Intelligent power distribution network management method and system for high-proportion access of new energy

Abstract

The invention relates to an intelligent power distribution network management method for high-proportion access of new energy, and the method comprises the following steps: S1, deploying an intelligent sensor and edge calculation equipment at each node and edge position of a power distribution network, and carrying out the data collection and preprocessing of the power distribution network; s2, on the basis of a deep learning model, according to the preprocessed power distribution network data, respectively predicting new energy supply and power demand, and carrying out multi-scenario simulation to evaluate the influence of new energy fluctuation on a power grid under different conditions; s3, constructing a multi-objective optimization model by using the preprocessed real-time data, the prediction and evaluation results of the S2 and the current operation condition of the power grid, and solving the multi-objective optimization model by using Pareto forward construction to obtain a reference scheduling strategy; s4, developing a dynamically adjusted Fibonacci strategy in combination with reinforcement learning, integrating an energy storage system and demand response to optimize power grid scheduling, optimizing a reference strategy, and obtaining a final scheduling strategy; the adaptive capacity of the power distribution network to new energy fluctuation and uncertainty is improved.

Description

Technical Field

[0001] The present invention relates to the field of smart grids, and in particular to a management method and system for a smart distribution network with high-proportion access of new energy. Background Art

[0002] In the context of energy transformation, the high proportion of access to new energy (such as wind power, solar energy, etc.) has become an important trend in the development of modern power systems. However, the intermittent and volatile characteristics of new energy have brought huge challenges to the dispatch and management of traditional power grids. This volatility requires the power grid to have higher flexibility and responsiveness to cope with the uncertainty of supply and demand. At the same time, with the popularization of distributed energy, smart devices and electric vehicles, the complexity of the power grid is increasing. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a smart distribution network management method and system with a high proportion of new energy access, which improves the distribution network's adaptability to new energy fluctuations and uncertainties and realizes efficient, stable and economical distribution network management.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for managing a smart distribution network with high-proportion access to new energy sources comprises the following steps:

[0006] S1: Deploy smart sensors and edge computing devices at each node and edge location of the distribution network to collect and preprocess distribution network data;

[0007] S2: Based on the deep learning model, according to the pre-processed distribution network data, the renewable energy supply and power demand are predicted respectively, and multi-scenario simulation is carried out to evaluate the impact of renewable energy fluctuations on the power grid under different circumstances;

[0008] S3: Using the preprocessed real-time data and the prediction and evaluation results of S2, as well as the current operation status of the power grid, a multi-objective optimization model is constructed, and the Pareto forward construction is used to solve the multi-objective optimization model and obtain a benchmark dispatch strategy;

[0009] S4: Develop a dynamically adjusted Fibonacci strategy with reinforcement learning, integrate energy storage systems with demand response to optimize grid dispatch, optimize the baseline strategy, and obtain the final dispatch strategy;

[0010] Furthermore, S1 is specifically as follows: smart sensors are installed at various nodes and key edge locations of the distribution network to measure voltage, current, and temperature parameters, and edge computing devices are deployed in substations to be responsible for localized data processing and storage; each smart sensor is connected through the interface of the edge device to collect power grid operation data in real time; the edge computing device denoises and processes missing values ​​for the collected data, and reduces the amount of data through aggregation.

[0011] Furthermore, S2 is specifically:

[0012] S21: Combine CNN and LSTM to build a prediction model with shared layers and task-specific layers that can simultaneously predict new energy supply and electricity demand;

[0013] S22: Conduct benchmark forecasts of new energy supply and power demand based on preprocessed distribution network data;

[0014] S23: On the basis of the baseline forecast, random disturbances are added to simulate the fluctuations of renewable energy, and the fluctuations are generated using the normal distribution model: extreme values ​​in historical data or extreme conditions predicted by meteorological models are used to construct extreme scenarios of electricity supply and demand; and possible bottlenecks and risks in each extreme scenario are measured.

