A multi-modal based microgrid scheduling optimization method and related equipment

Abstract

The present application relates to the technical field of microgrid scheduling, and provides a multi-modal based microgrid scheduling optimization method and related devices. The method includes: obtaining multi-modal data and environmental data of a target microgrid at multiple historical moments; performing pulse coding on all the multi-modal data to obtain pulse features of each modality, and fusing all the pulse features based on all the environmental data to obtain fused features; based on the fused features, using a prediction model to perform prediction to obtain multi-modal prediction data of the target microgrid; obtaining a target scheduling action set of the target microgrid according to the multi-modal prediction data; and performing scheduling optimization on the target microgrid according to the target scheduling action set. The method of the present application can reduce the computational resource consumption for microgrid scheduling optimization.

Description

Technical Field

[0001] The present application relates to the field of microgrid dispatching technology, and in particular to a multi-modal based microgrid dispatching optimization method and related equipment. Background Art

[0002] With the widespread application of renewable energy (such as photovoltaic and wind energy), microgrids, as an important part of the regional power system, can achieve self-sufficiency and interact with the main power grid through source-load prediction and intelligent scheduling. Among them, source-load prediction can grasp the power generation and load conditions in the microgrid in advance, providing a basis for formulating reasonable energy storage charging and discharging plans and market trading strategies; scheduling optimization can achieve efficient operation of microgrids in different time periods based on the prediction results, including peak shaving and valley filling, reducing power fluctuations and maximizing market benefits. However, due to the volatility of renewable energy and the uncertainty of load demand, it is difficult to achieve real-time balanced power generation and load demand prediction in microgrids.

[0003] In the field of micro-charge, the existing source-load prediction methods have weak ability to capture nonlinearity and long-term dependence, and are difficult to cope with complex multimodal data, application scenarios, and changes in special factors. In addition, the existing scheduling optimization methods require a large amount of computing resources and have high real-time requirements, which greatly increases the consumption of computing resources, resulting in resource waste and overall revenue decline. It can be seen that the current microgrid scheduling optimization methods have the problem of high computing resource consumption in microgrid scheduling optimization. Summary of the invention

[0004] The present application provides a multi-modal based microgrid scheduling optimization method and related equipment, which can solve the problem of high computing resource consumption in microgrid scheduling optimization.

[0005] In a first aspect, an embodiment of the present application provides a multi-modal microgrid scheduling optimization method, the microgrid scheduling optimization method comprising:

[0006] Acquire multimodal data and environmental data of the target microgrid at multiple historical moments; the multimodal data includes microgrid data of multiple modes;

[0007] Pulse encoding is performed on all multimodal data to obtain the pulse characteristics of each mode, and all pulse characteristics are fused based on all environmental data to obtain fusion characteristics; pulse characteristics are used to describe the characteristic information of microgrid data corresponding to the mode;

[0008] Based on the fusion features, the prediction model is used to make predictions and obtain the multi-modal prediction data of the target microgrid;

[0009] Obtain a target scheduling action set of the target microgrid according to the multimodal prediction data; the target scheduling action set is used to describe the operations that each microgrid node in the target microgrid needs to perform at a future moment;

[0010] The target microgrid is dispatched and optimized according to the target dispatch action set.

[0011] Optionally, the multiple modalities include multiple data modalities of the microgrid itself and data modalities outside the microgrid;

[0012] Pulse encoding is performed on all multimodal data to obtain the pulse characteristics of each mode, including:

[0013] Using a first feature extraction model, feature extraction is performed on all microgrid data corresponding to the data modalities of the plurality of microgrids themselves, to obtain a first data feature of the data modality of each microgrid itself;

[0014] Using a second feature extraction model, feature extraction is performed on all microgrid data corresponding to data modalities external to the plurality of microgrids to obtain a second data feature of the data modality external to each microgrid;

[0015] Pulse encoding is performed on each first data feature to obtain a pulse feature of the data mode of each microgrid itself;

[0016] Pulse encoding is performed on each second data feature to obtain a pulse feature of a data mode external to each microgrid.

[0017] Optionally, pulse encoding is performed on each first data feature to obtain a pulse feature of the data mode of each microgrid itself, including:

[0018] By formula:

[0019]

[0020]

[0021] Calculate the pulse characteristics of the microgrid's own data mode ;

[0022] in, represents the amplitude normalized value, represents the first data feature, represents the minimum value in the first data feature, represents the maximum value in the first data feature, , Indicates the last historical moment, Indicates the pulse emission frequency corresponding to the amplitude normalized value:

[0023]

[0024] in, Indicates the minimum frequency of pulse emission, Indicates the maximum frequency of pulse emission.

