Base station intelligent energy coordination method and system

Abstract

The application discloses a base station intelligent energy coordination method and system, and belongs to the technical field of energy management, and the technical solution points comprise the following steps: collecting a data set of each base station in a regional base station cluster; calculating energy constraint data based on energy data and environment data to obtain an energy scheduling sequence of each base station; obtaining a load prediction curve of each base station through a preset LSTM network model based on load data to determine a first base station; and obtaining a corresponding second base station and a third base station according to the load prediction curve of the first base station. The application obtains the load prediction curve of each base station through the preset LSTM network model and determines the first base station, obtains the second base station through the analytic hierarchy process, and obtains the third base station through multi-target reinforcement learning, so that the energy scheduling of the regional base station cluster is realized, the clean energy utilization efficiency and the energy scheduling precision are improved, and the stability of base station operation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, and more particularly to a base station intelligent energy collaborative scheduling method and system. BACKGROUND

[0002] With the increase of 5G base station density and energy consumption demand, the existing technology mainly adopts independent energy control mode taking a single base station as a unit, and the dispersed energy is in an isolated state, the utilization rate of clean energy is low, and the dependence on city power is high. In addition, the load prediction only depends on single dimension data, and the prediction result is out of touch with the actual energy scheduling demand, and the efficient use and safety guarantee of the regional base station cluster internal energy cannot be realized. Therefore, the existing technology has deficiencies in energy management. SUMMARY

[0003] In view of the deficiencies of the prior art, the purpose of the present application is to provide a base station intelligent energy collaborative scheduling method and system, by collecting the energy, load and environment data set of each base station in the regional base station cluster, calculating the energy constraint data and obtaining the energy scheduling sequence, and based on the preset LSTM network model, obtaining the load prediction curve, load prediction value, load peak valley period and first energy shortage of each base station. Based on the above load and energy parameters, the first base station is determined, and the second base station for data reception is obtained by the analytic hierarchy process, and the third base station for energy transmission is obtained by multi-objective reinforcement learning. According to the coordinated linkage of the first base station, the second base station and the third base station, the energy and load bidirectional collaborative scheduling of the regional base station cluster is realized, the clean energy utilization efficiency and the energy scheduling accuracy and the stability of the base station operation are improved, and the problems of high dependence on city power, scheduling decision relying on static experience, unreasonable energy configuration structure and high operating cost in traditional base station management are solved.

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] The base station intelligent energy collaborative scheduling method comprises:

[0006] Collecting the data set of each base station in the regional base station cluster, the data set comprising energy data, load data and environment data;

[0007] Based on the energy data and the environment data, the energy constraint data is calculated to obtain the energy scheduling sequence of each base station, the energy constraint data comprising photovoltaic prediction value, energy storage available capacity and city power data;

[0008] Based on the load data, the load prediction curve of each base station is obtained by the preset LSTM network model;

[0009] According to the load prediction curve, a load prediction value, a load peak valley period and a first energy shortage of each base station are obtained;

[0010] A first base station is determined according to the load prediction value, the load peak valley period and the first energy shortage of each base station;

[0011] According to the energy scheduling sequence, the load prediction value, the load peak valley period and the first energy shortage of the first base station, a corresponding second base station and a third base station are obtained, the second base station is used for receiving data transmitted by the first base station, and the third base station is used for transmitting energy to the first base station.

[0012] As a further improvement of the present application, the load prediction value, the load peak valley period and the first energy shortage of each base station are obtained according to the load prediction curve, comprising:

[0013] The load prediction value of each base station is obtained according to the load prediction curve and energy constraint data of each base station;

[0014] The load rate of each base station is calculated based on the load prediction curve, and the load peak valley period of each base station is determined according to the load rate of each base station;

[0015] The first energy shortage of each base station is obtained through the load prediction value and the energy constraint data.

[0016] As a further improvement of the present application, the first base station is determined according to the load prediction value, the load peak valley period and the first energy shortage of each base station, comprising:

[0017] The base station with a first energy shortage greater than 0 is determined as an evaluated base station according to the load prediction value, the load peak valley period and the first energy shortage of each base station;

[0018] The first base station is determined based on the load rate of the evaluated base station.

[0019] As a further improvement of the present application, the second base station and the third base station are obtained according to the energy scheduling sequence, the load prediction value, the load peak valley period and the first energy shortage of the first base station, the second base station is used for receiving data transmitted by the first base station, and the third base station is used for transmitting energy to the first base station, comprising:

[0020] The second base station is obtained by an analytic hierarchy process based on the load rate of a peripheral base station according to the load peak valley period and the load prediction value of the first base station;

[0021] obtaining a third base station through multi-objective reinforcement learning based on a second energy shortage of the first base station and an energy redundancy of a surrounding base station, the second energy shortage being obtained based on the first energy shortage and the second base station.

[0022] As a further improvement of the present application, the second base station is obtained through analytic hierarchy process based on a load rate of a surrounding base station according to a load peak-valley period and a load prediction value of the first base station, comprising:

[0023] obtaining a load redundancy of the first base station according to a load peak-valley period and a load prediction value of the first base station;

[0024] determining an evaluation index according to a load rate of the surrounding base station and a first distance from the first base station;

[0025] constructing a judgment matrix through analytic hierarchy process based on the evaluation index to obtain a ranking of candidate samples;

[0026] obtaining a second base station according to the ranking of candidate samples.

[0027] As a further improvement of the present application, the third base station is obtained through multi-objective reinforcement learning based on a second energy shortage of the first base station and an energy redundancy of a surrounding base station, the second energy shortage being obtained based on the first energy shortage and the second base station, comprising:

[0028] obtaining a second energy shortage of the first base station based on a first energy shortage of the first base station and the second base station;

[0029] obtaining an optimization target of multi-objective reinforcement learning according to an energy redundancy of a surrounding base station, a distance from the first base station, and a transmission path loss rate;

[0030] obtaining a third base station through multi-objective reinforcement learning according to the optimization target.

[0031] As a further improvement of the present application, if the energy redundancy of the third base station is less than the second energy shortage, a plurality of third base stations is obtained based on the second energy shortage, comprising:

[0032] obtaining a remaining energy shortage according to the second energy shortage of the first base station and an energy redundancy of a selected third base station;

[0033] obtaining a remaining candidate third base station based on the multi-objective reinforcement learning;

[0034] determining a plurality of third base stations according to the energy redundancy of the remaining candidate third base station and the remaining energy shortage.

