Decision scheduling method, system and equipment for virtual power plant and medium

Abstract

The invention discloses a decision scheduling method, system and device for a virtual power plant and a medium, and the method comprises the steps: obtaining multi-modal data and a current knowledge graph of the virtual power plant; the current knowledge graph comprises a plurality of pieces of first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph in the updating process of the previous knowledge graph, and the second knowledge is an existing knowledge triple reserved by the previous knowledge graph in the updating process of the previous knowledge graph; the knowledge generation time of the second knowledge is smaller than a knowledge elimination threshold; performing feature extraction and fusion on the multi-modal data to obtain multi-modal features; and inputting the multi-modal features and the current knowledge graph into the inference model subjected to fine tuning training for decision inference to obtain a scheduling strategy output by the inference model subjected to fine tuning training. The method can effectively improve the adaptive capacity of the virtual power plant scheduling decision. The invention relates to the technical field of power dispatching.

Description

Technical Field

[0001] This invention relates to the field of power dispatching technology, and in particular to a decision-making and dispatching method, system, equipment, and medium for a virtual power plant. Background Technology

[0002] With the large-scale integration of heterogeneous resources such as distributed energy, adjustable loads, and energy storage systems into virtual power plants, the decision-making and dispatching of virtual power plants has become one of the key areas of focus for power industry professionals.

[0003] Currently, related technologies typically use the knowledge learned by the model during the training phase to make decisions and inferences based on the multimodal data of virtual power plants. This approach is difficult to adapt to the high-frequency changes in the topology and operating rules of virtual power plants, and its adaptability in decision scheduling is limited.

[0004] Therefore, the problems with the relevant technologies still need to be solved and optimized. Summary of the Invention

[0005] The purpose of this invention is to at least partially solve one of the technical problems existing in the related art.

[0006] Therefore, one objective of this invention is to provide a decision-making and scheduling method, system, device, and medium for virtual power plants, wherein the method can effectively improve the adaptability of virtual power plant scheduling decisions.

[0007] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of this application include: In a first aspect, embodiments of this application provide a decision-making and scheduling method for a virtual power plant, including: Acquire multimodal data of a virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold. Feature extraction and fusion are performed on the multimodal data to obtain multimodal features; The multimodal features and the current knowledge graph are input into the fine-tuned inference model for decision-making and inference, and the scheduling strategy output by the fine-tuned inference model is obtained.

[0008] In addition, the method according to the above embodiments of this application may also have the following additional technical features: Furthermore, in one embodiment of this application, the method further includes: Obtain the update conditions and the update indication information corresponding to the scheduling strategy; Based on the update conditions, the update indication information is validated to obtain the update validation result; If the update verification result indicates that the update indication information meets the update condition, then the current knowledge graph is updated according to the multimodal features to obtain the updated knowledge graph.

[0009] Furthermore, in one embodiment of this application, updating the current knowledge graph based on the multimodal features to obtain an updated knowledge graph includes: The multimodal features and the current knowledge graph are input into the fine-tuned and trained reasoning model to perform knowledge reasoning and obtain reasoning knowledge triples; Based on the inference knowledge triplet, perform correlation analysis on each existing knowledge triplet in the current knowledge graph to obtain the new knowledge triplet; The current knowledge graph is subjected to knowledge elimination processing to obtain a knowledge graph after knowledge elimination; Based on the new knowledge triples, the knowledge graph after the knowledge has been eliminated is updated to obtain the updated knowledge graph.

[0010] Furthermore, in one embodiment of this application, the step of performing association analysis on each existing knowledge triple in the current knowledge graph based on the reasoning knowledge triple to obtain the new knowledge triple includes: Based on the inference knowledge triplet, weight analysis is performed on each of the existing knowledge triplets to obtain several association weights, and each association weight corresponds to one of the existing knowledge triplets. Based on all the aforementioned association weights, knowledge fusion is performed on all the existing knowledge triples to obtain the new knowledge triples.

