Model training method and base station energy-saving control method

Abstract

The invention discloses a model training method and a base station energy-saving control method. The model training method comprises the steps that a training data set is obtained, and the training data set comprises first time sequence data used for representing the number of lifting times of a barrier gate rod in an underground garage, second time sequence data used for representing the actual load capacity of a target elevator and third time sequence data used for representing the network load of a target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; a neural network model is trained based on the training data set, and the neural network model at least comprises a long and short term memory network and a full connection layer; in the training process, the output of the long and short term memory network is converted into a plurality of probabilities through a full connection layer. According to the method and the device, the technical problem that stable high-quality communication cannot be maintained due to the fact that a power regulation method of a related indoor distribution base station cannot be adaptively regulated based on environment change is solved.

Description

Technical Field

[0001] The present application relates to the field of mobile communication technology, and in particular to a model training method and a base station energy-saving control method. Background Art

[0002] In the field of mobile communications, especially in indoor distributed systems (IDS), power regulation of communication base stations is key to ensuring network service quality and achieving efficient energy utilization. Existing power regulation methods for indoor distributed base stations have significant limitations and are unable to adaptively adjust base station operating conditions based on environmental changes. This is particularly problematic in enclosed environments such as underground parking lots, creating a conflict between communication service quality and energy efficiency.

[0003] Specifically, related methods employ a fixed power operation mode around the clock, even maintaining high power consumption during off-peak hours. For example, in underground garages, user activity decreases significantly at night or during off-peak hours, but base stations continue to operate at peak power, resulting in energy waste. Energy-saving techniques proposed in these methods, such as hard shutdown and soft shutdown, can only respond to changes in network load after they are detected, but cannot predict impending changes. While hard shutdown saves energy during periods of low user activity, it can cause network service interruptions and impact user experience. While soft shutdown can adjust power consumption based on network load, it lacks the ability to predict environmental changes and cannot promptly address sudden increases in user numbers. Furthermore, related power adjustment strategies often rely on analyzing historical data and using predictive models to estimate future occupancy density and network demand. However, these models often assume that environmental changes are gradual and cannot effectively handle sudden events, such as the sudden increase in passenger flow before and after large-scale events. This leads to untimely power adjustments, which can affect communication quality.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The present application provides a model training method and a base station energy-saving control method to at least solve the technical problem of being unable to maintain stable high-quality communication due to the fact that the power adjustment method of the relevant indoor base station cannot be adaptively adjusted based on environmental changes.

[0006] According to one aspect of the present application, a model training method is provided, comprising: obtaining a training data set, wherein the training data set includes: first time series data for indicating the number of times a barrier pole in an underground garage is lifted, second time series data for indicating the actual load weight of a target elevator, and third time series data for indicating the network load of a target network, wherein the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; training a neural network model based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; during the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is a indoor base station serving the target network.

[0007] Optionally, the neural network model also includes: a temporal convolutional network and a spatial attention network, wherein the temporal convolutional network includes multiple channels, and the spatial attention network is connected to the temporal convolutional network.

[0008] Optionally, the method further includes: during the training process, performing sequence modeling and feature extraction processing on the training data set through a temporal convolutional network to obtain feature dimension values of each channel at multiple time steps.

[0009] Optionally, the method also includes: during the training process, performing global average pooling processing on the feature dimension values of each channel output by the temporal convolutional network at multiple time steps through the spatial attention network to obtain a first eigenvector; performing global maximum processing on the first eigenvector to obtain a second eigenvector; concatenating the first eigenvector and the second eigenvector to obtain a weight coefficient; and using the weight coefficient to perform feature enhancement on the training data set to obtain a target eigenvector.

[0010] Optionally, the method further includes: during the training process, capturing the temporal dependency in the target feature vector through the input gate, forget gate, and output gate in the long short-term memory network to obtain the features of all time steps and the state of the last time step.

[0011] Optionally, the method also includes: after converting the output of the long short-term memory network into multiple probabilities through a fully connected layer, determining the loss function based on the number of training samples, the number of categories, the true label of each training sample, and the predicted probability that each sample belongs to a different category, and stopping the training of the neural network model when the loss function meets the preset convergence condition.

[0012] Optionally, the first time series data is collected by a gate pole counter; the second time series data is collected by an elevator load sensor.

