Power distribution network carbon emission reduction optimization scheduling method and system

Abstract

The invention relates to the technical field of carbon emission management, in particular to a power distribution network carbon emission reduction optimization scheduling method and system, and the method comprises the steps: carrying out the association rule analysis of a power distribution network operation comprehensive data set through an association rule mining algorithm, and obtaining key influence factors which affect the energy consumption and carbon emission of a power distribution network; obtaining a carbon emission predicted value and an energy consumption predicted value by using the improved long-short-term memory neural network model; performing clustering analysis on the user power consumption behavior data by adopting a K-means clustering algorithm to obtain a power consumption portrait; generating a power distribution network carbon emission reduction scheduling strategy by using a reinforcement learning algorithm according to the power consumption portrait, the carbon emission prediction value and the energy consumption prediction value; and regulating and controlling the power resource configuration and the equipment operation state of each link of the power distribution network source network load storage in real time according to the power distribution network carbon emission reduction scheduling strategy. According to the method, a targeted power distribution network carbon emission reduction scheduling strategy is generated through the improved long-short-term memory neural network model and the reinforcement learning algorithm, and the efficiency and accuracy of power distribution network carbon emission management are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission management, and in particular to a method and system for optimizing and scheduling carbon emission reduction in a distribution network. Background Art

[0002] With the development of smart grids, distribution network systems need to process and analyze massive amounts of source, grid, load and storage data in real time. These source, grid, load and storage data include multiple dimensions such as carbon emissions, power consumption, and equipment operating parameters, showing high-dimensional and multi-scale characteristics. These data are not only huge in quantity, but also complex in structure, and include multiple types such as time series data and static parameter data, which brings great difficulty to data processing. When processing large-scale data sets, the efficiency of existing technologies is often difficult to meet real-time requirements, which makes it difficult for the system to efficiently identify key factors affecting energy consumption. At the same time, these algorithms often require a lot of computing resources and time when processing high-dimensional and multi-scale data, making it difficult to balance the accurate prediction and real-time performance of carbon emissions in distribution networks. In addition, the dynamic and uncertain nature of users' electricity consumption behavior further exacerbates the complexity of the problem, making it difficult to accurately capture users' real-time electricity consumption characteristics, resulting in the inability to formulate targeted emission reduction strategies.

[0003] To sum up, the existing technical means have obvious shortcomings in the accurate prediction of distribution network carbon emission management systems. How to balance the accurate prediction and control of distribution network carbon emissions with the efficiency of power grid operation under limited computing resources has become a technical problem that needs to be solved urgently in the current distribution network carbon emission management. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a distribution network carbon emission reduction optimization scheduling method and system.

[0005] In a first aspect, the present invention provides a method for optimizing and scheduling carbon emission reduction in a distribution network, the method comprising the following steps: collecting carbon emission data, electricity data, and real-time grid operation parameters of each link of the distribution network source, grid, load, and storage to form a comprehensive distribution network operation data set; Using association rule mining algorithm to analyze the distribution network operation comprehensive data set, the key influencing factors affecting the distribution network energy consumption and carbon emissions are obtained; Taking the key influencing factors as input features, using a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; Acquire user electricity consumption behavior data, and use K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; Generate a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the electricity consumption profile, the carbon emission forecast value and the energy consumption forecast value; According to the distribution network carbon emission reduction dispatching strategy, the power resource configuration and equipment operation status of each link of the source, grid, load and storage in the distribution network are regulated in real time.

[0006] In a further embodiment, the step of collecting carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive distribution network operation data set includes: Collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to obtain the original distribution network operation data set; Performing data cleaning on the original distribution network operation data set to obtain a preprocessed distribution network operation data set; The missing values ​​in the pre-processed distribution network operation data set are filled with valid data at the latest time point to obtain a filled distribution network operation data set; The infilled distribution network operation data set is normalized to obtain a distribution network operation comprehensive data set.

[0007] In a further embodiment, the step of using an association rule mining algorithm to perform association rule analysis on the distribution network operation comprehensive data set to obtain key influencing factors affecting distribution network energy consumption and carbon emissions includes: Discretizing the distribution network operation comprehensive data set by using an equal-width binning method to obtain discretized distribution network operation comprehensive data; The discrete distribution network operation comprehensive data is converted into a transactional data format to obtain transactional distribution network operation data; wherein each transaction represents a combination of distribution network operation states; Setting a minimum support threshold and a minimum confidence threshold, and using the Apriori algorithm to iteratively scan the transactional distribution network operation data to generate a candidate item set; Calculate the support of each candidate item set, and based on the support, select frequent item sets that meet the minimum support threshold from the candidate item set, and generate association rules based on the frequent item sets; Calculating the confidence of each association rule, and based on the confidence, selecting a set of strong association rules that meet a minimum confidence threshold from the association rules; The energy consumption factors of the distribution network that frequently appear in the strong association rule set are analyzed for their association strength, and the key factors affecting the energy consumption and carbon emissions of the distribution network are identified.

[0008] In a further embodiment, the step of using the key influencing factors as input features and using a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period includes: Performing principal component analysis on the comprehensive data set of the distribution network operation to extract the principal component feature vectors that affect carbon emissions and energy consumption; using a sliding window method to divide the principal component feature vectors and the key influencing factors according to time series to obtain principal component segments and key influencing factor segments; The principal component fragments and the key influencing factor fragments are input into a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value at each time point in the future time period.

