Power grid thermal stability dynamic adjustment method and system based on combination of large and small models

Abstract

The invention discloses a power grid thermal stability dynamic adjustment method and system based on combination of large and small models. The method comprises the following steps: training a reinforcement learning small model; obtaining the current operation state of the power grid, and matching a corresponding regulation and control plan from the knowledge base according to the current operation state of the power grid when the current operation state of the power grid is an abnormal state; and if the corresponding regulation and control plan does not exist in the knowledge base, outputting a corresponding regulation and control scheme through the trained reinforcement learning small model according to the current operation state of the power grid so as to realize dynamic regulation of the thermal stability of the power grid. The method can effectively utilize the text data in the power system, improves the decision accuracy when the thermal stability balance of the power grid is damaged, can quickly realize the dynamic adjustment of the thermal stability of the power grid, and improves the operation stability of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid regulation and control, and in particular to a method and system for dynamically adjusting thermal stability of a power grid based on the combination of large and small models. Background Art

[0002] In recent years, with the expansion of the proportion of distributed renewable energy and the deepening of market-oriented reforms, the topology and operation status of the power grid have become more and more complex; at the same time, with the continuous changes in the system load level, operation mode and the increase in scale, the impact of unknown factors in the power grid has also become more complex, so the operation of the power grid will face greater challenges. Due to the continuous expansion of power sources and loads in the power grid, as well as the transformation and construction of lines, the operation status and operation mode of the power grid system have changed significantly. With the increase in the proportion of distributed power sources and the increase in power at the receiving end, the load rate of power transmission facilities is also increasing. In this case, when the power of a line reaches the thermal stability limit or exits operation due to a fault, it will cause other lines to overload and exit operation, causing a chain reaction. Therefore, it is necessary to take measures according to the current operating status of the system to maintain the system in a safe and stable operating state. Reinforcement learning is usually used.

[0003] The basic idea of ​​reinforcement learning is that the agent perceives the state of the environment, takes certain actions according to the current state of the environment, changes the state of the environment, gives the agent certain feedback (reward) after the state of the environment changes, and the agent modifies the strategy according to the reward, and repeatedly iterates to find a strategy that maximizes the accumulated expected reward starting from the initial state. However, this traditional reinforcement learning method cannot utilize a large amount of text data in the power system, and cannot fit the model to the requirements in the regulation and adjustment targets, resulting in low data utilization and large decision-making deviation. Summary of the invention

[0004] The present invention aims to solve the technical problems in the related art at least to a certain extent. To this end, the first purpose of the present invention is to provide a method for dynamic adjustment of thermal stability of power grid based on the combination of large and small models, which can effectively utilize the language processing capabilities of text data and large language models in the power system to improve the rationality of decision-making, and can quickly realize dynamic adjustment of thermal stability of power grid when the thermal stability balance of power grid is destroyed, thereby improving the stability of power grid operation.

[0005] The second object of the present invention is to provide a power grid thermal stability dynamic adjustment system based on the combination of large and small models.

[0006] A third object of the present invention is to provide a computer-readable storage medium.

[0007] A fourth object of the present invention is to provide an electronic device.

[0008] To achieve the above object, the present invention is implemented through the following technical solutions:

[0009] A method for dynamic adjustment of power grid thermal stability based on a combination of large and small models, comprising:

[0010] Train a small reinforcement learning model;

[0011] The current operating state of the power grid is obtained. When the current operating state of the power grid is abnormal, a corresponding control plan is matched from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, the corresponding control plan is output through the trained reinforcement learning small model according to the current operating state of the power grid to achieve dynamic adjustment of thermal stability of the power grid.

[0012] Preferably, the training of the small reinforcement learning model includes:

[0013] Generate text descriptions containing grid state adjustment requirements through a language big model;

[0014] Constructing a power grid topology structure and acquiring power grid operation data, wherein the power grid operation data includes power generation output data and load consumption data;

[0015] The reinforcement learning small model is trained based on the text description generated by the language large model, the power grid topology and the power grid operation data.

