidays) or when there are significant changes in user behavior patterns.

[0080] (II) Network output includes:

[0081] The predicted power generation of new energy (such as solar energy and wind energy) in the future.

[0082] Forecast of grid load demand over a period of time in the future.

[0083] (III) The network structure includes:

[0084] 1. Convolutional Neural Network (CNN) part:

[0085] Input: Satellite imagery.

[0086] Processing: CNN is used to extract spatial features that can reveal the geographical distribution of solar panels, possible shadowing effects, etc.

[0087] Output: Feature representation of satellite imagery for predicting solar power generation.

[0088] 2. Long Short-Term Memory Network (LSTM) Part:

[0089] Input: historical grid load data, meteorological data, holiday information, industrial activity data, user behavior patterns.

[0090] Processing: LSTM is good at processing time series data and can capture the changing trends of power load and meteorological conditions over time.

[0091] Output: Deep feature representation of input data (historical grid load data, meteorological data, and other data such as holiday information, industrial activity data, user behavior patterns, etc.). These features capture the key patterns and trends of data over time. These deep feature representations are not directly predicted values ​​of grid load or meteorological conditions, but contain complex features of time dependency and correlation, which are crucial for subsequent prediction tasks.

[0092] 3. Data fusion and output layer:

[0093] The input is the spatial feature representation from CNN (involving the geographical distribution of solar panels, shadow effects, etc.) and the time series feature representation from LSTM (including historical grid load, meteorological data, etc.). The final output is the predicted values ​​of renewable energy generation capacity and grid load demand, which are based on a deep learning model that combines spatial and time series features.

[0094] First, the data fusion and output layers normalize the input features to ensure that features from different sources are comparable in magnitude and distribution.

[0095] Next, we can introduce the attention mechanism and assign different weights to different features through learning. This can automatically identify which features are more critical to the prediction task, thereby improving the accuracy of the prediction. The attention weight can be dynamically adjusted according to the contribution of the feature to the prediction result.

[0096] Then, the attention-weighted features are combined using fully connected layers or other fusion techniques (such as depth-wise separable convolutional layers). In this process, features from different sources are integrated into a unified feature space.

[0097] The conditional fusion strategy can be used to adjust the fusion weight distribution according to the current state of the power grid (such as season, special events), which enables the model to adapt to different operating conditions.

[0098] Finally, after feature fusion, the network will pass through a series of fully connected layers (or other appropriate output layer structures) to produce the final prediction results. These fully connected layers can be designed to output the predicted values ​​of renewable energy generation capacity and grid load demand. In order to improve the accuracy of the prediction, activation functions such as ReLU or sigmoid can be used in these layers to ensure the effectiveness and reliability of the output.

[0099] Furthermore, the data fusion and output layer includes a dynamic weight adjustment mechanism, which is specifically used to:

[0100] Encoding state information of the power grid to obtain a state code, wherein the state information includes season information or special event information;

[0101] Using the added input layer in the multimodal deep learning network, the receiving state encoding;

[0102] Calculate the weight of feature fusion based on the received state code;

[0103] According to the calculated feature fusion weight, the spatial feature representation and the time series feature representation are fused.

[0104] The data fusion and output layer in this embodiment dynamically adjusts the weights of the features output from CNN and LSTM during the fusion process according to the current state of the power grid (such as season, special events). This will ensure that the performance of the model is optimized under different conditions. The following is a specific implementation method:

[0105] (1) Data preparation and preprocessing

[0106] State coding: First, the state information of the power grid is coded. For example, a year is divided into four seasons, and each season is assigned a code (for example, spring = 1, summer = 2, autumn = 3, winter = 4). Similar coding is also performed for special events (such as holidays or major events).

[0107] Dataset annotation: In the training dataset, each sample is annotated with its corresponding state code. In this way, each training sample contains not only features from CNN and LSTM, but also state codes.

[0108] (2) Network structure adjustment

[0109] Introducing an additional input layer: An additional input layer is added to the multimodal deep learning network to receive the state encoding.

[0110] Design a conditional fusion layer: Before the data fusion layer, design a conditional fusion layer that uses the state encoding to adjust the weights of different features. This can be achieved by adding additional weight matrices that vary depending on the state encoding.

