Microgrid intelligent scheduling method and system based on neural network

Abstract

The invention discloses a micro-grid intelligent scheduling method and system based on a neural network, and the method comprises the steps: obtaining micro-grid load data, micro-grid energy storage data, micro-grid power generation data and electricity price data, inputting the micro-grid load data, the micro-grid energy storage data and the micro-grid power generation data into a pre-trained analysis neural network to obtain electric quantity profit and loss data, and inputting the electric quantity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data, and performing intelligent scheduling on the micro-grid according to the electric quantity profit and loss data and the decision data. According to the method, firstly, the electric quantity profit and loss data of the microgrid are extracted by analyzing the neural network, and dynamic changes of electric quantity supply and demand in the microgrid are effectively coped with; secondly, decision data are obtained through a prediction neural network, and efficient interaction between the micro-grid and an external power grid is achieved; and finally, intelligent scheduling is performed according to the data, so that the intelligent scheduling level of the micro-grid is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of neural networks, and in particular to a microgrid intelligent scheduling method and system based on neural networks. Background Art

[0002] With the acceleration of global energy transformation and the widespread adoption of distributed energy, the importance of intelligent scheduling for microgrids, a key component of modern power systems, has become increasingly prominent. Microgrids integrate multiple energy sources to achieve energy self-sufficiency and efficient utilization within a local area.

[0003] The operation of microgrids faces a number of challenges. First, the balance of power supply and demand within a microgrid is difficult to accurately predict. The intermittent and uncertain nature of renewable energy generation, coupled with frequent fluctuations in load demand, makes it difficult for traditional scheduling methods to effectively address these dynamic changes. Second, the demand for power exchange between microgrids and external power grids is becoming increasingly frequent and volatile. Traditional scheduling methods, with their static and fixed decision-making models, are unable to adapt to this rapidly changing dynamic rhythm, which in turn affects the economic operation and energy efficiency of microgrids. Summary of the Invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a microgrid intelligent scheduling method and system based on neural network. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] A microgrid intelligent scheduling method based on a neural network, comprising:

[0006] Obtaining microgrid load data, microgrid energy storage data, microgrid power generation data, and electricity price data; inputting the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analysis neural network to obtain power profit and loss data;

[0007] Inputting the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data;

[0008] The microgrid is intelligently dispatched according to the electricity profit and loss data and the decision data.

[0009] In a specific embodiment, the analysis neural network includes: an analysis input layer, a dense network layer, a probability analysis layer and an analysis output layer, wherein the input data of the analysis input layer includes the microgrid load data, the microgrid energy storage data and the microgrid power generation data; the dense network includes an initial convolution layer, a dense block layer and a transition layer in sequence; the probability analysis layer includes a probability residual convolution layer, a probability pooling activation layer and a probability full connection layer in sequence; the output data of the analysis output layer is power profit and loss data, wherein the power profit and loss data includes microgrid redundant power data, microgrid sustainable power supply data, microgrid power supply balance point data and microgrid income data.

[0010] In one embodiment, a method for analyzing neural network training includes:

[0011] Acquire data to be trained, wherein the data to be trained includes the microgrid load data to be trained, the microgrid energy storage data to be trained, the microgrid power generation data to be trained, and the power profit and loss data to be trained;

[0012] Performing data cleaning on the data to be trained to obtain cleaned data;

[0013] The cleaned data is input into the analysis neural network to be trained to obtain the analysis neural network, wherein the loss function of the analysis neural network is a first fusion loss function based on multi-weight KL divergence and cross entropy loss.

[0014] In a specific embodiment, inputting the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data includes:

[0015] Inputting the electricity price data into an external prediction sub-neural network to obtain first decision data;

[0016] Inputting the electricity profit and loss data into the internal prediction sub-neural network to obtain second decision data;

[0017] The electricity price data, the electricity profit and loss data, the first decision data and the second decision data are input into a weight fusion sub-neural network to obtain decision data.

[0018] In a specific embodiment, the external prediction sub-neural network includes: an external input layer, a feature dimensionality reduction layer, a feature extraction layer and an external output layer, wherein the input data of the external input layer is the electricity price data; the feature dimensionality reduction layer includes multiple convolutional layers and pooling layers connected in series; the feature extraction layer includes multiple fully connected layers; and the output data of the external output layer is the first decision data.