[0016] Preferably, the generating of a text description containing the power grid state adjustment requirement by means of a language macro model comprises:

[0017] Acquire power grid monitoring data, and generate power grid status alarm information when the power grid monitoring data is abnormal; input the power grid status alarm information into the language model so as to generate a text description containing power grid status adjustment requirements through the language model;

[0018] Construct the grid topology, including:

[0019] Extract the connection terminal identification information of different components and form a busbar and connection terminal identification form;

[0020] Connecting different components based on busbar and connection terminal identification forms to construct the power grid topology;

[0021] Training the reinforcement learning small model based on the text description generated by the language large model, the power grid topology and the power grid operation data includes:

[0022] Determining a power grid operation state based on the power grid topology and the power grid operation data;

[0023] Adjust the grid operation status to an abnormal state;

[0024] For abnormal conditions, the action strategy is adjusted multiple times through the output of the small reinforcement learning model to be trained until the thermal stability adjustment target is achieved, or voltage exceeds the limit, loops, continuous repeated actions, and the action step exceeds the preset step condition;

[0025] For each working condition, the reward value of the small reinforcement learning model to be trained is calculated, and for whether the power grid operation state meets the existing operation regulations and the text description, the corresponding reward value is generated through the large language model;

[0026] Sending the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained can adjust its own network parameters and complete one training;

[0027] Return to the step of adjusting the power grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than a preset threshold, and then end the training.

[0028] Preferably, the generating of a corresponding reward value by the language model for whether the power grid operation status meets the existing operation regulations and the text description includes:

[0029] Inputting the power grid operation status, the existing operation regulations and the text description into the language model to generate corresponding prompt words through the language model, and performing modeling based on the prompt words;

[0030] The model is built to generate new text related to the reward corresponding to the prompt word, and the new text is quantified to generate the corresponding reward value.

[0031] To achieve the above object, the second aspect of the present invention provides a power grid thermal stability dynamic adjustment system based on the combination of large and small models, comprising:

[0032] Training module, used to train small reinforcement learning models;

[0033] The decision-making module is used to obtain the current operating state of the power grid. When the current operating state of the power grid is abnormal, the corresponding control plan is matched from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, then according to the current operating state of the power grid, the corresponding control plan is output through the trained reinforcement learning small model to achieve dynamic adjustment of thermal stability of the power grid.

[0034] Preferably, the training module includes:

[0035] A generating unit, used for generating a text description including a power grid state adjustment requirement through a language large model;

[0036] A construction unit, used for constructing a power grid topology structure;

[0037] An acquisition unit, used to acquire power grid operation data, wherein the power grid operation data includes power generation output data and load consumption data;

[0038] A training unit is used to train the reinforcement learning small model based on the text description generated by the language large model, the power grid topology and the power grid operation data.

[0039] Preferably, when the generating unit generates the text description including the power grid state adjustment requirement through the language large model, it is specifically used to:

[0040] Acquire power grid monitoring data, and generate power grid status alarm information when the power grid monitoring data is abnormal; input the power grid status alarm information into the language model so as to generate a text description containing power grid status adjustment requirements through the language model;

[0041] When constructing the power grid topology structure, the construction unit is specifically used to:

[0042] Extract the connection terminal identification information of different components and form a busbar and connection terminal identification form;

[0043] Connecting different components based on busbar and connection terminal identification forms to construct the power grid topology;

[0044] When the training unit trains the reinforcement learning small model based on the text description generated by the language large model, the power grid topology structure and the power grid operation data, it is specifically used to:

[0045] Determining a power grid operation state based on the power grid topology and the power grid operation data;

[0046] Adjust the grid operation status to an abnormal state;

[0047] For abnormal conditions, the action strategy is adjusted multiple times through the output of the small reinforcement learning model to be trained until the thermal stability adjustment target is achieved, or voltage exceeds the limit, loops, continuous repeated actions, and the action step exceeds the preset step condition;

[0048] For each working condition, the reward value of the small reinforcement learning model to be trained is calculated, and for whether the power grid operation state meets the existing operation regulations and the text description, the corresponding reward value is generated through the large language model;

[0049] Sending the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained can adjust its own network parameters and complete one training;

[0050] Return to the step of adjusting the power grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than a preset threshold, and then end the training.

[0051] Preferably, when the training unit generates a corresponding reward value through the language model for determining whether the power grid operation status meets the existing operation regulations and the text description, the training unit is specifically used to:

[0052] Inputting the power grid operation status, the existing operation regulations and the text description into the language model to generate corresponding prompt words through the language model, and performing modeling based on the prompt words;

[0053] The model is built to generate new text related to the reward corresponding to the prompt word, and the new text is quantified to generate the corresponding reward value.

