Energy-saving parameter analysis report generation model training method and device, energy-saving parameter adjustment method and device, equipment and storage medium

Abstract

The invention relates to a training method and device of an energy-saving parameter analysis report generation model, an energy-saving parameter adjustment method and device, equipment and a storage medium. The method comprises the following steps: acquiring energy-saving reference information, energy-saving parameters and energy-saving feedback information of the energy-saving parameters of a sample base station, constructing a sample cue word according to the energy-saving reference information, and constructing an energy-saving parameter analysis report according to the energy-saving parameters; inputting the sample cue words into the large language model to obtain a predicted energy-saving parameter analysis report output by the large language model; determining a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determining a loss value of the large language model according to the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value; and performing fine tuning processing on the large language model according to the loss value to obtain an energy-saving parameter analysis report generation model. By adopting the method, energy-saving parameters with relatively high interpretability can be generated.

Description

Technical Field

[0001] The present application relates to the field of natural language processing, and in particular to a training method and device for an energy-saving parameter analysis report generation model, an energy-saving parameter adjustment method and device, equipment, and storage medium. Background Art

[0002] With the development of 5G technology (5th Generation Mobile Communication Technology), 5G base stations have been widely deployed. In order to save energy consumption of base stations, 5G base stations usually set energy-saving parameters related to energy saving, and achieve energy saving through energy-saving strategies such as deep sleep, channel shutdown, and symbol shutdown.

[0003] In the related technologies, reinforcement learning and other technologies have been widely used in the optimization of energy-saving parameters of base stations. However, due to the large scale of base stations, the scenarios and user groups involved may be different, and the training data of reinforcement learning methods often covers limited scenarios, resulting in limited generalization capabilities and poor online migration capabilities in some scenarios. Moreover, due to the black box characteristics of deep neural networks, methods based on deep reinforcement learning lack interpretability, making it difficult for operation and maintenance personnel to interpret the energy-saving parameters derived from the neural network and migrate these energy-saving parameters to base stations in other scenarios. Therefore, the related technology has the problem that the predicted energy-saving parameters can only be applied to specific scenarios, and it is difficult to transfer these energy-saving parameters to other scenarios. Summary of the invention

[0004] Based on this, it is necessary to provide a training method and device for an energy-saving parameter analysis report generation model, an energy-saving parameter adjustment method and device, equipment, and storage medium to address the above-mentioned technical problems.

[0005] In a first aspect, the present application provides a training method for an energy-saving parameter analysis report generation model. The method comprises:

[0006] Acquire energy-saving reference information, energy-saving parameters and energy-saving feedback information of the sample base stations, construct sample prompt words according to the energy-saving reference information, and construct an energy-saving parameter analysis report according to the energy-saving parameters;

[0007] Inputting the sample prompt words into a large language model to obtain a predicted energy-saving parameter analysis report output by the large language model;

[0008] determining a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determining a loss value of the large language model according to a difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value;

[0009] The large language model is fine-tuned according to the loss value to obtain an energy-saving parameter analysis report generation model.

[0010] In one embodiment, the energy-saving parameter corresponds to one or more energy-saving cycles, and the energy-saving feedback information includes energy-saving effect information corresponding to the energy-saving parameter in each of the energy-saving cycles;

[0011] The determining, according to the energy-saving feedback information, a reward value for the energy-saving parameter analysis report includes:

[0012] Determine a reward value for the number of energy-saving cycles according to the number of energy-saving cycles corresponding to the energy-saving parameter, wherein the reward value for the number of energy-saving cycles is positively correlated with the number of energy-saving cycles;

[0013] Determining an energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period;

[0014] A reward value for the energy-saving parameter analysis report is determined according to the energy-saving cycle quantity reward value and the energy-saving effect reward value.

[0015] In one embodiment, the energy-saving effect information includes a network quality score and an energy-saving duration score, and determining the energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period includes:

[0016] For any of the energy-saving cycles, weighted summing is performed on the network quality score and the energy-saving duration score corresponding to the energy-saving cycle to obtain an energy-saving effect score corresponding to the energy-saving cycle;

[0017] An energy-saving effect reward value is determined according to the energy-saving effect score corresponding to each energy-saving cycle.

[0018] In one embodiment, constructing a sample prompt word according to the energy-saving reference information includes:

[0019] Determining, according to the energy-saving reference information, candidate energy-saving parameters of the sample base station and an adjustment range of each of the candidate energy-saving parameters;

[0020] Constructing background description information according to the energy-saving reference information, and constructing task description information according to each of the candidate energy-saving parameters and the adjustment range of each of the candidate energy-saving parameters;

[0021] A sample prompt word is constructed according to the background description information and the task description information.

[0022] In one embodiment, constructing an energy-saving parameter analysis report according to the energy-saving parameters includes:

[0023] Constructing background understanding information corresponding to the background description information and task understanding information corresponding to the task description information;

[0024] constructing task conclusion information based on the background understanding information, the task understanding information and the energy-saving parameter;

[0025] An energy-saving parameter analysis report is constructed according to the background understanding information, the task understanding information and the task conclusion information.

[0026] In a second aspect, the present application also provides a training device for an energy-saving parameter analysis report generation model. The device comprises:

[0027] An acquisition module, used to acquire energy-saving reference information, energy-saving parameters and energy-saving feedback information of the sample base stations, construct sample prompt words according to the energy-saving reference information, and construct an energy-saving parameter analysis report according to the energy-saving parameters;

[0028] An input module, used for inputting the sample prompt words into a large language model to obtain a predicted energy-saving parameter analysis report output by the large language model;

[0029] a determination module, configured to determine a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determine a loss value of the large language model according to a difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value;

[0030] A fine-tuning module is used to fine-tune the large language model according to the loss value to obtain an energy-saving parameter analysis report generation model.

