Micro-grid multi-target real-time operation optimization scheduling method and system

Abstract

The invention provides a micro-grid multi-target real-time operation optimization scheduling method and system, and relates to the technical field of power scheduling. The TCN and GCN composite prediction model is adopted for prediction, and the influence of supply and demand uncertainty is reduced. The supply and demand prediction model combines a TCN and a GCN, effectively processes periodic and trend components in data, processes random components through a noise filtering technology, greatly improves the prediction precision, can better adapt to the power demand and the complexity of renewable energy output fluctuation, and improves the prediction accuracy. The influence of supply and demand uncertainty on microgrid optimization scheduling is better reduced, and the accuracy of microgrid optimization scheduling is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power dispatching, and in particular to a multi-objective real-time operation optimization dispatching method and system for a microgrid. Background Art

[0002] Microgrid, as a small power system that organically combines distributed power sources (such as solar photovoltaic power generation, wind power generation, etc.), energy storage devices and loads, has received widespread attention and application.

[0003] However, the output power of distributed power sources in microgrids is intermittent, volatile and uncertain, and the load demand also varies randomly, resulting in low accuracy in microgrid operation optimization. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides a microgrid multi-objective real-time operation optimization scheduling method and system, which solves the technical problem of low accuracy of microgrid operation optimization in the prior art.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] In a first aspect, the present invention provides a microgrid multi-objective real-time operation optimization scheduling method, comprising:

[0009] Obtain original load data, renewable energy output data, and meteorological data and pre-process them to obtain time series data;

[0010] Decompose the time series data, extract and optimize the features based on the decomposed data set, and obtain the periodic features, trend features and random features that are highly correlated with the target prediction variables;

[0011] The periodic features are processed by the pre-built time convolution network model to obtain the periodic prediction results; the trend features are processed by the pre-built graph convolution network model to obtain the trend prediction results; the random features are processed by using the noise filtering technology to obtain the random prediction results;

[0012] According to the periodic prediction results, trend prediction results and random prediction results, a microgrid operation optimization scheduling model is constructed and solved using a multi-objective particle swarm algorithm, and the microgrid is optimally scheduled based on the solution results.

[0013] Preferably, the performing sequence decomposition on the time series data includes:

[0014] Using the seasonal decomposition time series technique, the time series data is decomposed into three parts: trend component, seasonal component and random component, which are expressed as follows:

[0015] Y t =T t +S t +R t

[0016] Among them, Y t is the observed value of the time series data at time point t, T t is the trend component, S t is the seasonal component, R t It is the random component.

[0017] Preferably, the pre-constructed time convolutional network model includes: an input layer, multiple convolutional layers and an output layer; wherein the input layer is used to receive periodic features, the multiple convolutional layers are used to capture the time patterns in the periodic features, and the output layer is used to output periodic prediction results.

[0018] Preferably, the pre-built graph convolutional network model includes: an input layer, multiple graph convolutional layers and an output layer; wherein the input layer is used to receive trend features, the multiple graph convolutional layers are used to capture the interdependent relationship between trend features, and the output layer is used to output trend prediction results.

[0019] Preferably, the microgrid multi-objective real-time operation optimization scheduling method also includes: using a proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, and optimizing and updating the microgrid operation optimization scheduling model.

[0020] Preferably, the microgrid operation optimization scheduling model includes an objective function and constraints, and the objective function includes:

[0021] Cost minimization objective function:

[0022] Emission minimization objective function:

[0023] The energy efficiency maximization objective function is converted into a minimization form:

[0024]

[0025] Among them, N G is the number of generators, C i is the unit power generation cost of the i-th generator, P i is its output power, N s is the number of energy storage devices, M j is the operation and maintenance cost of the j-th energy storage device, Qj is its charge and discharge capacity, and the transaction cost C includes the cost of buying and selling electricity with the main power grid; E i is the unit carbon dioxide emission of the i-th generator; N R is the number of renewable energy generators, R k is the energy efficiency of the kth renewable energy generator;

[0026] Among them, P i =P per +P trend +P random , P per ,P trend ,P random ∈(P t ∪P s ∪P r ), that is, the output power includes the periodic prediction result of the energy output power obtained based on the time convolution network model, the corresponding trend prediction result obtained based on the graph convolution network model, and the corresponding random prediction result after noise reduction; P t is the periodic prediction result; P s is the trend prediction result; P r Predict outcomes for randomness;

[0027] Q j =Q per +Q trend +Q random , Q per ,Q trend ,Q random ∈(P t ∪P s ∪P r ), that is, the charge and discharge amount includes the periodic prediction results of the charge and discharge amount of the energy storage system obtained based on the time convolution network model, the corresponding trend prediction results obtained based on the graph convolution network model, and the corresponding random prediction results after noise reduction;