[0011] Furthermore, in one embodiment of this application, the step of performing knowledge elimination processing on the current knowledge graph to obtain a knowledge graph after knowledge elimination includes: Obtain the intermediate triplet, which is any existing knowledge triplet that has not been compared among all the existing knowledge triplets; Based on the knowledge elimination threshold, the knowledge generation time of the intermediate triplet is compared with the threshold to obtain the threshold comparison result. If the threshold comparison result indicates that the knowledge generation time of the middle triplet is less than the knowledge elimination threshold, then the middle triplet is retained; or, if the threshold comparison result indicates that the knowledge generation time of the middle triplet is greater than or equal to the knowledge elimination threshold, then the middle triplet is eliminated.

[0012] Furthermore, in one embodiment of this application, the step of extracting and fusing features from the multimodal data to obtain multimodal features includes: Feature extraction is performed on the multimodal data to obtain structured features and unstructured features, wherein the unstructured features include a first visual feature, a first semantic feature, and a first temporal feature; The unstructured features are subjected to unimodal self-attention processing to obtain second visual features, second semantic features, and second temporal features; Cross-modal mutual attention processing is performed on the second visual feature and the second semantic feature to obtain cross-modal fusion features; The multimodal features are obtained based on the structured features, the second temporal features, and the cross-modal fusion features.

[0013] Furthermore, in one embodiment of this application, the fine-tuned inference model is obtained through the following steps: Acquire multimodal training features, knowledge training graphs, and pre-trained models, and construct a reinforcement learning-large model hybrid decision-maker based on the pre-trained models; The multimodal training features and the knowledge training graph are input into the reinforcement learning-large model hybrid decision maker to obtain the output action; Based on the output action, the near-end policy of the reinforcement learning-large model hybrid decision maker is optimized and updated to obtain the fine-tuned inference model.

[0014] Secondly, embodiments of this application provide a decision-making and scheduling system for a virtual power plant, comprising: The first processing unit is used to acquire multimodal data of the virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold. The second processing unit is used to extract and fuse features from the multimodal data to obtain multimodal features; The third processing unit is used to input the multimodal features and the current knowledge graph into the fine-tuned and trained inference model to perform decision-making inference and obtain the scheduling strategy output by the fine-tuned and trained inference model.

[0015] Thirdly, embodiments of this application also provide an electronic device, including: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method described above.

[0016] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a processor-executable program, which, when executed by the processor, is used to implement the above-described method.

[0017] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application: This application discloses a decision-making and scheduling method, system, device, and medium for a virtual power plant. The method acquires multimodal data of the virtual power plant and a current knowledge graph. The current knowledge graph includes several first and second knowledge sets. The first knowledge sets are new knowledge triples added to the previous knowledge graph during its update, and the second knowledge sets are existing knowledge triples retained by the previous knowledge graph during its update. The knowledge generation time of the second knowledge is less than a knowledge obsolescence threshold. Feature extraction and fusion are performed on the multimodal data to obtain multimodal features. The multimodal features and the current knowledge graph are input into a finely tuned and trained inference model for decision-making and inference, resulting in a scheduling strategy output by the finely tuned and trained inference model. This method uses an inference model based on a continuously updated knowledge graph (i.e., the current knowledge graph) to perform decision-making and inference on multimodal features. This allows the inference model to dynamically adapt to the high-frequency changes in the virtual power plant's topology and operating rules, effectively improving the adaptability of subsequent decision-making and scheduling. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of this application or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart illustrating a decision-making and scheduling method for a virtual power plant provided in an embodiment of this application; Figure 2 A schematic diagram of the framework of a decision-making and scheduling system for a virtual power plant provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Currently, related technologies typically utilize the knowledge learned by the model during the training phase to make decisions based on multimodal data from virtual power plants. This multimodal data can specifically be the operational information of the virtual power plant, which exhibits characteristics such as multiple sources (power electronic equipment data, meteorological data, market electricity prices, etc.), cross-modal (structured data, images and videos, text commands), and high dynamism (second-level data updates). This approach lacks a dynamic evolution mechanism for knowledge and struggles to adapt to the high-frequency changes in the topology and operating rules of virtual power plants. For example, when new types of energy storage devices are connected, traditional models (such as reinforcement learning models) often require several weeks of retraining to converge, failing to meet real-time scheduling requirements, meaning their adaptability in decision-making and scheduling is limited.