[0013] According to another aspect of the present application, a base station energy-saving control method is also provided, including: obtaining time series data, wherein the time series data includes: first time series data for indicating the number of times a barrier pole in an underground garage is lifted, second time series data for indicating the actual load weight of a target elevator, and third time series data for indicating the network load of a target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; using a neural network model to analyze the time series data to obtain multiple probabilities, wherein the neural network model is obtained by training using the above-mentioned model training method, and the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset range, a second probability that the power consumption of the target indoor base station is within a second preset range, and a third probability that the power consumption of the target indoor base station is within a third preset range; adjusting the power consumption of the target indoor base station to the target range corresponding to the maximum probability among the multiple probabilities.

[0014] According to another aspect of the present application, a model training device is also provided, including: an acquisition module for acquiring a training data set, wherein the training data set includes: a first time series data for representing the number of times a barrier pole in an underground garage is lifted, a second time series data for representing the actual load weight of a target elevator, and a third time series data for representing the network load of a target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; a first training module for training a neural network model based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; a second training module for converting the output of the long short-term memory network into multiple probabilities through the fully connected layer during the training process, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is a indoor base station serving the target network.

[0015] According to another aspect of the present application, a non-volatile storage medium is also provided, which includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above model training method.

[0016] According to another aspect of the present application, an electronic device is provided, including: a memory and a processor, the processor being configured to run a program stored in the memory, wherein the above model training method is executed when the program is run.

[0017] According to another aspect of the present application, a computer program is also provided, wherein the above model training method is implemented when the computer program is executed by a processor.

[0018] According to another aspect of the present application, a computer program product is provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above model training method is implemented.

[0019] In the present application, a training data set is obtained, wherein the training data set includes: a first time series data for representing the number of times a barrier pole in an underground garage is lifted, a second time series data for representing the actual load weight of a target elevator, and a third time series data for representing the network load of a target network, wherein the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; a neural network model is trained based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; during the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset range, ... The second probability that the power consumption is within the second preset interval and the third probability that the power consumption of the target indoor base station is within the third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is the indoor base station serving the target network, which achieves the purpose of dynamically adjusting the working power consumption of the indoor base station according to the real-time situation of people entering and exiting the garage and the network usage status, thereby achieving the technical effect of avoiding the decline in communication quality due to the sudden increase in network demand while ensuring energy saving, and thus solving the technical problem of being unable to maintain stable high-quality communication due to the inability of the power adjustment method of the relevant indoor base station to adaptively adjust based on environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0021] Figure 1 is a flow chart of a model training method according to an embodiment of the present application;

[0022] Figure 2 is a flow chart of a base station energy-saving control method according to an embodiment of the present application;

[0023] Figure 3 is an architecture diagram of a system on which a base station energy-saving control method according to an embodiment of the present application relies;

[0024] Figure 4 is a structural diagram of a model training device according to an embodiment of the present application;

[0025] Figure 5 This is a hardware structure block diagram of a computer terminal according to a model training method of an embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] According to an embodiment of the present application, a method embodiment of a model training method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] Figure 1 is a flow chart of a model training method according to an embodiment of the present application, such as Figure 1 As shown, the method includes the following steps:

[0030] Step S102, obtain a training data set, wherein the training data set includes: first time series data for representing the number of times the barrier pole in the underground garage is lifted, second time series data for representing the actual load capacity of the target elevator, and third time series data for representing the network load of the target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage.

[0031] Preferably, the first time series data is collected by a gate pole counter; the second time series data is collected by an elevator load sensor.

[0032] In step S102 above, barrier lift counters are installed at the entrance and exit of the underground parking lot to record the number of vehicles entering and exiting per minute. Load sensors are also installed in all elevators leading to the underground parking lot to monitor real-time load capacity. Furthermore, a network monitoring system continuously collects load data on the target network, including metrics such as data flow and number of connections.

[0033] The collected data on the number of times the barrier pole is lifted (first time series data), the elevator load (second time series data), and the network load (third time series data) are integrated into a training dataset. The dataset should contain sufficient historical records to cover various possible environmental changes and network usage scenarios to ensure the generalization ability of the model.

[0034] Step S104: training the neural network model based on the training data set, wherein the neural network model at least includes: a long short-term memory network and a fully connected layer.