[0111] (3) Weight adjustment mechanism

[0112] Conditional weight calculation: In the conditional fusion layer, the weights of feature fusion are calculated based on the state encoding of the input. For example, if it is summer, the weights of features related to solar power generation may increase.

[0113] Furthermore, the data fusion and output layer includes an attention mechanism part for enhancing the prediction performance of the multimodal deep learning network; wherein the attention mechanism part is specifically used for:

[0114] Receiving the spatial feature representation and the time series feature representation;

[0115] Using a fully connected network, determining a first attention score of the spatial feature representation and a second attention score of the time series feature representation based on the spatial feature representation and the time series feature representation;

[0116] Normalizing the first attention score and the second attention score through a softmax layer to ensure that the sum of the first attention score and the second attention score is 1;

[0117] The spatial feature representation and the time series feature representation are weighted using the standardized first attention score and the standardized second attention score.

[0118] Description of the data fusion and output layer attention mechanism in this embodiment:

[0119] (1) Theoretical basis of attention mechanism:

[0120] First, we need to explain how the attention mechanism works. In simple terms, the attention mechanism can be seen as a trained network part that learns to assign different weights to different input features, thereby highlighting the most important features for the prediction task.

[0121] (2) Design of network structure:

[0122] After the existing CNN and LSTM output layers, add an attention layer. This layer contains a trainable weight matrix that is used to calculate the importance scores of the input features.

[0123] (3) Calculation of attention weight:

[0124] A small fully connected network is used to calculate the attention score for each feature. This network takes the output of CNN and LSTM as input and outputs the importance score of each feature.

[0125] These scores are then normalized through a softmax layer to ensure that the sum of all scores is 1, so that each score can be understood as the importance weight of that feature.

[0126] (4) Feature weighting and fusion:

[0127] Weight the input features using the calculated attention scores. Specifically, multiply each feature by its corresponding attention score.

[0128] Then, the weighted features are fused to form a comprehensive feature representation for subsequent prediction tasks.

[0129] Below is the reference implementation code for the multimodal deep learning network:

[0130]

[0131]

[0132]

[0133]

[0134] In summary, the prediction module 101 is indispensable in this power peak-shaving system based on the Internet of Things. It provides the system with powerful data analysis and prediction capabilities through advanced machine learning technology, ensuring that the power grid can operate efficiently and stably under various conditions.

[0135] In order to train the above multimodal deep learning network, several key steps need to be performed, including preparing data, defining loss functions and optimizers, writing training loops, and performing model evaluation. The following is a detailed description of these steps:

[0136] 1. Data preparation

[0137] Dataset loading: Load and preprocess the required datasets, including satellite images, historical grid load data, meteorological data, and grid status information.

[0138] Data preprocessing: Perform necessary preprocessing steps on different types of data. For example, satellite images may need to be resized and standardized; time series data such as power grid load and meteorological data may need to be normalized.

[0139] Data partitioning: Divide the data into training set, validation set, and test set.

[0140] 2. Define loss function and optimizer

[0141] Loss function: Choose an appropriate loss function based on the nature of the prediction task. For regression tasks, commonly used loss functions are mean square error (MSE) or root mean square error (RMSE).

[0142] Optimizer: Select an optimizer to adjust the network parameters. Common optimizers include SGD, Adam, etc. The choice of optimizer depends on the specific task and data.

[0143] 3. Write a training loop

[0144] Iterative training: The network is trained for multiple epochs, and each epoch includes the process of traversing the entire training set.

[0145] Forward Propagation: At each training step, the input data is propagated forward through the network to produce predicted outputs.

[0146] Calculate loss: Calculate the loss between the predicted output and the actual label.

[0147] Backpropagation: Calculate the gradient with respect to the loss and update the weights of the network through the optimizer.

[0148] Validation: At the end of each epoch, evaluate the performance of the model using the validation set, monitoring the loss and other metrics such as accuracy.

[0149] 4. Model evaluation and tuning

[0150] Performance evaluation: The trained model is evaluated using the validation set and the test set to ensure its generalization ability.

[0151] Hyperparameter tuning: Depending on the performance of the model on the validation set, you may need to tune hyperparameters such as learning rate, batch size, or network architecture.

[0152] Overfitting check: Monitor overfitting during training and, if necessary, apply techniques such as dropout or early stopping to reduce overfitting.