[0035] As a further improvement of the application, the energy constraint data is calculated based on the energy data and the environment data, and the energy scheduling sequence of each base station is obtained, wherein the energy constraint data comprises photovoltaic prediction value, energy storage available capacity and power grid data, which comprises:

[0036] The photovoltaic data, energy storage data and power grid data of each base station are obtained based on the energy data;

[0037] The photovoltaic prediction value is obtained according to the photovoltaic data and a preset photovoltaic prediction model;

[0038] The energy storage available capacity is obtained according to the energy storage data and a preset energy storage available capacity model;

[0039] The energy constraint data is obtained based on the photovoltaic prediction value, energy storage available capacity and power grid data, and the energy scheduling sequence of each base station is obtained according to the energy constraint data.

[0040] As a further improvement of the application, the preset LSTM network model is obtained by training, which comprises:

[0041] Based on the input feature set, the training set, the validation set and the test set are obtained by preset proportion division;

[0042] Based on the input feature set, the LSTM network structure is built, and the initial LSTM neural network model is obtained by initializing network parameters;

[0043] The initial LSTM neural network model is iteratively trained by the training set, the model performance is monitored by the validation set and the parameters are adjusted, and the optimized LSTM neural network model is obtained;

[0044] The prediction accuracy of the optimized LSTM neural network model is verified based on the test set, and the preset LSTM network model is obtained.

[0045] The application provides a base station intelligent energy collaborative scheduling system, which comprises a data processing module, a prediction module and an energy scheduling module:

[0046] The data processing module: collects the data set of each base station in the regional base station cluster, wherein the data set comprises energy data, load data and environment data; the energy constraint data is calculated based on the energy data and the environment data, and the energy scheduling sequence of each base station is obtained, wherein the energy constraint data comprises photovoltaic prediction value, energy storage available capacity and power grid data;

[0047] The prediction module: the load prediction curve of each base station is obtained by a preset LSTM network model based on the load data; the load prediction value, load peak valley period and first energy shortage of each base station are obtained according to the load prediction curve;

[0048] The energy deployment module determines a first base station according to the load prediction value, load peak valley period and first energy shortage of each base station; and obtains a corresponding second base station and third base station according to the energy scheduling sequence, load prediction value, load peak valley period and first energy shortage of the first base station, wherein the second base station is used to receive data transmitted by the first base station, and the third base station is used to transmit energy to the first base station.

[0049] The present application calculates photovoltaic prediction value, energy storage available capacity, and power grid data, etc. by collecting regional base station cluster energy, load and environmental data sets, and determines energy scheduling sequence based on the data sets. Meanwhile, the present application trains LSTM network model by using load data and energy scheduling sequence, and obtains load prediction curve, load prediction value, load peak valley period, first energy shortage, etc. of each base station. According to the above parameters, the present application determines the second base station which undertakes data unloading based on analytic hierarchy process, and obtains the third base station which provides energy supply by combining multi-objective reinforcement learning. Through the collaborative operation among the base stations in the regional base station cluster, the present application realizes intelligent energy deployment of the regional base station cluster, improves clean energy utilization rate, scheduling decision accuracy and base station operation reliability, and solves the problems of strong dependence on power grid, low load and energy matching degree, and high operation cost in traditional base station energy management. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 The step flow chart of the base station intelligent energy collaborative deployment method of the present application;

[0051] Figure 2 The step flow chart of predicting base station load by the preset LSTM network model;

[0052] Figure 3 The step flow chart of determining the second base station and the third base station;

[0053] Figure 4 The schematic diagram of determining the second base station and the third base station by the residential base station cluster according to the first base station;

[0054] Figure 5 The structural schematic diagram of the base station intelligent energy collaborative deployment system of the present application. DETAILED DESCRIPTION

[0055] The technical solutions of the present application will be described in detail below by means of the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical solutions of the present application, rather than limitations of the technical solutions of the present application.

[0056] Wherein the same parts are indicated by the same reference signs. The term "and / or" in the following description is merely an associated relationship between associated objects, which means that there can be three relationships, for example, A and / or B can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " generally represents that the associated objects before and after are in an "or" relationship.

[0057] As shown in Figure 1 The application provides a base station intelligent energy coordination method, comprising:

[0058] Collecting a data set of each base station in the regional base station cluster, the data set including energy data, load data, and environmental data;

[0059] Calculating energy constraint data based on the energy data and the environmental data to obtain an energy scheduling sequence of each base station, the energy constraint data including photovoltaic prediction value, energy storage available capacity, and mains data;

[0060] Obtaining a load prediction curve of each base station through a preset LSTM network model based on the load data;

[0061] Obtaining a load prediction value, a load peak-valley period, and a first energy shortage of each base station according to the load prediction curve;

[0062] Determining a first base station according to the load prediction value, the load peak-valley period, and the first energy shortage of each base station;

[0063] Obtaining a corresponding second base station and a third base station according to the energy scheduling sequence, the load prediction value, the load peak-valley period, and the first energy shortage of the first base station, the second base station being configured to receive data transmitted by the first base station, and the third base station being configured to transmit energy to the first base station.

[0064] The embodiment reduces the dependence of the base station on mains energy and improves the utilization efficiency of clean energy through the coordinated scheduling of the base station cluster in the region; improves the prediction and scheduling accuracy through the collaborative decision-making mechanism of the analytic hierarchy process and multi-objective reinforcement learning based on the LSTM network model to predict the base station load; compensates for the energy shortage of the base station through the bidirectional coordination of load data migration and energy supply, enhances the operation stability of the base station, optimizes the energy allocation structure, and reduces the operating cost.

[0065] Further, the embodiment provides a step of calculating energy constraint data based on energy data and environmental data to obtain an energy scheduling sequence of each base station, comprising:

[0066] Obtaining photovoltaic data, energy storage data, and mains data of each base station based on the energy data, and obtaining illumination intensity and environmental temperature of each base station based on the environmental data;

[0067] obtaining a photovoltaic prediction value according to photovoltaic data and a preset photovoltaic prediction model;

[0068] obtaining a storage available capacity according to storage data and a preset storage available capacity model;

[0069] obtaining an energy constraint data based on the photovoltaic prediction value, the storage available capacity and power grid data, and obtaining an energy scheduling sequence of each base station according to the energy constraint data.