[0025] Optionally, pulse encoding is performed on each second data feature to obtain a pulse feature of a data mode outside each microgrid, including:

[0026] By formula:

[0027]

[0028]

[0029]

[0030] Calculate the pulse characteristics of data modalities outside the microgrid ;

[0031] in, represents the normalized eigenvalue, represents the second data feature, represents the minimum value in the second data feature, represents the maximum value in the second data feature, Indicates the earliest pulse emission time, Indicates the latest pulse emission time, Indicates the pulse emission time.

[0032] Optionally, all pulse features are fused based on all environmental data to obtain fused features, including:

[0033] Perform preliminary fusion of all pulse features to obtain fused pulse features;

[0034] Based on all environmental data, the fused pulse features are updated using a pulse neural network to obtain fused features.

[0035] Optionally, all pulse features are preliminarily fused to obtain fused pulse features, including:

[0036] By formula:

[0037]

[0038] Calculate fusion pulse characteristics ;

[0039] in, , Both represent modal weights, represents the set of pulse features of all microgrid data modes, A collection of pulse signatures representing all microgrid external data modalities.

[0040] Optionally, based on all environmental data, the fused pulse feature is updated using a pulse neural network to obtain a fused feature, including:

[0041] By formula:

[0042]

[0043]

[0044]

[0045] Get fusion features ;

[0046] in, Indicates The membrane potential at a historical moment, Indicates The membrane potential at a historical moment, Indicates The neuron and The strength of the connection between neurons, Indicates the first The value corresponding to the historical moment, represents the time variation, Indicates The membrane potential reset value at the historical moment, represents the membrane potential threshold, represents the membrane potential reset value at the first historical moment, Indicates The membrane potential reset value at the historical moment, , , represents the number of neurons in the spiking neural network, Indicates Environmental constraint adjustment coefficient at a historical moment:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, represents the electricity price impact coefficient, represents the load fluctuation influence coefficient, represents the impact coefficient of renewable energy power generation, represents the energy storage equipment status influence coefficient, represents the electricity price coefficient, Indicates The electricity price data in the environmental data at each historical moment, represents the amplification factor of electricity price fluctuations, Indicates the intensity of electricity price fluctuations, represents the load fluctuation coefficient, Indicates Load demand data from environmental data at historical moments, Indicates the sensitivity of the load change rate to regulation, represents the load change rate, represents the renewable energy coefficient, Indicates Renewable energy generation data in environmental data at historical moments, represents the amplification factor of renewable energy fluctuations, Represents the fluctuation range of renewable energy power generation, Indicates Energy storage battery charge data in environmental data at historical moments, Indicates the sensitivity of energy storage state change to regulation, Indicates the charging status of the energy storage device.

[0053] Optionally, a target dispatch action set of a target microgrid is obtained according to the multimodal prediction data, including:

[0054] The multimodal prediction data is used as the state space, the preset multiple scheduling actions are used as the action space, and each microgrid node in the target microgrid is used as an intelligent agent;

[0055] Using the Q learning algorithm, based on the state space, the target scheduling action corresponding to each intelligent agent is determined from the action space;

[0056] The target dispatching actions corresponding to all intelligent agents are integrated to obtain the target dispatching action set of the target microgrid.

[0057] In a second aspect, an embodiment of the present application provides a multi-modal microgrid scheduling optimization device, comprising:

[0058] The first acquisition module acquires multimodal data and environmental data of the target microgrid at multiple historical moments; the multimodal data includes microgrid data of multiple modes;

[0059] The pulse coding module performs pulse coding on all multimodal data to obtain the pulse characteristics of each mode, and fuses all pulse characteristics based on all environmental data to obtain fusion characteristics; the pulse characteristics are used to describe the characteristic information of the microgrid data corresponding to the mode;

[0060] The prediction module uses the prediction model to make predictions based on the fusion features and obtain the multi-modal prediction data of the target microgrid;

[0061] The second acquisition module acquires a target scheduling action set of the target microgrid according to the multimodal prediction data; the target scheduling action set is used to describe the operations that each microgrid node in the target microgrid needs to perform at a future moment;

[0062] The scheduling optimization module optimizes the scheduling of the target microgrid according to the target scheduling action set.

[0063] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned multi-modal-based microgrid scheduling optimization method when executing the above-mentioned computer program.

[0064] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned multi-modal based microgrid scheduling optimization method.

