Power grid thermal stability dynamic adjustment method and system based on size model combination

Abstract

The application discloses a power grid thermal stability dynamic adjustment method and system based on a size model combination, and the method comprises the following steps: training a reinforcement learning small model; acquiring a current operation state of a power grid; when the current operation state of the power grid is an abnormal state, matching a corresponding regulation and control plan from a knowledge base according to the current operation state of the power grid; 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 the dynamic adjustment of the power grid thermal stability. The application can effectively utilize the text data in the power system, improve the accuracy of decision-making when the power grid thermal stability balance is destroyed, and quickly realize the dynamic adjustment of the power grid thermal stability, thereby improving the stability of the power grid operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid regulation, and particularly relates to a power grid thermal stability dynamic adjustment method and system based on size model combination. BACKGROUND

[0002] In recent years, with the expansion of the proportion of distributed new energy and the deepening of market-oriented reform, the control of the topology structure and the operation state of the power grid becomes more and more complex. At the same time, with the continuous change of system load level, operation mode and the increase of scale, the influence of unknown factors of the power grid becomes more complex, so the operation of the power grid will face greater challenges. Due to the continuous expansion of power supply and load in the power grid, as well as the reconstruction and new construction of lines, the operation state and operation mode of the power grid system have changed a lot. With the increase of the proportion of distributed power supply and the increase of power at the power 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 is out of operation due to failure, it will in turn cause other lines to overload and exit operation, causing a chain reaction. Therefore, measures need to be taken according to the current operation state of the system to maintain the system in a safe and stable operation state. Generally, reinforcement learning method is used.

[0003] The basic idea of reinforcement learning method 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, and gives the agent certain feedback (reward) after the state of the environment changes. The agent modifies the strategy according to the reward and iteratively finds a strategy that maximizes the cumulative expected reward from the initial state. However, this traditional reinforcement learning method cannot utilize the large amount of text data in the power system and cannot fit the model according to the requirements in the regulation and adjustment target, resulting in low data utilization and large decision bias. SUMMARY

[0004] The present application aims to at least partially solve the technical problems in the related art. To this end, the first object of the present application is to provide a power grid thermal stability dynamic adjustment method based on size model combination, which can effectively utilize the text data in the power system and the language processing capability of the language large model, improve the decision rationality, and quickly realize the power grid thermal stability dynamic adjustment when the power grid thermal stability balance is destroyed, thereby improving the stability of the power grid operation.

[0005] The second object of the present application is to provide a power grid thermal stability dynamic adjustment system based on size model combination.

[0006] The third object of the present application is to provide a computer readable storage medium.

[0007] The fourth object of the present application is to provide an electronic device.

[0008] To achieve the above object, the present application realizes the following technical solutions:

[0009] A power grid thermal stability dynamic adjustment method based on a large model combination, comprising:

[0010] Training a reinforcement learning small model;

[0011] Obtaining the current operating state of the power grid, and when the current operating state of the power grid is an abnormal state, matching a corresponding regulation and control plan from a knowledge base according to the current operating state of the power grid; if there is no corresponding regulation and control plan in the knowledge base, outputting a corresponding regulation and control scheme through the trained reinforcement learning small model according to the current operating state of the power grid, to realize the dynamic adjustment of the power grid thermal stability.

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

[0013] Generating a text description containing power grid state adjustment requirements through a language large model;

[0014] Constructing a power grid topology structure and obtaining power grid operation data, the power grid operation data including power generation output data and load consumption data;

[0015] Training 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.