[0028] E i =E per +E trend +E random , E per ,E trend ,E random ∈(P t ∪P s ∪P r ), that is, the unit power generation carbon dioxide emissions include the periodic prediction results of the generator carbon dioxide emissions based on the time convolution network model, the corresponding trend prediction results based on the graph convolution network model, and the corresponding random prediction results after noise reduction;

[0029] The constraints include:

[0030] Power balance constraint: P L (t) = P′ pv (t)+P′ W (t)+P grid (t)+P di (t)+P gas (t)+P ba (t)

[0031] Output constraints of diesel generators:

[0032] Output constraints of micro gas turbines:

[0033] Power constraints for mutual exchange and transmission between main microgrids:

[0034] Energy storage device service life constraints:

[0035] in, are the upper and lower limits of the diesel generator output and the micro gas turbine output, respectively, di , r gas are the upper and lower limits of the diesel generator output and the micro gas turbine climbing power, are the upper and lower limits of the power transmitted by the tie line, The output upper and lower limits of the energy storage device. The positive and negative values ​​represent the input and output of power. SOC min (t), SOC max (t) is the upper and lower limits of the energy storage capacity at time t.

[0036] In a second aspect, the present invention provides a microgrid multi-objective real-time operation optimization scheduling system, comprising:

[0037] The data processing module is used to obtain the original load data, renewable energy output data and meteorological data and perform preprocessing to obtain time series data;

[0038] The feature extraction and optimization module is used to decompose the time series data, and extract and optimize the features based on the decomposed data set to obtain the periodic features, trend features and random features that are highly correlated with the target prediction variables;

[0039] The prediction module is used to process the periodic features through the pre-built time convolution network model to obtain the periodic prediction results; process the trend features through the pre-built graph convolution network model to obtain the trend prediction results; and use the noise filtering technology to process the random features to obtain the random prediction results;

[0040] The optimization scheduling module is used to construct a microgrid operation optimization scheduling model based on periodic prediction results, trend prediction results and random prediction results, and solve it using a multi-objective particle swarm algorithm, and perform microgrid optimization scheduling based on the solution results.

[0041] Preferably, the microgrid multi-objective real-time operation optimization scheduling system further includes:

[0042] The model adjustment module is used to adopt the proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, optimize and update the microgrid operation optimization scheduling model.

[0043] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for a microgrid multi-objective real-time operation optimization scheduling method, wherein the computer program enables a computer to execute the microgrid multi-objective real-time operation optimization scheduling method as described above.

[0044] In a fourth aspect, the present invention provides an electronic device, comprising:

[0045] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include methods for executing the microgrid multi-objective real-time operation optimization scheduling method as described above.

[0046] (III) Beneficial effects

[0047] The present invention provides a microgrid multi-objective real-time operation optimization scheduling method and system. Compared with the prior art, it has the following beneficial effects:

[0048] The present invention adopts the TCN and GCN composite prediction model for prediction to reduce the impact of supply and demand uncertainty. The supply and demand prediction model combines TCN and GCN, effectively processes the periodicity and trend components in the data, and processes the random components through noise filtering technology, which greatly improves the prediction accuracy, making it more adaptable to the complexity of power demand and renewable energy output fluctuations, better reducing the impact of supply and demand uncertainty on microgrid optimization scheduling, and improving the accuracy of microgrid optimization scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A block diagram of a microgrid multi-objective real-time operation optimization scheduling method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] The embodiments of the present application solve the technical problem of low accuracy of microgrid operation optimization in the prior art by providing a microgrid multi-objective real-time operation optimization scheduling method and system, achieve accurate supply and demand forecasting, reduce the impact of supply and demand uncertainty on the optimization scheduling model, and improve the accuracy of microgrid operation optimization.

[0053] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0054] The output power of distributed power sources in microgrids is intermittent, volatile, and uncertain. At the same time, the load demand also changes randomly, which brings huge challenges to the stable operation and optimal scheduling of microgrids. The specific manifestations are as follows:

[0055] First, accurate supply and demand forecasting can better reduce the impact of supply and demand uncertainty and improve the accuracy of microgrid optimization scheduling. Common machine learning algorithms such as recurrent neural networks (RNN), support vector machines (SVM) and long short-term memory networks (LSTM) have certain capabilities in capturing the trend, seasonal changes and periodic characteristics of load data, and can output prediction results with high accuracy. However, these single prediction methods also have their own defects. When dealing with large-scale, high-dimensional and complex structural characteristics of microgrid data, their own defects become more prominent. These algorithms usually need to go through extremely cumbersome and time-consuming feature engineering processes in order to extract key information that is valuable for prediction, and their generalization ability is insufficient when dealing with nonlinear problems, making it difficult to capture the deep-level features and time-series dependencies of the data in all directions. Furthermore, their computational efficiency is relatively low, which makes it difficult to meet the strict real-time requirements of microgrid prediction tasks, and they are also prone to the dilemma of gradient vanishing or gradient exploding, which undoubtedly greatly increases the complexity of model training and the uncertainty of the results. Therefore, in order to better cope with the complexity, intermittency and volatility of multivariate heterogeneous data, different forecasting models can be combined to complement each other and reduce the errors caused by a single forecasting model. At the same time, composite forecasting can dynamically adjust the model combination or parameters according to different operating scenarios, so that the forecasting model can better adapt to various complex and changing operating environments and enhance the adaptability of the model.

[0056] Secondly, in terms of multi-objective optimization scheduling of microgrids, the optimization scheduling of microgrids involves multiple objectives such as economic cost and environmental governance cost, and there are often complex nonlinear relationships and conflicts between these objectives. Although the traditional linear weighted summation method can transform multi-objective problems into single-objective problems for solution, the allocation of weights often depends on the subjective judgment of decision makers and lacks objective basis, which easily leads to instability and deviation of optimization results. At the same time, information may be lost in the normalization process, further affecting the accuracy of the optimization results. On the other hand, although the Pareto optimal solution set method can provide a set containing multiple possible solutions and provide decision makers with a choice space, when the target space is too complex or high-dimensional, the solution set may become very large and difficult to choose. In addition, the Pareto optimal solution set method cannot assign specific weights to each solution, so that some solutions that are more meaningful to decision makers may not be highlighted. At the same time, its computational complexity is high, especially when dealing with large-scale or high-dimensional problems, the solution cost and time cost are high.

[0057] Finally, as time goes by and the operating environment of the microgrid continues to change, such as fluctuations in renewable energy generation and changes in load demand, the prediction accuracy of the original prediction model may decrease. At the same time, the optimization method relies on accurate prediction models to formulate effective optimization strategies. The inaccuracy of the prediction model will directly lead to the failure of the optimization strategy, thereby weakening the optimization effect. In addition, from an economic perspective, not adjusting the model may also cause the economic performance of the microgrid to be damaged, because the original prediction and optimization models may not accurately reflect the changes in key factors such as electricity prices, energy costs, and equipment maintenance costs. Finally, in the face of sudden disturbances or extreme situations, the prediction and optimization models that are not adjusted in time may lack sufficient robustness and cannot effectively cope with these challenges, thereby endangering the overall performance of the system. Therefore, regular model adjustment is essential to ensure the accuracy, economy, and robustness of the microgrid management system.

[0058] From the above description, it can be seen that the existing microgrid operation optimization scheduling mainly includes the following defects:

[0059] 1. The microgrid operation optimization scheduling model is insufficient in dealing with supply and demand uncertainty. The randomness and volatility of renewable energy and the uncertainty of load demand will increase the difficulty of microgrid operation optimization scheduling and have a negative impact on the accuracy of microgrid operation optimization.

[0060] 2. In the multi-objective optimization scheduling of microgrids, the goal of renewable energy utilization rate is rarely taken into consideration, which makes it difficult to achieve the goal of maximizing the utilization of renewable energy, further causing energy waste and affecting the sustainable application of renewable energy in microgrids.

[0061] 3. The existing microgrid optimization scheduling strategy lacks dynamic model updates and is not robust enough to deal with sudden disturbances or extreme situations. Fluctuations in renewable energy generation and constant changes in load demand will reduce the accuracy of microgrid optimization scheduling. At the same time, in the face of sudden disturbances or extreme situations, the microgrid optimization scheduling model cannot be adjusted in time.

[0062] In order to solve the above problems, the embodiment of the present invention proposes a multi-objective real-time operation optimization scheduling method and system for microgrids based on supply and demand forecasting. First of all, in order to reduce the impact of supply and demand uncertainty on the optimization scheduling model, accurate supply and demand forecasting is required. The embodiment of the present invention combines the advantages of temporal convolutional network (TCN) and graph convolutional network (GCN) to construct a composite prediction model for both supply and demand. The combination of TCN's deep mining capability for time series data and GCN's efficient extraction capability for spatial structure information significantly improves the accuracy of supply and demand forecasting, which provides a solid foundation for the coordinated optimization operation of source-grid-load-storage in microgrids.