[0031] In one embodiment, the energy-saving parameter corresponds to one or more energy-saving cycles, and the energy-saving feedback information includes energy-saving effect information corresponding to the energy-saving parameter in each of the energy-saving cycles;

[0032] The determining module is further used for:

[0033] Determine a reward value for the number of energy-saving cycles according to the number of energy-saving cycles corresponding to the energy-saving parameter, wherein the reward value for the number of energy-saving cycles is positively correlated with the number of energy-saving cycles;

[0034] Determining an energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period;

[0035] A reward value for the energy-saving parameter analysis report is determined according to the energy-saving cycle quantity reward value and the energy-saving effect reward value.

[0036] In one embodiment, the energy-saving effect information includes a network quality score and an energy-saving duration score, and the determination module is further used to:

[0037] For any of the energy-saving cycles, weighted summing is performed on the network quality score and the energy-saving duration score corresponding to the energy-saving cycle to obtain an energy-saving effect score corresponding to the energy-saving cycle;

[0038] An energy-saving effect reward value is determined according to the energy-saving effect score corresponding to each energy-saving cycle.

[0039] In one embodiment, the acquisition module is further used for:

[0040] Determining, according to the energy-saving reference information, candidate energy-saving parameters of the sample base station and an adjustment range of each of the candidate energy-saving parameters;

[0041] Constructing background description information according to the energy-saving reference information, and constructing task description information according to each of the candidate energy-saving parameters and the adjustment range of each of the candidate energy-saving parameters;

[0042] A sample prompt word is constructed according to the background description information and the task description information.

[0043] In one embodiment, the acquisition module is further used for:

[0044] Constructing background understanding information corresponding to the background description information and task understanding information corresponding to the task description information;

[0045] constructing task conclusion information based on the background understanding information, the task understanding information and the energy-saving parameter;

[0046] An energy-saving parameter analysis report is constructed according to the background understanding information, the task understanding information and the task conclusion information.

[0047] In a third aspect, the present application provides a method for adjusting energy-saving parameters. The method comprises:

[0048] Acquire energy-saving reference information of a target base station, and construct a target prompt word according to the energy-saving reference information;

[0049] Inputting the target prompt word into the energy-saving parameter analysis report generation model to obtain an energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained using the method described in any of the aforementioned embodiments;

[0050] According to the energy-saving parameter analysis report, current energy-saving parameters of the target base station are adjusted.

[0051] In a fourth aspect, the present application provides an energy-saving parameter adjustment device. The device comprises:

[0052] An acquisition module, used to acquire energy-saving reference information of a target base station, and construct a target prompt word according to the energy-saving reference information;

[0053] An input module, used for inputting the target prompt word into the energy-saving parameter analysis report generation model to obtain an energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained using the method described in any of the aforementioned embodiments;

[0054] An adjustment module is used to adjust the current energy-saving parameters of the target base station according to the energy-saving parameter analysis report.

[0055] In a fifth aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, any of the above methods is implemented.

[0056] In a sixth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements any of the above methods when executed by a processor.

[0057] In a seventh aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, any of the above methods is implemented.

[0058] The training method and device of the energy-saving parameter analysis report generation model, the energy-saving parameter adjustment method and device, the equipment, and the storage medium, construct sample prompt words according to the energy-saving reference information, construct the energy-saving parameter analysis report according to the energy-saving parameters, and use the sample prompt words, the energy-saving parameter analysis report, and the reward value calculated for the energy-saving parameter analysis report to fine-tune the large language model, so that the energy-saving parameter analysis report generation model obtained after fine-tuning can generate a report similar to the energy-saving parameter analysis report with a higher reward value according to the energy-saving reference information. The embodiment of the present application can generate an energy-saving parameter analysis report with high interpretability, so that the technician can refer to the energy-saving parameter analysis report to infer the reason why the large language model generates specific energy-saving parameters, and then can migrate the energy-saving parameters to other scenarios based on the energy-saving parameter analysis report, reducing the difficulty of converting the energy-saving parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flowchart of a training method for an energy-saving parameter analysis report generation model in one embodiment;

[0060] Figure 2 is a flow chart of step 106 in one embodiment;

[0061] Figure 3is a flow chart of step 206 in one embodiment;

[0062] Figure 4 is a flow chart of step 102 in one embodiment;

[0063] Figure 5 is a flow chart of step 102 in one embodiment;

[0064] Figure 6 A flowchart of a training method for an energy-saving parameter analysis report generation model in one embodiment;

[0065] Figure 7 A schematic diagram of a flow chart of an energy-saving parameter adjustment method in an embodiment;

[0066] Figure 8 A structural block diagram of a training device for generating an energy-saving parameter analysis report model in one embodiment;

[0067] Fig. 9 It is a structural block diagram of an energy-saving parameter adjustment device in an embodiment;

[0068] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0070] In one embodiment, Figure 1 As shown, a training method for an energy-saving parameter analysis report generation model is provided. This embodiment takes the method applied to a server as an example. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0071] Step 102, obtaining energy-saving reference information, energy-saving parameters and energy-saving feedback information of the sample base stations, constructing sample prompt words according to the energy-saving reference information, and constructing an energy-saving parameter analysis report according to the energy-saving reference information and the energy-saving parameters.