[0065] The above solution of the present application has the following beneficial effects:

[0066] In an embodiment of the present application, by acquiring multimodal data and environmental data of the target microgrid at multiple historical moments, and then pulse encoding all multimodal data, the pulse characteristics of each mode are obtained, and all pulse characteristics are fused based on all environmental data to obtain fusion characteristics, and then based on the fusion characteristics, prediction is performed using a prediction model to obtain multimodal prediction data of the target microgrid, and then the target scheduling action of the target microgrid is obtained according to the multimodal prediction data, and finally the target microgrid is scheduled and optimized according to the target scheduling action set. Among them, pulse encoding of multimodal data to obtain discrete pulse characteristics can efficiently integrate multimodal data, reduce the consumption of computing resources, and improve the performance in complex tasks. The pulse characteristics are fused based on environmental data to improve the accuracy of feature fusion. The accuracy of multimodal prediction data obtained by prediction based on accurate fusion characteristics is improved. Scheduling optimization is performed based on multimodal prediction data, and the scheduling action at the future moment can be obtained according to the predicted data at the current moment, without the need for strong real-time calculations, which effectively reduces the consumption of computing resources for scheduling optimization.

[0067] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application, 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 A flowchart of a multi-modal microgrid scheduling optimization method provided in an embodiment of the present application;

[0070] Figure 2 A schematic diagram of the structure of a multi-modal microgrid dispatch optimization device provided in one embodiment of the present application;

[0071] Figure 3 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0072] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0073] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0074] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0075] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0076] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0077] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0078] In view of the problem of high computing resource consumption in the existing microgrid scheduling optimization, the embodiment of the present application provides a microgrid scheduling optimization method based on multimodality, which obtains multimodal data and environmental data of the target microgrid at multiple historical moments, then pulse encodes all multimodal data to obtain pulse features of each mode, and fuses all pulse features based on all environmental data to obtain fusion features, and then predicts based on the fusion features using a prediction model to obtain multimodal prediction data of the target microgrid, and then obtains the target scheduling action of the target microgrid according to the multimodal prediction data, and finally optimizes the scheduling of the target microgrid according to the target scheduling action set. Among them, pulse encoding of multimodal data to obtain discrete pulse features can efficiently integrate multimodal data, reduce the consumption of computing resources, and improve the performance in complex tasks. Based on environmental data, pulse features are fused to improve the accuracy of feature fusion. The accuracy of multimodal prediction data obtained by predicting based on accurate fusion features is improved. Scheduling optimization is performed based on multimodal prediction data, and scheduling actions at future moments can be obtained based on predicted data at the current moment, without the need for strong real-time calculations, which effectively reduces the computing resource consumption of scheduling optimization.

[0079] Next, the multi-modal microgrid scheduling optimization method provided in this application is exemplified.

[0080] like Figure 1 As shown, the multi-modal microgrid scheduling optimization method provided in this application includes the following steps:

[0081] Step 11, obtaining multimodal data and environmental data of the target microgrid at multiple historical moments.

[0082] The above multimodal data includes microgrid data of multiple modes. The microgrid data of multiple modes include temperature data, power load record data, renewable energy generation data, time-of-use electricity price data, etc. The environmental data includes electricity price fluctuation data, load demand data, etc. Some microgrid data in the multimodal data may also be included in the environmental data.

[0083] In some embodiments of the present application, multimodal data and environmental data at multiple historical moments may be acquired by accessing a system of a target microgrid.

[0084] It should be noted that after obtaining multimodal data and environmental data, preprocessing is required. First, the sliding window technology is used to align data of different frequencies to a uniform time step. Suppose the time series of microgrid data of each mode at all historical moments is , the window size is W, then the sequence generated by the sliding window is:

[0085]

[0086] Then, the microgrid data is normalized and mapped to the [0,1] interval. The normalization formula is as follows: ;

[0087] in, represents the microgrid data, represents the normalized microgrid data, Indicates the minimum value of the microgrid data, Indicates the maximum value of the microgrid data.

[0088] Outliers (such as extreme data caused by sensor failure) and missing values ​​(such as data loss caused by network interruption) will affect the training and prediction accuracy of the model. Therefore, the box plot method is used to identify extreme points with obvious abnormalities, and then the Z score (Z-Score) is used to detect slight deviations. The K-nearest neighbor classification algorithm (KNN) is used to fill in the detected outliers. Specifically, the box plot method identifies outliers by calculating the quartile IQR of the data. The calculation formula for the quartile IQR is as follows:

[0089]

[0090] in, is the first quartile, which is the value at the lower 25% of the data set. is the third quartile, that is, the value of the 75% position of the data set. The formula for determining outliers is as follows:

[0091]

[0092] Calculate the Z-Score of each data, that is, the standard deviation distance between the data and the mean. , then the data is determined to be an outlier. The calculation method of Z-Score is as follows:

[0093] ;

[0094] in, is the mean of the data set, is the standard deviation.