[0054] To achieve the above objectives, the third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned method for dynamic adjustment of thermal stability of a power grid based on the combination of large and small models is implemented.

[0055] To achieve the above-mentioned purpose, the fourth aspect of the present invention provides a power grid control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned power grid thermal stability dynamic adjustment method based on the combination of large and small models is implemented.

[0056] The present invention has at least the following technical effects:

[0057] The present invention provides a power grid thermal stability dynamic adjustment scheme based on the combination of large and small models. The scheme adopts an algorithm design combining large and small models. The reinforcement learning small model is used to generate the switch action, i.e., the control scheme, under abnormal power grid conditions. The language large model converts the abstract text descriptions such as existing text regulations and adjustment targets into reward values ​​that can be understood by the reinforcement learning small model, so as to realize the training of the reinforcement learning small model and assist in judging the state of the power grid. When the thermal stability limit of the power grid is destroyed, it can be preferentially matched whether there is a relevant plan in the knowledge base. If not, the reinforcement learning small model agent is called to generate it quickly, so as to quickly realize the dynamic adjustment of the thermal stability of the power grid. Therefore, the present invention can effectively utilize the text data in the power system and the language processing capabilities of the language large model to improve the rationality of decision-making, and can quickly realize the dynamic adjustment of the thermal stability of the power grid when the thermal stability balance of the power grid is destroyed, thereby improving the stability of the power grid operation.

[0058] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The present invention is a flowchart of a method for dynamically adjusting thermal stability of a power grid based on a combination of large and small models according to an embodiment of the present invention.

[0060] Figure 2 This is a flowchart of the training of a small reinforcement learning model according to an embodiment of the present invention.

[0061] Figure 3 It is a structural block diagram of a power grid thermal stability dynamic adjustment system based on the combination of large and small models according to an embodiment of the present invention.

[0062] Figure 4 4 is a structural block diagram of a training module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The present embodiment is described in detail below, and examples of the embodiment are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0064] The following describes the method and system for dynamic adjustment of power grid thermal stability based on the combination of large and small models according to the present embodiment with reference to the accompanying drawings.

[0065] Figure 1 FIG. 1 is a flow chart of a method for dynamically adjusting thermal stability of a power grid based on a combination of large and small models according to an embodiment of the present invention. Figure 1 As shown, the power grid thermal stability dynamic adjustment method based on the combination of large and small models includes:

[0066] Step S1: Train a small reinforcement learning model.

[0067] In this embodiment, training the small reinforcement learning model includes: generating a text description containing grid status adjustment requirements through a large language model; constructing a grid topology structure and obtaining grid operation data, the grid operation data including power generation output data and load consumption data; and training the small reinforcement learning model based on the text description, grid topology structure and grid operation data generated by the large language model.

[0068] In one embodiment of the present invention, a text description containing the need for adjusting the power grid status is generated through a language big model, including: acquiring power grid monitoring data, and generating power grid status alarm information when the power grid monitoring data is abnormal; inputting the power grid status alarm information into the language big model, so as to generate a text description containing the need for adjusting the power grid status through the language big model.

[0069] Specifically, the grid monitoring data can be read, and then it can be determined whether the grid monitoring data is abnormal. If an abnormality occurs, such as the load rate of some lines is too high, in a heavy load or overload state, the operation mode of the grid needs to be adjusted to balance the load rates of different lines, maintain voltage balance, and make the grid meet stability requirements. Among them, when it is determined that the grid monitoring data is abnormal, a grid status alarm information will be generated, and then the grid status alarm information will be input into a language model such as ChatGPT (a language model), and a text description containing the grid status adjustment requirements and corresponding business requirements can be generated through the language model.

[0070] In one embodiment of the present invention, constructing a power grid topology structure includes: extracting connection terminal identification information of different components to form a bus and connection terminal identification form; connecting different components based on the bus and connection terminal identification form to construct the power grid topology structure.

[0071] Specifically, the physical node information of different components such as busbars or transformers, i.e., the connection terminal identification information of switches or other devices, can be extracted to form a busbar and connection terminal identification form, and then different components can be connected according to the busbar and connection terminal identification form to obtain the power grid topology. Among them, for a small number of components with missing physical node data, the physical nodes can be estimated through their topological nodes.