[0153] The Internet of Things module 102 includes sensors and intelligent metering devices installed at the grid measurement nodes, which are used to monitor the grid operation data in real time; and transmit the monitored grid operation data to the peak-shaving module; wherein the grid operation data includes energy flow, frequency, and voltage.

[0154] The IoT module 102 plays a vital role in this IoT-based power peak load regulation system. It is mainly composed of a variety of sensors and smart metering devices installed at key measurement nodes of the power grid, which work together to monitor the operation status of the power grid in real time.

[0155] First, the sensors and smart metering devices included in this module are deployed at key locations in the power grid, such as substations, distribution stations, and major power supply lines. These devices are highly sensitive and accurate, and can capture key parameters of the power grid in real time. For example, they can measure the direction and intensity of current flow, frequency fluctuations of the power grid, voltage levels, etc. These measurements are critical to understanding the real-time status of the power grid, especially when the load on the power grid changes greatly.

[0156] In addition, the IoT module 102 is not only a passive monitoring system, but also has the function of transmitting the collected data to the peak-shaving module 103 in real time. This real-time data transmission is crucial for the stable operation of the power system. Through the real-time transmitted data, the peak-shaving module can quickly learn the current state of the power grid, including the real-time flow of energy, the stability of power supply and any possible abnormal conditions.

[0157] With the help of the Internet of Things module 102, the peak load regulation module 103 can respond quickly based on the latest, real-time grid information. It can formulate or adjust the peak load regulation strategy based on the real-time monitored grid conditions, combined with the forecast data of the new energy generation capacity and grid load demand provided by the forecast module 101. These strategies may include adjusting the power output of the grid, optimizing the charging and discharging plan of the energy storage module, etc., to ensure the stable operation and power supply efficiency of the grid.

[0158] In short, the IoT module 102 is the "sensing organ" of the power peak load regulation system, which continuously monitors the lifeline of the power grid to ensure that the system can make decisions based on the latest and most accurate information. This high level of monitoring and data communication capabilities enables the power peak load regulation system to respond to changes in the power grid in real time, improve adaptability to fluctuations in new energy, and ensure the reliability and efficiency of the power grid.

[0159] The peak-shaving module 103 is used to receive the prediction results provided by the prediction module and the power grid operation data provided by the Internet of Things module; determine the peak-shaving strategy using a mixed integer linear programming model based on the prediction results and the power grid operation data; adjust the power output of the power grid and formulate a charging and discharging strategy for the energy storage module based on the peak-shaving strategy; and send charging and discharging instructions to the energy storage module based on the charging and discharging strategy.

[0160] The peak load module 103 plays a central control role in this power peak load system based on the Internet of Things. Its core function is to intelligently formulate and implement peak load strategies based on the data received from the prediction module 101 and the Internet of Things module 102 to optimize the overall operation efficiency and stability of the power grid.

[0161] First, the peak load module 103 receives data from the prediction module 101, which includes the renewable energy generation capacity and grid load demand predicted based on the machine learning algorithm. This prediction takes into account a variety of factors, such as historical data, meteorological conditions, seasonal changes, etc., thus providing forward-looking insights into future grid demand and supply.

[0162] At the same time, the peak load module also receives real-time grid operation data from the Internet of Things module 102. This includes key parameters such as the energy flow direction, frequency, and voltage of the grid, which are essential for real-time monitoring of the operation status of the grid.

[0163] With these two pieces of information, the peak load regulation module 103 uses a mixed integer linear programming (MILP) model to determine the peak load regulation strategy. MILP is an efficient mathematical optimization method that can handle large-scale and complex optimization problems. Here, it is used to analyze the current status and future forecasts of the power grid to develop the optimal power distribution and load regulation plan.

[0164] Based on these strategies, the peaking module will adjust the power output of the grid. This may involve increasing or decreasing the power generation of specific power plants, adjusting the output of renewable energy facilities (such as wind and solar) connected to the grid, and other similar adjustments. These adjustments are designed to ensure that the grid can operate at maximum efficiency at any given time, while meeting power demand and reducing energy waste.