[0035] In step S104, the preprocessed dataset is input into the model, and a long short-term memory network is used to extract time series features, including the periodic patterns of personnel entry and exit and the fluctuation trend of network load. These features will be used to predict future personnel density and network load. The model is trained using the training dataset, with the goal of optimizing the model parameters so that the model can accurately predict the relationship between changes in personnel density and network load in the underground garage. The training process includes adjusting the learning rate, the optimization algorithm (such as Adam or SGD), and the choice of loss function.

[0036] Step S106: During the training process, the output of the long short-term memory network (LSTM) is converted into multiple probabilities through a fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is the indoor base station serving the target network.

[0037] Indoor base stations have three power adjustment mechanisms: deep sleep (disabling 80% of RF channels while retaining the PDCCH), eco mode (activating 2T2R MIMO with 6dB power backoff), and full power (with carrier aggregation (CA) and beamforming optimization). Therefore, three preset power adjustment ranges are defined: the first range corresponds to deep sleep mode, the second range corresponds to eco mode, and the third range corresponds to full power mode. These ranges are set based on actual network requirements and energy-saving goals, and are non-overlapping and incremented sequentially.

[0038] During model training, the fully connected layer converts the LSTM output into a probability distribution of the three power adjustment states mentioned above. This means that for each set of input data, the model predicts the probability of belonging to each power adjustment range.

[0039] Taking into account changes in the environment and network requirements, the preset power adjustment interval threshold should be able to be dynamically adjusted to adapt to different time and scenarios. This can be achieved by regularly retraining the model or incorporating an adaptive learning mechanism into the model.

[0040] According to the above steps, a training data set is obtained, wherein the training data set includes: a first time series data for representing the number of times the gate pole in the underground garage is lifted, a second time series data for representing the actual load of the target elevator, and a third time series data for representing the network load of the target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; a neural network model is trained based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; during the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: the target room There is a first probability that the power consumption of the indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is the indoor base station serving the target network, which achieves the purpose of dynamically adjusting the working power consumption of the indoor base station according to the real-time situation of people entering and exiting the garage and the network usage status, thereby achieving the technical effect of avoiding the decline in communication quality due to the sudden increase in network demand while ensuring energy saving.

[0041] The following Figure 1 The steps shown are exemplified and explained.

[0042] According to some optional embodiments of the present application, the neural network model also includes: a temporal convolutional network and a spatial attention network, wherein the temporal convolutional network includes multiple channels, and the spatial attention network is connected to the temporal convolutional network.

[0043] The Temporal Convolutional Network (TCN) is a deep learning architecture specifically designed for processing time series data. Compared to traditional convolutional neural networks (CNNs) and recurrent neural networks (RNNs), TCNs have the following notable features: 1. No recurrent connections: TCNs use one-dimensional convolutional layers (commonly used for time series) instead of the recurrent connections found in RNNs. This avoids the vanishing and exploding gradient problems that plague RNNs during training. 2. Causal convolution: To ensure that the network only accesses information from the current time point and before (satisfying the causal nature of time series), TCNs employ causal convolutions, meaning that convolution operations only use data from the current and past time points. 3. Dilated convolution: The convolutional layers in TCNs incorporate a dilation factor, which effectively expands the time window and captures longer temporal dependencies without increasing the network's depth or number of parameters. 4. Parallel processing: Compared to RNNs, TCNs can perform computations in parallel, resulting in faster and more efficient training when processing large-scale time series data.

[0044] Spatial attention networks (SATs) are a technique that incorporates an attention mechanism into deep learning models, primarily used for processing images or spatially distributed data. Their core concept is to enable the model to focus on specific regions of the input data, thereby enhancing the features in those regions and improving model performance and accuracy. Spatial attention can be viewed as a local magnifying glass for the data, helping the model better understand and utilize key information within the input data. The implementation of a SAT involves several steps: First, a convolutional layer or other feature extractor generates a feature map that reflects the spatial characteristics of the input data. Next, operations such as global average pooling and global max pooling are used to generate a vector representing the importance of the entire feature map. This vector is then used to calculate an attention weight map, which has the same size as the feature map and whose value at each position reflects the relative importance of that position. Finally, the attention weight map is dot-multiplied with the original feature map to weight the features at each position. This allows the model to prioritize regions with higher attention weights in subsequent processing and ignore background or irrelevant regions with lower weights.