[0019] In a specific embodiment, the internal prediction sub-neural network includes: an internal input layer, a residual feature extraction layer and an internal output layer, wherein the input data of the internal input layer includes power surplus and deficit data; the residual feature extraction layer includes a convolution layer, a normalization layer, an activation layer and a jump connection layer in sequence; the output data of the internal output layer is the second decision data.

[0020] In a specific embodiment, the weight fusion sub-neural network includes: a weight input layer, an embedding layer, a weight link activation layer, an attention layer and a weight output layer, wherein the input data of the weight input layer includes the electricity price data, the electricity profit and loss data, the first decision data and the second decision data; the embedding layer includes a data splicing layer and a data alignment layer in sequence; the weight link activation layer includes multiple full-link layers and an activation layer; the attention layer includes a weight feature extraction layer and a weight feature cleaning layer in sequence; the output data of the weight output layer is decision data.

[0021] In a specific embodiment, the intelligent scheduling of the microgrid according to the power profit and loss data and the decision data includes:

[0022] Obtaining an electricity transaction strategy between the microgrid and the external power grid based on the electricity profit and loss data and the decision data;

[0023] Determine the electricity purchase and sale decision, the quantity of electricity purchased and sold, and the time of electricity purchase and sale according to the electricity transaction strategy;

[0024] The microgrid is externally intelligently dispatched according to the power purchase and sale decision, the power purchase and sale quantity, and the power purchase and sale time.

[0025] In a specific embodiment, the intelligent scheduling of the microgrid according to the power surplus and deficit data and the decision data includes:

[0026] Obtaining the power distribution ratio of the load units within the microgrid according to the power surplus and deficit data and the decision data;

[0027] Obtaining a power distribution ratio of power generation units within the microgrid according to the power surplus and deficit data and the decision data;

[0028] Obtaining a storage power distribution ratio of an energy storage unit within a microgrid according to the power surplus and deficit data and the decision data;

[0029] The microgrid is internally intelligently dispatched according to the power consumption allocation ratio, the power generation allocation ratio, and the power storage allocation ratio.

[0030] In one embodiment, a neural network-based microgrid intelligent dispatching system includes:

[0031] The acquisition unit is used to obtain microgrid load data, microgrid energy storage data, microgrid power generation data and electricity price data;

[0032] An analysis unit, configured to input the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analysis neural network to obtain power profit and loss data;

[0033] A prediction unit, configured to input the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data;

[0034] A dispatching unit is used to intelligently dispatch the microgrid according to the power profit and loss data and the decision data.

[0035] Beneficial effects of the present invention:

[0036] The present invention provides a neural network-based microgrid intelligent scheduling method and system. First, by analyzing the neural network, the method accurately extracts the microgrid's electricity profit and loss data, effectively responding to the dynamic changes in electricity supply and demand within the microgrid. Second, it obtains decision data through predictive neural networks to achieve efficient interaction between the microgrid and the external power grid. Finally, intelligent scheduling is performed based on this data, significantly improving the intelligent scheduling level of the microgrid.

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a microgrid intelligent scheduling method based on a neural network provided by an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of an analysis neural network of a neural network-based microgrid intelligent scheduling method provided by an embodiment of the present invention;

[0040] Figure 3 Schematic diagram of a prediction neural network of a neural network-based microgrid intelligent scheduling method provided by an embodiment of the present invention;

[0041] Figure 4 This is a module block diagram of a neural network-based microgrid intelligent dispatching system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described in detail below with reference to specific examples, but the embodiments of the present invention are not limited thereto.

[0043] Example 1

[0044] In one embodiment, see Figure 1 , Figure 1 This is a flow chart of a microgrid intelligent dispatching method based on neural network. The specific steps are as follows:

[0045] S1. Obtain microgrid load data, microgrid energy storage data, microgrid power generation data, and electricity price data:

[0046] The microgrid load data may include, for example, active power, reactive power, voltage, and current. As a preferred embodiment, the microgrid load data may also include historical data and peak and trough data of load points.

[0047] The microgrid energy storage data may include, for example, the battery's state of charge (SOC), state of health (SOH), charge and discharge current, and voltage. Preferably, the data may also include, for example, the battery's maintenance cycle.