[0015] Furthermore, CNN and LSTM are combined to build a prediction model with shared layers and task-specific layers, including input layer, feature extraction layer, task-specific network layer and output layer;

[0016] The input layer inputs various pre-processed data into the model, including time features, physical features, and weather data;

[0017] The feature extraction layer includes a convolutional neural network layer and an LSTM layer: first, a convolutional neural network is used to extract local time series patterns and identify patterns that appear repeatedly in a short time window, such as daily fluctuations; then an LSTM network is used to capture long-term dependencies in the time series. LSTM is good at processing sequence data and can remember dependency information for a longer period of time;

[0018] The dedicated task network layer includes a shared layer and a task-specific layer;

[0019] The features extracted by the feature extraction layer first enter a sharing layer to jointly learn the basic knowledge of supply and demand;

[0020] The task-specific layer: for supply forecasting, a fully connected layer is used to process and output the forecast value of new energy supply; another fully connected layer will process the specific features of demand forecasting and output the forecast value of power grid demand;

[0021] Output layer, the outputs of the two fully connected layers are used as the output layer prediction outputs to obtain the predicted new energy supply and grid demand.

[0022] Furthermore, the feature extraction layer includes a convolutional neural network layer and an LSTM layer, as follows:

[0023] Use a one-dimensional convolutional layer to extract short-term changing patterns from the input data by sliding the convolution kernel:

[0024]

[0025] Among them, z i，j is the output at row i and column j after the convolution operation; X i，j+k represents the value of the kth position starting from position j in the i-th row of the input sequence; W k is the weight of the convolution kernel, b is the bias term; K is the total number of positions;

[0026] After the convolutional layer, the LSTM layer is used to process the feature sequence:

[0027] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0028] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0029] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0030]

[0031] h t =o t *tanh(C t );

[0032] Among them, f t is the activation value of the forget gate, which determines the previous memory C t-1 The degree of retention; t is the activation value of the input gate, which determines the candidate memory of the current input The degree of increase; t is the activation value of the output gate; C t is the memory unit state at the current moment; ht is the hidden state; W f , W i , W o , W C is the weight; b f , b i , b o , b C is the bias; σ is the activation function; x t is the input at the current time t;

[0033] Introduce the attention mechanism to assign different importance weights to different time steps:

[0034]

[0035] Among them, a t represents the attention weight assigned to the hidden state h at time t t ; W a is the weight matrix; exp is the exponential function; is the representation of the weighted feature sequence, which is obtained by weighted summation of the hidden states of all time steps; T represents the total length of the time series; k′ is a traversal index variable.

[0036] Further, S23 is as follows:

[0037] Benchmark forecasts for new energy supply Add the disturbance ε g (t), based on the forecast of electricity demand Add the disturbance ε d (t):

[0038]

[0039] Using extreme weather and load values ​​from historical data, as well as extreme weather events predicted by meteorological models, we construct extreme supply and demand scenarios and run stochastic simulations using Monte Carlo simulations to generate boundary cases for power supply and demand through disturbances and extreme events;

[0040] In each scenario, the risk probability and potential loss are calculated, and the risk indicator VaR is used to quantify the risk:

[0041] VaR α = -inf{x|P(L>e)≤α};

[0042] Where L is the loss distribution, α is the confidence level, P(L>e) represents the probability that the loss L exceeds e, e is the preset threshold, and x is the input.

[0043] Furthermore, the S4 is specifically:

[0044] Construct a reinforcement learning model. The state space S is the current state of the power grid, including load demand, power generation status, and energy storage capacity. The action space A includes adjusting the charging and discharging of the energy storage system, adjusting demand response, and generator output. The reward function R is based on the optimization goal:

[0045] R=-a·f1(X)-b′·f1(X)-c·f3(X)+d·f4(X);

[0046] Among them, a, b′, c, d are weight coefficients;

[0047] The benchmark scheduling strategy is adjusted through the DQN algorithm, and the dynamically adjusted Fibonacci strategy is used during the scheduling process to allocate energy storage and response resources;

[0048] Continuously evaluate the effectiveness of the current strategy through real-time data, update the reward function, and dynamically update the strategy.

[0049] Furthermore, the multi-objective optimization model is constructed as follows:

[0050] The objectives of the multi-objective optimization model include minimizing investment f1(X), minimizing operating costs f2(X), minimizing power losses f3(X), and maximizing renewable energy utilization f4(X):

[0051]

[0052] Among them, C inv,i′ is the investment cost of equipment i′, x i′ is a decision variable, indicating whether equipment i′ is invested or installed; is the operating cost of equipment i′ at time t, is a binary variable, indicating whether device i′ is running at time t; L t (X) is the system power loss at time t as a function of the usage decision X; P renew,t is the power of renewable energy; P total,t is the total power;

[0053] Constraints include supply and demand balance and capacity limits:

[0054]

[0055] in, D is the output power of generator i″ at time t; t is the load demand at time t; is a binary variable, indicating whether generator i″ is running at time t; P min,i″ and P max,i″ are the minimum and maximum generable power of generator i″ respectively.