[0063] Afterwards, the embodiment of the present invention constructs a multi-objective operation optimization scheduling model, which comprehensively considers the three objectives of microgrid operation cost, emission level and renewable energy utilization rate, to ensure that the microgrid can maintain high efficiency, stability and low cost in various operation scenarios and maximize the use of renewable energy. Finally, in order to be able to monitor and analyze the operation status of the system in real time, so as to quickly respond to complex and changing environmental conditions and market demand changes, the embodiment of the present invention also adjusts the model. Through model adjustment, the operation efficiency of the system is further improved, and the resilience and rapid recovery ability to cope with emergencies during operation, as well as the sustainable benefits brought to the ecological environment and society, respond quickly to complex and changing environments, and enhance the adaptability and robustness of the system in complex and changing environments.

[0064] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0065] The embodiment of the present invention provides a multi-objective real-time operation optimization scheduling method for a microgrid. Figure 1 As shown, including:

[0066] S1, obtaining original load data, renewable energy output data and meteorological data and preprocessing them to obtain time series data;

[0067] S2. Decompose the time series data, and extract and optimize the features based on the decomposed data set to obtain the periodic features, trend features and random features that are strongly correlated with the target prediction variables;

[0068] S3. Process the periodic features through the pre-built time convolution network model to obtain the periodic prediction results; process the trend features through the pre-built graph convolution network model to obtain the trend prediction results; use the noise filtering technology to process the random features to obtain the random prediction results;

[0069] S4. Based on the periodic prediction results, trend prediction results and random prediction results, a microgrid operation optimization scheduling model is constructed and solved using a multi-objective particle swarm algorithm, and the microgrid is optimized and scheduled based on the solution results.

[0070] The embodiment of the present invention adopts a TCN and GCN composite prediction model for prediction to reduce the impact of supply and demand uncertainty. The supply and demand prediction model combines TCN and GCN, effectively processes the periodicity and trend components in the data, and processes the random components through noise filtering technology, which greatly improves the prediction accuracy, making it more adaptable to the complexity of power demand and renewable energy output fluctuations, better reducing the impact of supply and demand uncertainty on microgrid optimization scheduling, and improving the accuracy of microgrid optimization scheduling.

[0071] In one embodiment, S1, original load data, renewable energy output data and meteorological data are obtained and pre-processed to obtain time series data. Specifically, it includes:

[0072] Collect raw load data, renewable energy output data (such as solar energy, wind energy, etc.) and meteorological data (such as temperature, wind speed, humidity, etc.).

[0073] Data sources include but are not limited to distributed photovoltaic power generation systems, wind power generation systems, energy storage systems (such as batteries), load equipment (such as household electricity, commercial electricity) and grid connections.

[0074] The system regularly collects real-time data from photovoltaic panels, wind turbines, energy storage devices and load equipment, including power generation, power consumption, battery charge and discharge status, etc., and obtains weather information such as temperature, wind speed, humidity, etc. from the meteorological station.

[0075] The collected data is cleaned to remove outliers (such as data points outside the normal range) and duplicate values ​​(such as multiple records of the same data at the same time point) to obtain time series data. The cleaning process ensures the accuracy and consistency of the data and provides a reliable data basis for subsequent analysis.

[0076] In one embodiment, S2, sequence decomposition is performed on the time series data, and feature extraction and optimization are performed based on the decomposed data set to obtain periodic features, trend features, and random features that are highly correlated with the target prediction variable. Specifically, it includes:

[0077] Identify and separate the cyclical, trend, and random components in time series data. Use seasonal time series (STL) or other decomposition techniques to decompose the original data into three parts:

[0078] Periodic component: reflects the periodic change pattern of data over time.

[0079] Trend component: reflects the long-term changing trend of data over time.

[0080] Random component: represents the uncertainty in the data.

[0081] Decompose time series data into three parts: trend, seasonality and residual. The residual part is the random component.

[0082] Y t =T t +S t +R t

[0083] Among them, Yt is the observed value of the original time series data at time point t (Y t It can be the power load, renewable energy generation, energy storage system status or grid connection status at time point t), T t is the trend component, S t is the seasonal component, R t is the residual (random component).

[0084] Seasonal decomposition time series STL decomposition steps:

[0085] Use the Loess smoothing algorithm to estimate the trend component T from time series data t and seasonal component S t ,

[0086] R t The residual is calculated using the following formula:

[0087] R t =Y t -(T t +S t )

[0088] Use feature engineering techniques to select and optimize the extracted features. Select features that are highly correlated with the target prediction variable through methods such as correlation analysis. Focus on features that are closely related to load forecasting and renewable energy production forecasting, such as historical load, historical renewable energy production, temperature, wind speed, etc. Optimization steps:

[0089] Feature selection: Use feature selection algorithms (such as recursive feature elimination RFE) to determine the most important features. Among them, recursive feature elimination RFE is as follows:

[0090] The initial feature set F contains all features.