[0070] The energy data is obtained according to a data set, including photovoltaic data, storage data and power grid data of each base station; the photovoltaic data includes photovoltaic power, voltage and current, the storage data includes a state of charge (SOC) of a battery of a storage device, a state of health (SOH) of the battery and a charge and discharge current, and the power grid data includes an electricity consumption and a price period (peak, flat and valley); the environmental data is obtained through a data set, including the intensity of illumination and the environmental temperature of each base station; the preset photovoltaic prediction model is obtained by training through a BP neural network and an XGBoost algorithm based on historical photovoltaic data and historical environmental data of a corresponding period; the storage available capacity model is obtained according to the SOC and SOH of the storage device of each base station; and the energy scheduling sequence is the use sequence of photovoltaic energy, storage energy and power grid energy under different working conditions.

[0071] Specifically, first, historical photovoltaic data and historical environmental data of each base station within a first preset time are obtained based on the energy data, a prediction result of a BP neural network is obtained by learning a nonlinear relationship between the intensity of illumination, the environmental temperature and the photovoltaic data, the prediction result of the BP neural network is corrected by introducing a cloud cover obstruction coefficient to construct an objective function based on an XGBoost algorithm, and a photovoltaic prediction formula is obtained:

[0072]

[0073] wherein is the photovoltaic prediction value, is the photovoltaic prediction value of the BP neural network, is the cloud cover obstruction coefficient, which is calculated according to the cloud amount in meteorological data, is the intensity of illumination and the environmental temperature within a second preset time; is a weight coefficient calculated through the BP neural network, the photovoltaic prediction formula is obtained by combining the nonlinear fitting capability of the BP neural network and the error correction of the XGBoost algorithm to adapt to photovoltaic prediction requirements under complex weather conditions; and then a preset storage available capacity model is obtained based on historical storage data of each base station within the first preset time: wherein is a storage nominal capacity, is a storage state of charge, a temperature correction coefficient fitted by data association according to first energy storage data, is an emergency backup capacity; the energy storage available capacity of each base station is calculated according to the SOC in the second preset time through the energy storage available capacity model; finally, the photovoltaic energy is obtained according to the photovoltaic prediction value, and the energy scheduling order of each base station is obtained based on the photovoltaic energy, the energy storage available capacity and the grid energy: when the photovoltaic energy is greater than the base station load energy, the photovoltaic power is preferentially used, and the excess photovoltaic is used to charge the energy storage device, and the state of charge of the energy storage is monitored in real time during the charging process to avoid overcharging of the energy storage; when the photovoltaic energy is less than the base station load energy, the photovoltaic power is preferentially used in full, and the difference is supplemented by the grid, and the energy storage device remains in standby state to deal with sudden grid interruption; when the sudden grid interruption occurs, the energy storage device is preferentially used for power supply, and the state of charge of the energy storage is monitored in real time while the staff is pushed the grid interruption alarm. In the embodiment, the first preset time can be a period of time with the current time as the end point, and the second preset time can be a period of time with the current time as the starting point; the length of the first preset time and the second preset time is not limited in the embodiment, and can be set according to actual conditions by those skilled in the art; the first energy storage data in the embodiment is the historical energy storage data of each base station.

[0074] In the embodiment, the photovoltaic prediction value is obtained by the preset BP neural network, and the energy scheduling order is obtained based on the collaborative analysis of multiple source data, which breaks through the limitations of data dispersion and scheduling decision relying on static experience in traditional base station energy management, realizes the accurate quantization of energy constraint data and the scientific sorting of scheduling order, and significantly improves the global collaboration of base station cluster energy scheduling.

[0075] Further, the embodiment provides a step of obtaining a load prediction curve of each base station based on load data and an energy scheduling order through a preset LSTM network model, comprising:

[0076] constructing an input feature set according to historical load data, training the LSTM network model based on the input feature set, and obtaining the preset LSTM network model;

[0077] obtaining the load prediction curve of each base station in the second preset time through the preset LSTM network model.

[0078] The load data includes the device power consumption, the number of user connections and the traffic of the base station; the preset LSTM network model is trained through the historical load data in the first preset time; the load prediction curve is a continuous time series curve reflecting the dynamic change of the load of each base station in the second preset time, the abscissa of the load prediction curve is time, and the ordinate is the predicted value of the load power.

[0079] Specifically, as Figure 2As shown, first load information in a first preset time is obtained according to load data, the first load information is load time series data collected in the first preset time according to a preset time interval, the first load information is divided according to a sliding window method, and an input feature set is obtained according to an energy scheduling sequence; then, an LSTM network model is trained based on the input feature set, the input feature set is divided into a training set, a validation set and a test set according to a preset proportion, an LSTM network structure is built, the LSTM network model is iteratively trained through the training set, the mean absolute percentage error (MAPE) is monitored through the validation set and the network parameters are adjusted, the model generalization ability is verified through the test set, and a preset LSTM network model is obtained; finally, a load prediction curve is obtained through the preset LSTM network model based on second load data of each base station in a second preset time.

[0080] For example, assuming that for a base station B1 of a business circle base station cluster, a first preset time is first set to the past 30 days, load data of B1 is collected according to a time interval of 15 minutes, and an input feature set is constructed in combination with an energy scheduling priority of B1; then, an LSTM network architecture is built, and a preset LSTM network model is obtained through multiple iterative training; finally, current feature data of B1 is input, and a load prediction curve of the future 24 hours is obtained.

[0081] The embodiment combines load time series features and energy scheduling sequences to construct an input feature set, captures load dynamic change rules based on the time series modeling capability of an LSTM network, solves the problem that traditional load prediction ignores energy scheduling correlation factors, ensures the reliability of the prediction curve through multiple rounds of training and precision verification, and effectively improves the accuracy of base station energy deployment decisions.

[0082] Further, the embodiment provides a step of training an LSTM network based on an input feature set to obtain a preset LSTM network model, including:

[0083] Based on the input feature set, a training set, a validation set and a test set are obtained through a preset proportion;

[0084] Based on the input feature set, an LSTM network structure is built, and an initial LSTM neural network model is obtained through initialization of network parameters;

[0085] The initial LSTM neural network model is iteratively trained through the training set, the model performance is monitored through the validation set and the parameters are adjusted, and an optimized LSTM neural network model is obtained;

[0086] The prediction accuracy of the optimized LSTM neural network model is verified based on the test set, and a preset LSTM network model is obtained.

[0087] The preset proportion is determined by training requirements; and the prediction accuracy of the LSTM neural network model is determined by a mean absolute percentage error (MAPE), which is calculated according to the deviation between the predicted load value and the corresponding actual load value of the test set.