[0023] Furthermore, before inputting multimodal data into models for decision-making and inference, related technologies typically process multimodal data through manual feature engineering or machine learning methods. Manual feature engineering can only handle a small amount of structured data within multimodal datasets and struggles to handle the semantic understanding of unstructured data (such as equipment monitoring videos and maintenance log text). For example, rule-based fault diagnosis systems cannot identify novel equipment anomaly patterns, resulting in a high rate of missed diagnoses when atypical inverter faults occur in photovoltaic power plants. While machine learning methods can process some unstructured data, they suffer from insufficient cross-modal feature fusion capabilities. For instance, CNN networks cannot correlate equipment images with concurrent time-series operational parameters, leading to low anomaly localization accuracy in subsequent models.

[0024] It should be noted that the aforementioned related technologies are only used to assist in understanding the technical solutions of this application and do not mean that they belong to the publicly disclosed prior art.

[0025] In view of this, embodiments of this application provide a decision-making and scheduling method, system, device, and medium for a virtual power plant. The method uses a reasoning model based on a continuously updated knowledge graph (i.e., the current knowledge graph). Specifically, it updates the knowledge graph after knowledge elimination by new knowledge triples obtained through association analysis to obtain the current knowledge graph. It performs decision-making reasoning on multimodal features, which enables the reasoning model to dynamically adapt to the high-frequency changes in the topology and operating rules of the virtual power plant, effectively improving the adaptability of subsequent decision-making and scheduling.

[0026] Furthermore, this method performs intramodal self-attention processing on each feature in the unstructured features separately, and performs cross-modal mutual attention fusion on visual and semantic features. It can achieve cross-modal semantic-level fusion of multimodal data of virtual power plants, which is beneficial to improving the decision-making and scheduling capabilities of subsequent models.

[0027] Reference Figure 1 In this embodiment of the application, a decision-making and scheduling method for a virtual power plant includes: Step 110: Obtain multimodal data of the virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold; In this embodiment, multimodal data can be the operational information of a virtual power plant. This operational information, categorized by source, can include power electronic equipment data, meteorological data, market electricity prices, etc. When categorized by mode, the operational data can be structured data and unstructured data. Structured data can include power and voltage data collected by SCADA, while unstructured data can include image data, text data (such as operation log text, fault description text, etc.), and time-series data (such as power time-series data sampled at the second level). The current knowledge graph can be the knowledge graph used by the virtual power plant in the current decision-making and scheduling process. In some embodiments, the method further includes: Obtain the update conditions and the update indication information corresponding to the scheduling strategy; Based on the update conditions, the update indication information is validated to obtain the update validation result; In this embodiment, for any knowledge graph update process, the first step may be to obtain update conditions and update indication information output along with the scheduling strategy output by the inference model. Specifically, update conditions may be any one or more combinations of triggering conditions such as whether the device state significantly deviates from the normal mode, timing period, or new knowledge acquired manually, and update indication information may be indication information corresponding to the update conditions.

[0028] In this embodiment of the application, taking the update condition as whether the device status significantly deviates from the normal mode as an example, the specific judgment condition can be whether the anomaly score is greater than or equal to the anomaly threshold (such as 0.7). The corresponding update indication information can be the anomaly score of a certain device in the virtual power plant, which can be expressed as:

[0029] in, These are abnormal scores; It is a function for maximizing the value; Cosine similarity; Embed the current state of the device (such as embedding the current temperature of the inverter in a virtual power plant). Embed the normal state of the equipment (such as embedding the normal temperature of the inverter in a virtual power plant).

[0030] It is understandable that the anomaly score is calculated by the cosine similarity between the current state embedding and the normal state embedding of the device. When the device state deviates from the normal mode, the smaller the cosine similarity, the higher the corresponding anomaly score (e.g., when the cosine similarity between the inverter temperature embedding and the normal temperature embedding decreases, the anomaly score increases).