[0016] Preferably, the generation of a text description containing power grid state adjustment requirements through a language large model comprises:

[0017] Obtaining power grid monitoring data, and generating power grid state alarm information when the power grid monitoring data is abnormal; inputting the power grid state alarm information into the language large model, so as to generate a text description containing power grid state adjustment requirements through the language large model;

[0018] Constructing a power grid topology structure, comprising:

[0019] Extracting connection end identification information of different elements to form a bus and connection end identification table;

[0020] Connecting different elements based on the bus and connection end identification table to construct the power grid topology structure;

[0021] Training 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, comprising:

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

[0023] Adjusting the power grid operating state to an abnormal state;

[0024] In response to abnormal conditions, the action strategy output by the small reinforcement learning model to be trained is adjusted multiple times until the thermal stability adjustment target is achieved, or the voltage exceeds the limit, loop, continuous repetitive action, or action step size exceeds the preset step size occurs.

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

[0026] The calculated or generated reward value is sent to the reinforcement learning mini-model to be trained, so that the reinforcement learning mini-model to be trained can adjust its own network parameters and complete a training cycle.

[0027] Return to the steps of adjusting the power grid operation state to an abnormal state, and perform multiple training sessions 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, at which point the training ends.

[0028] Preferably, the step of generating a corresponding reward value based on whether the power grid operating status meets existing operating regulations and the text description, through the language big data model, includes:

[0029] The power grid operating status, the existing operating regulations, and the text description are input into the language model to generate corresponding prompt words, and modeling is performed based on the prompt words.

[0030] The model generates new text related to the reward corresponding to the prompt word, and then quantifies the new text to generate the corresponding reward value.

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

[0032] The training module is used to train small reinforcement learning models;

[0033] The decision-making module is used to obtain the current operating status of the power grid. When the current operating status of the power grid is abnormal, it matches the corresponding control plan from the knowledge base according to the current operating status of the power grid. If there is no corresponding control plan in the knowledge base, it outputs the corresponding control scheme through the trained reinforcement learning small model according to the current operating status of the power grid, so as to realize the dynamic adjustment of the power grid thermal stability.

[0034] Preferably, the training module includes:

[0035] The generation unit is used to generate text descriptions containing power grid state adjustment requirements through a large language model;

[0036] a construction unit, configured to construct a power grid topology;

[0037] an acquisition unit, configured to acquire power grid operation data, the power grid operation data including power generation output data and load consumption data;

[0038] a training unit, configured 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, the generation unit, when generating the text description containing power grid state adjustment requirements through the language large model, is specifically configured to:

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

[0041] The construction unit, when constructing the power grid topology, is specifically configured to:

[0042] extract connection end identification information of different elements to form a bus bar and connection end identification table;

[0043] connect different elements based on the bus bar and connection end identification table to construct the power grid topology;

[0044] The training unit, when 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, is specifically configured to:

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

[0046] adjust the power grid operation state to an abnormal state;

[0047] for the abnormal state, output an action strategy through the to-be-trained reinforcement learning small model for multiple adjustments until a thermal stability adjustment target is completed, or a voltage out-of-limit, a loop, a continuous repeated action, or an action step length exceeding a preset step length condition occurs;

[0048] for each condition, calculate a reward value of the to-be-trained reinforcement learning small model, and generate a corresponding reward value through the language large model for whether the power grid operation state meets an existing operation regulation and the text description;

[0049] send the calculated or generated reward value to the to-be-trained reinforcement learning small model, so that the to-be-trained reinforcement learning small model adjusts network parameters of itself to complete one training;

[0050] Return to the steps of adjusting the power grid operation state to an abnormal state, and perform multiple training sessions 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, at which point the training ends.

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

[0052] The power grid operating status, the existing operating regulations, and the text description are input into the language model to generate corresponding prompt words, and modeling is performed based on the prompt words.

[0053] The model generates new text related to the reward corresponding to the prompt word, and quantifies the new text to generate the corresponding reward value.

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

[0055] To achieve the above objectives, a 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, it implements the above-described method for dynamic adjustment of power grid thermal stability based on a combination of size models.