[0072] In an embodiment of the present application, the sample base station is one or more base stations selected from various base stations. For example, a base station with a longer operating time or a base station with more frequent energy-saving parameter adjustments can be selected as a sample base station. Energy-saving reference information is information that the sample base station needs to refer to when adjusting the energy-saving parameters. The specific information can be set by those skilled in the art according to actual needs. Exemplarily, the energy-saving reference information may include the energy consumption saved by the base station through the energy-saving state within a period of time before the energy-saving parameters are adjusted, the network quality information of the base station in the energy-saving state (which may include uplink transmission rate, downlink transmission rate, network delay, packet loss rate, etc.), the user's network quality perception of the base station in the non-energy-saving state and the network quality perception of the base station in the energy-saving state, the business volume of the base station in the energy-saving state, etc.

[0073] The energy-saving parameters are energy-saving parameters adjusted by the sample base station according to the energy-saving reference information. The energy-saving parameters may include energy-saving execution thresholds and energy-saving execution parameters. The energy-saving execution threshold represents the judgment condition for turning on the energy-saving mode, and may include, for example: an energy-saving execution threshold for business volume (if the business volume is less than the energy-saving execution threshold, the energy-saving mode is turned on), an energy-saving execution threshold for the current time (if the current time is later than the energy-saving execution threshold, the energy-saving mode is turned on), etc. The energy-saving execution parameters represent the relevant parameters when executing the energy-saving mode, and may include, for example: the execution duration of the channel shutdown, how many transmission channels are specifically closed when the channel is shut down, etc.

[0074] Energy-saving feedback information is feedback information on the energy-saving effect after the base station is saved according to the energy-saving parameters. The energy-saving feedback information may include the amount of energy saved by the base station after saving energy according to the updated energy-saving parameters compared to before the energy-saving parameters were updated, network quality information in the energy-saving state, and user ratings of network quality.

[0075] The energy-saving parameter analysis report generation model finally trained should use energy-saving reference information as input and output an energy-saving parameter analysis report containing energy-saving parameters. Since the embodiment of the present application constructs the energy-saving parameter analysis report generation model by fine-tuning the large language model, it is necessary to convert the energy-saving reference information into sample prompt words that the large language model can understand, and then input the sample prompt words into the large language model.

[0076] The embodiments of the present application do not specifically limit how to construct the sample prompt words, as long as the sample prompt words contain a task instruction requiring the large language model to generate an energy-saving parameter analysis report based on the energy-saving reference information, and the task instruction can be parsed by the large language model. For example, when the energy-saving reference information is the downlink transmission rate of the base station in the energy-saving state, the user's score for the network quality in the energy-saving state, and the target energy-saving value, and the energy-saving parameter is the specific number of channels that are shut down when the channel is shut down, the sample prompt words can be "Task: Based on the target energy-saving value, the downlink transmission rate of the base station in the energy-saving state in the previous cycle, and the user's score for the network quality in the energy-saving state, determine the number of channels that the base station will shut down when the channel is shut down in the next cycle. And based on the determined number of channels to be shut down, generate an energy-saving parameter analysis report that explains the reason for adopting this number."

[0077] The sample prompt words can be obtained by pre-building a template and filling the energy-saving reference information and adjustable energy-saving parameters for the base station into the template.

[0078] Furthermore, in order to enable the large language model to generate energy-saving parameters with higher accuracy as much as possible, rather than just inferring a possible reasonable energy-saving parameter based on the pre-trained corpus, the sample prompt words may also include adjustment prompts for the energy-saving parameters. The adjustment prompts are used to indicate to the large language model the basic principles of adjusting the energy-saving parameters according to the energy-saving reference information (for example, when a certain energy-saving reference information is lower than a threshold, whether the energy-saving parameter needs to be adjusted upward or downward, or whether the value of certain energy-saving reference information is positively correlated or negatively correlated with the energy-saving parameter), the restrictions on adjusting the energy-saving parameters (for example, when a certain energy-saving reference information is lower than a threshold, the energy-saving parameter cannot be adjusted upward), and the adjustable range of the energy-saving parameter. Still taking the above energy-saving reference information as the downlink transmission rate of the base station in the energy-saving state, the user's score for the network quality in the energy-saving state and the target energy-saving value, and the energy-saving parameter as the specific number of channels to be shut down when the channel is shut down, the adjustment prompt can be "Background information: (1) The more channels are shut down, the more significant the energy-saving effect. (2) The more channels are shut down, the lower the downlink transmission rate. (3) The more channels are shut down, the lower the user's score for network quality. (4) The reasonable downlink transmission rate is between A and B. (5) The reasonable score is between C and D. (6) When the user's score for network quality is lower than E, it is not allowed to increase the number of shutdown channels. (7) The number of shutdown channels should be between F and G."

[0079] Multiple adjustment prompts can be set for each energy-saving reference information, and the set adjustment prompts may include (1) the relationship between the energy-saving reference information and each energy-saving parameter; (2) the relationship between the energy-saving reference information and other energy-saving reference information; (3) the reasonable numerical range of the energy-saving reference information. Multiple adjustment prompts can also be set for each energy-saving parameter, and the set adjustment prompts may include (1) the relationship between the energy-saving parameter and the energy-saving value; (2) the relationship between the energy-saving parameter and other energy-saving parameters; (3) the reasonable numerical range of the energy-saving parameter. These adjustment prompts can be written by those skilled in the art based on experience. When generating adjustment prompts for a base station, adjustment prompts related to each energy-saving reference information and adjustment prompts related to each energy-saving parameter can be obtained based on the energy-saving reference information and adjustable energy-saving parameters of the base station, and these adjustment prompts can be spliced ​​together to obtain the final adjustment prompt.