[0095] After removing the outliers, the KNN algorithm is used to maintain the continuity of the data. That is, the K neighbors that are most similar to the missing value are found and filled according to the mean of these neighbors. The missing values ​​are filled as follows:

[0096]

[0097] in, represents the filling data for missing values, Indicates missing values Neighbor data.

[0098] It is worth mentioning that the quality and integrity of the data can be improved by performing preprocessing operations such as normalization, anomaly removal, and missing value filling on the acquired data.

[0099] Step 12: pulse encode all multimodal data to obtain the pulse features of each modality, and fuse all pulse features based on all environmental data to obtain fused features.

[0100] The above pulse characteristics are used to describe the characteristic information of the microgrid data corresponding to the modality (such as the change status of the microgrid data in time series, etc.). Multiple modalities include multiple microgrid data modalities (such as the power load record of the microgrid, the power generation of renewable energy, etc.) and microgrid external data modalities (such as temperature, time-of-use electricity price, etc.).

[0101] In some embodiments of the present application, the steps of pulse encoding all multimodal data to obtain pulse features of each modality, and fusing all pulse features based on all environmental data to obtain fused features include:

[0102] In the first step, a first feature extraction model is used to extract features from all microgrid data corresponding to the data modalities of the multiple microgrids themselves, so as to obtain a first data feature of the data modality of each microgrid itself.

[0103] Specifically, all microgrid data corresponding to the data modality of each microgrid are input into the first feature extraction model to obtain the first data feature of the data modality.

[0104] Exemplarily, the first feature extraction model can be a feature extraction model that can capture the long-term dependency of data, such as a long short-term memory network (LSTM). The calculation process of LSTM is:

[0105] First, calculate the forget gate to determine how much information in the current cell state needs to be discarded. The calculation formula is as follows:

[0106]

[0107] in, is the sigmoid function, and is a trainable parameter, is the hidden state at the previous time step, is the input of the current time step (i.e., all microgrid data corresponding to the microgrid’s own data modality).

[0108] Calculate input gate , determines how much new information is written to the cell state, and its calculation formula is as follows:

[0109]

[0110] Combine the results of the forget gate and the input gate to update the cell state , and its calculation formula is as follows:

[0111]

[0112] in, is the cell state at the previous time step, represents the Hadamard product, Represents the tanh activation function.

[0113] Calculate output gate , determines the value of the current hidden state, and its calculation formula is as follows:

[0114]

[0115] Combine the cell state and the result of the output gate to calculate the current hidden state , the hidden state of the last time step is the obtained data embedding (i.e., the first data feature). Hidden state The calculation formula is as follows:

[0116]

[0117] In the second step, a second feature extraction model is used to extract features from all microgrid data corresponding to data modalities external to multiple microgrids, and a second data feature of the data modality external to each microgrid is obtained.

[0118] Specifically, all microgrid data corresponding to the data modality outside each microgrid are input into the second feature extraction model to obtain the second data feature of the data modality.

[0119] Exemplarily, the second feature extraction model can be a feature extraction model such as a Transformer model that can capture complex dependencies. The calculation process of the Transformer model is:

[0120] First, the query, key, and value vectors of the structural features are calculated, and then the attention score between the structural features is calculated. The specific formula is as follows:

[0121]

[0122]

[0123] in, is the weight matrix of structural feature attention, is the dimension of the structural features, The data modality external to the microgrid is the microgrid data at a historical moment.

[0124] In order to capture the dependencies between different feature spaces, a multi-head attention mechanism is used:

[0125]

[0126] in, is the calculation result of attention head i, is the output weight matrix of multi-head attention.

[0127] To increase the expressiveness of the model, the output of the multi-head attention mechanism After two layers of feedforward neural network:

[0128]

[0129] in, Represents the encoding vector.

[0130] The encoding vectors corresponding to all historical moments are combined to obtain the second data feature.

[0131] The third step is to pulse encode each first data feature to obtain the pulse feature of the data mode of each microgrid itself.

[0132] Specifically, through the formula:

[0133]

[0134]

[0135] Calculate the pulse characteristics of the microgrid's own data mode .

[0136] in, represents the amplitude normalized value, represents the first data feature, represents the minimum value in the first data feature, represents the maximum value in the first data feature, , Indicates the last historical moment, Indicates the pulse emission frequency corresponding to the amplitude normalized value:

[0137]

[0138] in, Indicates the minimum frequency of pulse emission, Indicates the maximum frequency of pulse emission.

[0139] The fourth step is to pulse encode each second data feature to obtain the pulse feature of the data mode outside each microgrid.

[0140] Specifically, through the formula:

[0141]

[0142]

[0143]

[0144] Calculate the pulse characteristics of data modalities outside the microgrid .