[0009] In a further embodiment, the improved long short-term memory neural network model includes a sequentially connected input layer, a one-dimensional convolutional layer, a bidirectional LSTM layer, a multi-head attention mechanism layer, a stacked LSTM layer introducing residual connections, a regularization layer and an output layer.

[0010] In a further embodiment, the step of inputting the principal component fragment and the key influencing factor fragment into a pre-constructed improved long short-term memory neural network model to predict the carbon emission prediction value and energy consumption prediction value at each time point in the future time period comprises: Performing data fusion processing on the principal component fragment and the key influencing factor fragment to obtain characteristic sequence data; Extracting local features from the feature sequence data through a one-dimensional convolutional layer to obtain a local feature sequence; A bidirectional LSTM layer is used to capture the front-end dependency of the local feature sequence to obtain a bidirectional LSTM hidden state sequence; the bidirectional LSTM hidden state sequence is input into a multi-head attention mechanism layer, and the multi-head attention mechanism layer is used to perform attention weighting processing on different time steps to obtain a weighted hidden state sequence; According to the weighted hidden state sequence, a stacked LSTM layer with residual connection is used to learn feature representation to obtain a high-order feature sequence; Regularizing the high-order feature sequence using a regularization layer to obtain a regularized feature sequence; The regularized feature sequence is linearly transformed to obtain the carbon emission prediction value and energy consumption prediction value at each time point in the future time period.

[0011] In a further implementation scheme, the step of using the K-means clustering algorithm to perform cluster analysis on the user's electricity consumption behavior data to obtain different types of electricity consumption portraits includes: The electricity consumption period of each user is obtained according to the user's electricity consumption behavior data, and the standard deviation of the electricity consumption of each user in each electricity consumption period is calculated to obtain the electricity consumption time fluctuation value; Compare the power consumption time fluctuation value of each user with a preset fluctuation threshold, filter out the power consumption behavior data of users whose power consumption time fluctuation value is greater than the preset fluctuation threshold, and obtain the power consumption behavior data of users to be clustered; According to the electricity consumption time and electricity consumption in the electricity consumption behavior data of the users to be clustered, the average electricity consumption of each user is calculated; According to the average power consumption, the K-means clustering algorithm is used to cluster the power consumption behavior data of the users to be clustered to obtain user portraits of different categories.

[0012] In a further embodiment, the electricity usage behavior data includes historical electricity usage, electricity usage time period, and user identification; The electricity consumption profile includes electricity consumption time periods, average electricity consumption, and peak and valley characteristics.

[0013] In a further implementation scheme, the step of generating a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm based on the power consumption profile, the carbon emission forecast value and the energy consumption forecast value includes: Fuzzy processing is performed on the electricity consumption portrait, the carbon emission prediction value and the energy consumption prediction value to obtain a fuzzy input set; Using a fuzzy logic controller to perform fuzzy reasoning on the fuzzy input set to obtain a fuzzy output set; Defuzzifying the fuzzy output set to obtain fuzzy logic control strategy parameters; The distribution network is regarded as a reinforcement learning environment. The state space includes the current state of the power grid and the parameters of the fuzzy logic control strategy, and the action space is various distribution network carbon emission reduction scheduling strategies. With the goal of minimizing carbon emissions and energy consumption, the optimal action is selected according to the current grid state and fuzzy logic control strategy parameters, and the deep deterministic policy gradient algorithm is used to learn the optimal distribution network carbon emission reduction scheduling strategy under different grid states.

[0014] In a second aspect, the present invention provides a distribution network carbon emission reduction optimization scheduling system, the system comprising: The data acquisition module is used to collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive data set of distribution network operation; An association analysis module is used to perform association rule analysis on the distribution network operation comprehensive data set using an association rule mining algorithm to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network; A model prediction module is used to use the key influencing factors as input features and use a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; A clustering analysis module is used to obtain user electricity consumption behavior data, and use a K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; A strategy generation module, for generating a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the power consumption profile, the carbon emission forecast value and the energy consumption forecast value; The optimization scheduling module is used to adjust the power resource configuration and equipment operation status of each link of the source, grid, load and storage in the distribution network in real time according to the distribution network carbon emission reduction scheduling strategy.

[0015] The present invention provides a distribution network carbon emission reduction optimization scheduling method and system, the method collects carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a distribution network operation comprehensive data set; uses an association rule mining algorithm to perform association rule analysis on the distribution network operation comprehensive data set to obtain key influencing factors affecting the energy consumption and carbon emission of the distribution network; uses the key influencing factors as input features, and uses a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; obtains user electricity consumption behavior data, uses a K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data, and obtains electricity consumption portraits of different categories; uses a reinforcement learning algorithm to generate a distribution network carbon emission reduction scheduling strategy based on the electricity consumption portrait, the carbon emission forecast value and the energy consumption forecast value; and uses the distribution network carbon emission reduction scheduling strategy to real-time regulate the power resource configuration and equipment operation status of each link of the distribution network source, grid, load and storage. Compared with the existing technology, this method can accurately predict carbon emissions and energy consumption levels in future time periods through an improved long short-term memory neural network model and reinforcement learning algorithm, and generate targeted distribution network carbon emission reduction scheduling strategies, which significantly improves the efficiency and accuracy of distribution network carbon emission management, and achieves a dual improvement in energy conservation and emission reduction and grid operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the flow of the distribution network carbon emission reduction optimization scheduling method provided by an embodiment of the present invention; Figure 2 It is a block diagram of a distribution network carbon emission reduction optimization scheduling system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following specifically illustrates the implementation mode of the present invention in conjunction with the accompanying drawings. The embodiments are provided for illustrative purposes only and cannot be understood as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.