[0045] Furthermore, during the training process, the training dataset is subjected to sequence modeling and feature extraction processing through a temporal convolutional network to obtain the feature dimension values of each channel at multiple time steps.

[0046] The input layer of the neural network model includes three types of time series data: lift time T, elevator load W, and network load R. These data can cover underground garage entrance data and the periodic and sudden characteristics of network traffic. The specific representation is as follows:

[0047]

[0048] Wherein, n is the time series window length, which can be set according to actual conditions. In this embodiment, the sequence length is 1 minute and the sampling interval is 1 second.

[0049] Due to the sparsity of the data, a TCN network is used to extract features. It is superior to traditional CNN networks and can solve the problem of modeling long sequences with dilated convolution. It can process real-time data streams faster than RNN networks. Dilated causal convolution is used, and the specific formula is as follows:

[0050]

[0051] in, is the feature dimension value of the out_ch channel of the output sequence at time point t; is the convolution kernel parameter, the weight value of the out_ch-th convolution kernel at position k and output channel ch; is the feature dimension value of the chth channel of the output sequence at time point td·k (causality constraint: td·k≥0); bout_ch is the bias term of the out_chth convolution kernel; K is the convolution kernel size, and in this embodiment, K is 3; d is the expansion coefficient, specifically d = 1, 2, 4, 8.

[0052] Furthermore, during the training process, the spatial attention network is used to perform global average pooling on the feature dimension values of each channel output by the temporal convolutional network at multiple time steps to obtain the first eigenvector; the first eigenvector is subjected to global maximum processing to obtain the second eigenvector; the first eigenvector and the second eigenvector are concatenated to obtain a weight coefficient; the weight coefficient is used to perform feature enhancement on the training data set to obtain the target eigenvector.

[0053] The spatial attention layer first performs global average pooling on all channels. The specific formula is as follows:

[0054]

[0055] Then perform global maximum pooling, the specific formula is as follows:

[0056] Max out_ch =mac t z t,out_ch

[0057] Perform splicing and generate weights:

[0058] S=σ(Conv1D(Concat(Mean,Max)))

[0059] Perform feature enhancement:

[0060] Z att =Z⊙S

[0061] Among them, Z is the training data set.

[0062] Furthermore, during the training process, the temporal dependencies in the target feature vector are captured through the input gate, forget gate, and output gate in the long short-term memory network to obtain the features of all time steps and the state of the last time step. The specific formula is as follows:

[0063] i t =σ(W xi Z att,t +W hi h t-1 +b i )

[0064] f t =σ(W xf Z att,t +W hf h t-1 +b f )

[0065] o t =σ(W xo Z att,t +W ho h t-1 +b o )

[0066]

[0067] h t =o t ⊙tanh(C t )

[0068] Among them, i t is the input gate, f t For the forget gate, o t is the output gate, is the candidate memory, C t For memory update, h t In hidden state.

[0069] Then, a fully connected layer is used to merge the data. For indoor distribution, there are three power adjustment mechanisms: deep sleep, economic mode, and full power. Therefore, the output of the fully connected layer is a summary of the three modes, which is a three-classification problem. The formula is as follows:

[0070] y=Softmax(W f h T +b f )

[0071] The output of the neural network model is a vector, where each element corresponds to the probability of one of the above modes. For example, the vector might be [0.1, 0.3, 0.6], indicating that based on the current data, the model predicts a 10% probability that the indoor base station will be in deep sleep mode, a 30% probability in economy mode, and a 60% probability in full power mode. Based on this prediction, the mode with the highest probability is automatically selected and the base station's operating state is adjusted.

[0072] Preferably, after the output of the long short-term memory network is converted into multiple probabilities through a fully connected layer, the loss function is determined based on the number of training samples, the number of categories, the true label of each training sample, and the predicted probability that each sample belongs to a different category, and the training of the neural network model is stopped when the loss function meets the preset convergence condition.