[0072] It should be noted that after constructing the grid topology, the grid operation data is also obtained, in which the real power generation and load data are used as the output and consumption of the grid. The missing grid operation data can be supplemented by interpolation, state estimation and other methods.

[0073] In one embodiment of the present invention, the steps of training the reinforcement learning small model based on the text description generated by the language large model, the power grid topology and the power grid operation data are as follows: Figure 2 As shown, including:

[0074] S101: Determine a power grid operation state based on a power grid topology structure and power grid operation data.

[0075] S102: Adjust the operating state of the power grid to an abnormal state.

[0076] S103: For abnormal conditions, the action strategy is adjusted multiple times through the output of the small reinforcement learning model to be trained until the thermal stability adjustment target is achieved, or the voltage exceeds the limit, loops, continuous repeated actions, and the action step exceeds the preset step condition.

[0077] S104: For each operating condition, the reward value of the small reinforcement learning model to be trained is calculated, and the corresponding reward value is generated through the large language model for whether the power grid operation status meets the existing operation regulations and text descriptions.

[0078] S105: Send the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained can adjust its own network parameters and complete one training.

[0079] S106: Return to the step of adjusting the grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than a preset threshold, and then terminate the training.

[0080] Specifically, the small reinforcement learning model to be trained can read environmental parameters, including the grid topology and grid operation data, and then adjust according to the current grid operation state to make it enter an abnormal state. Subsequently, the small reinforcement learning model to be trained randomly makes some action explorations and gives corresponding switching actions, that is, makes an action strategy to adjust the grid operation state. In this embodiment, the switch action space function of the small reinforcement learning model to be trained is expressed as:

[0081] A=[a0,a1,a2,…,a n ](1)

[0082] Among them, a n As a whole, it represents the switch number that needs to be acted upon, and A represents the switch action space function.

[0083] The grid operation state is changed accordingly according to the given action. In this embodiment, the grid operation state space function is:

[0084] S=[B,P i ](2)

[0085] Among them, B represents the branch switch variable state set, P i represents the power of the ith branch, and S represents the grid operation state space function.

[0086] Perform actions multiple times until the termination conditions of a round of training are met: the thermal stability adjustment target is completed, or the following situations occur during the adjustment process: voltage exceeds the limit, loop, continuous repeated actions, action step exceeds the preset step, fault switch action, and duration exceeds the limit.

[0087] Then, for each working condition, the reward value of the small reinforcement learning model to be trained is calculated.

[0088] For example, for the condition where the line load rate is too high, the reward value of the small reinforcement learning model to be trained under the condition where the line load rate is too high can be calculated:

[0089]

[0090] Among them, R P Represents the reward value under the condition of excessive line load rate, Pr0 Represents the load rate of the branch with the highest line load rate before adjustment, P r1 Represents the load factor of the branch with the highest line load factor in the adjusted power grid.

[0091] For another example, the reward value of the small reinforcement learning model to be trained under the voltage over-limit condition is calculated as follows:

[0092]

[0093] Among them, R U represents the voltage over-limit reward (penalty), U represents the node voltage, and U max Represents the voltage upper limit per unit value, U min Represents the voltage lower limit per unit value, where the nodes are the physical nodes mentioned above, including switches and other devices.

[0094] Among them, the corresponding reward value can be generated by the language model for whether the power grid operation status meets the existing operation regulations and the text description generated by the language model.

[0095] It should be noted that regardless of whether the power grid operation status is satisfied, the corresponding reward value needs to be generated through the language model. That is, if the power grid operation status does not meet the existing operation regulations and text descriptions, a smaller reward value is generated. If the power grid operation status meets the existing operation regulations and text descriptions, a larger reward value is generated. For this case, the language model generates the reward value process as follows:

[0096] The grid operation status, existing operation regulations and text descriptions are input into the language model to generate corresponding prompt words through the language model, and modeling is performed based on the prompt words; new text related to rewards corresponding to the prompt words is generated through the modeling model, and the new text is quantified to generate corresponding reward values.