[0018] refer to Figure 1 , an embodiment of the present invention provides a method for optimizing the scheduling of carbon emission reduction in a distribution network, such as Figure 1 As shown, the method comprises the following steps: S1. Collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive data set of distribution network operation.

[0019] In this embodiment, the step of collecting carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive distribution network operation data set includes: Collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to obtain the original distribution network operation data set; Performing data cleaning on the original distribution network operation data set to obtain a preprocessed distribution network operation data set; The missing values ​​in the pre-processed distribution network operation data set are filled with valid data at the latest time point to obtain a filled distribution network operation data set; The infilled distribution network operation data set is normalized to obtain a distribution network operation comprehensive data set.

[0020] Specifically, this embodiment collects carbon emission data, electricity data and real-time grid operation parameters from each link of the power generation side (source), power grid side (grid), load side (load) and energy storage side (storage) of the distribution network, and summarizes the data collected from the above links to form an original distribution network operation data set, wherein the electricity data may include power generation and power consumption, and the real-time grid operation parameters may include voltage, current and power factor, etc. Then, this embodiment traverses the original distribution network operation data set, checks whether there are outliers in each data point, uses statistical methods to identify and process outliers, and for each missing value, uses linear interpolation or nearest neighbor interpolation to find the nearest valid data point within the time range before and after it, so as to fill the missing value according to the latest valid data, obtain the filled distribution network operation data set, and normalize the filled distribution network operation data set to obtain a comprehensive distribution network operation data set, which provides a reliable data basis for subsequent carbon emission reduction optimization scheduling.

[0021] S2. Use association rule mining algorithm to perform association rule analysis on the comprehensive data set of distribution network operation to obtain key influencing factors affecting energy consumption and carbon emissions of the distribution network.

[0022] In this embodiment, the step of using an association rule mining algorithm to perform association rule analysis on the distribution network operation comprehensive data set to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network includes: Discretizing the distribution network operation comprehensive data set by using an equal-width binning method to obtain discretized distribution network operation comprehensive data; The discrete distribution network operation comprehensive data is converted into a transactional data format to obtain transactional distribution network operation data; wherein each transaction represents a combination of distribution network operation states; Setting a minimum support threshold and a minimum confidence threshold, and using the Apriori algorithm to iteratively scan the transactional distribution network operation data to generate a candidate item set; Calculate the support of each candidate item set, and based on the support, select frequent item sets that meet the minimum support threshold from the candidate item set, and generate association rules based on the frequent item sets; Calculating the confidence of each association rule, and based on the confidence, selecting a set of strong association rules that meet a minimum confidence threshold from the association rules; The energy consumption factors of the distribution network that frequently appear in the strong association rule set are analyzed for their association strength, and the key factors affecting the energy consumption and carbon emissions of the distribution network are identified.

[0023] Specifically, this embodiment divides each continuous variable (such as power, voltage, current, etc.) in the distribution network operation comprehensive data set into several equal-width intervals through an equal-width binning method, and each interval is called a box. For example, the power data is divided into intervals such as 0-100kWh and 101-200kWh. Among them, for continuous data (such as load demand, power generation, etc.) in the distribution network operation comprehensive data set, this embodiment can set the interval width according to the distribution characteristics of the data, and allocate each data point in the distribution network operation comprehensive data set to the corresponding interval according to its value, so as to form a discretized data set, obtain the discretized distribution network operation comprehensive data, and then divide the distribution network at a certain point in time into the bins. The state (such as source power generation, network voltage, load power consumption, energy storage status, etc.) is regarded as a transaction, and the discretized distribution network operation comprehensive data is reorganized according to the transactional data format. Each transaction represents a combination of distribution network operation states at a time point. Each combination of distribution network operation states contains the discretized values ​​of all relevant parameters. For example, a transaction includes "load demand in the range of (100MW, 200MW], power generation in the range of [200MW, 300MW], carbon emissions in the range of [5 tons / hour, 15 tons / hour]", etc. The discretized distribution network operation comprehensive data is divided into multiple transactions according to time points, and each transaction is a set of discretized values ​​of all relevant parameters.

[0024] Next, this embodiment sets the minimum support threshold and the minimum confidence threshold according to actual needs, and uses the Apriori algorithm to perform multiple iterative scans on the transactional distribution network operation data. Each scan generates a candidate item set, and a new candidate item set is generated by connecting the frequent item sets generated by the previous scan. For each candidate item set, the frequency of its occurrence in all transactions (i.e., support) is calculated. The candidate item sets with support greater than or equal to the minimum support threshold are taken as frequent item sets. For each frequent item set, all possible association rules are extracted by combining its items, such as "If the load demand is in the range of (100MW, 200MW], then the power generation may be in the range of [200MW, 300MW] For each association rule, calculate its confidence (i.e., the probability of the rule consequent appearing under the condition that the rule antecedent appears). According to the confidence calculation result, the association rules with confidence greater than or equal to the minimum confidence threshold are regarded as strong association rules to form a strong association rule set. The rules in the strong association rule set are further analyzed to identify the frequently occurring distribution network energy consumption factors (such as load demand, power generation, carbon emissions, etc.). According to the association strength, determine the key influencing factors affecting the energy consumption of the distribution network. For example, if it is found that there is a strong correlation between load demand and power generation, and the change of load demand has a significant impact on power generation, it can be considered that load demand is one of the key factors affecting the energy consumption of the distribution network.