[0145] in, represents the normalized eigenvalue, represents the second data feature, represents the minimum value in the second data feature, represents the maximum value in the second data feature, Indicates the earliest pulse emission time, Indicates the latest pulse emission time, Indicates the pulse emission time.

[0146] The fifth step is to preliminarily fuse all pulse features to obtain fused pulse features.

[0147] Specifically, through the formula:

[0148]

[0149] Calculate fusion pulse characteristics .

[0150] in, , Both represent modal weights, represents the set of pulse features of all microgrid data modes, A collection of pulse signatures representing all microgrid external data modalities.

[0151] In the sixth step, based on all environmental data, the fused pulse features are updated using the pulse neural network to obtain the fused features.

[0152] Specifically, through the formula:

[0153]

[0154]

[0155] ;

[0156] Get fusion features .

[0157] in, Indicates The membrane potential at a historical moment, Indicates The membrane potential at a historical moment, Indicates The neuron and The strength of the connection between neurons, Indicates the first The value corresponding to the historical moment, represents the time variation, Indicates The membrane potential reset value at the historical moment, represents the membrane potential threshold, represents the membrane potential reset value at the first historical moment, Indicates The membrane potential reset value at the historical moment, , , represents the number of neurons in the spiking neural network, Indicates Environmental constraint adjustment coefficient at a historical moment:

[0158]

[0159]

[0160]

[0161]

[0162]

[0163] in, represents the electricity price impact coefficient, represents the load fluctuation influence coefficient, represents the impact coefficient of renewable energy power generation, represents the energy storage equipment status influence coefficient, represents the electricity price coefficient, Indicates The electricity price data in the environmental data at each historical moment, represents the amplification factor of electricity price fluctuations, Indicates the intensity of electricity price fluctuations, represents the load fluctuation coefficient, Indicates Load demand data from environmental data at historical moments, Indicates the sensitivity of the load change rate to regulation, Indicates the load change rate, represents the renewable energy coefficient, Indicates Renewable energy generation data in environmental data at historical moments, represents the amplification factor of renewable energy fluctuations, Represents the fluctuation range of renewable energy power generation, Indicates Energy storage battery charge data in environmental data at historical moments, Indicates the sensitivity of energy storage state change to regulation, Indicates the charging status of the energy storage device.

[0164] It should be noted that when the spiking neural network is used to update the fused pulse feature, the final fused feature is obtained by accumulating the pulse emission frequency of the input signal and the membrane potential update formula through time step. The calculation formula is:

[0165]

[0166] in, Indicates the basic issuance frequency.

[0167] It is understandable that the above fusion features The formula is the expression of the pulse neural network.

[0168] It is worth mentioning that pulse encoding of multimodal data can obtain discrete pulse features, which can efficiently integrate multimodal data, reduce the consumption of computing resources, and improve performance in complex tasks.

[0169] Step 13: Based on the fusion features, a prediction model is used to perform prediction to obtain multimodal prediction data of the target microgrid.

[0170] The multimodal prediction data is the multimodal data of the target microgrid at a future time.

[0171] Specifically, the fused features are input into the prediction model, and the prediction model outputs multimodal prediction data of the target microgrid.

[0172] Exemplarily, the above prediction model may be a support vector machine or other model.

[0173] It should be noted that before proceeding, the prediction model needs to be trained. Use the fusion features as training data and set multiple training tasks for the prediction model. (such as different weather, seasons, etc.), updated by gradient descent to obtain task-specific parameters . The update method is as follows:

[0174]

[0175] in, are the initial parameters of the model, is the intra-task learning rate, yes The gradient of Perform tasks based on training data The loss function after .

[0176] Use the gradient information of all tasks to update the meta parameters , so that it can be fine-tuned more quickly in new tasks, the formula is as follows:

[0177]

[0178] in, is the meta-learning rate and N is the total number of training tasks.

[0179] In new operating environments (such as sudden weather changes or market price fluctuations), the embedding vector is combined with the real-time data at the current moment, input into the meta-learning model for rapid fine-tuning, and the model parameters are updated. . The calculation formula is as follows:

[0180]

[0181] in, is the loss function in the new environment.

[0182] A combined loss function is designed to take into account both the prediction accuracy and scheduling performance of the model. The formula is as follows:

[0183]

[0184] in, , is the weight parameter, is the mean square error, which is used to measure the accuracy of the prediction results. is the scheduling loss function, which is used to balance profitability, stability and user satisfaction. The specific formula is as follows:

[0185]

[0186]

[0187] in, represents the cumulative benefit of the target microgrid, stability represents the stability, Indicates user satisfaction, , and are weights, represents multimodal prediction data, Represents real multimodal data. The evaluation of cumulative benefits is mainly measured by calculating the cumulative benefits over a period of time (such as a week or a month), the evaluation of stability is mainly measured by the standard deviation of microgrid load fluctuations, and the evaluation of user satisfaction is mainly measured by calculating the satisfaction rate of user needs. The calculation formulas of the three evaluation methods are as follows:

[0188]

[0189] in, For the Profits of a historical moment, For the The load of a historical moment, is the mean value of the load, is the amount of electricity actually supplied by the microgrid to users. The electricity demand of the user, Indicates the last historical moment, Indicates the total number of moments in the time period, that is, the total number of times the demand is evaluated (such as the number of time periods per day).