[0031] Furthermore, conditional verification can verify whether the update indication information meets the update conditions. For example, when the abnormal score is greater than or equal to the abnormal threshold (such as 0.7), an update verification result indicating that the update indication information meets the update conditions can be generated; or, when the abnormal score is less than the abnormal threshold, an update verification result indicating that the update indication information does not meet the update conditions can be generated, and in this case, the process can return to step 110.

[0032] If the update verification result indicates that the update indication information meets the update condition, then the current knowledge graph is updated according to the multimodal features to obtain the updated knowledge graph.

[0033] Further, updating the current knowledge graph based on the multimodal features to obtain the updated knowledge graph includes: The multimodal features and the current knowledge graph are input into the fine-tuned and trained reasoning model to perform knowledge reasoning and obtain reasoning knowledge triples; In this embodiment, knowledge reasoning can generate several triples corresponding to the current multimodal features (such as "energy storage battery - abnormal temperature - requires derating operation") through the decoder in the reasoning model based on the current multimodal features and knowledge graph. Each triple generated by reasoning is recorded as a reasoning knowledge triple, so as to realize the automatic recognition of the reasoning model in the face of topological structure and / or operating rules.

[0034] Based on the inference knowledge triplet, perform correlation analysis on each existing knowledge triplet in the current knowledge graph to obtain the new knowledge triplet; Further, the step of performing association analysis on each existing knowledge triple in the current knowledge graph based on the inference knowledge triple to obtain the new knowledge triple includes: Based on the inference knowledge triplet, weight analysis is performed on each of the existing knowledge triplets to obtain several association weights, and each association weight corresponds to one of the existing knowledge triplets. Based on all the aforementioned association weights, knowledge fusion is performed on all the existing knowledge triples to obtain the new knowledge triples.

[0035] In this embodiment, for any given inference knowledge triple, association analysis can be performed using a multilayer perceptron to calculate the association weights between the features of the inference knowledge triple and the entity embeddings in each existing knowledge triple. Then, based on each association weight and its corresponding existing knowledge triple, knowledge fusion is performed through a product-sum operation to obtain a new knowledge triple, which can be represented as:

[0036] in, For the first Entity embedding in existing knowledge triples Features in reasoning knowledge triples The association weight; It is a multilayer perceptron; This refers to the integrated knowledge, also known as the new knowledge triplet.

[0037] It is understood that the entity embedding in the embodiments of this application... This can be obtained through large model encoding. For example, for a triple of a certain device, it can be obtained based on the entity name in the triple of that device. (e.g., photovoltaic inverters) and attribute vectors (Such as rated power, operating temperature) After splicing, it is input into the Transformer network to generate entity embeddings. The embedding contains both semantic information (such as inverter type) and state information (such as temperature threshold).

[0038] The current knowledge graph is subjected to knowledge elimination processing to obtain a knowledge graph after knowledge elimination; Furthermore, the step of performing knowledge elimination processing on the current knowledge graph to obtain a knowledge graph after knowledge elimination includes: Obtain the intermediate triplet, which is any existing knowledge triplet that has not been compared among all the existing knowledge triplets; Based on the knowledge elimination threshold, the knowledge generation time of the intermediate triplet is compared with the threshold to obtain the threshold comparison result. If the threshold comparison result indicates that the knowledge generation time of the middle triplet is less than the knowledge elimination threshold, then the middle triplet is retained. Alternatively, if the threshold comparison result indicates that the knowledge generation time of the intermediate triplet is greater than or equal to the knowledge elimination threshold, then the intermediate triplet is eliminated.

[0039] In this embodiment, the knowledge generation time can be the cumulative time from the knowledge generation time of the intermediate triplet to the current time; the knowledge elimination process can be carried out in a cyclical manner, eliminating or retaining existing knowledge triplets in the current knowledge graph. Specifically, for any elimination cycle, the threshold comparison can be to compare whether the knowledge generation time of the intermediate triplet in the current elimination cycle is greater than or equal to the knowledge elimination threshold, and obtain the threshold comparison result.