[0080] The energy-saving parameter analysis report is used to indicate the value to which the energy-saving parameter should be adjusted based on the energy-saving reference information, as well as the reason for adjusting the energy-saving parameter to the value. If the sample prompt contains an adjustment prompt, the reason should be related to the adjustment prompt, for example, "The number of channels to be shut down should be H, for the following reasons: (1) H is between F and G, which has a significant energy-saving effect. (2) H is a lower value between F and G, which can increase the downlink transmission rate so that the downlink transmission rate remains between A and B. (3) H is a lower value between F and G, which can increase the user's score for network quality so that the downlink transmission rate remains between C and D. (4) Since the user's score for network quality in the previous cycle was lower than E, the number of channels to be shut down should not be increased in this cycle."

[0081] In the process of fine-tuning the large language model, the energy-saving parameter analysis report can be manually written by technical personnel in this field to obtain better training results.

[0082] Step 104: input the sample prompt words into the large language model to obtain a predicted energy-saving parameter analysis report output by the large language model.

[0083] In the embodiment of the present application, after obtaining the sample prompt word, the sample prompt word is input into the large language model to obtain the predicted energy saving parameter analysis report generated by the large language model. The embodiment of the present application does not limit the specific large language model used, and any large language model is applicable to the embodiment of the present application.

[0084] Step 106, determining a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determining a loss value of the large language model according to the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value.

[0085] In an embodiment of the present application, it is necessary to calculate the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report during fine-tuning training, and it is also necessary to determine the reward value for the energy-saving parameter analysis report based on the energy-saving feedback information, so that the energy-saving parameter analysis report generated by the large language model can be closer to those energy-saving parameter analysis reports generated based on energy-saving parameters with better energy-saving feedback information, thereby improving the accuracy of the energy-saving parameters predicted by the large language model.

[0086] For example, the energy-saving feedback information can be quantified by the energy-saving feedback score, for example, the better the energy-saving effect, the higher the energy-saving feedback score is set, the lower the user score is, the lower the energy-saving feedback score is set, etc. The reward value and the energy-saving feedback score can be further made positively correlated.

[0087] The difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report can be calculated by using any loss function when fine-tuning the large language model, such as cross entropy loss.

[0088] The final loss value can be made positively correlated with the above-mentioned difference, and the loss value can be made negatively correlated with the reward value, so that when the energy-saving parameter analysis report is constructed according to better energy-saving parameters, the reward value can amplify the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report, and when the energy-saving parameter analysis report is constructed according to poor energy-saving parameters, the reward value can reduce the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report.

[0089] Step 108, fine-tuning the large language model according to the loss value to obtain an energy-saving parameter analysis report generation model.

[0090] In the embodiment of the present application, the large language model is fine-tuned according to the loss value until the loss value is less than a threshold or the number of training rounds reaches a threshold, and a model capable of generating an energy-saving parameter analysis report can be obtained. The embodiment of the present application does not specifically limit the fine-tuning method, for example, LoRA (Low-Rank Adaptation) or various derivative methods of LoRA can be used to fine-tune the large language model.

[0091] The training method of the energy-saving parameter analysis report generation model provided in the embodiment of the present application constructs sample prompt words according to energy-saving reference information, constructs an energy-saving parameter analysis report according to energy-saving parameters, and uses the sample prompt words, the energy-saving parameter analysis report, and the reward value calculated for the energy-saving parameter analysis report to fine-tune the large language model, so that the energy-saving parameter analysis report generation model obtained after fine-tuning can generate a report similar to the energy-saving parameter analysis report with a higher reward value according to the energy-saving reference information. The embodiment of the present application can generate an energy-saving parameter analysis report with high interpretability, so that technical personnel can refer to the energy-saving parameter analysis report to infer the reason why the large language model generates specific energy-saving parameters, and then can migrate the energy-saving parameters to other scenarios based on the energy-saving parameter analysis report, reducing the difficulty of converting the energy-saving parameters.

[0092] In one embodiment, the energy-saving parameter corresponds to one or more energy-saving cycles, and the energy-saving feedback information includes energy-saving effect information corresponding to the energy-saving parameter in each energy-saving cycle; Figure 2 As shown, in step 106, the reward value for the energy-saving parameter analysis report is determined according to the energy-saving feedback information, including:

[0093] Step 202, determining an energy-saving cycle quantity reward value according to the energy-saving cycle quantity corresponding to the energy-saving parameter, wherein the energy-saving cycle quantity reward value is positively correlated with the energy-saving cycle quantity;

[0094] Step 204, determining an energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period;

[0095] Step 206: Determine a reward value for the energy-saving parameter analysis report according to the energy-saving cycle quantity reward value and the energy-saving effect reward value.

[0096] In the embodiment of the present application, the energy-saving period refers to the duration of a base station executing a certain energy-saving parameter, and the length of the energy-saving period can be set by those skilled in the art according to actual needs. If the base station does not adjust the energy-saving parameter after an energy-saving period ends, the energy-saving parameter will be postponed to the next energy-saving period, so an energy-saving parameter may correspond to one or more energy-saving periods.

[0097] The energy-saving feedback information is an array, which includes energy-saving effect information corresponding to multiple energy-saving cycles, such as the energy saving amount of the base station after saving energy according to the updated energy-saving parameters compared with before the energy-saving parameters were updated, network quality information in the energy-saving state, and user scores on network quality.