[0097] Specifically, the grid operation status, existing operation regulations and text descriptions can be input into the language model to generate prompt words, and then the prompt words are modeled using the concept of token (representing the smallest unit of data or text), so that new text related to rewards corresponding to the prompt words can be generated through the modeling model, and then the generated new text can be parsed into reward values. The calculation formula is as follows:

[0098] R(text)=parse{t1,t2,...,t m |a,s,c}(5)

[0099] Where R(text) is the reward value generated by the new text, t1~t mis a series of tokens obtained from the vocabulary by applying the probability chain rule, parse{} is the parsing function, a is the action of the power grid load transfer model; s is the environment of the target power grid system, and c is the prompt word context.

[0100] After obtaining the reward values ​​under various working conditions, the calculated or generated reward values ​​can be sent to the small reinforcement learning model to be trained, and the small reinforcement learning model to be trained changes its own network parameters according to the reward values, thereby completing this round of training. After that, return to reset the electrical network, reset the abnormal state, and conduct the next round of training until the reward value tends to be stable, completing the training process of the small reinforcement learning model to be trained, and obtaining the optimal small reinforcement learning model.

[0101] Step S2: Obtain the current operating state of the power grid. When the current operating state of the power grid is abnormal, match the corresponding control plan from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, then according to the current operating state of the power grid, output the corresponding control plan through the trained reinforcement learning model to achieve dynamic adjustment of the thermal stability of the power grid.

[0102] During the actual operation of the power grid, the current operating status of the power grid can be obtained. The large language model searches the knowledge base to determine whether the current operating status of the power grid conforms to the scenario described by a certain regulation plan. If there is a matching regulation plan, the disposal process in the regulation plan is directly output. If there is no matching regulation plan, the current operating status data of the power grid is further input into the trained reinforcement learning small model, and the reinforcement learning small model gives a regulation plan.

[0103] It should be noted that no matter it is the control plan matched in the knowledge base or the control plan given by the reinforcement learning model, after output, it is necessary to call the simulation API (application programming interface) to verify the disposal plan. After the verification is successful, the operation ticket can be generated and enter the action stage.

[0104] Among them, if the reinforcement learning model is called in the process of generating a disposal plan, the control plan newly generated by the reinforcement learning model and the corresponding abnormal state of the power grid need to be stored in the knowledge base.

[0105] Figure 3 FIG. 1 is a structural block diagram of a power grid thermal stability dynamic adjustment system based on a combination of large and small models according to an embodiment of the present invention. Figure 3 As shown, the power grid thermal stability dynamic adjustment system 100 based on the combination of large and small models includes a training module 10 and a decision module 20 that are interconnected.

[0106] Among them, the training module 10 is used to train the reinforcement learning model. The decision module 20 is used to obtain the current operating state of the power grid. When the current operating state of the power grid is abnormal, the corresponding control plan is matched from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, the corresponding control plan is output according to the current operating state of the power grid through the trained reinforcement learning model to achieve dynamic adjustment of thermal stability of the power grid.

[0107] Figure 4 FIG. 1 is a structural block diagram of a training module according to an embodiment of the present invention. Figure 4 As shown, the training module 10 includes a generating unit 11, a constructing unit 12, an acquiring unit 13 and a training unit 14. The generating unit 11, the constructing unit 12 and the acquiring unit 13 are all connected to the training unit 14.

[0108] The generation unit 11 is used to generate a text description containing grid status adjustment requirements through a large language model; the construction unit 12 is used to construct a grid topology structure; the acquisition unit 13 is used to acquire grid operation data, which includes power generation output data and load consumption data; the training unit 14 is used to train a small reinforcement learning model based on the text description, grid topology structure and grid operation data generated by the large language model.

[0109] In one embodiment of the present invention, when the generating unit 11 generates the text description including the power grid state adjustment requirement through the language large model, it is specifically used to:

[0110] The power grid monitoring data is obtained, and when the power grid monitoring data is abnormal, a power grid status alarm information is generated; the power grid status alarm information is input into the language big model so that a text description containing the power grid status adjustment requirements is generated through the language big model.

[0111] In one embodiment of the present invention, when constructing the power grid topology structure, the construction unit 12 is specifically used to:

[0112] The connection terminal identification information of different components is extracted to form a busbar and connection terminal identification form; different components are connected based on the busbar and connection terminal identification form to construct a power grid topology structure.