[0088] Specifically, first, the input feature set is divided by random sampling, and the load time sequence distribution characteristics of each data are kept consistent with the original input feature set during the division process to avoid model training deviation caused by data distribution deviation; then, an initial LSTM neural network model is obtained based on the dimension and the number of features of the input feature set, the initial LSTM neural network model includes two hidden layers, the activation function of the hidden layer is a ReLU function, the dimension of the output layer is 1, and the output layer adopts a linear activation function; then, the network parameters are initialized, including setting the learning rate to 0.001, the iteration number to 200, the optimizer to an Adam optimizer, and the loss function to a mean square error (MSE) function; then, the initial LSTM neural network model is iteratively trained by using the training set, the model performance is monitored by using the verification set, and the parameters are adjusted to obtain an optimized LSTM neural network model; finally, the predicted load sequence is obtained by using the optimized LSTM neural network model based on the test set, the MAPE of the predicted load sequence and the actual load sequence of the test set is calculated, and when the MAPE is less than or equal to 5%, the model is determined as the preset LSTM network model. The structure and parameter setting of the initial LSTM neural network in this embodiment are only examples, and this embodiment does not limit the same, and a person skilled in the art can set them according to actual conditions.

[0089] Further, the embodiment provides steps of obtaining a load prediction value, a load peak-valley period and a first energy shortage of each base station according to the load prediction curve, including:

[0090] Obtaining the load prediction value of each base station according to the load prediction curve and the energy constraint data of each base station;

[0091] Calculating the load rate of each base station based on the load prediction curve, and determining the load peak-valley period of each base station according to the load rate of each base station;

[0092] Obtaining the first energy shortage of each base station by the load prediction value and the energy constraint data.

[0093] The load prediction value is the average load power of each base station in the second preset time; the load rate is the ratio of the load prediction value to the rated load power of the base station, expressed in percentage (%); and the first energy shortage is the difference between the total energy consumed by each base station in the second preset time and the total energy provided by the energy constraint data.

[0094] Specifically, first, based on the load prediction curve, the total load consumption energy and the average load power of the base station in the second preset time are obtained by integral operation in the interval of the second preset time, that is, the load prediction value; then the load rate of the base station is obtained according to the ratio of the load prediction value and the rated load power of the base station, and the peak threshold of the preset load rate is 70% and the valley threshold is 50%, and the load peak valley period of the base station is determined according to the preset load rate threshold; then the first energy shortage is obtained by calculating the total load consumption energy of the base station in the second preset time and the total energy supply of the base station based on the energy constraint data, and the first energy shortage is the difference between the total load consumption energy and the total energy supply of the base station, if the difference is greater than zero, it represents that the base station has an energy gap, and if the difference is less than or equal to zero, it represents that the base station has sufficient energy, wherein the total energy supply of the base station is the sum of the photovoltaic energy, the available capacity of the energy storage and the commercial power energy. The peak threshold and the valley threshold of the preset load rate in the embodiment are only examples, and the embodiment does not limit them, and a person skilled in the art can set them according to the actual situation.

[0095] For example, assuming that for a base station A1 in a business circle base station cluster, the rated load power is 20kW, first, the total load consumption energy of A1 is obtained as 360kWh according to the load prediction curve, the load prediction value is 17kW, and the load peak segment is 18:00-22:00; then the load rate of the base station in the load peak segment is calculated as ; assuming that the photovoltaic energy of A1 is 144kWh, the available capacity of the energy storage is 16kWh, and the commercial power energy is 100kWh, the total energy supply of the base station is obtained as , and the first energy shortage of the base station A1 is , that is, A1 has an energy gap of 100kWh.

[0096] The embodiment quantitatively extracts the load core feature parameter based on the load prediction curve, accurately calculates the remaining energy shortage combined with the energy constraint data, not only realizes the matching analysis of the load state and the energy supply capacity, but also provides a quantitative basis for the subsequent determination of the first base station and the matching of the collaborative base station, and effectively improves the pertinence and reliability of the energy deployment of the base station cluster.

[0097] Further, the embodiment provides a step of determining the first base station according to the load prediction value, the load peak valley period and the first energy shortage of each base station, comprising:

[0098] According to the load prediction value, the load peak valley period and the first energy shortage of each base station, the base station with the first energy shortage greater than 0 is determined as the base station to be evaluated;

[0099] The first base station is determined based on the load rate of the base station to be evaluated.

[0100] The to-be-evaluated base station is a base station in the regional base station cluster whose energy supply cannot meet its load demand, i.e., the first energy shortage is greater than zero; the first base station is a base station with the largest load rate among the to-be-evaluated base stations, and needs to be prioritized for cross-base station collaborative scheduling.

[0101] Specifically, first, the first energy shortage of all base stations in the regional base station cluster is traversed to obtain a set of base stations with a first energy shortage value greater than 0, and the base stations in the set are to-be-evaluated base stations; then, the load rate of each to-be-evaluated base station is calculated, and a double-index sorting method is used to prioritize the to-be-evaluated base stations, the double-index sorting method including a first sorting index and a second sorting index, the first sorting index being the load rate and the second sorting index being the first energy shortage; first, all to-be-evaluated base stations are preliminarily sorted according to the load rate from high to low, and if there are base stations with the same load rate in the preliminary sorting, these base stations are further sorted according to the first energy shortage from large to small for secondary sorting to calibrate the priority; if there is a special case where the load rate and the first energy shortage are the same, the first base station is selected by calculating the overlap degree of the load peak-valley period of each base station and the overall high-load period of the region; finally, the to-be-evaluated base station with the highest priority is selected as the first base station, and if there are multiple to-be-evaluated base stations with the same load rate and first energy shortage, the final first base station is determined by combining the load peak-valley period overlap degree.

[0102] For example, the embodiment provides an industrial park base station cluster including eight base stations B1-B8, first, the first energy shortage of each base station is obtained to obtain the B1 base station (first energy shortage ), the B4 base station (first energy shortage ), and the B6 base station (first energy shortage ) with a first energy shortage greater than 0, and they are to-be-evaluated base stations; then, the load rate of the three to-be-evaluated base stations is calculated to obtain the load rate of the B1 base station , the load rate of the B4 base station , and the load rate of the B6 base station ; based on the double-index sorting method, first, the preliminary sorting is performed according to the first sorting index from high to low to obtain B1>B4>B6, and since there are no base stations with the same load rate in the preliminary sorting, secondary calibration by the second sorting index is not needed, and finally, the B1 base station with the highest priority is determined as the first base station.

[0103] Further, the embodiment provides a step of obtaining the corresponding second base station and third base station according to the energy scheduling sequence of the first base station, the load prediction value, the load peak-valley period, and the first energy shortage, including:

[0104] According to the load peak-valley period and the load prediction value of the first base station, a second base station is obtained based on the load rate of the surrounding base stations by analytic hierarchy process;

[0105] Based on the second energy shortage of the first base station and the energy redundancy of the surrounding base stations, a third base station is obtained by multi-objective reinforcement learning, and the second energy shortage is based on the first energy shortage and the second base station.