[0073] Among them, for the number of categories C: In this embodiment, there are three power adjustment states (deep sleep, economy mode, full power), so the number of categories C is 3. The number of training samples N: The number of samples in each batch of training data. The true label: For each training sample, there is a corresponding true power state label. Prediction probability: The fully connected layer converts the output of the LSTM into a probability distribution for each training sample belonging to a different category. Then, based on the number of training samples, the number of categories, the true label of each training sample, and the predicted probability that each sample belongs to a different category, a loss function, such as cross entropy loss, is determined.

[0074] The gradient of the loss function is calculated using the backpropagation algorithm and used to update the network parameters, aiming to minimize the loss function. The model is iteratively trained on multiple training examples, with each parameter update gradually reducing the loss function until convergence is reached or a predetermined number of training rounds are reached. Convergence can occur when the change in the training loss function falls below a preset threshold, or when the loss function on the validation set stops improving. Another strategy for stopping training is to reach a predetermined number of training rounds, even if the loss function hasn't fully converged, to prevent excessive training time.

[0075] Figure 2 is a flow chart of a base station energy saving control method according to an embodiment of the present application, such as Figure 2 As shown, the method includes the following steps:

[0076] Step S202, obtain timing data, wherein the timing data includes: first timing data for indicating the number of times the barrier pole in the underground garage is lifted, second timing data for indicating the actual load capacity of the target elevator, and third timing data for indicating the network load of the target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage.

[0077] First time series data: The number of times the barrier pole is lifted is collected through sensors installed at the entrances and exits of the underground garage. This data can reflect the number of vehicles and people entering and leaving the garage. This data helps to estimate the density of people in the garage. Second time series data: Obtain the actual load capacity of the target elevator, which provides another indirect way to measure the density of people in the garage, especially when the elevator serves as the main entrance to the garage. Third time series data: Monitor the network load of the target network (i.e., the indoor base station covering the underground garage), including data traffic, call volume, etc., which can reflect the current network usage.

[0078] The above time series data is collected dynamically and can reflect the environmental changes of the underground garage in real time, providing basic information for the analysis in the next step.

[0079] Step S204: Analyze the time series data using a neural network model to obtain multiple probabilities, wherein the neural network model is based on Figure 1 The multiple probabilities obtained by training using the model training method shown include: a first probability that the power consumption of the target indoor base station is within a first preset range, a second probability that the power consumption of the target indoor base station is within a second preset range, and a third probability that the power consumption of the target indoor base station is within a third preset range.

[0080] The collected time series data is analyzed using a pre-trained neural network model, which includes a temporal convolutional network (TCN), a spatial attention layer, a long short-term memory network (LSTM), and a fully connected layer.

[0081] The model outputs multiple probabilities, corresponding to the probabilities of the three power consumption adjustment states that the target indoor base station may be in: deep sleep mode, economy mode, and full power mode. The first probability indicates that the target indoor base station is in deep sleep mode (a low-power state with only essential channels retained). The second probability indicates that the target indoor base station is in economy mode (reduced power, activating a 2T2R MIMO configuration). The third probability indicates that the target indoor base station is in full power mode (all functions enabled, providing optimal communication performance).

[0082] After training, the neural network model can predict the occupant density and network load in the garage in the future based on the current time series data, and then predict the most suitable power consumption state of the indoor base station.

[0083] Step S206: Adjust the power consumption of the target indoor base station to a target interval corresponding to the maximum probability among the multiple probabilities.

[0084] Based on the multiple probabilities obtained in step S204, the state with the highest probability is selected as the current power consumption adjustment target for the target indoor base station. In other words, based on the output probability distribution, the intelligent system can make decisions and adjust the power consumption state of the target indoor base station. For example, if the first probability (deep sleep mode) is the highest, the system will enter deep sleep mode; if the second probability (economy mode) is the highest, the economy mode adjustment will be implemented; if the third probability (full power mode) is the highest, the base station will enter full power mode.

[0085] Through these steps, the system can automatically adapt to the dynamic environment of underground parking lots, predicting changes in occupancy density and network load in advance and intelligently adjusting the operating status of indoor base stations accordingly, achieving the dual goals of energy conservation and consumption reduction while ensuring communication quality. This approach overcomes the limitations of traditional energy-saving strategies and provides a new approach to intelligent energy-saving control of underground parking lot communication base stations.