[0025] S3. Taking the key influencing factors as input features, the pre-built improved long short-term memory neural network model is used to predict the carbon emission forecast value and energy consumption forecast value in the future time period.

[0026] In this embodiment, the step of using the key influencing factors as input features and using a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period includes: Performing principal component analysis on the comprehensive data set of the distribution network operation to extract the principal component feature vectors that affect carbon emissions and energy consumption; using a sliding window method to divide the principal component feature vectors and the key influencing factors according to time series to obtain principal component segments and key influencing factor segments; The principal component fragments and the key influencing factor fragments are input into a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value at each time point in the future time period.

[0027] In this embodiment, the improved long short-term memory neural network model includes an input layer, a one-dimensional convolution layer, a bidirectional LSTM layer, a multi-head attention mechanism layer, a stacked LSTM layer with residual connection, a regularization layer and an output layer connected in sequence. Therefore, the step of inputting the principal component fragment and the key influencing factor fragment into the pre-constructed improved long short-term memory neural network model to predict the carbon emission prediction value and energy consumption prediction value at each time point in the future time period includes: Performing data fusion processing on the principal component fragment and the key influencing factor fragment to obtain characteristic sequence data; Extracting local features from the feature sequence data through a one-dimensional convolutional layer to obtain a local feature sequence; A bidirectional LSTM layer is used to capture the front-end dependency of the local feature sequence to obtain a bidirectional LSTM hidden state sequence; the bidirectional LSTM hidden state sequence is input into a multi-head attention mechanism layer, and the multi-head attention mechanism layer is used to perform attention weighting processing on different time steps to obtain a weighted hidden state sequence; According to the weighted hidden state sequence, a stacked LSTM layer with residual connection is used to learn feature representation to obtain a high-order feature sequence; Regularizing the high-order feature sequence using a regularization layer to obtain a regularized feature sequence; The regularized feature sequence is linearly transformed to obtain the carbon emission prediction value and energy consumption prediction value at each time point in the future time period.

[0028] Specifically, this embodiment uses the principal component analysis (PCA) method to perform dimensionality reduction processing on the comprehensive data set of distribution network operation. PCA calculates the covariance matrix of the comprehensive data set of distribution network operation to identify the main components (i.e., eigenvectors) in the data. These principal components can retain the information in the original data set to the maximum extent. According to the analysis results of PCA, the first few principal components (usually determined according to the size of the eigenvalue or the proportion of explained variance) are selected as the key eigenvectors affecting carbon emissions and energy consumption. These eigenvectors represent the most important information in the original data set, and the principal component eigenvectors are obtained; then, the sliding window size (i.e., the length of the time period) is determined according to the prediction requirements and data characteristics, and the principal component eigenvectors and key influencing factors are divided into non-overlapping time segments according to the time series by using the sliding window method, thereby obtaining principal component segments and key influencing factor segments, each of which contains data within a period of time for subsequent model prediction.

[0029] It should be noted that, for the principal component eigenvector, principal component analysis (PCA) is a statistical method used to convert a set of possibly related variables into a set of linearly unrelated variables through orthogonal transformation, called principal components. These principal components can capture the main change trends in the data and are sorted according to the size of the variance. The principal component eigenvector represents the main information and structure in the data set. It is a dimensionality reduction technology that can reduce the dimension of the data while retaining the most important information; for key influencing factors, the association rule mining algorithm is used to discover interesting relationships between data items, such as frequent patterns, associations, correlations, etc. In the context of distribution network energy consumption and carbon emissions, variables or variable combinations that have a significant impact on energy consumption and carbon emissions can be identified. When the two are used in combination in this embodiment, the principal component eigenvector of PCA can help the model capture the overall trend and structure of the data; the key factors of association rule mining can help the model understand the relationship and interaction between specific variables. This multi-angle information input helps to improve the prediction accuracy and robustness of the model.

[0030] For the prediction process of the improved long short-term memory neural network model, this embodiment performs data fusion processing on the principal component segments and key influencing factor segments in each time segment to form feature sequence data containing all relevant information, inputs the feature sequence data into a one-dimensional convolution layer to capture local patterns and trends in the data, extracts local features of the data by performing convolution operations on the feature sequence data to obtain a local feature sequence, inputs the local feature sequence into a bidirectional LSTM layer, and uses the characteristics of the bidirectional LSTM to calculate hidden states through forward LSTM and backward LSTM respectively to capture the front and back dependencies of the local feature sequence, and concatenates the hidden states calculated by the forward LSTM and backward LSTM respectively to obtain a bidirectional LSTM hidden state sequence, which contains the global information of the time series. In this embodiment, combined with the one-dimensional convolution layer, LSTM can first extract local features of the data, and then capture the front and back dependencies between these local features through the bidirectional LSTM layer to form global features. This combination of local and global feature extraction can improve the prediction accuracy and generalization ability of the model.