[0091] Select a model M (linear regression) and an evaluation indicator E (mean square error).

[0092] Recursively remove one or more features and retrain the model M.

[0093] A feature subset that enables the model M to perform best is selected, and the feature subset includes periodic features, trend features, and random features.

[0094] The selected features are further processed, such as normalization, standardization and other optimization processes, to scale the features to the same scale and obtain the periodic features F that are strongly correlated with the target predictor variable. t , Trend characteristics F s and randomness characteristic F r .

[0095] In one embodiment, S3, through the pre-built time convolutional network model F t Processing is performed to obtain the periodic prediction result P t ; Through the pre-built graph convolutional network model, F s Processing is performed to obtain the trend prediction result P s ; Use noise filtering technology to r Processing is performed to obtain the random prediction result P r . Specifically include:

[0096] The pre-built temporal convolutional network (TCN) model includes an input layer, multiple convolutional layers, and an output layer, where the input layer is used to receive F t ,Multiple convolutional layers are used to capture the temporal patterns in periodic features, and the output layer is used to output periodic prediction results.,This model can effectively capture the temporal dependencies in time series data and is suitable for,prediction of periodic data.

[0097] The input layer receives F t , where F t The periodic features at time t that are strongly correlated with the target predictor variable are obtained through feature selection and optimization.

[0098] In the convolution operation of layer l, causal convolution is used to ensure that the model only sees data before the current time point. The convolution operation can be expressed as:

[0099]

[0100] in, is the output of layer l, is the convolution kernel weight of the lth layer, F t (tk) represents the periodic characteristics of the target predictor variable with strong correlation at time tk (i.e., a historical value of the periodic characteristics of the target predictor variable with strong correlation at time t), b (l) is the bias term of the lth layer, and K is the size of the convolution kernel.

[0101] To alleviate the gradient vanishing problem, TCN uses residual connections. Assume is the input of the lth layer, and the output after residual connection can be expressed as:

[0102]

[0103] Among them, Y t (l) is the final output of layer l, is the feature input to this layer.

[0104] The output after the residual connection is transformed nonlinearly through an activation function (such as ReLU):

[0105]

[0106] The final output layer outputs the prediction result P through the stacking of multiple convolutional layers. t :

[0107]

[0108] Where L is the index of the last layer and f is the activation function of the output layer (such as a linear activation function).

[0109] The pre-built graph convolutional network (GCN) model can capture the dependencies between nodes and is suitable for predicting data with trend changes. The model includes an input layer, multiple graph convolution layers, and an output layer.

[0110] The input layer receives F s , where F s It is a trend feature that is strongly correlated with the target prediction variable obtained through feature selection and optimization.

[0111] Let A be the adjacency matrix of the graph, representing the connection relationship between nodes, and D be the degree matrix, representing the number of connections of each node.

[0112] In the graph convolution operation at layer l, the representation update of the node can be expressed as:

[0113]

[0114] Among them, H (l) is the node representation matrix of the lth layer, H (l-1) is the node representation matrix of the l-1th layer, W (l) is the weight matrix of the lth layer, A ～ =A+I is the adjacency matrix plus the self-loop (I is the identity matrix), D ～ Yes A ～ is the degree matrix, and σ is the activation function (such as ReLU).

[0115] The final output layer will pass through the stacking of multiple graph convolutional layers to output the prediction result P s :

[0116] P s =f(H (L) )

[0117] Where L is the index of the last layer and f is the activation function of the output layer (such as a linear activation function).

[0118] F r It is treated as a noise term in the model to reduce its impact on the prediction results.

[0119] Use noise filtering techniques (such as smoothing filters) to deal with random characteristics.

[0120] Use Savitzky-Golay filter to smooth F r , the formula is:

[0121]

[0122] Among them, F r is a random feature with strong correlation with the target prediction variable obtained through feature selection and optimization, c j are the filter coefficients (usually calculated by the least squares method), and m is half the size of the filter window (which determines the degree of smoothing)

[0123] In one embodiment, S4, based on the periodic prediction results, trend prediction results and random prediction results, a microgrid operation optimization scheduling model is constructed and solved by a multi-objective particle swarm algorithm, and the microgrid optimization scheduling is performed according to the solution results. Specifically, it includes:

[0124] Cost minimization objective function:

[0125] Among them, N G is the number of generators, C i is the unit power generation cost of the i-th generator, P i is its output power, N s is the number of energy storage devices, M j is the operation and maintenance cost of the j-th energy storage device, Q j is its charge and discharge capacity, and the transaction cost C includes the cost of buying and selling electricity with the main grid.