[0086] Figure 3is an architecture diagram of a system on which a base station energy-saving control method according to an embodiment of the present application relies, such as Figure 3 As shown, the system includes: perception layer, analysis layer, execution layer and feedback layer, among which,

[0087] The perception layer is responsible for collecting real-time environmental data, which is used for subsequent analysis and decision-making. In the underground parking scenario, it mainly includes the following two key parts:

[0088] 1. Basement Barrier Lift Counter: Installed at the entrances of underground parking garages, such as vehicle entrances and elevator entrances, this device records the number of times the barrier lifts. This indirectly reflects the number of vehicles and people entering the underground parking lot and is an important input for estimating changes in occupancy density.

[0089] 2. Elevator load sensors: These sensors are deployed in elevators leading to underground parking lots to monitor elevator loads. Changes in elevator loads can also provide trend information on occupancy density.

[0090] It should be noted that compared with other technical solutions, this embodiment adds a basement gate lift counter and an elevator load sensor to the perception layer to facilitate the calculation of the number of people entering the garage entrance. In addition, new network gates can be modified for API docking without the need for new equipment. Traditional RS485 gates can be equipped with DTU modules, and pure relay gates can be equipped with optoelectronic isolation collectors. The modification costs are relatively low, and the data that needs to be collected includes the timing of the lift. Elevator load sensors are currently basically equipped in all residential areas. Data can be obtained only through the interface without the need for equipment modification. The collected data is the elevator load.

[0091] The main function of the analysis layer is to conduct in-depth analysis of the data collected from the perception layer to make reasonable power consumption adjustment decisions. The following core components are used here: 1. Behavior prediction model (i.e. Figure 1 1. Neural network model in the method shown): It consists of a temporal convolutional network, a spatial attention layer, and a long short-term memory network. These neural network components work together to extract features from time series data and predict changes in the density of people and network load in the underground garage over a period of time. 2. Energy consumption decision model: Based on the output of the behavior prediction model, the analysis layer also includes a decision model to determine the appropriate power consumption level of the communication base station. Based on the predicted density of people and network load, it decides whether to enter deep sleep mode, economy mode, or full power mode to achieve effective energy management and optimize user experience.

[0092] The execution layer implements the decisions made by the analysis layer and directly adjusts the power state of communication base stations. It receives instructions from the analysis layer. For example, if the decision model predicts low occupancy density and light network load, it instructs the base station to enter deep sleep mode, shutting down most RF channels to maintain the minimum necessary communication capacity. If medium occupancy density and network load are predicted, it switches to economy mode, activating the 2T2R MIMO configuration and moderately reducing power consumption. If high occupancy density and heavy network load are predicted, it commands the base station to enter full power mode, enabling all functions to ensure optimal communication service.

[0093] The feedback layer is key to the system's closed-loop control. It monitors the operating status of base stations after adjustments made by the execution layer, collecting data on actual occupancy density, network load, and communication quality to evaluate execution results. This feedback information is fed back to the analysis layer to continuously optimize the model's predictive capabilities and decision-making strategies, forming a continuous self-learning and self-adjustment process. If actual results deviate significantly from predictions, the feedback layer can trigger the system to re-enter the perception and analysis phase, promptly revising decisions and ensuring system stability and accuracy.

[0094] Figure 3 The overall system architecture shown here integrates IoT sensors and deep learning technology to achieve intelligent energy-saving control of communication base stations in underground parking lots. This system not only senses environmental changes in real time but also makes accurate predictions using a complex neural network model. It then intelligently adjusts the base station's power consumption based on these predictions, ultimately forming a closed-loop feedback control mechanism that effectively balances energy conservation with communication service quality.

[0095] Figure 4 is a structural diagram of a model training device according to an embodiment of the present application, such as Figure 4 As shown, the device includes:

[0096] The acquisition module 42 is used to obtain a training data set, wherein the training data set includes: first time series data for representing the number of times the gate pole in the underground garage is lifted, second time series data for representing the actual load capacity of the target elevator, and third time series data for representing the network load of the target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage.

[0097] The first training module 44 is used to train the neural network model based on the training data set, wherein the neural network model at least includes: a long short-term memory network and a fully connected layer.