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

[0057] This invention provides a dynamic adjustment scheme for power grid thermal stability based on a combination of large and small models. The scheme employs an algorithm design combining large and small models. A reinforcement learning small model generates switching actions (control schemes) under abnormal power grid conditions, while a language large model transforms abstract textual descriptions such as existing textual regulations and adjustment targets into reward values ​​that the reinforcement learning small model can understand, thus enabling the training of the reinforcement learning small model and assisting in judging the state of the power grid. When the power grid's thermal stability limit is breached, the system prioritizes checking if there are relevant contingency plans in the knowledge base. If not, the reinforcement learning small model is invoked to quickly generate one, thereby rapidly achieving dynamic adjustment of the power grid's thermal stability. Therefore, this invention effectively utilizes textual data in the power system and the language processing capabilities of the language large model to improve decision-making rationality and rapidly achieve dynamic adjustment of the power grid's thermal stability when the thermal stability balance is disrupted, thereby improving the stability of power grid operation.

[0058] Additional aspects and advantages of the present application will be apparent from the following description, the drawings, and the appended claims. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A flowchart of the power grid thermal stability dynamic adjustment method based on size model combination of an embodiment of the present application.

[0060] Figure 2 A flowchart of the reinforcement learning small model training of an embodiment of the present application.

[0061] Figure 3 A structural block diagram of the power grid thermal stability dynamic adjustment system based on size model combination of an embodiment of the present application.

[0062] Figure 4 A structural block diagram of the training module of an embodiment of the present application. DETAILED DESCRIPTION

[0063] The embodiments are described in detail below with reference to the accompanying drawings, examples of which are shown in the drawings, wherein the same or similar notations represent the same or similar elements throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0064] The power grid thermal stability dynamic adjustment method and system based on size model combination of the present embodiment are described below with reference to the accompanying drawings.

[0065] Figure 1 A flowchart of the power grid thermal stability dynamic adjustment method based on size model combination of an embodiment of the present application. As shown in Figure 1 the power grid thermal stability dynamic adjustment method based on size model combination includes:

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

[0067] In the present embodiment, the reinforcement learning small model is trained, including: generating a text description containing power grid state adjustment requirements through a language large model; constructing a power grid topology and obtaining power grid operation data, the power grid operation data including power generation output data and load consumption data; 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.

[0068] In an embodiment of the present application, the text description containing power grid state adjustment requirements is generated through a language large model, including: obtaining power grid monitoring data, and generating power grid state alarm information when the power grid monitoring data is abnormal; inputting the power grid state alarm information into the language large model so as to generate the text description containing power grid state adjustment requirements through the language large model.

[0069] Specifically, the power grid monitoring data can be read, and then it is judged whether the power grid monitoring data is abnormal. If it is abnormal, such as the load rate of part of the line being too high, being in a heavy load or overload state, the operation mode of the power grid needs to be adjusted to balance the load rate of different lines, maintain voltage balance, and make the power grid meet the stability requirements. When the power grid monitoring data is judged to be abnormal, the power grid state alarm information is generated, and then the power grid state alarm information is input into a language large model such as ChatGPT (a language model), and then the language large model can generate a text description containing power grid state adjustment requirements and corresponding business requirements.

[0070] In an embodiment of the present application, the power grid topology structure is constructed, including: extracting the connection end identification information of different elements to form a bus and connection end identification table; connecting different elements based on the bus and connection end identification table to construct the power grid topology structure.

[0071] Specifically, the physical node information of different elements such as bus or transformer equipment, i.e. the connection end identification information of switches or other equipment, can be extracted to form a bus and connection end identification table, and then different elements can be connected according to the bus and connection end identification table to obtain the power grid topology structure. For elements with a small amount of missing physical node data, the physical node can be estimated through the topology node.

[0072] It should be noted that after the power grid topology structure is constructed, the power grid operation data is also obtained, wherein the real power generation and load data are used as the output and consumption of the power grid, and the missing power grid operation data can be filled in through interpolation, state estimation and other methods.

[0073] In an embodiment of the present application, the steps of training 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 are as shown in Figure 2 , including:

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

[0075] S102: adjusting the power grid operation state to an abnormal state.