[0031] In this embodiment, the bidirectional LSTM hidden state sequence is input into the multi-head attention mechanism layer, and the attention weights between different time steps are calculated to perform weighted attention processing on different time steps to obtain a weighted hidden state sequence, so that the model pays more attention to the time steps that have an important impact on the prediction results, further improving the accuracy of the prediction, and the weighted hidden state sequence is input into the stacked LSTM layer with residual connection. The stacked LSTM layer with residual connection learns high-order feature representation through a multi-layer LSTM structure, and introduces residual connection between each layer of LSTM structure, and adds the input and output to obtain a high-order feature sequence. In this embodiment, residual connection is introduced to alleviate the gradient vanishing problem in the deep network, and the high-order feature sequence is regularized (such as L2 regularization) to reduce the overfitting risk of the model to obtain a regularized feature sequence. The regularized feature sequence is input into the output layer, and the carbon emission prediction value and energy consumption prediction value at each time point in the future time period are obtained through linear transformation. In this embodiment, the improved long short-term memory neural network model is used to accurately predict the carbon emissions and energy consumption of the distribution network in the future time period, providing strong support for the carbon emission reduction optimization scheduling of the distribution network.

[0032] S4. Obtain user electricity usage behavior data, and use K-means clustering algorithm to perform cluster analysis on the user electricity usage behavior data to obtain electricity usage portraits of different categories.

[0033] In this embodiment, the step of using the K-means clustering algorithm to perform cluster analysis on the user's electricity consumption behavior data to obtain electricity consumption portraits of different categories includes: The electricity consumption period of each user is obtained according to the user's electricity consumption behavior data, and the standard deviation of the electricity consumption of each user in each electricity consumption period is calculated to obtain the electricity consumption time fluctuation value; Compare the power consumption time fluctuation value of each user with a preset fluctuation threshold, filter out the power consumption behavior data of users whose power consumption time fluctuation value is greater than the preset fluctuation threshold, and obtain the power consumption behavior data of users to be clustered; According to the electricity consumption time and electricity consumption in the electricity consumption behavior data of the users to be clustered, the average electricity consumption of each user is calculated; According to the average power consumption, the K-means clustering algorithm is used to cluster the power consumption behavior data of the users to be clustered to obtain user portraits of different categories.

[0034] Specifically, in this embodiment, the user's electricity consumption behavior can be recorded in real time through the smart meter system to obtain the user's electricity consumption behavior data, which includes historical electricity consumption, electricity consumption time period and user identification, etc. The obtained user electricity consumption behavior data is cleaned to remove abnormal values ​​and duplicate data, and the data is formatted into a form suitable for subsequent analysis. For example, the electricity consumption is converted into a unified unit (such as kilowatt-hour), and the electricity consumption time period is converted into a timestamp or a standard time period. Then, according to the electricity consumption behavior data of each user, the electricity consumption time period is determined. For example, a day can be divided into multiple time periods (such as one time period per hour), and the electricity consumption in each time period is counted. The standard deviation of the electricity consumption of each user in each electricity consumption time period is calculated as the electricity consumption time fluctuation value. The standard deviation reflects the degree of discreteness of the time distribution of the user's electricity consumption. The specific calculation formula is: In the formula, σ is the standard deviation; N is the number of electricity consumption periods; x i is the electricity consumption in the ith period; μ is the average electricity consumption.

[0035] In this embodiment, a power consumption time fluctuation value threshold is set according to the actual situation of the distribution network, and the power consumption time fluctuation value of each user is compared with the preset fluctuation threshold, and users whose power consumption time fluctuation value is greater than the preset fluctuation threshold are screened out. The power consumption behavior of these users is relatively unstable and they are the main objects of cluster analysis. For the screened users to be clustered, the average power consumption of each user is calculated according to the power consumption time and power consumption in their power consumption behavior data. The number of clusters is determined according to the actual situation and the needs of cluster analysis. In the K-means clustering algorithm, K users to be clustered are randomly selected as initial cluster centers. For each user to be clustered, the Euclidean distance between it and each cluster center is calculated according to the average power consumption, and the average power consumption of each user is calculated. It is assigned to the category to which the nearest cluster center belongs. In each iterative step, the K-means clustering algorithm will recalculate the cluster center of each category and update the cluster center of each category to the average of the average electricity consumption of all users in the category. The above steps are repeated until the cluster center no longer changes or the preset number of iterations is reached. The user's electricity consumption characteristics are extracted according to the final clustering results, and the users to be clustered are divided into K different categories. Each category has corresponding user electricity consumption characteristics, which include electricity consumption time period, average electricity consumption, and electricity peak and valley characteristics. The information such as electricity consumption time period, average electricity consumption, and electricity peak and valley characteristics are integrated into electricity consumption portraits, providing a basis for subsequent carbon emission reduction optimization scheduling.

[0036] S5. Generate a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm based on the electricity consumption profile, the carbon emission forecast value and the energy consumption forecast value.