[0048] The microgrid power generation data may include, for example, the number, power, voltage, and current of power generation equipment. As a preferred embodiment, the data may also include, for example, time data and geographic data of the power generation equipment.

[0049] The electricity price data may be, for example, real-time electricity prices, time-of-use electricity prices, and peak-valley electricity prices in the external market. As a preferred embodiment, the electricity price trend data and electricity price fluctuation data may also be included.

[0050] S2. Input the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analytical neural network to obtain power profit and loss data; this process not only improves the accuracy of power profit and loss analysis, but also provides a solid data foundation for subsequent intelligent scheduling, ensuring that the microgrid can maintain a balance between power supply and demand under different operating conditions.

[0051] Specifically, see Figure 2 , Figure 2 The schematic diagram of the analysis neural network of the microgrid intelligent dispatching method based on neural network is shown, wherein the analysis neural network includes: an analysis input layer, a dense network layer, a probability analysis layer and an analysis output layer.

[0052] The input data of the analysis input layer includes the microgrid load data, the microgrid energy storage data and the microgrid power generation data, specifically:

[0053] Input data includes microgrid load data, such as active power, reactive power, voltage, and current; microgrid energy storage data, such as battery state of charge (SOC), state of health (SOH), charge and discharge current, voltage, and temperature; and microgrid power generation data, such as the output power of renewable energy power generation units, environmental parameters such as light intensity and wind speed, and the operating status of traditional power generation units.

[0054] The dense network includes an initial convolutional layer, a dense block layer and a transition layer in sequence.

[0055] Specifically, the initial convolutional layer uses multiple convolution kernels to convolve the input data and extract local features. For example, a 3×3 convolution kernel is used with a stride of 1 and a padding of “same” to keep the input and output feature maps of the same size.

[0056] A dense block layer consists of multiple densely connected convolutional layers, each followed by a batch normalization layer and a ReLU activation layer. For example, take two convolutional layers, the first convolutional layer has 32 3×3 convolution kernels, and the second convolutional layer has 64 3×3 convolution kernels.

[0057] Transition layers are used to reduce the dimensionality of feature maps and lower computational complexity. They consist of convolutional layers and pooling layers. For example, a convolutional layer using a 1×1 convolution kernel reduces the number of channels in a feature map by half, followed by a 2×2 max pooling layer with a stride of 2, thereby reducing the feature map resolution by 75%.

[0058] The probability analysis layer includes a probability residual convolution layer, a probability pooling activation layer and a probability full connection layer in sequence.

[0059] Specifically, the probabilistic residual convolution layer combines residual connections and convolution operations to enable the network to learn the probabilistic distribution of features. For example, the convolution layer uses a 3×3 convolution kernel, a stride of 1, and a "same" padding, followed by a batch normalization layer and a ReLU activation layer.

[0060] The probabilistic pooling activation layer uses pooling to reduce the feature dimensionality while also introducing nonlinearity by applying an activation function. For example, a 2×2 max pooling layer with a stride of 2 is used, followed by a ReLU activation layer.

[0061] The probabilistic fully connected layer maps the feature vector to a probability space related to power surplus or deficit. For example, the fully connected layer has 128 nodes and is followed by a ReLU activation layer.

[0062] The output data of the analysis output layer is power profit and loss data, wherein the power profit and loss data includes microgrid redundant power data, microgrid sustainable power supply data, microgrid power supply balance point data and microgrid income data.

[0063] Specifically, the output layer has 4 nodes, each node corresponds to a specific power profit and loss indicator. The activation function is selected according to the specific task. For example, the linear activation function is used for redundant power data and sustainable power supply data, the Sigmoid activation function is used for power supply balance point data, and the linear activation function is used for income data.

[0064] Based on the above analysis of the neural network architecture, the training process of the neural network is analyzed, including:

[0065] Acquire data to be trained, wherein the data to be trained includes the microgrid load data to be trained, the microgrid energy storage data to be trained, the microgrid power generation data to be trained, and the power profit and loss data to be trained.

[0066] The data to be trained is cleaned to obtain cleaned data. Specifically, the data cleaning includes:

[0067] Filtering: Use low-pass, high-pass, or band-pass filtering to remove high-frequency noise or low-frequency drift from the data, preserving the valid features of the data. For example, high-frequency noise in load and power generation data can be filtered to smooth the data curve.