[0056] Furthermore, the benchmark scheduling strategy is adjusted through the DQN algorithm, and the dynamically adjusted Fibonacci strategy is adopted in the scheduling process to allocate energy storage and response resources, as follows:

[0057] The DQN algorithm determines the optimal strategy by approximating the Q-value function:

[0058]

[0059] Where η is the learning rate; δ is the discount factor; Q(S t , A t ) is in state S t Next take action A t Q value estimation of R t For state S t Take action A t Immediate rewards after To go from the next state S t+1 The maximum Q value at the beginning;

[0060] A fixed-size experience replay buffer D is used to store the experienced state transition tuples, and the past experience (S t ,A t ,R t ,S t+1 ), randomly sample and update the Q network to reduce data correlation;

[0061] When updating the Q value, a delayed target network is used to stabilize the training, and the parameters of the target network are updated every certain number of steps;

[0062] In the dispatching process, the Fibonacci sequence is used to guide the energy storage charging and discharging strategy, and the control amount ΔE of some charging and discharging is dynamically adjusted. t :

[0063]

[0064] Among them, F g is the gth Fibonacci number, ΔE max is the maximum possible charge and discharge change of the energy storage system; n is the number of Fibonacci numbers, and i″′ is an index variable.

[0065] A smart distribution network management system with a high proportion of new energy access includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically executes the steps in the smart distribution network management method with a high proportion of new energy access as described above.

[0066] The present invention has the following beneficial effects:

[0067] 1. The present invention improves the adaptability of the power grid to the fluctuation and uncertainty of new energy, and realizes efficient, stable and economical distribution network management;

[0068] 2. The present invention learns the common features of the two tasks in the shared layer, and then performs incremental learning for each task in the task-specific layer. This method effectively utilizes the features of the input data, improves the model's ability to simultaneously predict new energy supply and grid demand, improves resource efficiency, and reduces development and maintenance costs. At the same time, the accuracy and robustness of the prediction are improved because the two tasks share information;

[0069] 3. The present invention combines multi-objective optimization with real-time reinforcement learning, adopts the Fibonacci strategy for strategy adjustment, realizes the intelligent scheduling of energy storage and demand response, optimizes the operating efficiency of the power grid, and maximizes the use of renewable energy. Through DQN combined with the dynamically adjusted Fibonacci strategy, it can adaptively optimize the scheduling strategy in the ever-changing power grid environment to maximize the use of renewable energy and cost-effectiveness. Real-time evaluation and experience learning enable the system to have powerful decision-making and scheduling capabilities in uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0071] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0072] refer to Figure 1 In this embodiment, a method for managing a smart distribution network with a high proportion of new energy access is provided, comprising the following steps:

[0073] S1: Deploy smart sensors and edge computing devices at each node and edge location of the distribution network to collect and preprocess distribution network data;

[0074] S2: Based on the deep learning model, according to the pre-processed distribution network data, the renewable energy supply and power demand are predicted respectively, and multi-scenario simulation is carried out to evaluate the impact of renewable energy fluctuations on the power grid under different circumstances;

[0075] S3: Using the preprocessed real-time data and the prediction and evaluation results of S2, as well as the current operation status of the power grid, a multi-objective optimization model is constructed, and the Pareto forward construction is used to solve the multi-objective optimization model and obtain a benchmark dispatch strategy;

[0076] S4: Develop a dynamically adjusted Fibonacci strategy using reinforcement learning, integrate energy storage systems with demand response to optimize grid dispatch, optimize the baseline strategy, and obtain the final dispatch strategy.

[0077] In this embodiment, S1 specifically includes: installing smart sensors at various nodes and key edge locations of the distribution network to measure voltage, current, and temperature parameters, and deploying edge computing devices at substations to be responsible for localized data processing and storage; each smart sensor is connected through the interface of the edge device (such as MODBUS, CAN bus) to collect power grid operation data in real time; the edge computing device denoises and processes missing values ​​of the collected data, and reduces the amount of data through aggregation.