[0098] Since the more energy-saving cycles executed by the energy-saving parameter, the better the energy-saving effect of the energy-saving parameter, an energy-saving cycle number reward value can be calculated according to the number of energy-saving cycles corresponding to the energy-saving parameter. This value can be proportional to the number of energy-saving cycles, or can be related to the cumulative sum of energy-saving effect reward values ​​corresponding to each energy-saving cycle.

[0099] Furthermore, the energy-saving effect reward value can also be determined according to the energy-saving effect information corresponding to each energy-saving cycle. The energy-saving effect information corresponding to each energy-saving cycle can be scored respectively, and then the energy-saving effect reward value is obtained by taking the average of the scores or taking the average after weighting. The weight of the score can be related to the energy-saving cycle to which the energy-saving effect information belongs. For example, when the similarity between the energy-saving cycle and the current energy-saving cycle is high, the weight of the score can be greater. The similarity between two energy-saving cycles can be determined based on the similarity of the energy-saving reference information of the two energy-saving cycles and the similarity of the time range covered by the energy-saving cycles.

[0100] In one embodiment, Figure 3 As shown, the energy-saving effect information includes a network quality score and an energy-saving duration score. In step 206, the energy-saving effect reward value is determined according to the energy-saving effect information corresponding to each energy-saving period, including:

[0101] Step 302: for any energy-saving period, a weighted summation process is performed on the network quality score and the energy-saving duration score corresponding to the energy-saving period to obtain an energy-saving effect score corresponding to the energy-saving period;

[0102] Step 304: determine the energy-saving effect reward value according to the energy-saving effect score corresponding to each energy-saving cycle.

[0103] In an embodiment of the present application, the network quality score is used to characterize the network quality during the energy-saving period after energy-saving parameters are used for energy saving. The higher the score, the better the network quality. The network quality score can be determined based on the uplink transmission rate (this value may be positively correlated with the network quality score), the downlink transmission rate (this value may be positively correlated with the network quality score), the network delay (this value may be negatively correlated with the network quality score), the packet loss rate (this value may be negatively correlated with the network quality score), the number of service migrations after entering the energy-saving mode (this value may be negatively correlated with the network quality score, and the reduction in total services after entering the energy-saving mode (representing that the service is migrated to other base stations or the user cannot use the service) can be used to characterize whether the user is aware of the energy-saving mode), etc.

[0104] The energy-saving duration score represents the actual duration of the base station in energy-saving mode after the energy-saving parameters are adopted. The duration is positively correlated with the energy-saving duration score.

[0105] The technicians can adjust the weights of these two scores according to actual needs. The weighted sum of these two scores can be used to obtain the energy-saving effect score. Further, the energy-saving effect scores of each energy-saving cycle can be averaged or weighted averaged to obtain the energy-saving effect reward value. The energy-saving effect reward value and the energy-saving cycle number reward value with weights are added together to obtain the reward value, as shown in formula (1):

[0106] Formula (I)

[0107] in, is the reward value, represents the expected value, for, is the weight of the network quality score, To save energy time, is the weight of the energy-saving duration score, is the weight of the energy-saving cycle reward value, is the total number of energy-saving cycles corresponding to the energy-saving parameters, Represents the sum of the energy-saving duration scores generated in all energy-saving cycles.

[0108] The negative logarithmic function is used to calculate the difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report. The loss value of the large language model is shown in formula (II):

[0109] Formula (II)

[0110] Among them, L is the loss value, T is the total number of samples, The sample prompt word is In the case of large prediction models, energy-saving parameter analysis reports are generated probability.

[0111] In one embodiment, Figure 4 As shown, in step 102, constructing a sample prompt word according to the energy-saving reference information includes:

[0112] Step 402, determining candidate energy-saving parameters of the sample base station and an adjustment range of each candidate energy-saving parameter according to the energy-saving reference information;

[0113] Step 404, constructing background description information according to the energy-saving reference information, and constructing task description information according to each candidate energy-saving parameter and the adjustment range of each candidate energy-saving parameter;

[0114] Step 406: construct sample prompt words according to the background description information and the task description information.

[0115] In an embodiment of the present application, in order to avoid the situation where the energy-saving parameters recommended for adjustment by the large language model have no actual correlation with the energy-saving reference information, before generating the sample prompt words, the alternative energy-saving parameters related to the energy-saving reference information can be first screened out from the adjustable energy-saving parameters of the sample base station based on the energy-saving reference information.

[0116] Exemplarily, the method of the aforementioned embodiment may be referred to, and the correlation between each energy-saving parameter and each energy-saving reference information may be determined in advance based on experience. Furthermore, the strength of the correlation may be set. For example, the correlation between the number of channel shutdowns and the downlink transmission rate may be stronger, and the correlation with the packet loss rate may be weaker. When screening the alternative energy-saving parameters, the energy-saving parameters may be sorted according to the strength of the correlation for each energy-saving reference information, and the energy-saving parameters with a preset number of top rankings may be screened out as the alternative energy-saving parameters. After obtaining the alternative energy-saving parameters, a pre-set adjustment range may be obtained for the alternative energy-saving parameters.

[0117] Furthermore, background description information can be constructed based on the energy-saving reference information, and task description information can be constructed based on each alternative energy-saving parameter and the adjustment range of each alternative energy-saving parameter. The background description information is used to describe the background information that needs to be considered when the base station adjusts the energy-saving parameters, such as what each energy-saving reference information is, the meaning of each energy-saving reference information, the relationship between each energy-saving reference information and the energy-saving parameter, the target range of each energy-saving reference information, etc. For details, please refer to the relevant description of the energy-saving reference information in the "Adjustment Prompt" in the aforementioned embodiment, which will not be repeated in the embodiments of this application. The task description information is used to describe the tasks that the large language model needs to perform, as well as the information necessary to perform the task. For details, please refer to the relevant description of the energy-saving parameters in the "Task Instructions" and "Adjustment Prompts" in the aforementioned embodiments, which will not be repeated in the embodiments of this application.