[0068] Wavelet transform: Use wavelet transform to perform multi-resolution analysis on data, decompose the data and remove noise components at different scales, and then reconstruct the denoised data.

[0069] The cleaned data is input into the analysis neural network to be trained to obtain the analysis neural network, wherein the loss function of the analysis neural network is a first fusion loss function based on multi-weight KL divergence and cross entropy loss, and the first fusion loss function is:

[0070] L FUS =a*L KL +b*L CE ,

[0071] Among them, LKL and LCE are multi-weight KL divergence and cross entropy loss indicators, respectively, and a and b are the fusion coefficients of multi-weight KL divergence and cross entropy loss, respectively.

[0072] Analyzing the neural network training process includes:

[0073] Forward propagation: The cleaned data is input into the initialized analysis neural network, and passes through the analysis input layer, dense network layer, probability analysis layer and analysis output layer in sequence to obtain the network's predicted output.

[0074] Loss calculation: Use the defined first fusion loss function to calculate the loss value between the predicted output and the true label to evaluate the prediction performance of the network.

[0075] Backpropagation and optimization: Based on the calculated loss value, the gradients of each layer of the network are calculated using the backpropagation algorithm. The network parameters are then updated using an optimization algorithm (such as stochastic gradient descent or Adam) to minimize the loss function. During training, these steps are iterated until the network converges, meaning the loss value stabilizes within a certain range or the preset number of training rounds is reached. This results in a trained analytical neural network.

[0076] S3. Input the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data. Figure 3 , Figure 3 It is a schematic diagram of a predictive neural network for a microgrid intelligent scheduling method based on a neural network.

[0077] S31. Inputting the electricity price data into an external prediction sub-neural network to obtain first decision data;

[0078] The external prediction sub-neural network includes: an external input layer, a feature dimension reduction layer, a feature extraction layer and an external output layer, wherein:

[0079] The input data of the external input layer is electricity price data;

[0080] Specifically, the feature dimensionality reduction layer first extracts local features of the electricity price data using a first set of 16 convolutional layers with 3×3 kernels, a stride of 1, and a "same" padding to preserve as much original information as possible. This is followed by a 2×2 max pooling layer with a stride of 2 to reduce the dimensionality of the features. A second set of 8 convolutional layers with a 5×5 kernel and a stride of 3 is then used to extract more abstract features. This is followed by a 4×4 max pooling layer with a stride of 2 to further reduce the feature dimensionality. Finally, a third set of convolutional layers uses a 7×7 kernel, 4 kernels, and a stride of 5 to further extract high-level features. This is followed by an 8×8 max pooling layer with a stride of 4 to reduce the feature dimensionality to a dimension suitable for subsequent processing.

[0081] The feature extraction layer consists of three groups of fully connected layers. The first group of fully connected layers contains 64 nodes and is followed by a ReLU activation layer, which introduces nonlinearity and enables the network to learn complex feature relationships. The second group of fully connected layers has 32 nodes and is followed by a tanh activation layer to further enhance the expressiveness of features. The third group of fully connected layers has 16 nodes and is followed by a softmax activation layer, which maps features to the probability distribution of different trading decision categories.

[0082] The external output layer includes two output nodes: one indicating whether to buy or sell (0 or 1), and the other indicating the buy or sell quantity (using a linear activation function). This provides clear decision-making recommendations for the microgrid's external electricity transactions. For example, if the model determines that the current electricity price is high and the microgrid has a surplus of electricity, the output node may indicate a buy or sell decision of 1 (sell) and the corresponding amount of electricity sold.

[0083] S32. Inputting the power profit and loss data into the internal prediction sub-neural network to obtain second decision data;

[0084] The internal prediction sub-neural network includes: an internal input layer, a residual feature extraction layer and an internal output layer, wherein,

[0085] The internal input layer receives power profit and loss data, including key indicators such as microgrid redundant power, sustainable power supply time, power supply balance point, and microgrid revenue.