[0098] The second training module 46 is used to convert the output of the long short-term memory network into multiple probabilities through a fully connected layer during the training process, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset range, a second probability that the power consumption of the target indoor base station is within a second preset range, and a third probability that the power consumption of the target indoor base station is within a third preset range; the maximum value of the first preset range is less than the minimum value of the second preset range, and the maximum value of the second preset range is less than the minimum value of the third preset range; the target indoor base station is the indoor base station serving the target network.

[0099] Optionally, the neural network model also includes: a temporal convolutional network and a spatial attention network, wherein the temporal convolutional network includes multiple channels, and the spatial attention network is connected to the temporal convolutional network.

[0100] Optionally, during the training process, the training dataset is subjected to sequence modeling and feature extraction processing through a temporal convolutional network to obtain feature dimension values of each channel at multiple time steps.

[0101] Optionally, during the training process, the spatial attention network is used to perform global average pooling processing on the feature dimension values of each channel output by the temporal convolutional network at multiple time steps to obtain a first eigenvector; the first eigenvector is subjected to global maximum processing to obtain a second eigenvector; the first eigenvector and the second eigenvector are concatenated to obtain a weight coefficient; the weight coefficient is used to perform feature enhancement on the training data set to obtain a target eigenvector.

[0102] Optionally, during the training process, the temporal dependencies in the target feature vector are captured through the input gate, forget gate, and output gate in the long short-term memory network to obtain the features of all time steps and the state of the last time step.

[0103] Optionally, after converting the output of the long short-term memory network into multiple probabilities through a fully connected layer, the loss function is determined based on the number of training samples, the number of categories, the true label of each training sample, and the predicted probability that each sample belongs to a different category, and the training of the neural network model is stopped when the loss function meets the preset convergence condition.

[0104] Optionally, the first time series data is collected by a gate pole counter; the second time series data is collected by an elevator load sensor.

[0105] It should be noted that the above Figure 4The modules in the embodiment can be program modules (for example, a set of program instructions that implement a specific function) or hardware modules. For the latter, they can be expressed in the following forms, but are not limited to these: the expression form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0106] It should be noted that Figure 4 The preferred implementation of the embodiment shown can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0107] Figure 5 FIG1 shows a hardware structure block diagram of a computer terminal for implementing a model training method. Figure 5 As shown, the computer terminal 50 may include one or more (502a, 502b, ..., 502n are shown in the figure) processors 502 (the processor 502 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 504 for storing data, and a transmission module 506 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 5 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 5 More or fewer components than shown, or with Figure 5 Different configurations shown.

[0108] It should be noted that the one or more processors 502 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 50. As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0109] The memory 504 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the model training method in the embodiment of the present application. The processor 502 executes various functional applications and data processing by running the software programs and modules stored in the memory 504, that is, realizing the above-mentioned model training method. The memory 504 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 504 may further include a memory remotely arranged relative to the processor 502, and these remote memories can be connected to the computer terminal 50 via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0110] The transmission module 506 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 50. In one embodiment, the transmission module 506 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission module 506 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0111] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 50 .

[0112] It should be noted that, in some optional embodiments, the above Figure 5 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. Figure 5 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0113] It should be noted that Figure 5 The computer terminal shown is used to execute Figure 1 The model training method shown, therefore the relevant explanations in the execution method of the above commands are also applicable to the electronic device and will not be repeated here.

[0114] An embodiment of the present application further provides a non-volatile storage medium, which includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above model training method.

[0115] A program for a non-volatile storage medium to perform the following functions: obtaining a training data set, wherein the training data set includes: first time series data for indicating the number of times a barrier pole in an underground garage is lifted, second time series data for indicating the actual load weight of a target elevator, and third time series data for indicating the network load of a target network, wherein the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; training a neural network model based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; during the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of a target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is a indoor base station serving the target network.

[0116] An embodiment of the present application also provides an electronic device, including: a memory and a processor, wherein the processor is used to run a program stored in the memory, wherein the above model training method is executed when the program is running.

[0117] The processor is used to run a program that performs the following functions: obtaining a training data set, wherein the training data set includes: a first time series data for representing the number of times a barrier pole in an underground garage is lifted, a second time series data for representing the actual load weight of a target elevator, and a third time series data for representing the network load of a target network, the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; training a neural network model based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; during the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is a indoor base station serving the target network.