[0078] In this embodiment, S2 is specifically:

[0079] S21: Combine CNN and LSTM to build a prediction model with shared layers and task-specific layers that can simultaneously predict new energy supply and electricity demand;

[0080] S22: Conduct benchmark forecasts of new energy supply and power demand based on preprocessed distribution network data;

[0081] S23: On the basis of the baseline forecast, random disturbances are added to simulate the fluctuations of renewable energy, and the fluctuations are generated using the normal distribution model: extreme values ​​in historical data or extreme conditions predicted by meteorological models are used to construct extreme scenarios of power supply and demand; bottlenecks and risks that may arise in each extreme scenario are measured, such as overloaded lines or frequency deviations.

[0082] In this embodiment, CNN and LSTM are combined to construct a prediction model with shared layers and task-specific layers, including an input layer, a feature extraction layer, a dedicated task network layer, and an output layer;

[0083] The input layer inputs various pre-processed data into the model, including time features such as hours, days of the week, holidays, etc.; physical features such as voltage, current, power factor, etc.; weather data such as temperature, wind speed and solar radiation, which are particularly important for new energy supply;

[0084] The feature extraction layer includes a convolutional neural network layer and an LSTM layer: first, a convolutional neural network is used to extract local time series patterns and identify patterns that appear repeatedly in a short time window, such as daily fluctuations; then an LSTM network is used to capture long-term dependencies in the time series. LSTM is good at processing sequence data and can remember dependency information for a longer period of time;

[0085] The dedicated task network layer includes a shared layer and a task-specific layer;

[0086] The features extracted by the feature extraction layer first enter a sharing layer to jointly learn the basic knowledge of supply and demand;

[0087] The task-specific layer: for supply forecasting, a fully connected layer (Dense Layer) is used to process and output the forecast value of new energy supply; another fully connected layer will process the specific features of demand forecasting and output the forecast value of power grid demand;

[0088] Output layer, the outputs of the two fully connected layers are used as the output layer prediction outputs to obtain the predicted new energy supply and grid demand.

[0089] In this embodiment, the feature extraction layer includes a convolutional neural network layer and an LSTM layer, as follows:

[0090] Use a one-dimensional convolution (Conv1D) layer to extract short-term changing patterns from the input data by sliding the convolution kernel:

[0091]

[0092] Among them, z i，j is the output at row i and column j after the convolution operation; X i，j+k represents the value of the kth position starting from position j in the i-th row of the input sequence; W k is the weight of the convolution kernel, b is the bias term; K is the total number of positions;

[0093] After the convolutional layer, the LSTM layer is used to process the feature sequence:

[0094] f t =σ(W f ·h t-1 ,x t ]+b f )

[0095] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0096] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0097]

[0098] h t =o t *tanh(C t );

[0099] Among them, f t is the activation value of the forget gate, which determines the previous memory C t-1The degree of retention; t is the activation value of the input gate, which determines the candidate memory of the current input The degree of increase; t is the activation value of the output gate; C t is the memory unit state at the current moment; h t is the hidden state; W f , W i , W o , W C is the weight; b f , b i , b o , b C is the bias; σ is the activation function; x t is the input at the current time t;

[0100] Introduce the attention mechanism to assign different importance weights to different time steps:

[0101]

[0102] Among them, a t represents the attention weight assigned to the hidden state h at time t t ; W a is the weight matrix; exp is the exponential function; is the representation of the weighted feature sequence, which is obtained by weighted summation of the hidden states of all time steps; T represents the total length of the time series; k′ is a traversal index variable.

[0103] In this embodiment, S23 is specifically as follows:

[0104] Benchmark forecasts for new energy supply Add the disturbance ε g (t), based on the forecast of electricity demand Add the disturbance ε d (t):

[0105]

[0106] Using extreme weather and load values ​​from historical data, as well as extreme weather events predicted by meteorological models, we construct extreme supply and demand scenarios and run stochastic simulations using Monte Carlo simulations to generate boundary cases for power supply and demand through disturbances and extreme events;

[0107] Preferably, in this embodiment, specifically, by collecting the impact of extreme weather events on power demand and power generation capacity in the past, such as heavy rain, heavy snow, heat waves, etc., historical load data is obtained, especially load changes during extreme weather periods.