[0086] The residual feature extraction layer uses 32 3×3 convolution kernels with a stride of 1 and "same" padding to extract local features of the power surplus and deficit data. This is followed by a batch normalization layer to standardize the data and accelerate network convergence. A Reluctant Unit (ReLU) activation layer introduces nonlinearity, enabling the model to learn complex feature relationships in the data. Finally, skip connections are used to directly add the input to the output, effectively alleviating the vanishing gradient problem in deep networks and improving model training efficiency and performance.

[0087] The output data of the internal output layer is the second decision data.

[0088] S33. Input the electricity price data, the electricity profit and loss data, the first decision data and the second decision data into a weight fusion sub-neural network to obtain decision data.

[0089] The weight fusion sub-neural network includes: a weight input layer, an embedding layer, a weight link activation layer, an attention layer and a weight output layer, wherein,

[0090] The weight input layer is used to receive multi-source information such as electricity price data, electricity profit and loss data, first decision data and second decision data.

[0091] The embedding layer consists of a data concatenation layer and a data alignment layer. The data concatenation layer is used to concatenate all input data into a complete vector, integrating information from different sources. The data alignment layer ensures that all data dimensions are consistent, providing a unified data format for subsequent processing.

[0092] The weighted link activation layer consists of two fully connected layers. The first fully connected layer has 256 nodes and is followed by a ReLU activation layer. This layer performs feature transformation and nonlinear mapping on the input data to extract fused features. The second fully connected layer has 128 nodes and continues to use a ReLU activation layer to further optimize feature representation, highlight important information, and provide a more accurate basis for the final decision.

[0093] The attention layer consists of a weighted feature extraction layer and a weighted feature cleaning layer. The weighted feature extraction layer uses a fully connected layer to extract features and capture key information from the data. The weighted feature cleaning layer uses a dropout layer to remove noise, prevent overfitting, and improve the model's generalization capabilities. For example, during data fusion, the attention mechanism determines the importance of electricity price data and electricity profit and loss data in decision-making, highlighting the impact of high electricity prices and electricity surpluses on selling decisions.

[0094] The weighted output layer includes several output nodes. Different output nodes correspond to decision results such as whether to buy or sell external electricity, the buying and selling quantity, and the internal load distribution ratio. The activation function is selected according to the specific task. For example, the Sigmoid function is used to output the probability value for the buying and selling decision, the linear activation function is used to output the specific value for the buying and selling quantity, and the Softmax function is used to make the sum of the output values ​​1 for the internal load distribution ratio to ensure the rationality and effectiveness of the decision results.

[0095] S4. Perform intelligent scheduling on the microgrid based on the electricity profit and loss data and the decision data, wherein the intelligent scheduling includes external intelligent scheduling and internal intelligent scheduling.

[0096] For external intelligent scheduling, for example, the following steps may be included:

[0097] S411. Obtaining an electricity transaction strategy between the microgrid and the external power grid based on the electricity profit and loss data and the decision data, wherein the electricity transaction strategy includes:

[0098] Analyze power surplus and deficit: This system evaluates the microgrid's power surplus and deficit data in real time, including whether the current power is sufficient and the degree of power redundancy or shortage. For example, if the power surplus exceeds a set threshold and it is predicted that power generation will continue to exceed load demand over the next period of time, then it is determined that there is a power surplus that can be sold. Conversely, if the power is insufficient and the gap is expected to be large, then it is necessary to consider purchasing power.

[0099] Combining decision-making data with electricity price information to formulate trading strategies: This involves integrating information such as buying and selling trends and quantity ranges from decision-making data, while also considering real-time electricity prices and forecasted price trends. For example, when electricity prices are high, the strategy may be to sell excess electricity to generate higher profits; when prices are low, the strategy may consider buying required electricity to reduce costs.

[0100] Consider grid interaction constraints, such as access capacity limits and trading time limits for the external grid. For example, if the external grid has limited access capacity during a certain time period, even if the microgrid has a surplus of electricity, the trading strategy must be adjusted based on the capacity limit to avoid exceeding the grid's carrying capacity.

[0101] S412. Determine the power purchase and sale decision, power purchase and sale quantity, and power purchase and sale time according to the power transaction strategy, specifically including:

[0102] Energy purchase and sale decisions: Based on the energy trading strategy, the microgrid determines whether to buy or sell energy in the current and future time periods. For example, if it is predicted that the microgrid's power generation will drop significantly and load demand will increase in the next few hours, and the current energy surplus is limited, the decision will be made to buy energy at this time to meet the subsequent demand.