[0040] Understandably, if the threshold comparison result shows that the knowledge generation time is greater than or equal to the knowledge elimination threshold, it means that the knowledge of the intermediate triplet is relatively old and the intermediate triplet is outdated and invalid. In this case, the intermediate triplet can be eliminated and deleted from the current knowledge graph. Alternatively, if the threshold comparison result shows that the knowledge generation time is less than the knowledge elimination threshold, it means that the knowledge of the intermediate triplet is still valid. In this case, the intermediate triplet can be retained in the current knowledge graph.

[0041] The same logic applies to the intermediate triples in the remaining elimination cycles. After the elimination cycle concludes, all retained intermediate triples are identified as the knowledge graph after knowledge elimination. For example, any intermediate triple can be represented as:

[0042] in, For the first Entity Embedded The function expression corresponding to the middle triplet; For knowledge generation time; This is the knowledge obsolescence threshold.

[0043] Based on the new knowledge triples, the knowledge graph after the knowledge has been eliminated is updated to obtain the updated knowledge graph.

[0044] In this embodiment of the application, the knowledge graph update may involve adding several new knowledge triples to the knowledge graph after knowledge elimination to obtain an updated knowledge graph. This updated knowledge graph can be used as the current knowledge graph in the next decision scheduling process.

[0045] Step 120: Extract and fuse features from the multimodal data to obtain multimodal features; In the embodiments of this application, features of each modality in multimodal data can be extracted and fused to obtain multimodal features.

[0046] In some embodiments, the step of extracting and fusing features from the multimodal data to obtain multimodal features includes: Feature extraction is performed on the multimodal data to obtain structured features and unstructured features, wherein the unstructured features include a first visual feature, a first semantic feature, and a first temporal feature; The unstructured features are subjected to unimodal self-attention processing to obtain second visual features, second semantic features, and second temporal features; Cross-modal mutual attention processing is performed on the second visual feature and the second semantic feature to obtain cross-modal fusion features; The multimodal features are obtained based on the structured features, the second temporal features, and the cross-modal fusion features.

[0047] In the embodiments of this application, multimodal data includes structured data and unstructured data. Structured data may be power, voltage, etc., acquired by SCADA; while unstructured data may be image data, text data, time series data, etc.

[0048] Understandably, for structured data, feature extraction can involve standardizing the structured data to eliminate scale differences between data of different dimensions, thereby obtaining structured features. For unstructured data, feature extraction can involve using a Vision Transformer network to extract visual features from image data, thus obtaining the first visual features; using a BERT network to extract semantic features from text data, thus obtaining the first semantic features; and using a Time Transformer network to extract temporal features from time-series data, thus obtaining the first temporal features.

[0049] Single-modal self-attention processing can be based on the self-attention mechanism, which processes each modal feature (such as visual features, semantic features, or temporal features) in the structured features separately. By calculating the self-attention weight of each modal feature, and then weighting and summing the calculated self-attention weight with the corresponding modal feature, a second visual feature, a second semantic feature, and a second temporal feature that strengthen long-distance dependencies within the same modality can be obtained.

[0050] It should be noted that cross-modal mutual attention processing can be based on the cross-attention mechanism, fusing the second visual feature and the second semantic feature across modally to obtain the cross-modal fused feature, which can be represented as:

[0051] in, For cross-modal fusion features; The attention weight of the second visual feature to the second semantic feature; This is a second semantic feature; The mapping weight matrix for the second semantic feature; It is a secondary visual feature; The mapping weight matrix for the second visual features; For vector dimensions.

[0052] Step 130: Input the multimodal features and the current knowledge graph into the fine-tuned inference model for decision-making and inference, and obtain the scheduling strategy output by the fine-tuned inference model.

[0053] In this embodiment of the application, for the current decision-making and scheduling process, multimodal features and the current knowledge graph can be input into the inference model of fine-tuning the training number. The inference model is used to generate the scheduling strategy for several devices of the virtual power plant, thereby obtaining the scheduling strategy, so that the power grid system can perform scheduling control of the virtual power plant based on the determined scheduling strategy.