[0106] The second base station is a base station surrounding the first base station that has data receiving and processing capabilities and sufficient load redundancy, which receives data unloaded by the first base station to reduce the load pressure of the first base station. The third base station is a base station surrounding the first base station that has energy supply exceeding its own demand and has an energy transmission interface condition, which is used to transport energy to the first base station to make up for the gap of the first base station. The surrounding base station refers to a base station that belongs to the same regional base station cluster as the first base station and has a straight-line distance of no more than a first preset distance. The second energy shortage is the remaining energy supply-demand gap of the first base station after unloading data through the second base station. The analytic hierarchy process refers to a decision-making method for sorting and screening candidate second base stations by constructing a multi-dimensional evaluation index system. Multi-objective reinforcement learning refers to an algorithm that optimizes energy supply efficiency, transmission cost, and loss by model training and iterative decision-making to obtain the optimal energy supply base station.

[0107] Specifically, as shown in Figure 3 the load redundancy of the first base station is calculated based on the load peak-valley period and the load prediction value of the first base station, which is the amount of load data that needs to be transferred by the first base station. According to the candidate base stations surrounding the first base station, the load rate of the surrounding base stations and the first distance from the first base station are used as evaluation indexes, and the analytic hierarchy process is used for weight distribution and quantitative processing of the evaluation indexes. By comparing the performance of the candidate base stations in each index, a judgment matrix is constructed to obtain the priority ranking of the candidate base stations. According to the ranking result, the candidate base station that best meets the load transfer demand of the first base station is selected as the final second base station. Then, according to the first energy shortage of the first base station and the actual load capacity of the second base station, the second energy shortage, which is the remaining energy gap of the first base station after data transfer, is obtained. If the second energy shortage is not positive, the process is terminated. If it is positive, a comprehensive optimization objective function of multi-objective reinforcement learning is constructed based on the energy redundancy of the surrounding base stations, the distance from the first base station, and the transmission path loss rate according to the energy supply scenario. The energy redundancy of the surrounding base stations, the distance from the first base station, and the transmission path loss rate are used as the state space of the reinforcement learning model input, and the selected supply base station or excluded candidate base station is used as the action space. The value of the comprehensive optimization objective function is used as the reward signal. Then the model parameters are initialized and the reward value is converged through multiple rounds of environment interaction and iterative training until the model finds the optimal decision strategy. Finally, the reward value of each candidate third base station is obtained from the converged model, and the base station with the highest reward value is selected as the final third base station.

[0108] The embodiment realizes accurate matching of data transfer target base stations and energy supply target base stations through the collaborative decision of the analytic hierarchy process and multi-objective reinforcement learning, solves the limitation of only relying on single dimension screening of collaborative base stations in traditional base station energy deployment, and improves the scientificity and efficiency of decision-making through quantitative indexes and intelligent algorithms, thereby effectively enhancing the energy collaborative scheduling capability of the regional base station cluster.

[0109] Further, the embodiment provides a step of obtaining a second base station based on the load rate of a first surrounding base station through the analytic hierarchy process according to the load peak-valley period and the load prediction value of a first base station, including:

[0110] obtaining a load redundancy of the first base station according to the load peak-valley period and the load prediction value of the first base station;

[0111] determining an evaluation index according to the load rate of the first surrounding base station and the first distance from the first base station;

[0112] constructing a judgment matrix based on the evaluation index through the analytic hierarchy process to obtain the ranking of the candidate sample;

[0113] obtaining the second base station according to the ranking of the candidate sample.

[0114] The load redundancy is the amount of load data that needs to be transferred by the first base station beyond the safe operation load threshold in the load peak period. The first surrounding base station is a base station in the same regional base station cluster as the first base station and with a straight-line distance of no more than a second preset distance. The first distance is the straight-line distance between the second base station and the first base station. The candidate sample is a set of second base stations that meet the basic acceptance condition. The second base station is a base station with the optimal comprehensive acceptance capability and capable of receiving the unloaded data of the first base station among the candidate samples. The first preset distance and the second preset distance in the embodiment can be a distance from the first base station as a starting point, and the length of the first preset distance and the second preset distance is not limited in the embodiment, and can be set according to the actual situation by those skilled in the art.

[0115] Specifically, based on the peak load period of the first base station, the predicted load value for that period is obtained and used as the average load during the peak period. Assuming that the load threshold for safe operation of the base station is 80% of the rated load, the load redundancy of the first base station is calculated by subtracting 80% of the rated load from the average load during the peak period. This redundancy represents the amount of load data that needs to be transferred. If the calculation result is non-positive, it is determined that the first base station has no data offloading requirement, and this step is terminated. Then, load redundancy and transmission convenience are selected as evaluation indicators. The load redundancy is calculated based on the load rate of the first surrounding base stations. The calculation method is: load redundancy = 1 - (load rate / 100%). The larger the load redundancy value, the greater the load redundancy. The stronger the receiving capacity of the first surrounding base station, the better. Transmission convenience is obtained based on the first distance; the shorter the distance, the higher the transmission convenience. The weights of the two evaluation indicators are set to obtain a comprehensive evaluation index system. The candidate samples are used as decision-making layer elements, and the comprehensive evaluation index is used as criterion layer elements. The 1-9 scaling method is used to compare the relative importance of the candidate samples under each index to obtain a judgment matrix. By calculating the maximum eigenvalue and corresponding eigenvector of the judgment matrix, the comprehensive ranking of each candidate sample is determined after consistency verification. The candidate samples are ranked from high to low to obtain the ranking of the candidate samples. Finally, the second base station that receives the data offloaded by the first base station is determined based on the ranking of the candidate samples.

[0116] For example, such as Figure 4 As shown, this embodiment provides a residential base station cluster, which includes 8 base stations C1-C8. The first base station has been identified as C1, and its rated load is... The peak load period is The load forecast value for this period is Then the load redundancy of the first base station C1 is If the calculated load redundancy is greater than 0, then the first base station is determined to need to perform data transfer. The preset load rate threshold is used; firstly, the area around the first base station C1 with a first preset distance is selected. C2, C3, and C4 within the first perimeter base station are assumed to have the following load rates: , , The first distances are respectively , , Calculate the load redundancy of each base station as follows: , and These three base stations were selected as candidate samples for the second base station. Then, the analytic hierarchy process was used to construct a judgment matrix to obtain a comprehensive ranking of the candidate samples as C2 > C4 > C3. Finally, by further verifying the transmission bandwidth of C2, it was determined that C2 met the conditions for receiving data, so the C2 base station was selected as the second base station.