[0165] In addition, the peak-shaving module is responsible for developing charging and discharging strategies for energy storage modules. This includes the management of energy storage solutions such as battery energy storage systems and pumped storage. Based on the demand and supply of the grid, the peak-shaving module instructs these energy storage facilities when to charge and discharge to further balance the load of the grid and improve the efficient use of energy.

[0166] Finally, the peak load module also sends charging and discharging instructions to the energy storage module. These instructions are based on previously formulated strategies to ensure that the energy storage facilities can provide or store electricity at the best time to support the load and supply balance of the grid.

[0167] Furthermore, the objective function Objective of the mixed integer linear programming model adopts the following formula 1:

[0168] Objective=w1·C total +w2·U renew -w3·E carbon +w4·R system Formula 1

[0169] Among them, C total is the economic cost function; U renew is the renewable energy utilization function; E carbon is the carbon emission function; R system is the system reliability function; w1, w2, w3, w4 are weight coefficients, which can be obtained through expert knowledge or statistically obtained through historical data; minimizing this comprehensive objective function helps achieve multiple goals of the power system, including cost efficiency, environmental sustainability, and system stability. This approach allows grid operators to optimize operating costs and reduce environmental impact while ensuring power supply reliability.

[0170] Economic cost function C total Use the following formula 2:

[0171] ∑ t (C gen (t)·P gen (t)-C sell (t)·F grid (t)+C store (S bat (t),S pump (t))) Formula 2

[0172] Among them, C gen (t) is the cost of electricity generation at time t; this may include fuel costs, operation and maintenance costs, etc. For renewable energy, this cost may be low or close to zero.

[0173] C sell (t) is the electricity selling price at time t, which may change according to market conditions and changes in electricity demand.

[0174] P gen (t) is the predicted power generation at time t, which includes the power generation of all energy sources, such as fossil fuels, wind power, solar power, etc., which can be obtained through real-time monitoring or prediction models;

[0175] F grid (t) is the energy flowing from the grid to the storage facility at time t, which is a positive number; it can also represent the energy flowing from the storage facility to the grid at time t, which is a negative number.

[0176] In the power system, F grid The (t) parameter indicates the energy flow between the grid and the energy storage facility at a specific time point t. The positive and negative values ​​of this parameter reflect the direction of energy flow, and the specific meanings are as follows:

[0177] The energy F flowing from the grid to the energy storage facility at time t is grid (t) Description:

[0178] 1. When F grid When (t) is a positive number:

[0179] This means that at time t, energy is flowing from the grid to the energy storage facility. In other words, the grid is delivering energy to the energy storage facility (such as a battery storage system or pumped storage system).

[0180] This typically occurs when grid load is low and generation is high, for example when renewable energy sources such as solar or wind are producing more power than the grid currently needs.

[0181] The process involves storing excess electricity for use during times of peak demand, helping to balance grid loads and improve energy efficiency.

[0182] 2. When F grid When (t) is negative:

[0183] This means that energy is flowing from the storage facility to the grid. In this case, the storage facility is releasing energy back to the grid.

[0184] This typically occurs during times of high load on the grid, such as during peak electricity demand or when renewable energy generation is insufficient to meet immediate grid demand.

[0185] By releasing stored energy, energy storage facilities help maintain grid stability and meet the immediate needs of the grid.

[0186] F grid (t) It can be measured through a smart metering system. Specifically, smart meters or other advanced measuring equipment can be installed between the power grid and the energy storage facilities to monitor and record the direction and amount of energy flow in real time.

[0187] Power operators can also use data analysis tools to predict and plan F grid (t) changes to ensure optimal energy flow between the grid and energy storage facilities.

[0188] F grid (t) is a key parameter in power system management. It not only helps grid operators effectively manage power supply and demand, but also is an important tool for optimizing the operation of energy storage facilities and improving the overall efficiency of the grid. grid (t), can make better use of renewable energy, reduce dependence on traditional energy, and improve the reliability and stability of the power grid.

[0189] S bat (t) and S pump (t) represent the status of battery energy storage and pumped storage at time t, respectively. These status information may include current energy storage level, historical usage pattern, etc.

[0190] S bat (t) represents the energy stored in the battery energy storage system at time t. This can include the battery’s current charge level in kilowatt-hours and may also include the battery’s state of health and available capacity.