[0076] S103: for the abnormal state, outputting the action strategy through the to-be-trained reinforcement learning small model for multiple adjustments until the thermal stability adjustment target is completed, or the voltage exceeds the limit, the loop, the continuous repeated action, or the action step length exceeds the preset step length working condition.

[0077] S104: for each working condition, calculating the reward value of the to-be-trained reinforcement learning small model, and for whether the power grid operation state meets the existing operation regulations and the text description, generating the corresponding reward value through the language large model.

[0078] S105: send the reward value calculated or generated to the to-be-trained reinforcement learning sub-model, so that the to-be-trained reinforcement learning sub-model adjusts the network parameters of itself, and completes one training.

[0079] S106: return to the step of adjusting the power grid operation state to the abnormal state, and perform multiple training until the difference between the reward value after the current training and the reward value after the last training is less than a preset threshold, and then end the training.

[0080] Specifically, the to-be-trained reinforcement learning sub-model can read environment parameters, including power grid topology and power grid operation data, and then adjust the current power grid operation state to enter an abnormal state. Subsequently, the to-be-trained reinforcement learning sub-model randomly makes some actions to explore and gives corresponding switch actions, that is, makes action strategies to adjust the power grid operation state. In this embodiment, the switch action space function of the to-be-trained reinforcement learning sub-model is represented as:

[0081] A=[a0,a1,a2,…,a n ](1)

[0082] Wherein, a n represents the switch number that needs to be acted, and A represents the switch action space function.

[0083] According to the given action, the power grid operation state is changed, and in this embodiment, the power grid operation state space function is:

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

[0085] Wherein, B represents a branch switch variable state set, P i represents the power of the i-th branch, and S represents the power grid operation state space function.

[0086] Multiple actions are made until the termination condition of one round of training is met: the thermal stability adjustment target is completed, or the following situations occur during adjustment: voltage out-of-limit, loop, continuous repeated action, action step length exceeding the preset step length, fault switch action, time length out-of-limit working condition.

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

[0088] For example, for the line load rate too high working condition, the reward value of the to-be-trained reinforcement learning sub-model under the line load rate too high working condition can be calculated as:

[0089]

[0090] Wherein, R P represents the reward value under the line load rate too high working condition, Pr0 representing the load rate of the branch with the highest line load rate of the power grid before adjustment, P r1 representing the load rate of the branch with the highest line load rate of the power grid after adjustment.

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

[0092]

[0093] wherein, R U representing the voltage out-of-limit reward (penalty), U representing the node voltage, U max representing the upper limit of the voltage unit value, U min representing the lower limit of the voltage unit value, wherein the node is the physical node as described above, including switches and other devices.

[0094] wherein, for whether the power grid operating state meets the existing operating regulations and the text description generated by the language large model, the language large model can generate the corresponding reward value.

[0095] It should be noted that whether the power grid operating state meets or not, the language large model needs to generate the corresponding reward value, that is, if the power grid operating state does not meet the existing operating regulations and the text description, a smaller reward value is generated, and if the power grid operating state meets the existing operating regulations and the text description, a larger reward value is generated. For this case, the language large model generates the reward value as follows:

[0096] The power grid operating state, the existing operating regulations and the text description are input into the language large model to generate the corresponding prompt words through the language large model, and the prompt words are modeled based on the prompt words; the new text related to the reward corresponding to the prompt words is generated through the modeling model, and the new text is quantified to generate the corresponding reward value.

[0097] Specifically, the power grid operating state, the existing operating regulations and the text description can be input into the language large 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) to generate new text related to the reward corresponding to the prompt words through the modeling model, and then the generated new text is parsed into a reward value, and the calculation formula is as follows:

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

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

[0100] After obtaining the reward values under various working conditions, the calculated or generated reward values can be sent to the trained reinforcement learning sub-model, and the trained reinforcement learning sub-model changes its network parameters according to the reward values, thereby completing the current training. Subsequently, the power grid is reset, the abnormal state is reconfigured, and the next round of training is performed until the reward value tends to be stable, thereby completing the training process of the trained reinforcement learning sub-model and obtaining the optimal reinforcement learning sub-model.