[0113] In one embodiment of the present invention, when training the reinforcement learning small model based on the text description, power grid topology and power grid operation data generated by the language large model, the training unit 14 is specifically used to:

[0114] Determine the grid operation state based on the grid topology and grid operation data; adjust the grid operation state to an abnormal state; for abnormal states, adjust the action strategy output by the small reinforcement learning model to be trained multiple times until the thermal stability adjustment target is achieved, or voltage exceeds the limit, loop, continuous repeated action, and action step exceeds the preset step condition; for each condition, calculate the reward value of the small reinforcement learning model to be trained, and generate a corresponding reward value for whether the grid operation state meets the existing operation regulations and text description through the language large model; send the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained adjusts its own network parameters to complete a training; return to the step of adjusting the grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than the preset threshold, and then end the training.

[0115] In one embodiment of the present invention, when the training unit 14 generates a corresponding reward value through the language model for determining whether the power grid operation status meets the existing operation regulations and text descriptions, it is specifically used to:

[0116] The grid operation status, existing operation regulations and text descriptions are input into the language model to generate corresponding prompt words through the language model, and modeling is performed based on the prompt words; new text related to rewards corresponding to the prompt words is generated through the modeling model, and the new text is quantified to generate corresponding reward values.

[0117] It should be noted that the specific implementation of the power grid thermal stability dynamic adjustment system based on the combination of large and small models in the embodiment of the present invention can refer to the specific implementation of the power grid thermal stability dynamic adjustment method based on the combination of large and small models mentioned above. To avoid redundancy, it will not be repeated here.

[0118] In summary, the present invention adopts a method that combines a large language model intelligent agent with a small reinforcement learning model. The large language model is used to realize direct modeling of language constraints and rewards. The small reinforcement learning model is used as a decision generation module. A corresponding integrated framework is proposed to generate end-to-end thermal stability adjustment decision plans that meet the text requirements. Finally, a good thermal stability adjustment of the power grid that meets the power grid operating status requirements and dispatching procedures is achieved, and it has high stability, thereby reducing the risk of heavy or even overloaded operation of the power grid to a certain extent, and providing a more reliable guarantee for the safe and stable operation of the power system.

[0119] Furthermore, the present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned power grid thermal stability dynamic adjustment method based on the combination of large and small models is implemented.

[0120] Furthermore, the present invention also provides a power grid control device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned power grid thermal stability dynamic adjustment method based on the combination of large and small models is implemented.

[0121] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0122] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A method for dynamic adjustment of power grid thermal stability based on the combination of large and small models, characterized in that: include: Train a small reinforcement learning model; The current operating state of the power grid is obtained. When the current operating state of the power grid is abnormal, a corresponding control plan is matched from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, the corresponding control plan is output through the trained reinforcement learning small model according to the current operating state of the power grid to achieve dynamic adjustment of thermal stability of the power grid.

2. The method for dynamic adjustment of power grid thermal stability based on the combination of large and small models as claimed in claim 1, characterized in that: The training of the small reinforcement learning model includes: Generate text descriptions containing grid state adjustment requirements through a language big model; Constructing a power grid topology structure and acquiring power grid operation data, wherein the power grid operation data includes power generation output data and load consumption data; The reinforcement learning small model is trained based on the text description generated by the language large model, the power grid topology and the power grid operation data.

3. The method for dynamic adjustment of power grid thermal stability based on the combination of large and small models as claimed in claim 2, characterized in that: The generating of a text description containing the power grid state adjustment requirements by means of a language large model includes: Acquire power grid monitoring data, and generate power grid status alarm information when the power grid monitoring data is abnormal; input the power grid status alarm information into the language model so as to generate a text description containing power grid status adjustment requirements through the language model; Construct the grid topology, including: Extract the connection terminal identification information of different components and form a busbar and connection terminal identification form; Connecting different components based on busbar and connection terminal identification forms to construct the power grid topology; Training the reinforcement learning small model based on the text description generated by the language large model, the power grid topology and the power grid operation data includes: Determining a power grid operation state based on the power grid topology and the power grid operation data; Adjust the grid operation status to an abnormal state; For abnormal conditions, the action strategy is adjusted multiple times through the output of the small reinforcement learning model to be trained until the thermal stability adjustment target is achieved, or voltage exceeds the limit, loops, continuous repeated actions, and the action step exceeds the preset step condition; For each working condition, the reward value of the small reinforcement learning model to be trained is calculated, and for whether the power grid operation state meets the existing operation regulations and the text description, the corresponding reward value is generated through the large language model; Sending the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained can adjust its own network parameters and complete one training; Return to the step of adjusting the power grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than a preset threshold, and then end the training.