[0108] Use meteorological models to predict possible extreme weather events in the future and estimate their probability and characteristics.

[0109] Constructing extreme supply and demand scenarios:

[0110] Use historical extreme weather data and weather forecast data to identify potential extreme weather scenarios. For each scenario, determine the likely changes in demand (e.g., peak load) and supply (e.g., generator failures, reduced renewable energy output).

[0111] Introduce random disturbances to simulate weather uncertainty and its impact on the power system. Set disturbance parameters such as temperature change, wind speed reduction, etc.

[0112] Monte Carlo simulation:

[0113] Simulation process:

[0114] A large number of simulations (e.g., 1,000 or 10,000) are run for each scenario to evaluate how the power system responds to extreme conditions. In each simulation, weather parameters and load conditions are randomly sampled.

[0115] Output analysis:

[0116] The simulation results are statistically analyzed to obtain reliability indicators of the power system under extreme scenarios, such as the probability of power outage and maximum load loss.

[0117] In each scenario, the risk probability and potential loss are calculated, and the risk indicator VaR is used to quantify the risk:

[0118] VaR α = -inf{x|P(L>e)≤α};

[0119] Where L is the loss distribution, α is the confidence level, P(L>e) represents the probability that the loss L exceeds e, e is the preset threshold, and x is the input.

[0120] In this embodiment, the multi-objective optimization model is constructed as follows:

[0121] The objectives of the multi-objective optimization model include minimizing investment f1(X), minimizing operating costs f2(X), minimizing power losses f3(X), and maximizing renewable energy utilization f4(X):

[0122]

[0123] Among them, C inv,i′ is the investment cost of equipment i′, x i′ is a decision variable, indicating whether equipment i′ is invested or installed; is the operating cost of equipment i′ at time t, is a binary variable, indicating whether device i′ is running at time t; L t (X) is the system power loss at time t as a function of the usage decision X; P renew,t is the power of renewable energy; P total,t is the total power;

[0124] Constraints include supply and demand balance and capacity limits:

[0125]

[0126] in, D is the output power of generator i″ at time t; t is the load demand at time t; is a binary variable, indicating whether generator i″ is running at time t; P min,i″ and P max,i″ are the minimum and maximum generable power of generator i″ respectively.

[0127] In this embodiment, S4 is specifically:

[0128] Construct a reinforcement learning model. The state space S is the current state of the power grid, including load demand, power generation status, and energy storage capacity. The action space A includes adjusting the charging and discharging of the energy storage system, adjusting demand response, and generator output. The reward function R is based on the optimization goal:

[0129] R=-a·f1(X)-b′·f1(X)-c·f3(X)+d·f4(X);

[0130] Among them, a, b′, c, d are weight coefficients;

[0131] The benchmark scheduling strategy is adjusted through the DQN algorithm, and the dynamically adjusted Fibonacci strategy is used during the scheduling process to allocate energy storage and response resources;

[0132] Continuously evaluate the effectiveness of the current strategy through real-time data, update the reward function, and dynamically update the strategy.

[0133] In this embodiment, the benchmark scheduling strategy is adjusted by the DQN algorithm, and the dynamically adjusted Fibonacci strategy is adopted in the scheduling process to allocate energy storage and response resources, as follows:

[0134] The DQN algorithm determines the optimal strategy by approximating the Q-value function:

[0135]

[0136] Where η is the learning rate; δ is the discount factor; Q(S t , A t) is in state S t Next take action A t Q value estimation of R t For state S t Take action A t Immediate rewards after To go from the next state S t+1 The maximum Q value at the beginning;

[0137] A fixed-size experience replay buffer D is used to store the experienced state transition tuples, and the past experience (S t ,A t ,R t ,S t+1 ), randomly sample and update the Q network to reduce data correlation;

[0138] When updating the Q value, a delayed target network is used to stabilize the training, and the parameters of the target network are updated every certain number of steps;

[0139] In the dispatching process, the Fibonacci sequence is used to guide the energy storage charging and discharging strategy, and the control amount ΔE of some charging and discharging is dynamically adjusted. t :

[0140]

[0141] Among them, F g is the gth Fibonacci number, ΔE max is the maximum possible charge and discharge change of the energy storage system; n is the number of Fibonacci numbers, and i″′ is an index variable.