[0101] Step S2: Obtain the current operating state of the power grid, and when the current operating state of the power grid is an abnormal state, match a corresponding regulation and control plan from the knowledge base according to the current operating state of the power grid; if there is no corresponding regulation and control plan in the knowledge base, output a corresponding regulation and control scheme through the trained reinforcement learning sub-model according to the current operating state of the power grid, so as to realize the dynamic adjustment of the power grid thermal stability.

[0102] In the actual operation process of the power grid, the current operating state of the power grid can be obtained, and the language large model searches in the knowledge base to determine whether the current operating state of the power grid meets the scenario described in a regulation and control plan. If there is a matching regulation and control plan, the disposal process in the regulation and control plan is directly output, and if there is no matching regulation and control plan, the current operating state data of the power grid is further input to the trained reinforcement learning sub-model, and the regulation and control scheme is given by the reinforcement learning sub-model.

[0103] It should be noted that, whether the regulation and control plan matched in the knowledge base or the regulation and control scheme given by the reinforcement learning sub-model, after output, the simulation API (application programming interface) needs to be called to verify the disposal scheme, and after successful verification, the operation ticket can be generated and the action stage is entered.

[0104] If the reinforcement learning sub-model is called in the process of generating the disposal scheme, the newly generated regulation and control scheme of the reinforcement learning sub-model and the corresponding power grid abnormal state also need to be stored in the knowledge base.

[0105] Figure 3 The structure block diagram of the power grid thermal stability dynamic adjustment system based on the combination of large and small models according to the embodiment of the application is shown in FIG. 1. Figure 3 As shown in FIG. 1, the power grid thermal stability dynamic adjustment system based on the combination of large and small models 100 includes a training module 10 and a decision module 20 connected to each other.

[0106] The training module 10 is configured to train the reinforcement learning sub-model.

[0107] Figure 4 The training module 10 is configured to train the reinforcement learning sub-model. Figure 4 The training module 10 includes a generation unit 11, a construction unit 12, an acquisition unit 13, and a training unit 14.

[0108] The generation unit 11 is configured to generate a text description containing power grid state adjustment requirements through a language large model. The construction unit 12 is configured to construct a power grid topology. The acquisition unit 13 is configured to acquire power grid operation data, including power generation output data and load consumption data. The training unit 14 is configured to train a reinforcement learning sub-model based on the text description generated by the language large model, the power grid topology, and the power grid operation data.

[0109] In one embodiment of the present application, when the generation unit 11 generates a text description containing power grid state adjustment requirements through a language large model, it is specifically configured to:

[0110] acquire power grid monitoring data and generate power grid state alarm information when the power grid monitoring data is abnormal.

[0111] In one embodiment of the present application, when the construction unit 12 constructs a power grid topology, it is specifically configured to:

[0112] extract connection end identification information of different elements to form a bus and connection end identification table.

[0113] In one embodiment of the present application, when the training unit 14 trains a reinforcement learning sub-model based on the text description generated by the language large model, the power grid topology, and the power grid operation data, it is specifically configured to:

[0114] determine a power grid operation state based on a power grid topology structure and power grid operation data; adjust the power grid operation state to an abnormal state; for the abnormal state, output an action strategy by a to-be-trained reinforcement learning small model for multiple adjustments until a thermal stability adjustment target is completed, or a voltage out-of-limit, a loop, a continuous repeated action, or an action step length exceeding a preset step length working condition occurs; for each working condition, calculate a reward value of the to-be-trained reinforcement learning small model, and for whether the power grid operation state meets existing operation regulations and text descriptions, generate a corresponding reward value by a language large model; send the calculated or generated reward value to the to-be-trained reinforcement learning small model, so that the to-be-trained reinforcement learning small model adjusts network parameters of itself to complete one training; return to the step of adjusting the power grid operation state to the abnormal state, and perform multiple trainings until a difference between a reward value after current training and a reward value after last training is less than a preset threshold, and the training is ended.