[0054] In some embodiments, the fine-tuned inference model is obtained through the following steps: Acquire multimodal training features, knowledge training graphs, and pre-trained models, and construct a reinforcement learning-large model hybrid decision-maker based on the pre-trained models; The multimodal training features and the knowledge training graph are input into the reinforcement learning-large model hybrid decision maker to obtain the output action; Based on the output action, the near-end policy of the reinforcement learning-large model hybrid decision maker is optimized and updated to obtain the fine-tuned inference model.

[0055] In this embodiment, the pre-trained model can be a large language model; the multimodal training features and knowledge training graph are the multimodal features and knowledge graph used for training, respectively. The reinforcement learning-large model hybrid decision-maker can be constructed based on reinforcement learning techniques and the pre-trained model. After inputting the multimodal training features and knowledge training graph into the reinforcement learning-large model hybrid decision-maker, an output action can be obtained, which can be represented as:

[0056] in, For output actions; For pre-trained models; To reinforce deep Q-network functions in learning; For knowledge training graphs; The state space includes multimodal training features and other relevant feature parameters (such as market electricity price, power grid topology, etc.). These are the parameters of the pre-trained model; To enhance the behavioral network function in reinforcement learning.

[0057] The reward function corresponding to this output action can be expressed as:

[0058] in, For instant rewards; , and Weighting coefficients (e.g.) =0.6, =0.3, =0.1); For changes in earnings; For reliability changes; for the number of constraint violations; Let t be the electricity price at time t; This refers to the power regulation amount; To adjust costs; N is the total number of equipment; Let be the health index of the i-th device at time t; Let be the health index of the i-th device at time (t-1); This is an indicator function; it takes the value 1 when a constraint is violated, and 0 otherwise (e.g., 1 when power exceeds the limit).

[0059] The health index in the reward function can be expressed as:

[0060] in, The health index of the equipment (0 indicates a serious malfunction, 1 indicates normal operation); It is a multimodal feature; For knowledge graphs; To embed mappings into knowledge graphs, multimodal features are incorporated. With knowledge graph embedding (Such as equipment health rules) After splicing, it is processed by a multilayer sensor. Output Health Index (HI); A multilayer perceptron for calculating health indices.

[0061] Understandably, after obtaining the output action, the model parameters can be updated using the Proximal Policy Optimization (PPO) algorithm to obtain a finely tuned inference model. Furthermore, in practical applications, knowledge distillation can be performed periodically on the inference model to generate a lightweight inference model that ensures inference accuracy. Using this lightweight model for decision-making can effectively improve the efficiency of inference. Additionally, the Elastic Weight Consolidation (EWC) method can be used to incrementally learn the finely tuned inference model, enabling it to handle new tasks.

[0062] The following describes in detail, with reference to the accompanying drawings, a decision-making and scheduling system for a virtual power plant according to an embodiment of this application.

[0063] Reference Figure 2 The decision-making and scheduling system for a virtual power plant proposed in this application includes: The first processing unit 101 is used to acquire multimodal data of the virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold. The second processing unit 102 is used to extract and fuse features from the multimodal data to obtain multimodal features; The third processing unit 103 is used to input the multimodal features and the current knowledge graph into the fine-tuned and trained inference model to perform decision-making inference and obtain the scheduling strategy output by the fine-tuned and trained inference model.

[0064] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0065] Reference Figure 3 This application also provides an electronic device, including: At least one processor 201; At least one memory 202 is used to store at least one program; When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the method embodiment described above.

[0066] Similarly, it can be understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0067] This application also provides a computer-readable storage medium storing a program executable by a processor 201, which, when executed by the processor 201, is used to implement the above-described method embodiments.

[0068] Similarly, the content of the above method embodiments is applicable to the present computer-readable storage medium embodiments. The specific functions implemented by the present computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0069] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0070] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0071] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0072] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0073] If a function is implemented as 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 technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0075] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0076] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0077] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0078] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0079] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A decision-making and scheduling method for a virtual power plant, characterized in that, include: Acquire multimodal data of a virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold. Feature extraction and fusion are performed on the multimodal data to obtain multimodal features; The multimodal features and the current knowledge graph are input into the fine-tuned inference model for decision-making and inference, and the scheduling strategy output by the fine-tuned inference model is obtained.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the update conditions and the update indication information corresponding to the scheduling strategy; Based on the update conditions, the update indication information is validated to obtain the update validation result; If the update verification result indicates that the update indication information meets the update condition, then the current knowledge graph is updated according to the multimodal features to obtain the updated knowledge graph.