[0190] When the value of the combined loss function is less than the preset value of the loss function, the training is completed, and the prediction model at this time is used to predict the fusion model to obtain multimodal prediction data for subsequent steps. If the value of the combined loss function is greater than or equal to the preset value of the loss function, it returns to update through the gradient descent method to obtain task-specific parameters. steps.

[0191] It is worth mentioning that by setting multiple training tasks to train the prediction model, the accuracy of the prediction model for different scenarios can be improved, and the versatility and reliability of the prediction model can be improved.

[0192] Step 14: Obtain a target scheduling action set of the target microgrid based on the multimodal prediction data.

[0193] The above target scheduling action set is used to describe the operations that each microgrid node in the target microgrid needs to perform at a future moment, such as charging or discharging energy storage equipment, and making decisions on the amount of electricity involved in market transactions.

[0194] Specifically, the multimodal prediction data is used as the state space, the preset multiple scheduling actions are used as the action space, and each microgrid node in the target microgrid is regarded as an intelligent agent. Then, the Q learning algorithm is used to determine the target scheduling action corresponding to each intelligent agent from the action space based on the state space. Finally, the target scheduling actions corresponding to all intelligent agents are integrated to obtain the target scheduling action set of the target microgrid.

[0195] Exemplarily, each microgrid node in the target microgrid is regarded as an intelligent agent, as an independent unit participating in the scheduling optimization, and each intelligent agent has an independent target scheduling action.

[0196] The update formula of the Q value function in the above Q learning algorithm is:

[0197]

[0198] in, Indicates the i-th agent in the current state Execute the scheduling action The Q value, represents the learning rate, Indicates the i-th agent in the current state Execute the scheduling action Instant rewards, represents the discount factor, Indicates the state at the next moment. It represents the set of actions that may be performed in the next state. This means that the agent is in the next state The best action will be selected, which can maximize the Q value (that is, the action that can bring the most future rewards).

[0199] After each interaction, the agent adopts a The greedy strategy chooses the next action as follows:

[0200]

[0201] in, is the expected return after taking action C, To explore the probability, the value is [0,1]. As the training progresses, it gradually decreases The value of makes the model gradually transition from exploration to utilizing existing knowledge.

[0202] Each agent receives an immediate reward after executing the target scheduling action. , which reflects the effect of the current target scheduling action. At the same time, the agent uses long-term rewards Guide its strategy optimization to maximize long-term benefits. When a stable state is reached and the strategy is no longer significantly updated, the model is considered to have converged, and the target scheduling action at this time is used as the final target scheduling action for subsequent steps. The specific formula is as follows:

[0203]

[0204] in, is the discount factor used to balance short-term and long-term benefits, represents the reward for the first k actions.

[0205] Step 15: Optimize the scheduling of the target microgrid according to the target scheduling action set.

[0206] Specifically, the microgrid nodes in the target microgrid are scheduled and optimized according to the target scheduling action in the target scheduling action set. For example, if the target scheduling action is to charge the energy storage device, the energy storage device of the corresponding microgrid node is controlled to charge at a future moment.

[0207] In some embodiments of the present application, in order to improve the overall benefits of the system, a cooperation-competition game model based on Nash equilibrium can also be used to achieve coordination among various intelligent agents. Among them, each intelligent agent can cooperate (peak shaving and valley filling) or compete (maximize its own benefits). Each microgrid node that is an energy storage device acts as an intelligent agent and executes a charging and discharging strategy based on the current market electricity price and load level. Charging is performed when the electricity price is low, and discharging is performed when the load is peak or the electricity price is high, so as to achieve peak shaving and valley filling.

[0208] In the market, each agent increases its own income by seizing trading opportunities to maximize profits while satisfying the system stability constraints;

[0209] The double descent method is used to find the Nash equilibrium, that is, all agents cannot obtain higher returns by unilaterally changing their strategies under a given strategy. The formula for the Nash equilibrium is as follows:

[0210] ;

[0211] in, For the current agent The optimal strategy, is the strategy combination of other agents, For intelligent agents The profit function of .

[0212] It should be noted that during the actual operation of the microgrid, the real-time data of the system is continuously collected and the status of the intelligent agent is updated. At the same time, new data is continuously used for online learning and strategy adjustment to ensure the optimality of the scheduling plan.