[0115] In an embodiment of the present application, when the training unit 14 generates a corresponding reward value by a language large model for whether the power grid operation state meets existing operation regulations and text descriptions, it is specifically used for:

[0116] input the power grid operation state, the existing operation regulations and the text descriptions into the language large model to generate a corresponding prompt word by the language large model, and model based on the prompt word; generate a new text about rewards corresponding to the prompt word by the modeling model, and quantize the new text to generate a corresponding reward value.

[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 embodiments of the present application can refer to the specific implementation of the power grid thermal stability dynamic adjustment method based on the combination of large and small models described above. To avoid redundancy, it will not be repeated here.

[0118] In summary, the present application adopts the method of combining a language large model agent with a reinforcement learning small model, realizes direct modeling of language constraints and rewards through a language large model, uses a reinforcement learning small model as a decision generation module, proposes a corresponding integrated framework, and generates a thermal stability adjustment decision scheme that meets text regulations in an end-to-end manner. Finally, a good power grid thermal stability adjustment that meets the requirements of the power grid operation state and the dispatching regulations is realized, and has high stability, thereby reducing the risk of power grid overload or even overload operation to a certain extent, and providing a more reliable guarantee for the safe and stable operation of the power system.

[0119] Further, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the power grid thermal stability dynamic adjustment method based on the combination of large and small models.

[0120] Further, the application further provides a power grid regulation device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the power grid thermal stability dynamic adjustment method based on the size model combination is realized.

[0121] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... " does not, without more constraints, preclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the recited element.

[0122] Although the present application has been described in detail through the preferred embodiments, it should be appreciated that the above description should not be considered as limiting the present application. Various modifications and substitutions to the present application will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present application should be defined by the appended claims.

Claims

1. A power grid thermal stability dynamic adjustment method based on size model combination, characterized in that, The method comprises the following steps: training a reinforcement learning sub-model, comprising: generating a text description containing power grid state adjustment requirements through a language large model; constructing a power grid topology structure and obtaining power grid operation data, wherein the power grid operation data comprises power generation output data and load consumption data; training the reinforcement learning sub-model based on the text description generated by the language large model, the power grid topology structure, and the power grid operation data; obtaining a current power grid operation state, and when the current power grid operation state is an abnormal state, matching a corresponding regulation and control plan from a knowledge base according to the current power grid operation state; if there is no corresponding regulation and control plan in the knowledge base, outputting a corresponding regulation and control scheme through the trained reinforcement learning sub-model according to the current power grid operation state to realize dynamic adjustment of the power grid thermal stability; the step of generating a text description containing power grid state adjustment requirements through a language large model comprises: obtaining power grid monitoring data, and generating power grid state alarm information when the power grid monitoring data is abnormal; inputting the power grid state alarm information into the language large model so as to generate a text description containing power grid state adjustment requirements through the language large model; constructing a power grid topology structure, comprising: extracting connection end identification information of different elements to form a bus and connection end identification table; connecting different elements based on the bus and connection end identification table to construct the power grid topology structure; the step of training the reinforcement learning sub-model based on the text description generated by the language large model, the power grid topology structure, and the power grid operation data comprises: determining a power grid operation state based on the power grid topology structure and the power grid operation data; adjusting the power grid operation state to an abnormal state; for the abnormal state, outputting an action strategy through the reinforcement learning sub-model to be trained for multiple times until a thermal stability adjustment target is completed, or a voltage out-of-limit, a loop, a continuous repeated action, or an action step length exceeding a preset step length condition occurs; for each condition, calculating a reward value of the reinforcement learning sub-model to be trained, and generating a corresponding reward value through the language large model for whether the power grid operation state meets an existing operation regulation and the text description; sending the calculated or generated reward value to the reinforcement learning sub-model to be trained, so that the reinforcement learning sub-model to be trained adjusts its network parameters to complete one training; returning to the step of adjusting the power grid operation state to the abnormal state, and performing multiple training until a difference between a current reward value after training and a reward value after last training is less than a preset threshold, and then ending the training.