[0037] In this embodiment, the step of generating a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the power consumption profile, the carbon emission prediction value and the energy consumption prediction value includes: Fuzzy processing is performed on the electricity consumption portrait, the carbon emission prediction value and the energy consumption prediction value to obtain a fuzzy input set; Using a fuzzy logic controller to perform fuzzy reasoning on the fuzzy input set to obtain a fuzzy output set; Defuzzifying the fuzzy output set to obtain fuzzy logic control strategy parameters; The distribution network is regarded as a reinforcement learning environment. The state space includes the current state of the power grid and the parameters of the fuzzy logic control strategy, and the action space is various distribution network carbon emission reduction scheduling strategies. With the goal of minimizing carbon emissions and energy consumption, the optimal action is selected according to the current grid state and fuzzy logic control strategy parameters, and the deep deterministic policy gradient algorithm is used to learn the optimal distribution network carbon emission reduction scheduling strategy under different grid states.

[0038] Specifically, this embodiment standardizes the power consumption profile (such as time series data of user power consumption, distribution of power consumption peak hours, etc.), carbon emission prediction value and energy consumption prediction value to ensure that the data are on the same scale, and then uses fuzzy set theory to convert these standardized data into fuzzy sets. For example, the power consumption in the power consumption profile can be divided into three fuzzy sets of "low", "medium" and "high"; the carbon emission prediction value and energy consumption prediction value can also be similarly divided into "low", "medium" and "high" to convert these data into fuzzy inputs suitable for fuzzy logic processing. The rule base of the fuzzy logic controller is designed based on expert experience and historical data. These rule definitions The fuzzy control strategy that should be output under different fuzzy input combinations is determined. For example, if the electricity consumption profile is "high" and the carbon emission forecast value is "high", a fuzzy control strategy indicating the need to reduce electricity supply or increase the use of renewable energy is output. The fuzzy input set is input into the fuzzy logic controller, and fuzzy reasoning is performed according to the rule base to obtain a fuzzy output set. The fuzzy output set contains a series of fuzzy control strategies for the current input situation, and a defuzzification method such as the centroid method is used to convert the fuzzy output set into specific values ​​or parameters. These parameters are fuzzy logic control strategy parameters, such as the percentage of reducing electricity supply, increasing the priority of renewable energy use, etc.

[0039] Next, this embodiment regards the distribution network as a reinforcement learning environment, wherein the state space includes the current state of the power grid (such as load level, renewable energy generation, energy storage device status, etc.) and fuzzy logic control strategy parameters, and the action space includes various possible distribution network carbon emission reduction scheduling strategies, such as adjusting generator output, scheduling energy storage device charging and discharging, adjusting user electricity consumption plans, etc. The current state of the power grid and the fuzzy logic control strategy parameters are encoded into a form that can be processed by the reinforcement learning algorithm as the input of the algorithm. At the same time, the objective function is defined as minimizing the weighted sum of carbon emissions and energy consumption. The deep deterministic policy gradient (DDPG) algorithm is used for reinforcement learning. The DDPG algorithm combines deep neural networks and deterministic policy gradient methods, and can handle problems in high-dimensional state spaces and continuous action spaces. The DDPG algorithm continuously tries and iteratively updates the policy network to find The optimal distribution network carbon emission reduction scheduling strategy under the same power grid state. During the training process, the DDPG algorithm selects actions according to the current power grid state and fuzzy logic control strategy parameters. After executing the action, the reward (i.e., the negative value of the objective function) and the new state of the environmental feedback are observed, and then the strategy network is updated using this information. After the training is completed, the learned optimal distribution network carbon emission reduction scheduling strategy is applied to the actual distribution network. The optimal action (i.e., scheduling strategy) is selected according to the current power grid state and fuzzy logic control strategy parameters, and the power resource configuration and equipment operation status of each link of the distribution network source (generator), network (transmission line), load (user load), and storage (energy storage equipment) are regulated in real time. Through the above steps, this embodiment realizes the distribution network carbon emission reduction optimization scheduling method based on reinforcement learning algorithm, which can effectively reduce the carbon emissions and energy consumption of the distribution network and improve the utilization efficiency of power resources.

[0040] S6. According to the distribution network carbon emission reduction dispatching strategy, the power resource configuration and equipment operation status of each link of the source, grid, load and storage in the distribution network are controlled in real time.

[0041] Traditional methods have problems such as large consumption of computing resources and long processing time when processing high-dimensional and multi-scale source, grid, load and storage data. This embodiment uses an improved long short-term memory neural network model to achieve accurate prediction of carbon emissions and energy consumption levels in future time periods. At the same time, through a reinforcement learning algorithm, a distribution network carbon emission reduction scheduling strategy is generated that can effectively reduce carbon emissions and improve energy utilization efficiency, thereby achieving real-time processing and accurate prediction of data. According to the generated distribution network carbon emission reduction scheduling strategy, the power resource configuration and equipment operating status of each link of the distribution network source, grid, load and storage are adjusted in real time, ensuring the safe and stable operation of the power grid while improving the operation efficiency of the power grid.