[0118] The training method of the energy-saving parameter analysis report generation model provided in the embodiment of the present application, after selecting the alternative energy-saving parameters related to the energy-saving reference information according to the energy-saving reference information, constructs the sample prompt words according to the alternative energy-saving parameters. Since the large language model does not have real reasoning ability, by selecting the alternative energy-saving parameters and requiring the large language model to select the parameters that need to be adjusted from the alternative energy-saving parameters, it is possible to avoid the large language model giving energy-saving parameters that actually have no effect on the energy-saving reference information in the energy-saving parameter analysis report, thereby improving the generation accuracy of the energy-saving parameter analysis report.

[0119] In one embodiment, Figure 5 As shown, an energy-saving parameter analysis report is constructed based on the energy-saving parameters, including:

[0120] Step 502, constructing background understanding information corresponding to the background description information and task understanding information corresponding to the task description information;

[0121] Step 504, constructing task conclusion information based on the background understanding information, the task understanding information and the energy-saving parameters;

[0122] Step 506: construct an energy-saving parameter analysis report based on the background understanding information, the task understanding information, and the task conclusion information.

[0123] In the embodiment of the present application, in order to make the energy-saving parameter analysis report and the sample prompt words have a strong correlation, the structure of the energy-saving parameter analysis report can be related to the structure of the sample prompt words. The background understanding information can be constructed based on the background description information, and the background understanding information is used to indicate which energy-saving parameters need to be adjusted based on the background description information in the sample prompt words, and what impact these energy-saving parameters have on the energy-saving reference information after adjustment. The task understanding information can also be constructed based on the task description information, and the task understanding information is used to indicate how the energy-saving parameters can be adjusted based on the task description information in the sample prompt words.

[0124] After obtaining the background understanding information and task understanding information, the task conclusion information can be constructed. The task conclusion information includes the energy-saving parameters that the large language model should predict, and the reasoning process of obtaining the energy-saving parameters based on the background understanding information and task understanding information. Then, the background understanding information, task understanding information and task conclusion information can be spliced ​​to obtain an interpretable energy-saving parameter analysis report.

[0125] In one embodiment, Figure 6 As shown, a training method for an energy-saving parameter analysis report generation model is provided, comprising:

[0126] S1, clearly define the energy-saving parameters that need to be optimized, and define the upper and lower limits of the thresholds for each energy-saving parameter that are in line with the actual business situation; determine the energy-saving reference information that needs to be collected to calculate the energy-saving parameter reward and punishment function, such as the energy-saving parameters actually implemented by the base station, energy-saving duration, energy consumption saved, and corresponding network indicators during energy saving.

[0127] S2, based on the energy-saving reference information collected by S1, constructs a sample prompt word X. The sample prompt word contains four parts: (1) description of the optimizable energy-saving parameters, (2) description of the optimization task, (3) description of expert experience, and (4) space for suggesting threshold values ​​for energy-saving parameters.

[0128] The expert experience is used to characterize the relationship between each energy-saving reference information and each energy-saving parameter, how each energy-saving reference information can adjust the energy-saving parameter under what conditions, etc. The expert experience can be set by those skilled in the art based on experience.

[0129] S3, taking the energy-saving parameter analysis report Y including the energy-saving parameter a of the historical execution as the expected output of the large language model.

[0130] S4 combines X and Y collected in S3 into a sample pair for fine-tuning training, and uses the loss function including the energy-saving parameter reward and punishment function (ie, R in the formula) shown in the above formula (ii) to perform fine-tuning training on the large language model.

[0131] S5, based on the model fine-tuned in S4, interacts with the current energy-saving status, network quality and user perception data of the cell where the target base station is located, obtains an energy-saving parameter analysis report, and obtains recommended energy-saving parameters from the energy-saving parameter analysis report.

[0132] The training method of the energy-saving parameter analysis report generation model mentioned above uses large model fine-tuning learning to realize automatic tuning of base station energy-saving parameters, combines the reward and punishment function of base station energy-saving parameter self-optimization to improve the fine-tuning method, and enables the model to have the ability of self-optimization of base station energy-saving parameters through fine-tuning learning.

[0133] In one embodiment, Figure 7 As shown, a method for adjusting energy-saving parameters is provided, comprising:

[0134] Step 702, obtaining energy-saving reference information of the target base station, and constructing a target prompt word according to the energy-saving reference information;

[0135] Step 704: input the target prompt word into the energy-saving parameter analysis report generation model to obtain the energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained by using the method of any of the above embodiments;

[0136] Step 706: Adjust the current energy-saving parameters of the target base station according to the energy-saving parameter analysis report.

[0137] In the embodiment of the present application, for the target base station that needs to adjust the energy-saving parameters, the energy-saving reference information of the target base station can be obtained. Then, the method for constructing sample prompt words in the above embodiment can be referred to, and the target prompt words can be constructed according to the energy-saving reference information of the target base station.

[0138] Furthermore, the target prompt word can be input into the energy-saving parameter analysis report generation model to obtain the energy-saving parameter analysis report. The energy-saving parameter analysis report includes adjustment suggestions for each energy-saving parameter and the reasons for generating the adjustment suggestions.