[0117] This embodiment calculates the offloadable scale by combining the load characteristics of the first base station, and constructs a hierarchical analysis decision model with quantitative evaluation indicators as the core, realizing the scientific selection of the second base station. It effectively solves the problem that the selection of base stations in traditional data offloading relies on experience and lacks quantitative basis, ensuring that the selected second base station has sufficient load redundancy and efficient transmission conditions, improving the pertinence and efficiency of load diversion of the first base station, and providing reliable decision support for the collaborative scheduling of base station clusters.

[0118] Furthermore, this embodiment provides a step for obtaining a third base station based on the second energy deficit of the first base station and the energy redundancy of the second surrounding base stations through multi-objective reinforcement learning, including:

[0119] The second energy deficit of the first base station is obtained based on the first energy deficit of the first base station and the second base station;

[0120] The optimization objective for multi-objective reinforcement learning is obtained based on the energy redundancy of the second surrounding base station, the distance to the first base station, and the transmission path loss rate.

[0121] The third base station is obtained through multi-objective reinforcement learning based on the optimization objective.

[0122] Among them, the second energy deficit is the energy supply and demand difference remaining after the data transfer from the first base station to the second base station; the second surrounding base station is a base station that belongs to the same regional energy coordination network as the first base station, is within a straight-line distance of no more than 5km, and has an energy transmission interface; the energy redundancy is the amount of remaining energy that the second surrounding base station can supply to the first base station after meeting its own load requirements; the transmission path loss rate is the proportion of power loss of energy during transmission; and the third base station is the optimal base station among the second surrounding base stations that can meet the energy supply requirements of the first base station.

[0123] Specifically, firstly, based on the actual amount of transferred data received by the second base station. Combined with the unit data processing energy consumption of the first base station Energy saving factor for data unloading The energy consumption reduction of the first base station through data transfer was obtained. The calculation formula is:

[0124]

[0125] Then through the formula Obtaining a second energy deficit ,in The first energy deficit for the first base station, if If the value is non-positive, it is determined that the first base station does not require power replenishment, and this step is terminated; based on the power redundancy of the second surrounding base station... Distance from the first base station Transmission path loss rate The comprehensive optimization objective function of the multi-objective reinforcement learning model is constructed by setting weights for the sub-objectives of supply sufficiency, transmission efficiency, and energy utilization. :

[0126]

[0127] in Based on constraints to ensure sufficient energy replenishment, through Optimize transmission distance to reduce loss costs, through Weights are assigned to adapt to different scheduling priorities, and the optimal power replenishment base station selection under multi-objective conditions is achieved through comprehensive optimization of the objective function. Then, the energy redundancy of the second surrounding base station, the distance to the first base station, and the transmission path loss rate are used as the state space of the reinforcement learning agent, the excluded candidate third base station is used as the action space, and the comprehensive optimization objective function value is used as the reward signal. Next, the parameters of the multi-objective reinforcement learning model are initialized, and multiple rounds of iterative training are used to find the optimal decision strategy until the reward value converges. Finally, the converged reward values ​​are sorted from high to low, and the candidate third base station with the highest reward value is determined as the third base station. If multiple candidate third base stations have similar reward values ​​after convergence... If the energy storage capacity ratios are the same, the final third base station needs to be determined. First, based on the percentage of available energy storage capacity of each base station, the base station with the highest percentage is selected. The percentage of available energy storage capacity is the ratio of available energy storage capacity to total energy storage capacity. If the percentages of available energy storage capacity of all candidate third base stations are the same, then the photovoltaic energy stability is further compared. Photovoltaic energy stability is calculated as the ratio of the standard deviation of photovoltaic power of a base station over the past 72 hours to its mean. The base station with the lower ratio is selected. If the above two indicators are still the same, then the historical energy replenishment response speed of the candidate third base stations is compared. The historical energy replenishment response speed is the average time taken from startup to transmission in the last 10 energy replenishments. The base station with the shortest time, i.e., the highest collaborative efficiency, is selected as the final third base station. The unit data processing energy consumption and data offloading energy consumption saving coefficients in this embodiment are merely examples and are not intended to limit the implementation. Those skilled in the art can set them according to actual conditions.

[0128] For example, such as Figure 4 As shown, this embodiment assumes a residential base station cluster, which includes 8 base stations C1-C8. The first energy deficit of the first base station C1 is... The actual amount of transferred data received by the second base station C2 is The energy consumption reduction of the first base station C1 through data transfer is calculated, and then the second energy deficit of the first base station C1 is obtained; then, the second preset distance around the first base station C1 is selected. C5, C6, and C7, which have transmission conditions, are candidate base stations of the third base station, and their energy redundancy amounts are , , , distances from C1 are , , , and transmission path loss rates are , , ; the calculated comprehensive optimization objective function values of each base station are , and ; the reward value of the C5 base station obtained through reinforcement learning is the highest, and its energy redundancy amount is greater than the second energy shortage of the first base station C1, so the C5 base station is determined as the third base station.

[0129] The embodiment accurately calculates the remaining energy shortage by combining the data transfer effect, realizes intelligent selection of the energy supply base station through multi-objective reinforcement learning, solves the resource waste problem caused by single target decision in traditional energy deployment, and improves the scientificity and adaptability of the decision through the quantitative model and iterative training, so as to ensure that the third base station can efficiently and economically make up for the energy gap of the first base station, and enhance the energy collaborative supply capability of the regional base station cluster.

[0130] Further, the embodiment provides a method for obtaining a plurality of third base stations based on the second energy shortage, if the energy redundancy amount of the third base station is less than the second energy shortage, comprising the steps of:

[0131] obtaining a remaining energy shortage based on the second energy shortage of the first base station and the energy redundancy amount of the selected third base station;

[0132] obtaining a remaining candidate third base station based on multi-objective reinforcement learning;

[0133] determining a plurality of third base stations based on the energy redundancy amount of the remaining candidate third base station and the remaining energy shortage.

[0134] Wherein, the remaining energy shortage is the difference between the second energy shortage of the first base station and the energy redundancy amount of the selected third base station; the remaining candidate third base station is a further energy-supplyable base station set obtained by sorting the reward values from high to low among the candidate third base stations in the multi-objective reinforcement learning model, excluding the selected third base station and the invalid base station without an energy transmission interface; and the plurality of third base stations are energy-supplyable base stations whose energy redundancy amounts can meet the remaining energy shortage.