[0191] Among them, the current charge level of the battery can be monitored in real time by the battery management system (BMS). The BMS can accurately record the charge and discharge amount of the battery to determine the current energy storage level.

[0192] The battery health status and available capacity can be evaluated by analyzing the battery's historical usage data and performing regular performance tests.

[0193] S pump (t) refers to the amount of water stored in the pumped storage system at time t, usually measured as the amount of water in the reservoir, which is directly related to the energy storage capacity of the system.

[0194] The water level sensor is used to monitor the reservoir water level of the pumped storage system in real time. The water level can be used to obtain the amount of water in the reservoir. The water level reflects the energy storage status of the system.

[0195] Analyzing historical operational data, such as records of pumping and generation activities, can help estimate the system’s average energy storage level and operating efficiency.

[0196] The status of battery energy storage and pumped storage systems reflects the energy storage status of the energy storage facilities at a specific point in time. This status information is crucial for calculating energy storage costs because it affects the operating efficiency and cost of energy storage facilities. Through real-time monitoring, historical data analysis and performance evaluation, these states can be accurately determined to effectively manage energy storage resources, optimize grid operating costs and improve the overall efficiency of the grid.

[0197] Cstore is the storage cost function, which can be calculated using the following formula:

[0198] C store (S bat (t),S pump (t)) = α·ΔS bat (t)+β·S bat (t)+γ·ΔS pump (t)+δ·S pump (t)

[0199] Among them, α is the cost coefficient of unit energy loss; β is the fixed cost coefficient of battery storage; ΔS bat (t) is the charge and discharge amount of the battery at time t; in this function, S bat (t) is the current battery storage level; γ is the energy loss cost coefficient per unit of water conversion; δ is the fixed cost coefficient of pumped storage; ΔS pump (t) is the amount of water charged and discharged by pumped storage at time t; in this function, S pump (t) is the current pumped storage level.

[0200] Renewable energy utilization function U renew Using the following formula 3:

[0201]

[0202] Among them, P renew (t) represents the renewable energy power generation at time t;

[0203] Carbon emission function E carbon Using the following formula 4:

[0204] E carbon =∑ t E factor (P gen (t)-P renew (t)) Formula 4

[0205] Among them, E factor is the carbon emission factor, which represents the carbon emission per unit of non-renewable energy generation;

[0206] System reliability function R system Using the following formula 5:

[0207] R system =∑ t (w5·R storage (t)+w6·R match (t)+w7·R renewable (t)) Formula 5

[0208] Among them, R storage (t) is the energy storage responsiveness score, which can be evaluated based on the difference between the remaining capacity of the energy storage facility and the predicted load; specifically, if the energy storage facility has sufficient capacity to meet the demand during peak hours, this score will be higher.

[0209] R match (t) is the load matching score, which can be evaluated by comparing the difference between dispatchable generation capacity and forecasted load; this item measures the degree of match between dispatchable generation resources (such as traditional thermal power generation and dispatchable renewable energy) and the forecasted load of the grid. The score calculation can be determined based on the consistency between the output capacity of dispatchable generation resources and the forecasted load. If the generation capacity can match the load demand well, the score will be high.

[0210] ·R renewable (t) is the renewable energy utilization efficiency score, which can be evaluated based on the ratio of renewable energy output to total electricity generation; this item reflects the utilization efficiency of renewable energy (such as wind power, solar power) at a specific time t.

[0211] w5, w6, w7 are weight coefficients, which can be obtained from empirical data.

[0212] Furthermore, the constraints of the mixed integer linear programming model include an energy balance constraint, and the energy balance constraint is expressed using the following formula:

[0213] P gen (t) = P dem (t)+F grid (t)

[0214] Among them, P dem (t) is the grid demand at time t, which refers to the sum of electricity demand from all consumers (households, industry, commerce, etc.) at that point in time.

[0215] Furthermore, the constraints of the mixed integer linear programming model include energy storage device constraints, and the energy storage device constraints are expressed using the following formula:

[0216] S min ≤S bat (t)

[0217] S pump (t)≤S max

[0218] Among them, S min and S maxThey refer to the minimum and maximum storage capacity limits of energy storage facilities in operation. These values ​​are usually provided by the manufacturer of the energy storage equipment or determined according to the design specifications and safety standards of the equipment.