[0213] It is worth mentioning that pulse encoding of multimodal data can obtain discrete pulse features, which can efficiently integrate multimodal data, reduce the consumption of computing resources, and improve the performance in complex tasks. The pulse features are fused based on environmental data to improve the accuracy of feature fusion. The accuracy of multimodal prediction data obtained by prediction based on accurate fusion features is improved. Scheduling optimization is performed based on multimodal prediction data. The scheduling actions at future moments can be obtained based on the predicted data at the current moment, without the need for strong real-time calculations, which effectively reduces the computing resource consumption of scheduling optimization.

[0214] In addition, the method of the present application converts the multimodal data of the microgrid into pulse signals after being processed by traditional neural networks such as LSTM and Transformer, and then inputs them into the pulse neural network. This combined method can efficiently integrate multiple types of data and improve the performance of the network in complex tasks;

[0215] Environmental data is introduced to adjust feature fusion, dynamically adjusting the frequency of pulse emission and membrane potential threshold according to real-time environmental changes (such as electricity price fluctuations, load demand, etc.), ensuring that the model can respond quickly to different market conditions and load fluctuations;

[0216] Different from traditional neural network methods, the method of the present application utilizes pulse coding technology to convert important characteristic signals into sparse pulse sequences, which not only reduces the consumption of computing resources but also improves the learning efficiency of time series data.

[0217] The following is an exemplary description of the multi-modal microgrid scheduling optimization device provided in this application.

[0218] like Figure 2 As shown, the embodiment of the present application provides a multi-modal microgrid dispatch optimization device, and the multi-modal microgrid dispatch optimization device 200 includes:

[0219] The first acquisition module 201 acquires multimodal data and environmental data of the target microgrid at multiple historical moments; the multimodal data includes microgrid data of multiple modes;

[0220] The pulse coding module 202 performs pulse coding on all multimodal data to obtain the pulse characteristics of each mode, and fuses all pulse characteristics based on all environmental data to obtain fused characteristics; the pulse characteristics are used to describe the characteristic information of the microgrid data corresponding to the mode;

[0221] The prediction module 203 performs prediction based on the fusion features and uses the prediction model to obtain multi-modal prediction data of the target microgrid;

[0222] The second acquisition module 204 acquires a target scheduling action set of the target microgrid according to the multimodal prediction data; the target scheduling action set is used to describe the operations that each microgrid node in the target microgrid needs to perform at a future moment;

[0223] The scheduling optimization module 205 performs scheduling optimization on the target microgrid according to the target scheduling action set.

[0224] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0225] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0226] like Figure 3 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 3 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0227] Specifically, when the processor D100 executes the computer program D102, it obtains multimodal data and environmental data of the target microgrid at multiple historical moments, then pulse codes all multimodal data to obtain pulse features of each mode, and fuses all pulse features based on all environmental data to obtain fusion features, and then predicts based on the fusion features using a prediction model to obtain multimodal prediction data of the target microgrid, and then obtains the target scheduling action of the target microgrid based on the multimodal prediction data, and finally optimizes the scheduling of the target microgrid based on the target scheduling action set. Among them, pulse coding multimodal data to obtain discrete pulse features can efficiently integrate multimodal data, reduce the consumption of computing resources, and improve performance in complex tasks. Fusion of pulse features based on environmental data improves the accuracy of feature fusion, and the accuracy of multimodal prediction data obtained by prediction based on accurate fusion features is improved. Scheduling optimization is performed based on multimodal prediction data, and scheduling actions at future moments can be obtained based on predicted data at the current moment, without the need for strong real-time calculations, which effectively reduces the consumption of computing resources for scheduling optimization.

[0228] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0229] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0230] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0231] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0232] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the multi-modal microgrid dispatch optimization method device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0233] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0234] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. 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.

[0235] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A multi-modal microgrid dispatch optimization method, characterized in that: include: Acquire multimodal data and environmental data of a target microgrid at multiple historical moments; the multimodal data includes microgrid data of multiple modes; Pulse encoding is performed on all multimodal data to obtain the pulse features of each modality, and all pulse features are fused based on all environmental data to obtain fused features; The pulse feature is used to describe the characteristic information of the microgrid data corresponding to the mode; Based on the fusion features, a prediction model is used to perform prediction to obtain multimodal prediction data of the target microgrid; Acquire a target scheduling action set of the target microgrid according to the multimodal prediction data; the target scheduling action set is used to describe the operations that each microgrid node in the target microgrid needs to perform at a future moment; The target microgrid is scheduled and optimized according to the target scheduling action set.