[0139] Then, the current energy-saving parameters of the target base station can be adjusted with reference to the energy-saving parameter analysis report. For example, the energy-saving parameters recommended in the energy-saving parameter analysis report can be used as new energy-saving parameters. Alternatively, the step of inputting the target prompt words into the energy-saving parameter analysis report generation model can be repeated multiple times, and then according to the actual performance of each energy-saving parameter recommended in the energy-saving parameter analysis report on the target base station in the past, a parameter with the best actual performance is selected as the new energy-saving parameter.

[0140] The energy-saving parameter adjustment method provided in the embodiment of the present application constructs a target prompt word according to the energy-saving reference information, inputs the target prompt word into the energy-saving parameter analysis report generation model, obtains the energy-saving parameter analysis report, and then adjusts the energy-saving parameters according to the energy-saving parameter analysis report. The embodiment of the present application can generate an energy-saving parameter analysis report with high interpretability, so that the technician can refer to the energy-saving parameter analysis report to infer the reason why the large language model generates specific energy-saving parameters, and then can migrate the energy-saving parameters to other scenarios based on the energy-saving parameter analysis report, reducing the difficulty of converting the energy-saving parameters.

[0141] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0142] Based on the same inventive concept, the embodiment of the present application also provides a training device for an energy-saving parameter analysis report generation model for implementing the training method for the energy-saving parameter analysis report generation model involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the training device for one or more energy-saving parameter analysis report generation models provided below can be referred to the limitations of the training method for the energy-saving parameter analysis report generation model above, and will not be repeated here.

[0143] In one embodiment, Figure 8 As shown, a training device 800 for generating an energy-saving parameter analysis report model is provided, comprising: an acquisition module 802, an input module 804, a determination module 806, and a fine-tuning module 808, wherein:

[0144] An acquisition module 802 is used to acquire energy-saving reference information, energy-saving parameters and energy-saving feedback information of the energy-saving parameters of the sample base station, construct a sample prompt word according to the energy-saving reference information, and construct an energy-saving parameter analysis report according to the energy-saving parameters;

[0145] An input module 804, used to input the sample prompt words into a large language model to obtain a predicted energy-saving parameter analysis report output by the large language model;

[0146] A determination module 806, configured to determine a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determine a loss value of the large language model according to a difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value;

[0147] The fine-tuning module 808 is used to fine-tune the large language model according to the loss value to obtain an energy-saving parameter analysis report generation model.

[0148] The training device of the energy-saving parameter analysis report generation model provided in the embodiment of the present application constructs sample prompt words according to energy-saving reference information, constructs an energy-saving parameter analysis report according to energy-saving parameters, and uses the sample prompt words, the energy-saving parameter analysis report, and the reward value calculated for the energy-saving parameter analysis report to fine-tune the large language model, so that the energy-saving parameter analysis report generation model obtained after fine-tuning can generate a report similar to the energy-saving parameter analysis report with a higher reward value according to the energy-saving reference information. The embodiment of the present application can generate an energy-saving parameter analysis report with high interpretability, so that technical personnel can refer to the energy-saving parameter analysis report to infer the reason why the large language model generates specific energy-saving parameters, and then can migrate the energy-saving parameters to other scenarios based on the energy-saving parameter analysis report, reducing the difficulty of converting the energy-saving parameters.

[0149] In one embodiment, the energy-saving parameter corresponds to one or more energy-saving cycles, and the energy-saving feedback information includes energy-saving effect information corresponding to the energy-saving parameter in each of the energy-saving cycles;

[0150] The determining module 806 is further configured to:

[0151] Determine a reward value for the number of energy-saving cycles according to the number of energy-saving cycles corresponding to the energy-saving parameter, wherein the reward value for the number of energy-saving cycles is positively correlated with the number of energy-saving cycles;

[0152] Determining an energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period;

[0153] A reward value for the energy-saving parameter analysis report is determined according to the energy-saving cycle quantity reward value and the energy-saving effect reward value.

[0154] In one embodiment, the energy saving effect information includes a network quality score and an energy saving duration score, and the determination module 806 is further used to:

[0155] For any of the energy-saving cycles, weighted summing is performed on the network quality score and the energy-saving duration score corresponding to the energy-saving cycle to obtain an energy-saving effect score corresponding to the energy-saving cycle;

[0156] An energy-saving effect reward value is determined according to the energy-saving effect score corresponding to each energy-saving cycle.

[0157] In one embodiment, the acquisition module 802 is further used to:

[0158] Determining, according to the energy-saving reference information, candidate energy-saving parameters of the sample base station and an adjustment range of each of the candidate energy-saving parameters;

[0159] Constructing background description information according to the energy-saving reference information, and constructing task description information according to each of the candidate energy-saving parameters and the adjustment range of each of the candidate energy-saving parameters;

[0160] A sample prompt word is constructed according to the background description information and the task description information.

[0161] In one embodiment, the acquisition module 802 is further used to:

[0162] Constructing background understanding information corresponding to the background description information and task understanding information corresponding to the task description information;

[0163] constructing task conclusion information based on the background understanding information, the task understanding information and the energy-saving parameter;

[0164] An energy-saving parameter analysis report is constructed according to the background understanding information, the task understanding information and the task conclusion information.

[0165] In one embodiment, Fig. 9 As shown, an energy-saving parameter adjustment device 900 is provided, including: an acquisition module 902, an input module 904, and an adjustment module 906, wherein:

[0166] An acquisition module 902 is used to acquire energy-saving reference information of a target base station and construct a target prompt word according to the energy-saving reference information;

[0167] An input module 904 is used to input the target prompt word into the energy-saving parameter analysis report generation model to obtain an energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained using the method described in any of the above embodiments;

[0168] The adjustment module 906 is configured to adjust the current energy-saving parameters of the target base station according to the energy-saving parameter analysis report.