[0219] S bat S (t) represents the actual storage capacity of the battery energy storage system at a specific time t. pump (t) represents the actual water storage (or energy storage) of the pumped storage system at a specific time t.

[0220] Furthermore, the constraints of the mixed integer linear programming model include power generation resource allocation optimization constraints, and the power generation resource allocation optimization constraints are expressed using the following formula:

[0221] ΔP gen (t)≤Rate change ·P gen (t-1)

[0222] Where ΔP gen (t) is the change in power generation between time t and t-1; Rate change The maximum rate of change allowed in power generation, usually set based on the technical characteristics of the grid and safety standards. This may be determined by the grid operator, planners or relevant industry standards.

[0223] Furthermore, the constraints of the mixed integer linear programming model include an environmental impact minimization constraint, and the environmental impact minimization constraint is expressed using the following formula:

[0224] E impact (t) = EnvImpactFactor·(P gen (t)-P renew (t))

[0225] Among them, E impact (t) is the environmental impact at time t; EnvImpactFactor is the environmental impact factor, which is used to quantify the environmental impact of each unit of non-renewable energy generation. This factor can be obtained through experimental data. The environmental impact factor is a quantitative indicator that indicates the degree of environmental impact generated per unit of electricity when generating electricity from non-renewable energy sources (such as coal, natural gas, and oil). This may include emissions such as carbon emissions, sulfides, nitrogen oxides, and particulate matter.

[0226] Furthermore, the objective function of the mixed integer linear programming model further includes a network loss minimization factor C loss , which is expressed using the following formula:

[0227] C loss =∑ tLossFactor·(P gen (t)-P dem (t)) 2

[0228] Among them, LossFactor is the network loss factor, which is used to quantify the energy loss in the power grid. The determination of LossFactor needs to take into account the physical characteristics of the power grid, such as the length of the transmission line, the material of the line (such as copper, aluminum), the cross-sectional area of ​​the line, etc. These characteristics affect the resistivity, which in turn affects the loss of energy during transmission. In addition, the power industry usually has a series of standards and models for calculating and predicting power grid losses. These models may be based on detailed power grid simulations and long-term industry research.

[0229] The network loss factor is a parameter used to measure the efficiency of power transmission in the power system. It takes into account the loss of energy in the power grid due to factors such as resistance during transmission.

[0230] Factors influencing the network loss factor include:

[0231] Length of transmission line: The longer the transmission distance, the greater the losses generally.

[0232] Line material: Different materials have different resistivities. For example, copper and aluminum are common transmission line materials, and they have different resistivities.

[0233] Line cross-sectional area: Lines with larger cross-sectional areas generally have lower resistance, resulting in reduced losses.

[0234] The design and maintenance of the grid also affects losses.

[0235] Calculation method of network loss factor:

[0236] Based on physical models: LossFactor can be calculated using transmission line models from electrical engineering. These models estimate energy losses based on the physical properties of the line (e.g., length, cross-sectional area, material).

[0237] Use standards and models: The power industry usually has standardized models and methods to estimate the average losses of different types of power grids. These models are often based on a large amount of power grid data and long-term studies.

[0238] Experimental data: A more accurate LossFactor can be obtained by actually measuring the loss of a specific power grid or line.

[0239] At this time, the objective function of the mixed integer linear programming model can be expressed as:

[0240] Objective=w1·C total +w2·Urenew -w3·E carbon +w4·R system +w8·C loss

[0241] Among them, w8 is the weight coefficient, which can be obtained through experimental data.

[0242] In summary, the peak load regulation module 103 is the intelligent brain of the power peak load regulation system. It comprehensively utilizes forecast data and real-time monitoring data, applies advanced mathematical models, and intelligently adjusts the operation of the power grid to ensure efficient use of energy and the stability of the power grid.

[0243] The energy storage module 104 includes a battery energy storage submodule and a pumped energy storage submodule; it is used to adjust the charge and discharge according to the charge and discharge instructions sent by the peak load regulation module to support the load and supply balance of the power grid.

[0244] The energy storage module 104 plays a vital role in this power peak load regulation system based on the Internet of Things. It consists of two key submodules: a battery energy storage submodule and a pumped storage submodule. These submodules not only serve as a buffer for power supply, but also play a key role in regulating and balancing the load of the power grid.