[0135] Specifically, first, the second energy shortage of the first base station and the actual output energy redundancy amount of the selected third base station are obtained., the remaining energy shortage is calculated by the formula , wherein The emergency backup capacity of the selected third base station needs to be subtracted. Secondly, according to the initial candidate third base station data obtained by the multi-objective reinforcement learning model, the candidate third base stations excluding the selected third base station and the base station without a bidirectional energy transmission interface are obtained. According to the reward values obtained by the model, the remaining candidate third base stations are sorted from high to low. Then, according to the order of the remaining candidate third base stations, the energy redundancy of the base station is matched with the remaining energy shortage in turn. If the energy redundancy of a single base station is greater than or equal to the remaining energy shortage, the base station is included in the multiple third base station set, and the matching is terminated. If the energy redundancy of a single base station is less than the remaining energy shortage, the base station is included in the multiple third base station set and the remaining energy shortage is updated, and the , wherein is the updated remaining energy shortage, is the energy redundancy of the base station after subtracting its emergency backup capacity, and then the next base station is matched until the remaining energy shortage is less than or equal to 0 or all candidate third base stations are traversed. Finally, if the remaining energy shortage is still greater than zero after traversing all candidate third base stations, the first base station is supplemented with energy through the power supply, and a multi-base station energy supplement deficiency warning is issued.

[0136] For example, it is assumed that in a base station cluster of an industrial park, the second energy shortage of the first base station D1 is , the actual output energy redundancy of the selected third base station D5 is , and the remaining energy shortage of the first base station D1 is calculated to be . Based on the initial candidate third base stations obtained by the multi-objective reinforcement learning model, after excluding D5 and D8 without a bidirectional energy transmission interface, the remaining candidate third base stations D6, D7, and D9 are obtained according to the reward value sorting. According to the sorting, D6 base station is matched first. The actual output energy redundancy of D6 base station is , which is greater than , so D6 base station is included in the multiple third base station set and the matching is terminated. Finally, the multiple third base stations are determined to be D5 and D6.

[0137] In this embodiment, the remaining energy shortage is calculated by the difference. The candidate sorting based on the multi-objective reinforcement learning model realizes efficient screening of base stations, obtains a multiple third base station combination that meets the energy supplement demand, realizes full-fledged filling of the energy shortage of the first base station, avoids the risk of power supply interruption caused by single base station energy supplement deficiency, improves the redundancy and operation reliability of regional base station cluster energy deployment, and further reduces the dependence on the power supply.

[0138] Further, as Figure 5 ​As shown, the embodiment of the present application provides a base station intelligent energy coordination scheduling system, which comprises a data processing module, a prediction module and an energy scheduling module:

[0139] The data processing module: collects the data set of each base station in the regional base station cluster, the data set comprising energy data, load data and environmental data; calculates energy constraint data based on the energy data and the environmental data, to obtain the energy scheduling sequence of each base station, the energy constraint data comprising photovoltaic prediction value, energy storage available capacity and mains data;

[0140] The prediction module: obtains the load prediction curve of each base station through a preset LSTM network model based on the load data; obtains the load prediction value, load peak valley period and first energy shortage of each base station according to the load prediction curve;

[0141] The energy scheduling module: determines the first base station according to the load prediction value, load peak valley period and first energy shortage of each base station; obtains the corresponding second base station and third base station according to the energy scheduling sequence, load prediction value, load peak valley period and first energy shortage of the first base station, the second base station being used for receiving the data transmitted by the first base station, and the third base station being used for transmitting energy to the first base station.

[0142] The data processing module converts the dispersed energy, load and environmental original data into standardized data sets through centralized collection and integration of multi-source data, and at the same time, calculates the energy constraint data and schedules the sequence order, to provide accurate basic data support for the subsequent modules; the prediction module takes the load data and energy scheduling sequence output by the data processing module as input, and obtains the load prediction curve through the time series prediction capability of the LSTM network, to achieve the purpose of predicting the future state from historical data statistics, the load prediction value, load peak valley period and first energy shortage output by the module provide the basis for the energy scheduling module to determine the base station that needs to be preferentially scheduled; the energy scheduling module takes the load characteristic parameters output by the prediction module as the basis, determines the first base station, matches the second base station and the third base station, and converts the results of data processing and prediction analysis into a specific coordinated scheduling scheme, at the same time, the scheduling requirements of the module also reversely propose optimization guidance for the collection accuracy of the data processing module and the prediction accuracy of the prediction module, to promote the whole system to form a dynamic optimization closed loop.

[0143] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the variations disclosed herein illustrate the architecture, functionality, and operation of possible implementations of apparatuses and computer program products according to various embodiments. In this regard, each flowchart block and / or block in the flowcharts and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that, in some alternative implementations, the flowchart blocks and / or blocks in the flowcharts and / or block diagrams can represent a Figure 1 The apparatus can be specially constructed for the required purposes, or it can be a general purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may

[0145] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the variations disclosed herein illustrate the architecture, functionality, and operation of possible implementations of apparatuses and computer program products according to various embodiments. In this regard, each flowchart block and / or block in the flowcharts and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that, in some alternative implementations, the flowchart blocks and / or blocks in the flowcharts and / or block diagrams can represent a Figure 1 The apparatus can be specially constructed for the required purposes, or it can be a general purpose computer selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may

[0146] The above description is only preferred embodiments of the present application, the protection scope of the present application is not limited to the above-mentioned embodiments, and any technical scheme falling within the idea of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled in the art, some improvements and refinements without departing from the principle of the present application shall be considered as the protection scope of the present application.