[0103] The amount of electricity to buy or sell is precisely calculated by combining profit and loss data, decision-making data, and the buying and selling trends and quantity ranges in the trading strategy. For example, based on the current surplus, the available capacity of the energy storage system, and predicted power generation and load changes, the amount of electricity to sell can be determined to be X kilowatt-hours, ensuring that the microgrid still has a safe operating reserve after the sale.

[0104] The optimal time or period for buying and selling electricity is determined based on factors such as electricity price fluctuations, the microgrid's own generation and load forecast curves, and grid interaction time constraints. For example, electricity can be sold during peak electricity price periods and purchased during low-price periods, while avoiding external grid transaction restrictions.

[0105] S413. Performing external intelligent scheduling on the microgrid based on the power purchase and sale decision, the power purchase and sale quantity, and the power purchase and sale time, specifically including:

[0106] Sending a trading instruction: Based on the determined electricity purchase and sale decision, quantity, and time, the microgrid's communication system sends a trading instruction to the external grid's trading platform. For example, at the selected selling time, an instruction to sell X kWh of electricity is sent to the grid trading platform.

[0107] Monitor and adjust transaction execution: Real-time monitoring of transaction execution status, including power delivery and real-time changes in transaction prices. If transaction execution deviations occur, such as sudden external power grid failures leading to transaction interruptions or real-time electricity price fluctuations exceeding expected ranges, timely adjustments to trading strategies will be made based on actual conditions, such as renegotiation of transaction quantity or time, and adjustment of selling or buying prices.

[0108] Update the trading plan: Dynamically update the subsequent electricity trading plan based on the transaction execution results and the real-time operating status of the microgrid. For example, if the current electricity sale is successful and the microgrid's electricity surplus decreases, the subsequent trading plan will adjust the amount of electricity sold or the time interval accordingly.

[0109] For internal intelligent scheduling, for example, the following steps may be included:

[0110] S421. Obtaining the power distribution ratio of the load units within the microgrid based on the power surplus and deficit data and the decision data, specifically including:

[0111] Assessing power load priorities: Prioritize various load units within the microgrid based on their importance and power demand characteristics. For example, critical loads such as hospitals and data centers can be given the highest priority to ensure their power supply stability and power requirements, while interruptible loads or non-critical production loads can be assigned a lower priority.

[0112] Calculate power allocation ratios based on power surplus and deficit data and decision-making data: Based on the current power surplus and deficit situation and the internal load adjustment trends in the decision-making data, the power allocation ratio for each load unit is calculated according to load priority. For example, when power is surplus, the power allocation ratio for low-priority loads is appropriately increased to improve energy efficiency. When power is insufficient, power supply to high-priority loads is prioritized, and power allocation to low-priority loads is proportionally reduced.

[0113] S422. Determine the power distribution ratio of the power generation units within the microgrid based on the power surplus and deficit data and the decision data, specifically including:

[0114] Analyze the operating status and available capacity of power generation units: Real-time monitoring of the operating parameters of each power generation unit within the microgrid (such as solar photovoltaic panels, wind turbines, diesel generators, etc.), including output power, fuel reserves (for fossil fuel power generation units), equipment health status, etc., to determine its available power generation capacity and stable operating range.

[0115] Optimize power allocation based on electricity surplus and deficit and decision-making data: The power allocation for each power unit is calculated by comprehensively considering the power surplus and deficit data, the available capacity of each power generation unit, and the power adjustment strategy in the decision-making data. For example, when renewable energy generation is sufficient and stable, its proportion in the power allocation is increased to reduce reliance on fossil fuel generation. When renewable energy generation is insufficient, the output proportion of controllable power generation units such as diesel generators is appropriately increased according to the scheduling instructions in the decision-making data to ensure a stable power supply for the microgrid.

[0116] S423. Obtaining the storage power distribution ratio of the energy storage unit within the microgrid based on the power surplus and deficit data and the decision data, specifically including:

[0117] Evaluate the current status of energy storage units: Monitor parameters such as the state of charge (SOC), state of health (SOH), and charge and discharge power limits of energy storage systems (such as battery energy storage) to determine their current energy storage level and charge and discharge capabilities.