[0245] The battery energy storage submodule mainly uses large-capacity batteries to store electrical energy. This energy storage method is very suitable for storing and releasing electricity in the short term, especially when it is necessary to respond quickly to changes in grid load. For example, when solar or wind power generation suddenly increases, excess electricity can be stored in these batteries. Conversely, when the grid load increases and renewable energy generation is not enough to meet demand, these batteries can be used to quickly supply power to help maintain the stability of the grid.

[0246] The pumped hydro submodule uses a different approach to energy storage, using electricity to pump water uphill during low-load periods and then releasing the water flow to generate electricity during high-load periods. This type of energy storage is well suited for long-term energy storage because it can provide large amounts of power when needed. This approach is particularly useful for responding to long-term grid load fluctuations or providing additional power during peak hours.

[0247] The operation of the energy storage module 104 is directly guided and controlled by the peak shaving module 103. Based on the forecast data and real-time grid operation data analyzed by the peak shaving module, and the peak shaving strategy developed using the mixed integer linear programming model, the energy storage module receives charging and discharging instructions. These instructions clearly indicate when to charge and when to release the stored energy, so as to support the balance of grid load and supply.

[0248] For example, when the forecast module indicates that solar power generation will increase significantly in the next few hours, the peak-shaving module may instruct the battery storage submodule to charge to store excess power. Similarly, if it is predicted that the grid will face high loads during a specific period of time, the peak-shaving module may instruct the pumped hydro storage submodule to release energy to help meet the increased power demand.

[0249] In general, the energy storage module 104 is an indispensable part of the power peak load regulation system. It ensures the stable operation of the power grid under various conditions through flexible charging and discharging operations, especially in the face of the instability of renewable energy and daily load fluctuations. In this way, the energy storage module not only improves the reliability of the power grid, but also optimizes the efficiency of energy use, making the operation of the entire power grid more economical and environmentally friendly.

[0250] Although the present application is disclosed as above in the form of a preferred embodiment, it is not intended to limit the present application. Any technical personnel in this field may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A power peak load regulation system based on the Internet of Things, characterized in that: include: The prediction module is used to predict the renewable energy generation capacity and grid load demand using machine learning algorithms, and pass the prediction results to the peak load regulation module; The Internet of Things module includes sensors and intelligent metering devices installed at the power grid measurement nodes, which are used to monitor the power grid operation data in real time; and transmit the monitored power grid operation data to the peak load module; wherein the power grid operation data includes energy flow direction, frequency, and voltage; A peak-shaving module is used to receive the prediction results provided by the prediction module and the power grid operation data provided by the Internet of Things module; determine the peak-shaving strategy using a mixed integer linear programming model according to the prediction results and the power grid operation data; adjust the power output of the power grid and formulate a charging and discharging strategy for the energy storage module according to the peak-shaving strategy; and send charging and discharging instructions to the energy storage module according to the charging and discharging strategy; An energy storage module, including a battery energy storage submodule and a pumped energy storage submodule; used to adjust charging and discharging according to the charging and discharging instructions sent by the peak load regulation module to support the load and supply balance of the power grid; The objective function Objective of the mixed integer linear programming model adopts the following formula 1: Objective=w1·C total +w2·U renew -w3·E carbon +w4·R system Formula 1 Among them, C total is the economic cost function; U renew is the renewable energy utilization rate function; E carbon is the carbon emission function; R system is the system reliability function; w1, w2, w3, w4 are weight coefficients; Economic cost function C total Use the following formula 2: Σ t (C gen (t)·P gen (t)-C sell (t)·F grid (t)+C store (S bat (t), S pump (t))) Formula 2 Among them, C gen (t) is the electricity generation cost at time t; C sell (t) is the electricity price at time t; C store is the storage cost function; P gen (t) is the predicted total power generation at time t; F grid (t) is the energy flowing from the grid to the storage facility at time t; S bat (t) and S pump (t) represent the states of battery energy storage and pumped storage at time t, respectively; Renewable energy utilization function U renew Using the following formula 3: Among them, P renew (t) represents the renewable energy power generation at time t; Carbon emission function E carbon Using the following formula 4: E carbon = ∑ t E factor · (P gen (t) - P renew (t)) Formula 4 Among them, E factor is the carbon emission factor, which represents the carbon emission per unit of non-renewable energy electricity generation; System reliability function R system Using the following formula 5: R system = Σ t (w5·R storage (t) + w6·R match (t) + w7·R renewable (t)) Equation 5 Among them, R storage (t) is the energy storage responsiveness score, which is evaluated based on the difference between the remaining capacity of the energy storage facility and the predicted load; R match (t) is the load matching score, which is evaluated by comparing the difference between the dispatchable generation capacity and the forecast load; R renewable (t) is the renewable energy utilization efficiency score, which is evaluated based on the ratio of renewable energy output to total power generation; w5, w6, and w7 are weight coefficients.