Claims

1. A method for intelligent energy coordinated allocation of base stations, characterized in that, include: The dataset is collected from each base station within the regional base station cluster. The dataset includes energy data, load data, and environmental data. Based on the energy data and the environmental data, energy constraint data is calculated to obtain the energy scheduling order of each base station. The energy constraint data includes photovoltaic forecasts, available energy storage capacity, and mains power data. Based on the load data, a load prediction curve for each base station is obtained through a preset LSTM network model; The load forecast value, peak and valley periods, and first energy deficit of each base station are obtained based on the load forecast curve. The first base station is determined based on the load forecast value, peak and valley periods of load, and first energy deficit of each base station; Based on the energy scheduling order, load forecast value, load peak and valley periods and first energy shortage of the first base station, the corresponding second base station and third base station are obtained. The second base station is used to receive data transmitted by the first base station, and the third base station is used to transmit energy to the first base station. The step of obtaining corresponding second and third base stations based on the energy scheduling order, load forecast, peak and valley periods, and first energy deficit of the first base station, wherein the second base station is used to receive data transmitted by the first base station, and the third base station is used to transmit energy to the first base station, including: Based on the peak and valley periods of the first base station and the predicted load value, the second base station is obtained through the analytic hierarchy process based on the load rate of the surrounding base stations. Based on the second energy deficit of the first base station and the energy redundancy of surrounding base stations, a third base station is obtained through multi-objective reinforcement learning. The second energy deficit is obtained based on the first energy deficit and the second base station. The third base station is obtained through multi-objective reinforcement learning based on the second energy deficit of the first base station and the energy redundancy of surrounding base stations. The second energy deficit is obtained based on the first energy deficit and the second base station, and includes: The second energy deficit of the first base station is obtained based on the first energy deficit of the first base station and the second base station; The optimization objective for multi-objective reinforcement learning is obtained based on the energy redundancy of surrounding base stations, the distance to the first base station, and the transmission path loss rate. The third base station is obtained through multi-objective reinforcement learning based on the optimization objective.

2. The base station intelligent energy coordinated allocation method according to claim 1, characterized in that, The step of obtaining the load forecast value, peak and valley periods, and first energy deficit for each base station based on the load forecast curve includes: The load prediction value for each base station is obtained based on the load prediction curve and the energy constraint data of each base station. The load rate of each base station is calculated based on the load prediction curve, and the peak and valley periods of the load of each base station are determined according to the load rate of each base station. The first energy deficit of each base station is obtained by using the load forecast value and the energy constraint data.

3. The intelligent energy coordinated allocation method for base stations according to claim 1, characterized in that, The step of determining the first base station based on the load forecast value, peak and off-peak periods, and first energy deficit of each base station includes: Based on the load forecast value, load peak and valley periods and first energy deficit of each base station, base stations with a first energy deficit greater than 0 are identified as base stations to be evaluated. The first base station is determined based on the load rate of the base station to be evaluated.

4. The base station intelligent energy coordinated allocation method according to claim 1, characterized in that, The step of obtaining the second base station based on the load peak and valley periods and load prediction values ​​of the first base station, and using the analytic hierarchy process (AHP) based on the load rates of surrounding base stations, includes: The load redundancy of the first base station is obtained based on the peak and valley periods of the first base station and the load prediction value. Evaluation indicators are determined based on the load rate of the surrounding base stations and the first distance from the first base station; Based on the evaluation indicators, a judgment matrix is ​​constructed using the analytic hierarchy process (AHP) to obtain the ranking of candidate samples. The second base station is obtained based on the sorting of the candidate samples.

5. The intelligent energy coordinated allocation method for base stations according to claim 1, characterized in that, If the energy redundancy of the third base station is less than the second energy deficit, multiple third base stations are obtained based on the second energy deficit, including: The remaining energy deficit is obtained based on the second energy deficit of the first base station and the energy redundancy of the selected third base station; The remaining candidate third base stations are obtained based on the multi-objective reinforcement learning; Multiple third base stations are determined based on the energy redundancy of the remaining candidate third base stations and the remaining energy deficit.

6. The intelligent energy coordinated allocation method for base stations according to claim 1, characterized in that, The energy constraint data is calculated based on the energy data and the environmental data to obtain the energy scheduling order of each base station. The energy constraint data includes photovoltaic forecasts, available energy storage capacity, and mains power data, including: Based on the energy data, photovoltaic data, energy storage data, and mains power data for each base station are obtained; The photovoltaic prediction value is obtained based on the photovoltaic data and the preset photovoltaic prediction model; The available energy storage capacity is obtained based on the energy storage data and the preset available energy storage capacity model; Energy constraint data is obtained based on the photovoltaic forecast, available energy storage capacity, and grid power data. The energy scheduling order of each base station is then determined based on the energy constraint data.

7. The base station intelligent energy coordinated allocation method according to claim 1, characterized in that, The preset LSTM network model is obtained through training, including: Based on the input feature set, the training set, validation set, and test set are obtained by dividing the set according to a preset ratio. An LSTM network structure is built based on the input feature set, and an initial LSTM neural network model is obtained by initializing the network parameters. The initial LSTM neural network model is iteratively trained using the training set, and the model performance is monitored and parameters are adjusted using the validation set to obtain an optimized LSTM neural network model. The prediction accuracy of the optimized LSTM neural network model is verified based on the test set, and a preset LSTM network model is obtained.

8. A base station intelligent energy collaborative allocation system, used to implement the base station intelligent energy collaborative allocation method according to any one of claims 1-7, characterized in that, It includes a data processing module, a forecasting module, and an energy allocation module: The data processing module collects datasets from each base station within the regional base station cluster. These datasets include energy data, load data, and environmental data. Based on the energy data and environmental data, it calculates energy constraint data to obtain the energy scheduling order for each base station. The energy constraint data includes photovoltaic forecasts, available energy storage capacity, and mains power data. The prediction module obtains the load prediction curve for each base station based on the load data using a preset LSTM network model; and obtains the load prediction value, peak and valley periods, and first energy deficit for each base station based on the load prediction curve. The energy allocation module determines the first base station based on the load forecast value, load peak and valley periods, and first energy shortage of each base station; Based on the energy scheduling order, load forecast value, load peak and valley periods and first energy shortage of the first base station, the corresponding second base station and third base station are obtained. The second base station is used to receive data transmitted by the first base station, and the third base station is used to transmit energy to the first base station. The step of obtaining corresponding second and third base stations based on the energy scheduling order, load forecast, peak and valley periods, and first energy deficit of the first base station, wherein the second base station is used to receive data transmitted by the first base station, and the third base station is used to transmit energy to the first base station, including: Based on the peak and valley periods of the first base station and the predicted load value, the second base station is obtained through the analytic hierarchy process based on the load rate of the surrounding base stations. Based on the second energy deficit of the first base station and the energy redundancy of surrounding base stations, a third base station is obtained through multi-objective reinforcement learning. The second energy deficit is obtained based on the first energy deficit and the second base station. The third base station is obtained through multi-objective reinforcement learning based on the second energy deficit of the first base station and the energy redundancy of surrounding base stations. The second energy deficit is obtained based on the first energy deficit and the second base station, and includes: The second energy deficit of the first base station is obtained based on the first energy deficit of the first base station and the second base station; The optimization objective for multi-objective reinforcement learning is obtained based on the energy redundancy of surrounding base stations, the distance to the first base station, and the transmission path loss rate. The third base station is obtained through multi-objective reinforcement learning based on the optimization objective.

Images