[0118] Determine the storage power allocation ratio based on power surplus and deficit and decision data: Based on the microgrid's power surplus and deficit, the energy storage dispatch instructions in the decision data, and the current state of the energy storage unit, the storage power allocation ratio is determined. For example, when the microgrid has a power surplus and the energy storage unit is not full, the charging power ratio of the energy storage unit is increased to store the excess power. When power is insufficient but the energy storage unit has a certain reserve and the SOC allows discharge, the discharge power ratio of the energy storage unit is increased to release power to the microgrid.

[0119] S424. Performing internal intelligent scheduling on the microgrid according to the power consumption allocation ratio, the power generation allocation ratio, and the power storage allocation ratio, specifically including:

[0120] Executing power allocation instructions: Based on the calculated power allocation ratios for each load unit, power generation unit, and energy storage unit, the microgrid control system sends specific power adjustment instructions to each device. For example, it sends instructions to the solar photovoltaic inverter to adjust its output power to supply power to different loads according to predetermined ratios; it sends charge and discharge power instructions to the energy storage converter to control the charging and discharging process of the energy storage system.

[0121] Real-time monitoring and dynamic adjustment: Continuously monitor the operating status and power flow of each device within the microgrid, as well as the overall power balance of the microgrid. If abnormal conditions such as equipment failure, sudden load changes, or power generation fluctuations occur, the power allocation ratio is promptly recalculated based on real-time data, and the operating status of each device is quickly adjusted to ensure balanced power supply and demand within the microgrid and stable operation.

[0122] Coordinate the operation of various units to ensure power supply quality and stability: During internal intelligent scheduling, focus on coordinating the operational relationships between load units, power generation units, and energy storage units to avoid problems such as excessive power fluctuations and voltage instability. For example, by rationally allocating the charge and discharge power of the energy storage system and the output power of the power generation units, power fluctuations within the microgrid are smoothed, power supply quality is improved, and stable and reliable operation of the microgrid is ensured.

[0123] A neural network-based microgrid intelligent scheduling method in this embodiment first accurately extracts the microgrid's electricity profit and loss data by analyzing the neural network, effectively responding to the dynamic changes in electricity supply and demand within the microgrid; secondly, it obtains decision data through a predictive neural network to achieve efficient interaction between the microgrid and the external power grid; finally, it performs intelligent scheduling based on this data, significantly improving the intelligent scheduling level of the microgrid.

[0124] This embodiment also discloses a microgrid intelligent dispatching system based on neural network, see Figure 4 ,include:

[0125] The acquisition unit is used to obtain microgrid load data, microgrid energy storage data, microgrid power generation data and electricity price data;

[0126] An analysis unit, configured to input the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analysis neural network to obtain power profit and loss data;

[0127] A prediction unit, configured to input the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data;

[0128] A dispatching unit is used to intelligently dispatch the microgrid according to the power profit and loss data and the decision data.

[0129] Although the present application is described herein with reference to various embodiments, those skilled in the art will be able to understand and implement other variations of the disclosed embodiments in practicing the claimed application by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

[0130] The above is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, without departing from the concept of the present invention, several simple deductions or substitutions can be made, which should be considered to fall within the scope of protection of the present invention.

Claims

1. A microgrid intelligent scheduling method based on neural network, characterized in that: include: Obtain microgrid load data, microgrid energy storage data, microgrid power generation data and electricity price data; Inputting the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analysis neural network to obtain power profit and loss data; Inputting the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data; The microgrid is intelligently dispatched according to the electricity profit and loss data and the decision data.

2. A microgrid intelligent scheduling method based on neural network according to claim 1, characterized in that: The analysis neural network includes: an analysis input layer, a dense network layer, a probability analysis layer and an analysis output layer, wherein the input data of the analysis input layer includes the microgrid load data, the microgrid energy storage data and the microgrid power generation data; the dense network includes an initial convolution layer, a dense block layer and a transition layer in sequence; the probability analysis layer includes a probability residual convolution layer, a probability pooling activation layer and a probability full connection layer in sequence; the output data of the analysis output layer is power profit and loss data, wherein the power profit and loss data includes microgrid redundant power data, microgrid sustainable power supply data, microgrid power supply balance point data and microgrid income data.