2. The power peak load regulation system according to claim 1, characterized in that: The prediction module is specifically used to use a pre-trained multimodal deep learning network to predict the renewable energy generation capacity and grid load demand; wherein the multimodal deep learning network includes a convolutional neural network part, a long short-term memory network part and a data fusion and output layer; the convolutional neural network part receives satellite images, extracts spatial features using the convolutional neural network, and obtains the spatial feature representation of the satellite image for predicting solar power generation; the long short-term memory network part receives historical grid load data, meteorological data, holiday information, industrial activity data, and user behavior patterns, and obtains the time series feature representation of the input data using the long short-term memory network; the data fusion and output layer fuses the spatial feature representation and the time series feature representation, and outputs the predicted value of the renewable energy generation capacity and grid load demand.

3. The power peak load regulation system according to claim 2, characterized in that: The data fusion and output layer includes an attention mechanism part for enhancing the prediction performance of the multimodal deep learning network; wherein the attention mechanism part is specifically used for: Receiving the spatial feature representation and the time series feature representation; Using a fully connected network, determining a first attention score of the spatial feature representation and a second attention score of the time series feature representation based on the spatial feature representation and the time series feature representation; Normalizing the first attention score and the second attention score through a softmax layer to ensure that the sum of the normalized first attention score and the normalized second attention score is 1; The spatial feature representation and the time series feature representation are weighted using the standardized first attention score and the standardized second attention score.

4. The power peak load regulation system according to claim 3, characterized in that: The data fusion and output layer includes a dynamic weight adjustment mechanism, which is specifically used for: Encoding state information of the power grid to obtain a state code, wherein the state information includes season information or special event information; Using the added input layer in the multimodal deep learning network, the receiving state encoding; Calculate the weight of feature fusion based on the received state code; According to the calculated feature fusion weight, the spatial feature representation and the time series feature representation are fused.

5. The power peak load regulation system according to claim 1, characterized in that: The constraints of the mixed integer linear programming model include an energy balance constraint, which is expressed using the following formula: P gen (t)=P dem (t)+F grid (t) Among them, P dem (t) is the grid demand at time t.

6. The power peak load regulation system according to claim 1, characterized in that: The constraints of the mixed integer linear programming model include energy storage device constraints, which are expressed using the following formula: S min ≤S bat (t) S pump (t)≤S max Among them, S min and S max They respectively refer to the minimum and maximum storage capacity limits of energy storage facilities in operation.

7. The power peak load regulation system according to claim 1, characterized in that: The constraints of the mixed integer linear programming model include power generation resource allocation optimization constraints, which are expressed using the following formula: ΔP gen (t)≤Rate change ·P gen (t-1) Among them, ΔP gen (t) is the change in power generation between time t and t-1; Rate change is the maximum rate of change allowed in power generation.

8. The power peak load regulation system according to claim 1, characterized in that: The constraints of the mixed integer linear programming model include an environmental impact minimization constraint, which is expressed by the following formula: E impact (t)=EnvImpactFactor·(P gen (t)-P renew (t)) Among them, E impact (t) is the environmental impact at time t; EnvImpactFactor is the environmental impact factor used to quantify the environmental impact of each unit of non-renewable energy generation.

9. The power peak load regulation system according to claim 5, characterized in that: The objective function of the mixed integer linear programming model further includes a network loss minimization factor C loss , which is expressed using the following formula: C loss =∑ t LossFactor·(P gen (t)-P dem (t)) 2 Among them, LossFactor is the network loss factor, which is used to quantify the energy loss in the power grid.

Images