2. The method of claim 1, wherein the size model-based combined grid thermal steady dynamic adjustment method is characterized by, the step of generating a corresponding reward value through the language large model for whether the power grid operation state meets an existing operation regulation and the text description comprises: inputting the power grid operation state, the existing operation regulation, and the text description into the language large model to generate a corresponding prompt word through the language large model, and modeling based on the prompt word; generating a new text related to the reward corresponding to the prompt word through the modeling model, and quantifying the new text to generate a corresponding reward value.

3. A power grid thermal stability dynamic adjustment system based on size model combination, characterized in that, The method comprises the following steps: training a reinforcement learning sub-model through a training module; the training module comprises: The generating unit is configured to generate a text description containing power grid state adjustment requirements by using a language large model; The constructing unit is configured to construct a power grid topology; The obtaining unit is configured to obtain power grid operation data, which includes power generation output data and load consumption data; The training unit is configured 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; The decision module is configured to obtain a current operation state of the power grid, and when the current operation state of the power grid is an abnormal state, match a corresponding regulation and control plan from a knowledge base according to the current operation state of the power grid; if there is no corresponding regulation and control plan in the knowledge base, output 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 thermal stability dynamic adjustment of the power grid; When the generating unit generates the text description containing the power grid state adjustment requirements by using the language large model, the generating unit is specifically configured to: obtain power grid monitoring data, and when the power grid monitoring data is abnormal, generate power grid state alarm information; input the power grid state alarm information into the language large model, so as to generate the text description containing the power grid state adjustment requirements by using the language large model; When the constructing unit constructs the power grid topology, the constructing unit is specifically configured to: extract connection end identification information of different elements to form a bus and connection end identification table; connect different elements based on the bus and connection end identification table 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, and the power grid operation data, the training unit is specifically configured to: determine a power grid operation state based on the power grid topology and the power grid operation data; adjust the power grid operation state to an abnormal state; for the abnormal state, output an action strategy through the to-be-trained reinforcement learning small model for multiple times of adjustment, until a thermal stability adjustment target is completed, or a voltage out-of-limit, a loop, a continuous repeated action, or an action step length exceeding a preset step length condition occurs; for each condition, calculate a reward value of the to-be-trained reinforcement learning small model, and for whether the power grid operation state meets an existing operation regulation and the text description, generate a corresponding reward value through the language large model; send the calculated or generated reward value to the to-be-trained reinforcement learning small model, so that the to-be-trained reinforcement learning small model adjusts network parameters of itself to complete one training; return to the step of adjusting the power grid operation state to the abnormal state, and perform multiple times of training, until a difference between a current reward value after training and a reward value after last training is less than a preset threshold, and then end the training.

4. The size model based combined grid thermal steady dynamic adjustment system of claim 3, wherein, When the training unit generates a corresponding reward value through the language large model for whether the power grid operation state meets the existing operation regulation and the text description, the training unit is specifically configured to: input the power grid operation state, the existing operation regulation, and the text description into the language large model, so as to generate a corresponding prompt word through the language large model, and model based on the prompt word. A new text related to the reward corresponding to the prompt word is generated by modeling the model, and the new text is quantified to generate a corresponding reward value.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the power grid thermal stability dynamic adjustment method based on size model combination as claimed in any one of claims 1-2.

6. A power grid regulating device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the power grid thermal stability dynamic adjustment method based on size model combination as claimed in any one of claims 1-2.

Images