[0126] Among them, P i =P per +P trend +P random , P per ,P trend ,P random ∈(P t ∪P s ∪P r ), that is, the output power includes the periodic prediction result of the energy output power obtained based on the time convolution network model, the corresponding trend prediction result obtained based on the graph convolution network model, and the corresponding random prediction result after noise reduction;

[0127] Q j =Q per +Q trend +Q random , Q per ,Q trend ,Qrandom ∈(P t ∪P s ∪P r ), that is, the charge and discharge amount includes the periodic prediction results of the charge and discharge amount of the energy storage system obtained based on the time convolution network model, the corresponding trend prediction results obtained based on the graph convolution network model, and the corresponding random prediction results after noise reduction;

[0128] Emission minimization objective function:

[0129] Among them, E i is the unit carbon dioxide emission of the i-th generator.

[0130] illustrate:

[0131] E i =E per +E trend +E random , E per ,E trend ,E random ∈(P t ∪P s ∪P r ), that is, the unit power generation carbon dioxide emissions include the periodic prediction results of the generator carbon dioxide emissions based on the time convolution network model, the corresponding trend prediction results based on the graph convolution network model, and the corresponding random prediction results after noise reduction;

[0132] Energy efficiency maximization objective function (converted to minimization form):

[0133] Among them, N R is the number of renewable energy generators, R k is the energy efficiency of the kth renewable energy generator.

[0134] Optimize scheduling constraints:

[0135] Power balance constraint: P L (t) = P′ pv (t)+P′ W (t)+P grid (t)+P di (t)+P gas (t)+P ba (t)

[0136] Output constraints of diesel generators:

[0137] Output constraints of micro gas turbines:

[0138] Power constraints for mutual exchange and transmission between main microgrids:

[0139] Energy storage device service life constraints:

[0140] in, are the upper and lower limits of the diesel generator output and the micro gas turbine output, respectively, di , r gas are the upper and lower limits of the diesel generator output and the micro gas turbine climbing power, are the upper and lower limits of the power transmitted by the tie line, The output upper and lower limits of the energy storage device. The positive and negative values ​​represent the input and output of power. SOC min (t), SOC max (t) is the upper and lower limits of the energy storage capacity at time t to ensure its service life.

[0141] In an embodiment of the present invention, the MOPSO algorithm is used to solve the microgrid operation optimization scheduling model, which specifically includes:

[0142] In the MOPSO algorithm, updating the position and velocity of particles is a key step, and the fitness value of each particle needs to be calculated according to the objective function.

[0143] Position: For each particle, its position can be represented as a vector, where each element corresponds to a decision variable (such as the output power of a generator, the charge and discharge amount of an energy storage device, etc.).

[0144] Let the position of particle p be X p ,but

[0145] Speed: The speed of a particle is also a vector, which indicates the change of each decision variable in each iteration. Let the speed of particle p be V p ,but

[0146] Location Updates:

[0147] Velocity update: Velocity update is performed according to the standard formula of PSO algorithm, such as inertia weight, cognitive part and social part influence.

[0148] Update individual optimal position and global optimal position (Pareto frontier):

[0149] For each particle, compare its current position with the individual optimal position. If the current position is better, update the individual optimal position.

[0150] For the global optimal position, the concept of Pareto front is used for updating, that is, all non-dominated solutions are retained to form a new Pareto front.

[0151] It should be noted that, in the specific implementation process, in order to enable the microgrid operation optimization scheduling model to adapt quickly when facing new data distribution and environmental changes, and maintain its prediction and optimization accuracy, the embodiment of the present invention also includes: S5, using a proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, optimize and update the microgrid operation optimization scheduling model.

[0152] Specifically include:

[0153] In terms of model updating and optimization, the proximal policy optimization (PPO) algorithm is used for real-time dynamic optimal energy management (OEM). With its ability to learn from historical data, the PPO algorithm can accurately capture the uncertainty characteristics of renewable energy generation and load consumption, thereby continuously updating and optimizing the model to ensure accurate and efficient energy management. At the same time, incremental learning and transfer learning techniques are also used to enable the model to adapt quickly when faced with new data distribution and environmental changes, maintaining the accuracy of its prediction and optimization. In terms of performance evaluation, key indicators such as expected insufficient energy (EENS) and loss of load probability (LOLP) are used to comprehensively test the operation effect of the microgrid. In addition, the predictive performance of the model will be carefully evaluated regularly, including multiple dimensions such as prediction error and stability, to ensure continuous optimization and improvement of the system.

[0154] The optimization aspects mainly include:

[0155] Hyperparameter tuning: Adjust the model’s hyperparameters, such as learning rate, regularization parameter, tree depth, etc., through grid search, random search, or Bayesian optimization.

[0156] Feature selection: Select the features that have the greatest impact on the prediction results and remove noise and irrelevant features.