[0042] An embodiment of the present invention provides a method for optimizing and scheduling carbon emission reduction in a distribution network. The method collects carbon emission data, electricity data, and real-time grid operation parameters of each link of the distribution network source, grid, load, and storage to form a comprehensive data set of distribution network operation; uses an association rule mining algorithm to perform association rule analysis on the comprehensive data set of distribution network operation to obtain key influencing factors affecting the energy consumption and carbon emission of the distribution network; uses the key influencing factors as input features, and uses a pre-built improved long short-term memory neural network model to predict carbon emission forecast values ​​and energy consumption forecast values ​​in a future time period; obtains user electricity consumption behavior data, and uses a K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; uses a reinforcement learning algorithm to generate a distribution network carbon emission reduction scheduling strategy based on the electricity consumption portrait, the carbon emission forecast value, and the energy consumption forecast value; and uses the distribution network carbon emission reduction scheduling strategy to regulate the power resource configuration and equipment operation status of each link of the distribution network source, grid, load, and storage in real time. Compared with the existing technology, this method generates a targeted distribution network carbon emission reduction scheduling strategy through an improved long short-term memory neural network model and reinforcement learning algorithm, realizes the accurate prediction and real-time optimization scheduling of distribution network carbon emissions and energy consumption, significantly improves the efficiency and accuracy of distribution network carbon emission management, and achieves a dual improvement in energy conservation and emission reduction and grid operation efficiency.

[0043] It should be noted that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0044] In one embodiment, Figure 2 As shown, an embodiment of the present invention provides a distribution network carbon emission reduction optimization scheduling system, the system comprising: The data collection module 101 is used to collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive data set of distribution network operation; An association analysis module 102 is used to perform association rule analysis on the distribution network operation comprehensive data set using an association rule mining algorithm to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network; The model prediction module 103 is used to use the key influencing factors as input features and use the pre-built improved long short-term memory neural network model to predict the carbon emission prediction value and energy consumption prediction value in the future time period; The cluster analysis module 104 is used to obtain the user's electricity consumption behavior data, and perform cluster analysis on the user's electricity consumption behavior data using a K-means clustering algorithm to obtain electricity consumption portraits of different categories; A strategy generation module 105, configured to generate a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the power consumption profile, the carbon emission prediction value and the energy consumption prediction value; The optimization scheduling module 106 is used to adjust the power resource configuration and equipment operation status of each link of the distribution network source, network, load and storage in real time according to the distribution network carbon emission reduction scheduling strategy.

[0045] For the specific definition of a distribution network carbon emission reduction optimization scheduling system, please refer to the above-mentioned definition of a distribution network carbon emission reduction optimization scheduling method, which will not be repeated here. A person of ordinary skill in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in this application can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0046] The embodiment of the present invention provides a distribution network carbon emission reduction optimization scheduling system, wherein the system collects carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage through a data acquisition module to form a comprehensive distribution network operation data set; the association analysis module uses an association rule mining algorithm to perform association rule analysis on the comprehensive distribution network operation data set to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network; the model prediction module uses the key influencing factors as input features, and uses a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; the clustering analysis module obtains user electricity consumption behavior data, and uses a K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; the strategy generation module uses a reinforcement learning algorithm to generate a distribution network carbon emission reduction scheduling strategy based on the electricity consumption portrait, the carbon emission forecast value and the energy consumption forecast value; the optimization scheduling module adjusts the power resource configuration and equipment operation status of each link of the distribution network source, grid, load and storage in real time according to the distribution network carbon emission reduction scheduling strategy. Compared with the existing technology, this system generates targeted distribution network carbon emission reduction scheduling strategies through an improved long short-term memory neural network model and reinforcement learning algorithm, realizes accurate prediction and real-time optimization scheduling of distribution network carbon emissions and energy consumption, significantly improves the efficiency and accuracy of distribution network carbon emission management, and achieves a dual improvement in energy conservation and emission reduction and grid operation efficiency.

[0047] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for optimizing the scheduling of carbon emission reduction in a distribution network, characterized in that: The following steps are involved: Collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive data set of distribution network operation; Using association rule mining algorithm to analyze the distribution network operation comprehensive data set, the key influencing factors affecting the distribution network energy consumption and carbon emissions are obtained; Taking the key influencing factors as input features, using a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; Acquire user electricity consumption behavior data, and use K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; Generate a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the electricity consumption profile, the carbon emission forecast value and the energy consumption forecast value; According to the distribution network carbon emission reduction dispatching strategy, the power resource configuration and equipment operation status of each link of the source, grid, load and storage in the distribution network are regulated in real time.

2. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 1, characterized in that: The step of collecting carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive distribution network operation data set includes: Collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to obtain the original distribution network operation data set; Performing data cleaning on the original distribution network operation data set to obtain a preprocessed distribution network operation data set; The missing values ​​in the pre-processed distribution network operation data set are filled with valid data at the latest time point to obtain a filled distribution network operation data set; The infilled distribution network operation data set is normalized to obtain a distribution network operation comprehensive data set.

3. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 1, characterized in that: The step of using an association rule mining algorithm to perform association rule analysis on the distribution network operation comprehensive data set to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network includes: Discretizing the distribution network operation comprehensive data set by using an equal-width binning method to obtain discretized distribution network operation comprehensive data; The discrete distribution network operation comprehensive data is converted into a transactional data format to obtain transactional distribution network operation data; wherein each transaction represents a combination of distribution network operation states; Setting a minimum support threshold and a minimum confidence threshold, and using the Apriori algorithm to iteratively scan the transactional distribution network operation data to generate a candidate item set; Calculate the support of each candidate item set, and based on the support, select frequent item sets that meet the minimum support threshold from the candidate item set, and generate association rules based on the frequent item sets; Calculating the confidence of each association rule, and based on the confidence, selecting a set of strong association rules that meet a minimum confidence threshold from the association rules; The energy consumption factors of the distribution network that frequently appear in the strong association rule set are analyzed for their association strength, and the key factors affecting the energy consumption and carbon emissions of the distribution network are identified.

4. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 1, characterized in that: The step of using the key influencing factors as input features and using a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period includes: Performing principal component analysis on the comprehensive data set of the distribution network operation to extract principal component feature vectors that affect carbon emissions and energy consumption; using a sliding window method to divide the principal component feature vectors and the key influencing factors according to time series to obtain principal component segments and key influencing factor segments; The principal component fragments and the key influencing factor fragments are input into a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value at each time point in the future time period.

5. A method for optimizing and dispatching carbon emission reduction in a distribution network as claimed in claim 4, characterized in that: The improved long short-term memory neural network model includes an input layer, a one-dimensional convolution layer, a bidirectional LSTM layer, a multi-head attention mechanism layer, a stacked LSTM layer introducing residual connections, a regularization layer and an output layer connected in sequence.

6. A method for optimizing and dispatching carbon emission reduction in a distribution network as claimed in claim 5, characterized in that: The step of inputting the principal component fragment and the key influencing factor fragment into a pre-built improved long short-term memory neural network model to predict the carbon emission prediction value and energy consumption prediction value at each time point in the future time period includes: Performing data fusion processing on the principal component fragment and the key influencing factor fragment to obtain characteristic sequence data; Extracting local features from the feature sequence data through a one-dimensional convolutional layer to obtain a local feature sequence; A bidirectional LSTM layer is used to capture the front-end dependency of the local feature sequence to obtain a bidirectional LSTM hidden state sequence; the bidirectional LSTM hidden state sequence is input into a multi-head attention mechanism layer, and the multi-head attention mechanism layer is used to perform attention weighting processing on different time steps to obtain a weighted hidden state sequence; According to the weighted hidden state sequence, a stacked LSTM layer with residual connection is used to learn feature representation to obtain a high-order feature sequence; Regularizing the high-order feature sequence using a regularization layer to obtain a regularized feature sequence; The regularized feature sequence is linearly transformed to obtain the carbon emission prediction value and energy consumption prediction value at each time point in the future time period.

7. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 1, characterized in that: The step of using the K-means clustering algorithm to perform cluster analysis on the user's electricity consumption behavior data to obtain electricity consumption portraits of different categories includes: The electricity consumption period of each user is obtained according to the user's electricity consumption behavior data, and the standard deviation of the electricity consumption of each user in each electricity consumption period is calculated to obtain the electricity consumption time fluctuation value; Compare the power consumption time fluctuation value of each user with the preset fluctuation threshold, filter out the power consumption behavior data of users whose power consumption time fluctuation value is greater than the preset fluctuation threshold, and obtain the power consumption behavior data of users to be clustered; According to the electricity consumption time and electricity consumption in the electricity consumption behavior data of the users to be clustered, the average electricity consumption of each user is calculated; According to the average power consumption, the K-means clustering algorithm is used to cluster the power consumption behavior data of the users to be clustered to obtain user portraits of different categories.

8. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 7, characterized in that: Electricity consumption behavior data includes historical electricity consumption, electricity consumption time period and user identification; The electricity consumption profile includes electricity consumption time periods, average electricity consumption, and peak and valley characteristics.

9. A method for optimizing and dispatching carbon emission reduction in a distribution network according to claim 1, characterized in that: The step of generating a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the power consumption profile, the carbon emission prediction value and the energy consumption prediction value comprises: Fuzzy processing is performed on the electricity consumption portrait, the carbon emission prediction value and the energy consumption prediction value to obtain a fuzzy input set; Using a fuzzy logic controller to perform fuzzy reasoning on the fuzzy input set to obtain a fuzzy output set; Defuzzifying the fuzzy output set to obtain fuzzy logic control strategy parameters; The distribution network is regarded as a reinforcement learning environment, the state space includes the current state of the power grid and the parameters of the fuzzy logic control strategy, and the action space is various distribution network carbon emission reduction scheduling strategies; With the goal of minimizing carbon emissions and energy consumption, the optimal action is selected according to the current grid state and fuzzy logic control strategy parameters, and the deep deterministic policy gradient algorithm is used to learn the optimal distribution network carbon emission reduction scheduling strategy under different grid states.

10. A distribution network carbon emission reduction optimization scheduling system, characterized in that: The system comprises: The data acquisition module is used to collect carbon emission data, electricity data and real-time grid operation parameters of each link of the distribution network source, grid, load and storage to form a comprehensive data set of distribution network operation; An association analysis module is used to perform association rule analysis on the distribution network operation comprehensive data set using an association rule mining algorithm to obtain key influencing factors affecting the energy consumption and carbon emissions of the distribution network; A model prediction module is used to use the key influencing factors as input features and use a pre-built improved long short-term memory neural network model to predict the carbon emission forecast value and energy consumption forecast value in the future time period; A clustering analysis module is used to obtain user electricity consumption behavior data, and use a K-means clustering algorithm to perform cluster analysis on the user electricity consumption behavior data to obtain electricity consumption portraits of different categories; A strategy generation module, for generating a distribution network carbon emission reduction scheduling strategy using a reinforcement learning algorithm according to the power consumption profile, the carbon emission forecast value and the energy consumption forecast value; The optimization scheduling module is used to adjust the power resource configuration and equipment operation status of each link of the source, grid, load and storage in the distribution network in real time according to the distribution network carbon emission reduction scheduling strategy.

Images