3. The method according to claim 2, characterized in that, The step of updating the current knowledge graph based on the multimodal features to obtain the updated knowledge graph includes: The multimodal features and the current knowledge graph are input into the fine-tuned and trained reasoning model to perform knowledge reasoning and obtain reasoning knowledge triples; Based on the inference knowledge triplet, perform correlation analysis on each existing knowledge triplet in the current knowledge graph to obtain the new knowledge triplet; The current knowledge graph is subjected to knowledge elimination processing to obtain a knowledge graph after knowledge elimination; Based on the new knowledge triples, the knowledge graph after the knowledge has been eliminated is updated to obtain the updated knowledge graph.

4. The method according to claim 3, characterized in that, The step of performing association analysis on each existing knowledge triple in the current knowledge graph based on the inferred knowledge triple to obtain the new knowledge triple includes: Based on the inference knowledge triplet, weight analysis is performed on each of the existing knowledge triplets to obtain several association weights, and each association weight corresponds to one of the existing knowledge triplets. Based on all the aforementioned association weights, knowledge fusion is performed on all the existing knowledge triples to obtain the new knowledge triples.

5. The method according to claim 3, characterized in that, The step of performing knowledge elimination processing on the current knowledge graph to obtain a knowledge graph after knowledge elimination includes: Obtain the intermediate triplet, which is any existing knowledge triplet that has not been compared among all the existing knowledge triplets; Based on the knowledge elimination threshold, the knowledge generation time of the intermediate triplet is compared with the threshold to obtain the threshold comparison result. If the threshold comparison result indicates that the knowledge generation time of the middle triplet is less than the knowledge elimination threshold, then the middle triplet is retained; or, if the threshold comparison result indicates that the knowledge generation time of the middle triplet is greater than or equal to the knowledge elimination threshold, then the middle triplet is eliminated.

6. The method according to claim 1, characterized in that, The step of extracting and fusing features from the multimodal data to obtain multimodal features includes: Feature extraction is performed on the multimodal data to obtain structured features and unstructured features, wherein the unstructured features include a first visual feature, a first semantic feature, and a first temporal feature; The unstructured features are subjected to unimodal self-attention processing to obtain second visual features, second semantic features, and second temporal features; Cross-modal mutual attention processing is performed on the second visual feature and the second semantic feature to obtain cross-modal fusion features; The multimodal features are obtained based on the structured features, the second temporal features, and the cross-modal fusion features.

7. The method according to claim 1, characterized in that, The fine-tuned and trained inference model is obtained through the following steps: Acquire multimodal training features, knowledge training graphs, and pre-trained models, and construct a reinforcement learning-large model hybrid decision-maker based on the pre-trained models; The multimodal training features and the knowledge training graph are input into the reinforcement learning-large model hybrid decision maker to obtain the output action; Based on the output action, the near-end policy of the reinforcement learning-large model hybrid decision maker is optimized and updated to obtain the fine-tuned inference model.

8. A decision-making and dispatching system for a virtual power plant, characterized in that, include: The first processing unit is used to acquire multimodal data of the virtual power plant and the current knowledge graph; the current knowledge graph includes several first knowledge and second knowledge; the first knowledge is a new knowledge triple added to the previous knowledge graph during the previous knowledge graph update process, and the second knowledge is an existing knowledge triple retained by the previous knowledge graph during the previous knowledge graph update process; the knowledge generation time of the second knowledge is less than the knowledge elimination threshold. The second processing unit is used to extract and fuse features from the multimodal data to obtain multimodal features; The third processing unit is used to input the multimodal features and the current knowledge graph into the fine-tuned and trained inference model to perform decision-making inference and obtain the scheduling strategy output by the fine-tuned and trained inference model.

9. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the method as described in any one of claims 1-7.

Images