2. The microgrid dispatch optimization method according to claim 1, characterized in that: The multiple modalities include multiple data modalities of the microgrid itself and data modalities outside the microgrid; The pulse encoding is performed on all multimodal data to obtain the pulse characteristics of each mode, including: Using a first feature extraction model, feature extraction is performed on all microgrid data corresponding to the data modalities of the plurality of microgrids themselves, to obtain a first data feature of the data modality of each microgrid itself; Using a second feature extraction model, feature extraction is performed on all microgrid data corresponding to data modalities external to the plurality of microgrids to obtain a second data feature of the data modality external to each microgrid; Pulse encoding is performed on each of the first data features to obtain a pulse feature of the data mode of each microgrid itself; Pulse encoding is performed on each of the second data features to obtain a pulse feature of the data mode outside each microgrid.

3. The microgrid dispatch optimization method according to claim 2, characterized in that: The pulse encoding of each of the first data features to obtain the pulse features of the data modality of each microgrid itself includes: By formula: Calculate the pulse characteristics of the microgrid's own data mode ; in, represents the amplitude normalized value, represents the first data feature, represents the minimum value in the first data feature, represents the maximum value of the first data feature, , Indicates the last historical moment, Indicates the pulse emission frequency corresponding to the amplitude normalized value: in, Indicates the minimum frequency of pulse emission, Indicates the maximum frequency of pulse emission.

4. The microgrid dispatch optimization method according to claim 3, characterized in that: The pulse encoding of each of the second data features to obtain a pulse feature of a data modality outside each microgrid includes: By formula: Calculate the pulse characteristics of data modalities outside the microgrid ; in, represents the normalized eigenvalue, represents the second data feature, represents the minimum value in the second data feature, represents the maximum value of the second data feature, Indicates the earliest pulse emission time, Indicates the latest pulse emission time, Indicates the pulse emission time.

5. The microgrid dispatch optimization method according to claim 1, characterized in that: The method of fusing all pulse features based on all environmental data to obtain fused features includes: Perform preliminary fusion of all pulse features to obtain fused pulse features; Based on all environmental data, the fused pulse feature is updated using a pulse neural network to obtain a fused feature.

6. The microgrid dispatch optimization method according to claim 5, characterized in that: The preliminary fusion of all pulse features to obtain fused pulse features includes: By formula: Calculate fusion pulse characteristics ; in, , Both represent modal weights, represents the set of pulse features of all microgrid data modes, A collection of pulse signatures representing all microgrid external data modalities.

7. The microgrid dispatch optimization method according to claim 6, characterized in that: The method of updating the fused pulse feature by using a pulse neural network based on all environmental data to obtain a fused feature includes: By formula: Get fusion features ; in, Indicates The membrane potential at each historical moment, Indicates The membrane potential at a historical moment, Indicates The neuron and The strength of the connection between neurons, represents the first The value corresponding to the historical moment, represents the time variation, Indicates The membrane potential reset value at the historical moment, represents the membrane potential threshold, represents the membrane potential reset value at the first historical moment, Indicates The membrane potential reset value at the historical moment, , , represents the number of neurons in the spiking neural network, Indicates Environmental constraint adjustment coefficient at a historical moment: in, represents the electricity price impact coefficient, represents the load fluctuation influence coefficient, represents the impact coefficient of renewable energy power generation, represents the energy storage equipment status influence coefficient, represents the electricity price coefficient, Indicates The electricity price data in the environmental data at each historical moment, represents the amplification factor of electricity price fluctuations, Indicates the intensity of electricity price fluctuations, represents the load fluctuation coefficient, Indicates Load demand data from environmental data at historical moments, Indicates the sensitivity of the load change rate to regulation, Indicates the load change rate, represents the renewable energy coefficient, Indicates Renewable energy generation data in environmental data at historical moments, represents the amplification factor of renewable energy fluctuations, Represents the fluctuation range of renewable energy power generation, Indicates Energy storage battery charge data in environmental data at historical moments, Indicates the sensitivity of energy storage state change to regulation, Indicates the charging status of the energy storage device.

8. The microgrid dispatch optimization method according to claim 1, characterized in that: The step of acquiring a target scheduling action set of the target microgrid according to the multimodal prediction data includes: The multimodal prediction data is used as a state space, the preset multiple scheduling actions are used as an action space, and each microgrid node in the target microgrid is used as an intelligent agent; Using a Q-learning algorithm, based on the state space, a target scheduling action corresponding to each intelligent agent is determined from the action space; The target scheduling actions corresponding to all intelligent agents are integrated to obtain the target scheduling action set of the target microgrid.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the multi-modal based microgrid scheduling optimization method as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multi-modal based microgrid scheduling optimization method as described in any one of claims 1 to 8 is implemented.

Images