[0157] Feature optimization: Create new features or transform existing features to better capture patterns in the data.

[0158] The embodiment of the present invention further provides a microgrid multi-objective real-time operation optimization scheduling system, comprising:

[0159] The data processing module is used to obtain the original load data, renewable energy output data and meteorological data and perform preprocessing to obtain time series data;

[0160] The feature extraction and optimization module is used to decompose the time series data, and extract and optimize the features based on the decomposed data set to obtain the periodic features, trend features and random features that are highly correlated with the target prediction variables;

[0161] The prediction module is used to process the periodic features through the pre-built time convolution network model to obtain the periodic prediction results; process the trend features through the pre-built graph convolution network model to obtain the trend prediction results; and use the noise filtering technology to process the random features to obtain the random prediction results;

[0162] The optimization scheduling module is used to construct a microgrid operation optimization scheduling model based on periodic prediction results, trend prediction results and random prediction results, and solve it using a multi-objective particle swarm algorithm, and perform microgrid optimization scheduling based on the solution results.

[0163] In the specific implementation process, the microgrid multi-objective real-time operation optimization and scheduling system also includes a model adjustment module, which is used to adopt a proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, optimize and update the microgrid operation optimization scheduling model.

[0164] It can be understood that the microgrid multi-objective real-time operation optimization scheduling system provided in the embodiment of the present invention corresponds to the above-mentioned microgrid multi-objective real-time operation optimization scheduling method. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the microgrid multi-objective real-time operation optimization scheduling method, and will not be repeated here.

[0165] An embodiment of the present invention also provides a computer-readable storage medium storing a computer program for microgrid multi-objective real-time operation optimization scheduling, wherein the computer program enables a computer to execute the microgrid multi-objective real-time operation optimization scheduling method as described above.

[0166] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the program includes a method for executing the microgrid multi-objective real-time operation optimization scheduling method as described above.

[0167] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0168] 1. The embodiment of the present invention adopts the TCN and GCN composite prediction model for prediction to reduce the impact of supply and demand uncertainty. The supply and demand prediction model combines TCN and GCN, effectively processes the periodicity and trend components in the data, and processes the random components through noise filtering technology, which greatly improves the prediction accuracy, making it more adaptable to the complexity of power demand and renewable energy output fluctuations, better reducing the impact of supply and demand uncertainty on microgrid optimization scheduling, and improving the accuracy of microgrid optimization scheduling.

[0169] 2. The embodiment of the present invention proposes a microgrid operation optimization scheduling model that minimizes comprehensive costs, minimizes emissions, and maximizes renewable energy utilization. It can better ensure that the microgrid can maintain high efficiency, stability, and low cost in various operating scenarios and maximize the use of renewable energy.

[0170] 3. The embodiment of the present invention continuously updates and optimizes the model through the proximal strategy optimization algorithm to ensure accurate and efficient energy management. In the face of sudden disturbances or extreme situations, the prediction and optimization model adjusted in real time can respond quickly and show stronger robustness, ensuring that the overall performance of the system is not affected.

[0171] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0172] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective real-time operation optimization scheduling method for a microgrid, characterized in that: include: Obtain original load data, renewable energy output data, and meteorological data and pre-process them to obtain time series data; Decompose the time series data, extract and optimize the features based on the decomposed data set, and obtain the periodic features, trend features and random features that are highly correlated with the target prediction variables; The periodic features are processed by the pre-built time convolutional network model to obtain the periodic prediction results; the trend features are processed by the pre-built graph convolutional network model to obtain the trend prediction results; Use noise filtering technology to process random features and obtain random prediction results; According to the periodic prediction results, trend prediction results and random prediction results, a microgrid operation optimization scheduling model is constructed and solved using a multi-objective particle swarm algorithm, and the microgrid is optimally scheduled based on the solution results.

2. The microgrid multi-objective real-time operation optimization scheduling method according to claim 1, characterized in that: The step of performing sequence decomposition on the time series data includes: Using the seasonal decomposition time series technique, the time series data is decomposed into three parts: trend component, seasonal component and random component, which are expressed as follows: Y t =T t +S t +R t Among them, Y t is the observed value of the time series data at time point t, T t is the trend component, S t is the seasonal component, R t It is the random component.

3. The microgrid multi-objective real-time operation optimization scheduling method according to claim 1, characterized in that: The pre-built time convolutional network model includes: an input layer, multiple convolutional layers and an output layer; wherein the input layer is used to receive periodic features, the multiple convolutional layers are used to capture the time patterns in the periodic features, and the output layer is used to output periodic prediction results.