[0142] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0144] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0146] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for managing a smart distribution network with high proportion of new energy access, characterized in that: The following steps are involved: S1: Deploy smart sensors and edge computing devices at each node and edge location of the distribution network to collect and preprocess distribution network data; S2: Based on the deep learning model, according to the pre-processed distribution network data, the renewable energy supply and power demand are predicted respectively, and multi-scenario simulation is carried out to evaluate the impact of renewable energy fluctuations on the power grid under different circumstances; S3: Using the preprocessed real-time data and the prediction and evaluation results of S2, as well as the current operation status of the power grid, a multi-objective optimization model is constructed, and the Pareto forward construction is used to solve the multi-objective optimization model and obtain a benchmark dispatch strategy; S4: Develop a dynamically adjusted Fibonacci strategy using reinforcement learning, integrate energy storage systems with demand response to optimize grid dispatch, optimize the baseline strategy, and obtain the final dispatch strategy.

2. The method for managing a smart distribution network with high-proportion access to new energy according to claim 1 is characterized in that: Specifically, S1 includes: installing smart sensors at various nodes and key edge locations of the distribution network to measure voltage, current, and temperature parameters, and deploying edge computing devices at substations to be responsible for localized data processing and storage; each smart sensor is connected through the interface of the edge device to collect power grid operation data in real time; the edge computing device denoises and processes missing values ​​of the collected data, and reduces the amount of data through aggregation.

3. The method for managing a smart distribution network with high-proportion access to new energy according to claim 1 is characterized in that: The S2 is specifically: S21: Combine CNN and LSTM to build a prediction model with shared layers and task-specific layers that can simultaneously predict new energy supply and electricity demand; S22: Conduct benchmark forecasts of new energy supply and power demand based on preprocessed distribution network data; S23: On the basis of the baseline forecast, random disturbances are added to simulate the fluctuations of renewable energy, and the fluctuations are generated using the normal distribution model: extreme values ​​in historical data or extreme conditions predicted by meteorological models are used to construct extreme scenarios of electricity supply and demand; and possible bottlenecks and risks in each extreme scenario are measured.

4. The method for managing a smart distribution network with high-proportion access to new energy according to claim 3 is characterized in that: The CNN and LSTM are combined to construct a prediction model with shared layers and task-specific layers, including an input layer, a feature extraction layer, a dedicated task network layer, and an output layer; The input layer inputs various pre-processed data into the model, including time features, physical features, and weather data; The feature extraction layer includes a convolutional neural network layer and an LSTM layer: first, a convolutional neural network is used to extract local time series patterns and identify patterns that appear repeatedly in a short time window; then an LSTM network is used to capture long-term dependencies in the time series. LSTM is good at processing sequence data and can remember dependency information for a longer period of time; The dedicated task network layer includes a shared layer and a task-specific layer; The features extracted by the feature extraction layer first enter a sharing layer to jointly learn the basic knowledge of supply and demand; The task-specific layer: for supply forecasting, a fully connected layer is used to process and output the forecast value of new energy supply; another fully connected layer will process the specific features of demand forecasting and output the forecast value of power grid demand; Output layer, the outputs of the two fully connected layers are used as the output layer prediction outputs to obtain the predicted new energy supply and grid demand.

5. The method for managing a smart distribution network with high-proportion access to new energy according to claim 4 is characterized in that: The feature extraction layer includes a convolutional neural network layer and an LSTM layer, as follows: Use a one-dimensional convolutional layer to extract short-term changing patterns from the input data by sliding the convolution kernel: Among them, z i，j is the output at row i and column j after the convolution operation; X i，j+k represents the value of the kth position starting from position j in the i-th row of the input sequence; W k is the weight of the convolution kernel, b is the bias term; K is the total number of positions; After the convolutional layer, the LSTM layer is used to process the feature sequence: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t *tanh(C t ); Among them, f t is the activation value of the forget gate, which determines the previous memory C t-1 The degree of retention; t is the activation value of the input gate, which determines the candidate memory of the current input The degree of increase; t is the activation value of the output gate; C t is the memory unit state at the current moment; h t is the hidden state; W f , W i , W o , W C is the weight; b f , b i , b o , b C is the bias; σ is the activation function; x t is the input at the current time t; Introduce the attention mechanism to assign different importance weights to different time steps: Among them, a t represents the attention weight assigned to the hidden state h at time t t ; W a is the weight matrix; exp is the exponential function; is the representation of the weighted feature sequence, which is obtained by weighted summation of the hidden states of all time steps; T represents the total length of the time series; k′ is a traversal index variable.