4. The method for dynamic adjustment of power grid thermal stability based on the combination of large and small models as claimed in claim 3, characterized in that: The generation of a corresponding reward value by the language model for whether the power grid operation status meets the existing operation regulations and the text description includes: Inputting the power grid operation status, the existing operation regulations and the text description into the language model to generate corresponding prompt words through the language model, and performing modeling based on the prompt words; The model is built to generate new text related to the reward corresponding to the prompt word, and the new text is quantified to generate the corresponding reward value.

5. A power grid thermal stability dynamic adjustment system based on the combination of large and small models, characterized in that: include: Training module, used to train small reinforcement learning models; The decision-making module is used to obtain the current operating state of the power grid. When the current operating state of the power grid is abnormal, the corresponding control plan is matched from the knowledge base according to the current operating state of the power grid; if there is no corresponding control plan in the knowledge base, then according to the current operating state of the power grid, the corresponding control plan is output through the trained reinforcement learning small model to achieve dynamic adjustment of thermal stability of the power grid.

6. The power grid thermal stability dynamic adjustment system based on the combination of large and small models as claimed in claim 5, characterized in that: The training module includes: A generating unit, used for generating a text description including a power grid state adjustment requirement through a language large model; A construction unit, used for constructing a power grid topology structure; An acquisition unit, used to acquire power grid operation data, wherein the power grid operation data includes power generation output data and load consumption data; A training unit is used to train the reinforcement learning small model based on the text description generated by the language large model, the power grid topology and the power grid operation data.

7. The power grid thermal stability dynamic adjustment system based on the combination of large and small models as claimed in claim 6 is characterized in that: When the generating unit generates a text description including the power grid state adjustment requirement through the language large model, it is specifically used to: Acquire power grid monitoring data, and generate power grid status alarm information when the power grid monitoring data is abnormal; input the power grid status alarm information into the language model so as to generate a text description containing power grid status adjustment requirements through the language model; When constructing the power grid topology structure, the construction unit is specifically used to: Extract the connection terminal identification information of different components and form a busbar and connection terminal identification form; Connecting different components based on busbar and connection terminal identification forms to construct the power grid topology; When the training unit trains the reinforcement learning small model based on the text description generated by the language large model, the power grid topology structure and the power grid operation data, it is specifically used to: Determining a power grid operation state based on the power grid topology and the power grid operation data; Adjust the grid operation status to an abnormal state; For abnormal conditions, the action strategy is adjusted multiple times through the output of the small reinforcement learning model to be trained until the thermal stability adjustment target is achieved, or voltage exceeds the limit, loops, continuous repeated actions, and the action step exceeds the preset step condition; For each working condition, the reward value of the small reinforcement learning model to be trained is calculated, and for whether the power grid operation state meets the existing operation regulations and the text description, the corresponding reward value is generated through the large language model; Sending the calculated or generated reward value to the small reinforcement learning model to be trained, so that the small reinforcement learning model to be trained can adjust its own network parameters and complete one training; Return to the step of adjusting the power grid operation state to an abnormal state, and perform multiple trainings until the difference between the reward value after the current training and the reward value after the previous training is less than a preset threshold, and then end the training.

8. The power grid thermal stability dynamic adjustment system based on the combination of large and small models as claimed in claim 7 is characterized in that: When the training unit generates a corresponding reward value through the language model for determining whether the power grid operation state meets the existing operation regulations and the text description, the training unit is specifically used to: Inputting the power grid operation status, the existing operation regulations and the text description into the language model to generate corresponding prompt words through the language model, and performing modeling based on the prompt words; The model is built to generate new text related to the reward corresponding to the prompt word, and the new text is quantified to generate the corresponding reward value.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for dynamic adjustment of thermal stability of a power grid based on the combination of large and small models as described in any one of claims 1 to 4 is implemented.

10. A power grid control 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 method for dynamic adjustment of thermal stability of a power grid based on the combination of large and small models as described in any one of claims 1 to 4 is implemented.

Images