[0169] The energy-saving parameter adjustment device provided in the embodiment of the present application constructs a target prompt word according to the energy-saving reference information, inputs the target prompt word into the energy-saving parameter analysis report generation model, obtains the energy-saving parameter analysis report, and then adjusts the energy-saving parameters according to the energy-saving parameter analysis report. The embodiment of the present application can generate an energy-saving parameter analysis report with high interpretability, so that technicians can refer to the energy-saving parameter analysis report to infer the reason why the large language model generates specific energy-saving parameters, and then can migrate the energy-saving parameters to other scenarios based on the energy-saving parameter analysis report, reducing the difficulty of converting the energy-saving parameters.

[0170] Each module in the above device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module above.

[0171] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a training method for generating an energy-saving parameter analysis report is implemented.

[0172] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0173] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0174] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0175] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0176] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0177] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0178] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0179] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for generating an energy-saving parameter analysis report model, characterized in that: The method comprises: Acquire energy-saving reference information, energy-saving parameters and energy-saving feedback information of the sample base stations, construct sample prompt words according to the energy-saving reference information, and construct an energy-saving parameter analysis report according to the energy-saving parameters; Inputting the sample prompt words into a large language model to obtain a predicted energy-saving parameter analysis report output by the large language model; determining a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determining a loss value of the large language model according to a difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value; The large language model is fine-tuned according to the loss value to obtain an energy-saving parameter analysis report generation model.

2. The method according to claim 1, characterized in that The energy-saving parameter corresponds to one or more energy-saving cycles, and the energy-saving feedback information includes energy-saving effect information corresponding to the energy-saving parameter in each of the energy-saving cycles; The determining, according to the energy-saving feedback information, a reward value for the energy-saving parameter analysis report includes: Determine a reward value for the number of energy-saving cycles according to the number of energy-saving cycles corresponding to the energy-saving parameter, wherein the reward value for the number of energy-saving cycles is positively correlated with the number of energy-saving cycles; Determining an energy-saving effect reward value according to the energy-saving effect information corresponding to each energy-saving period; A reward value for the energy-saving parameter analysis report is determined according to the energy-saving cycle quantity reward value and the energy-saving effect reward value.

3. The method according to claim 2, characterized in that The energy-saving effect information includes a network quality score and an energy-saving duration score, and the energy-saving effect reward value is determined according to the energy-saving effect information corresponding to each energy-saving period, including: For any of the energy-saving cycles, weighted summing is performed on the network quality score and the energy-saving duration score corresponding to the energy-saving cycle to obtain an energy-saving effect score corresponding to the energy-saving cycle; An energy-saving effect reward value is determined according to the energy-saving effect score corresponding to each energy-saving cycle.

4. The method according to claim 1, characterized in that: The step of constructing a sample prompt word according to the energy-saving reference information includes: Determining, according to the energy-saving reference information, candidate energy-saving parameters of the sample base station and an adjustment range of each of the candidate energy-saving parameters; Constructing background description information according to the energy-saving reference information, and constructing task description information according to each of the candidate energy-saving parameters and the adjustment range of each of the candidate energy-saving parameters; A sample prompt word is constructed according to the background description information and the task description information.

5. The method according to claim 4, characterized in that The step of constructing an energy-saving parameter analysis report according to the energy-saving parameters includes: Constructing background understanding information corresponding to the background description information and task understanding information corresponding to the task description information; constructing task conclusion information based on the background understanding information, the task understanding information and the energy-saving parameter; An energy-saving parameter analysis report is constructed according to the background understanding information, the task understanding information and the task conclusion information.

6. A method for adjusting energy-saving parameters, characterized in that: The method comprises: Acquire energy-saving reference information of a target base station, and construct a target prompt word according to the energy-saving reference information; Inputting the target prompt word into the energy-saving parameter analysis report generation model to obtain an energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained using the method described in any one of claims 1 to 5; According to the energy-saving parameter analysis report, current energy-saving parameters of the target base station are adjusted.

7. A training device for generating an energy-saving parameter analysis report model, characterized in that: The device comprises: An acquisition module, used to acquire energy-saving reference information, energy-saving parameters and energy-saving feedback information of the sample base stations, construct sample prompt words according to the energy-saving reference information, and construct an energy-saving parameter analysis report according to the energy-saving parameters; An input module, used for inputting the sample prompt words into a large language model to obtain a predicted energy-saving parameter analysis report output by the large language model; a determination module, configured to determine a reward value for the energy-saving parameter analysis report according to the energy-saving feedback information, and determine a loss value of the large language model according to a difference between the predicted energy-saving parameter analysis report and the energy-saving parameter analysis report and the reward value; A fine-tuning module is used to fine-tune the large language model according to the loss value to obtain an energy-saving parameter analysis report generation model.

8. An energy-saving parameter adjustment device, characterized in that: The device comprises: An acquisition module, used to acquire energy-saving reference information of a target base station, and construct a target prompt word according to the energy-saving reference information; An input module, used to input the target prompt word into the energy-saving parameter analysis report generation model to obtain an energy-saving parameter analysis report for the target base station; wherein the energy-saving parameter analysis report generation model is trained using the method described in any one of claims 1 to 5; An adjustment module is used to adjust the current energy-saving parameters of the target base station according to the energy-saving parameter analysis report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images