6. The method for managing a smart distribution network with high-proportion access to new energy according to claim 3 is characterized in that: The S23 is specifically as follows: Benchmark forecasts for new energy supply Add the disturbance ε g (t), based on the forecast of electricity demand Add the disturbance ε d (t): Using extreme weather and load values ​​from historical data, as well as extreme weather events predicted by meteorological models, we construct extreme supply and demand scenarios and run stochastic simulations using Monte Carlo simulations to generate boundary cases for power supply and demand through disturbances and extreme events; In each scenario, the risk probability and potential loss are calculated, and the risk indicator VaR is used to quantify the risk: VaR α =-inf{x∣P(L>e)≤α}; Where L is the loss distribution, α is the confidence level, P(L>e) represents the probability that the loss L exceeds e, e is the preset threshold, and x is the input.

7. According to claim 1, a smart distribution network management method with high proportion of new energy access, It is characterized in that The S4 is specifically: Construct a reinforcement learning model. The state space S is the current state of the power grid, including load demand, power generation status, and energy storage capacity. The action space A includes adjusting the charging and discharging of the energy storage system, adjusting demand response, and generator output. The reward function R is based on the optimization goal: R=-a·f1(X)-b′·f1(X)-c·f3(X)+d·f4(X); Among them, a, b′, c, d are weight coefficients; The benchmark scheduling strategy is adjusted through the DQN algorithm, and the dynamically adjusted Fibonacci strategy is used during the scheduling process to allocate energy storage and response resources; Continuously evaluate the effectiveness of the current strategy through real-time data, update the reward function, and dynamically update the strategy.

8. According to claim 1, a smart distribution network management method with high proportion of new energy access, It is characterized in that The multi-objective optimization model is constructed as follows: The objectives of the multi-objective optimization model include minimizing investment f1(X), minimizing operating costs f2(X), minimizing power losses f3(X), and maximizing renewable energy utilization f4(X): Among them, C inv,i′ is the investment cost of equipment i′, x i′ is a decision variable, indicating whether equipment i′ is invested or installed; is the operating cost of equipment i′ at time t, is a binary variable, indicating whether device i′ is running at time t; L t (X) is the system power loss at time t as a function of the usage decision X; P renew,t is the power of renewable energy; P total,t is the total power; Constraints include supply and demand balance and capacity limits: in, D is the output power of generator i″ at time t; t is the load demand at time t; is a binary variable, indicating whether generator i″ is running at time t; P min,i″ and P max,i″ are the minimum and maximum generable power of generator i″ respectively.

9. The method for managing a smart distribution network with high-proportion access to new energy according to claim 1, characterized in that: The benchmark scheduling strategy is adjusted by the DQN algorithm, and the dynamically adjusted Fibonacci strategy is adopted in the scheduling process to allocate energy storage and response resources, as follows: The DQN algorithm determines the optimal strategy by approximating the Q-value function: Where η is the learning rate; δ is the discount factor; Q(S t , A t ) is in state S t Next take action A t Q value estimation of R t For state S t Take action A t Immediate rewards after To go from the next state S t+1 The maximum Q value at the beginning; A fixed-size experience replay buffer D is used to store the experienced state transition tuples, and the past experience (S t ,A t ,R t ,S t+1 ), randomly sample and update the Q network to reduce data correlation; When updating the Q value, a delayed target network is used to stabilize the training, and the parameters of the target network are updated every certain number of steps; In the dispatching process, the Fibonacci sequence is used to guide the energy storage charging and discharging strategy, and the control amount ΔE of some charging and discharging is dynamically adjusted. t : Among them, F g is the gth Fibonacci number, ΔE max is the maximum possible charge and discharge change of the energy storage system; n is the number of Fibonacci numbers, and i″′ is an index variable.

10. An intelligent distribution network management system with high proportion of new energy access, characterized in that: It includes a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, it specifically executes the steps in the smart distribution network management method with a high proportion of new energy access as described in any one of claims 1 to 9.

Images