[0118] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0119] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0120] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0122] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0123] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0124] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the relevant technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0125] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A model training method, characterized in that: include: Obtaining a training data set, wherein the training data set includes: first time series data representing the number of times a barrier pole in an underground garage is lifted, second time series data representing the actual load of a target elevator, and third time series data representing a network load of a target network, where the target elevator is an elevator leading to the underground garage and the target network is a network covering the underground garage; Training a neural network model based on the training data set, wherein the neural network model includes at least: a long short-term memory network and a fully connected layer; During the training process, the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is a indoor base station serving the target network.

2. The method according to claim 1, characterized in that The neural network model also includes: a temporal convolutional network and a spatial attention network, wherein the temporal convolutional network includes multiple channels, and the spatial attention network is connected to the temporal convolutional network.

3. The method according to claim 2, characterized in that The method further comprises: During the training process, the training data set is subjected to sequence modeling and feature extraction processing through the temporal convolutional network to obtain feature dimension values of each channel at multiple time steps.

4. The method according to claim 3, characterized in that The method further comprises: During the training process, the spatial attention network performs global average pooling processing on the feature dimension values of each channel output by the temporal convolutional network at multiple time steps to obtain a first feature vector; Performing global maximum processing on the first eigenvector to obtain a second eigenvector; performing concatenation processing on the first eigenvector and the second eigenvector to obtain a weight coefficient; The weight coefficient is used to perform feature enhancement on the training data set to obtain a target feature vector.

5. The method according to claim 4, characterized in that The method further comprises: During the training process, the temporal dependency in the target feature vector is captured through the input gate, forget gate, and output gate in the long short-term memory network to obtain the features of all time steps and the state of the last time step.

6. The method according to claim 1 or 5, characterized in that The method further comprises: After the output of the long short-term memory network is converted into multiple probabilities through the fully connected layer, the loss function is determined based on the number of training samples, the number of categories, the true label of each training sample, and the predicted probability that each sample belongs to a different category. When the loss function meets the preset convergence condition, the training of the neural network model is stopped.

7. The method according to claim 1, characterized in that The first time series data is collected by a gate pole counter; the second time series data is collected by an elevator load sensor.

8. A base station energy-saving control method, characterized in that: include: Acquiring time series data, wherein the time series data includes: first time series data indicating the number of times a barrier pole in an underground garage is lifted; second time series data indicating the actual load of a target elevator; and third time series data indicating a network load of a target network, where the target elevator is an elevator leading to the underground garage and the target network is a network covering the underground garage; Analyzing the time series data using a neural network model to obtain multiple probabilities, wherein the neural network model is obtained by training using the model training method according to any one of claims 1 to 7, and the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset range, a second probability that the power consumption of the target indoor base station is within a second preset range, and a third probability that the power consumption of the target indoor base station is within a third preset range; The power consumption of the target indoor base station is adjusted to a target interval corresponding to the maximum probability among the multiple probabilities.

9. A model training device, characterized in that: include: an acquisition module, configured to acquire a training data set, wherein the training data set includes: first time series data representing the number of times a barrier pole in an underground garage is lifted, second time series data representing the actual load of a target elevator, and third time series data representing a network load of a target network, wherein the target elevator is an elevator leading to the underground garage, and the target network is a network covering the underground garage; A first training module is configured to train a neural network model based on the training data set, wherein the neural network model includes at least a long short-term memory network and a fully connected layer; The second training module is used to convert the output of the long short-term memory network into multiple probabilities through the fully connected layer during the training process, wherein the multiple probabilities include: a first probability that the power consumption of the target indoor base station is within a first preset interval, a second probability that the power consumption of the target indoor base station is within a second preset interval, and a third probability that the power consumption of the target indoor base station is within a third preset interval; the maximum value of the first preset interval is less than the minimum value of the second preset interval, and the maximum value of the second preset interval is less than the minimum value of the third preset interval; the target indoor base station is the indoor base station serving the target network.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the model training method described in any one of claims 1 to 7 and the base station energy-saving control method described in claim 8.

11. An electronic device, characterized in that: include: A memory and a processor, wherein the processor is used to run a program stored in the memory, wherein when the program is run, the model training method described in any one of claims 1 to 7 and the base station energy saving control method described in claim 8 are executed.

12. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the model training method according to any one of claims 1 to 7 and the base station energy-saving control method according to claim 8 are implemented.

Images