3. A microgrid intelligent scheduling method based on neural network according to claim 1, characterized in that: Methods for analyzing neural network training include: Acquire data to be trained, wherein the data to be trained includes the microgrid load data to be trained, the microgrid energy storage data to be trained, the microgrid power generation data to be trained, and the power profit and loss data to be trained; Performing data cleaning on the data to be trained to obtain cleaned data; The cleaned data is input into the analysis neural network to be trained to obtain the analysis neural network, wherein the loss function of the analysis neural network is a first fusion loss function based on multi-weight KL divergence and cross entropy loss.

4. A microgrid intelligent scheduling method based on neural network according to claim 1, characterized in that: The step of inputting the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data includes: Inputting the electricity price data into an external prediction sub-neural network to obtain first decision data; Inputting the electricity profit and loss data into the internal prediction sub-neural network to obtain second decision data; The electricity price data, the electricity profit and loss data, the first decision data and the second decision data are input into a weight fusion sub-neural network to obtain decision data.

5. A neural network-based microgrid intelligent scheduling method according to claim 4, characterized in that: The external prediction sub-neural network includes: an external input layer, a feature dimensionality reduction layer, a feature extraction layer and an external output layer, wherein the input data of the external input layer is the electricity price data; the feature dimensionality reduction layer includes multiple convolutional layers and pooling layers connected in series; the feature extraction layer includes multiple fully connected layers; and the output data of the external output layer is the first decision data.

6. A microgrid intelligent dispatching method based on neural network according to claim 4, characterized in that: The internal prediction sub-neural network includes: an internal input layer, a residual feature extraction layer and an internal output layer, wherein the input data of the internal input layer includes power surplus and deficit data; the residual feature extraction layer includes a convolution layer, a normalization layer, an activation layer and a jump connection layer in sequence; the output data of the internal output layer is the second decision data.

7. A neural network-based microgrid intelligent dispatching method according to claim 4, characterized in that: The weight fusion sub-neural network includes: a weight input layer, an embedding layer, a weight link activation layer, an attention layer and a weight output layer, wherein the input data of the weight input layer includes the electricity price data, the electricity profit and loss data, the first decision data and the second decision data; the embedding layer includes a data splicing layer and a data alignment layer in sequence; the weight link activation layer includes multiple full-link layers and an activation layer; the attention layer includes a weight feature extraction layer and a weight feature cleaning layer in sequence; the output data of the weight output layer is decision data.

8. The neural network-based microgrid intelligent scheduling method according to claim 1, characterized in that: The intelligent scheduling of the microgrid according to the power profit and loss data and the decision data includes: Obtaining an electricity transaction strategy between the microgrid and the external power grid based on the electricity profit and loss data and the decision data; Determine the electricity purchase and sale decision, the quantity of electricity purchased and sold, and the time of electricity purchase and sale according to the electricity transaction strategy; The microgrid is externally intelligently dispatched according to the power purchase and sale decision, the power purchase and sale quantity, and the power purchase and sale time.

9. The neural network-based microgrid intelligent scheduling method according to claim 1, characterized in that: The intelligent scheduling of the microgrid according to the power profit and loss data and the decision data includes: Obtaining the power distribution ratio of the load units within the microgrid according to the power surplus and deficit data and the decision data; Obtaining a power distribution ratio of power generation units within the microgrid according to the power surplus and deficit data and the decision data; Obtaining a storage power distribution ratio of an energy storage unit within a microgrid according to the power surplus and deficit data and the decision data; The microgrid is internally intelligently dispatched according to the power consumption allocation ratio, the power generation allocation ratio, and the power storage allocation ratio.

10. A microgrid intelligent dispatching system based on neural network, characterized in that: include: The acquisition unit is used to obtain microgrid load data, microgrid energy storage data, microgrid power generation data and electricity price data; An analysis unit, configured to input the microgrid load data, the microgrid energy storage data, and the microgrid power generation data into a pre-trained analysis neural network to obtain power profit and loss data; A prediction unit, configured to input the electricity profit and loss data and the electricity price data into a pre-trained prediction neural network to obtain decision data; A dispatching unit is used to intelligently dispatch the microgrid according to the power profit and loss data and the decision data.

Images