4. The microgrid multi-objective real-time operation optimization scheduling method according to claim 1, characterized in that: The pre-built graph convolutional network model includes: an input layer, multiple graph convolutional layers and an output layer; wherein the input layer is used to receive trend features, the multiple graph convolutional layers are used to capture the interdependent relationship between trend features, and the output layer is used to output trend prediction results.

5. The microgrid multi-objective real-time operation optimization scheduling method according to claim 1, characterized in that: The microgrid multi-objective real-time operation optimization scheduling method also includes: using a proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, and optimizing and updating the microgrid operation optimization scheduling model.

6. The multi-objective real-time operation optimization scheduling method for a microgrid according to any one of claims 1 to 5, characterized in that: The microgrid operation optimization scheduling model includes an objective function and constraints. The objective function includes: Cost minimization objective function: Emission minimization objective function: The energy efficiency maximization objective function is converted into a minimization form: Among them, N G is the number of generators, C i is the unit power generation cost of the i-th generator, P i is its output power, N s is the number of energy storage devices, M j is the operation and maintenance cost of the j-th energy storage device, Q j is its charge and discharge capacity, and the transaction cost C includes the cost of buying and selling electricity with the main power grid; E i is the unit carbon dioxide emission of the i-th generator; N R is the number of renewable energy generators, R k is the energy efficiency of the kth renewable energy generator; Where Pi = Pper + Ptrend + Prandom, P per ,P trend ,P random ∈(P t ∪P s ∪P r ), that is, the output power includes the periodic prediction result of the energy output power obtained based on the time convolution network model, the corresponding trend prediction result obtained based on the graph convolution network model, and the corresponding random prediction result after noise reduction; P t is the periodic prediction result; P s is the trend prediction result; P r Predict outcomes for randomness; Qj=Qper+Qtrend+Qrandom,Q per ,Q trend ,Q random ∈(P t ∪P s ∪P r ), that is, the charge and discharge amount includes the periodic prediction results of the charge and discharge amount of the energy storage system obtained based on the time convolution network model, the corresponding trend prediction results obtained based on the graph convolution network model, and the corresponding random prediction results after noise reduction; E i =E per +E trend +E random , E per ,E trend ,E random ∈(P t ∪P s ∪P r ), that is, the unit power generation carbon dioxide emissions include the periodic prediction results of the generator carbon dioxide emissions based on the time convolution network model, the corresponding trend prediction results based on the graph convolution network model, and the corresponding random prediction results after noise reduction; The constraints include: Power balance constraint: P L (t) = Pp′v(t) + P W ′(t)+Pgrid(t)+Pdi(t)+Pgas(t)+Pba(t) Output constraints of diesel generators: Output constraints of micro gas turbines: Power constraints for mutual exchange and transmission between main microgrids: Energy storage device service life constraints: in, are the upper and lower limits of the diesel generator output and the micro gas turbine output, respectively, di , r gas are the upper and lower limits of the diesel generator output and the micro gas turbine climbing power, are the upper and lower limits of the power transmitted by the tie line, The output upper and lower limits of the energy storage device. The positive and negative values ​​represent the input and output of power. SOC min (t), SOC max (t) is the upper and lower limits of the energy storage capacity at time t.

7. A microgrid multi-objective real-time operation optimization scheduling system, characterized in that: include: The data processing module is used to obtain the original load data, renewable energy output data and meteorological data and perform preprocessing to obtain time series data; The feature extraction and optimization module is used to decompose the time series data, and extract and optimize the features based on the decomposed data set to obtain the periodic features, trend features and random features that are highly correlated with the target prediction variables; The prediction module is used to process the periodic features through the pre-built time convolution network model to obtain the periodic prediction results; process the trend features through the pre-built graph convolution network model to obtain the trend prediction results; and use the noise filtering technology to process the random features to obtain the random prediction results; The optimization scheduling module is used to construct a microgrid operation optimization scheduling model based on periodic prediction results, trend prediction results and random prediction results, and solve it using a multi-objective particle swarm algorithm, and perform microgrid optimization scheduling based on the solution results.

8. The microgrid multi-objective real-time operation optimization and dispatching system according to claim 7, characterized in that: The microgrid multi-objective real-time operation optimization dispatching system also includes: The model adjustment module is used to adopt the proximal strategy optimization algorithm to perform real-time dynamic optimal energy management, optimize and update the microgrid operation optimization scheduling model.

9. A computer-readable storage medium, characterized in that: It stores a computer program for a microgrid multi-objective real-time operation optimization scheduling method, wherein the computer program enables a computer to execute the microgrid multi-objective real-time operation optimization scheduling method as claimed in any one of claims 1 to 6.

10. An electronic device, characterized in that: include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the microgrid multi-objective real-time operation optimization scheduling method according to any one of claims 1 to 6.

Images