Base station dormancy method based on flow prediction

Abstract

The invention relates to the technical field of communication, and provides a base station dormancy method based on flow prediction, and the method comprises the steps: carrying out the feature extraction of historical network flow data, obtaining a flow feature vector, and calculating the time sequence similarity between space units based on the flow feature vector; constructing a space-time dependency graph based on the time sequence similarity; obtaining static environment characteristics and dynamic environment characteristics of the target area according to the spatial environment data, and calculating model parameters based on the static environment characteristics and the dynamic environment characteristics; based on the space-time dependency graph and the model parameters, performing traffic prediction on the target area by using a traffic prediction model to obtain predicted network traffic of all space units; and according to the predicted network traffic, constructing an optimization objective function used for describing all base station states of the target area, and carrying out minimization solution on the optimization objective function to obtain a base station dormancy control scheme of the target area. The method provided by the invention can improve the accuracy of dormancy control on the base station.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a base station sleep method based on traffic prediction. Background Art

[0002] As the core infrastructure of modern communications, communication base stations provide a variety of services such as voice, SMS and data transmission to terminal users within the coverage of base stations by receiving, demodulating, processing and forwarding signals. In recent years, the new generation of communication networks represented by 5G is rapidly integrating into and promoting the development of various fields of the economy and society. However, with the widespread popularization of smart mobile terminals and the rapid rise of emerging applications such as virtual reality and Internet of Vehicles, the current energy consumption of 5G base stations has reached 70% of the total energy consumption of wireless communication networks, and higher standards and requirements are put forward for the power consumption and operating costs of communication base stations. In addition, the uneven distribution of communication traffic in time and space has further aggravated the energy consumption problem by causing waste of network resources. For example, urban areas often face tight network resources during peak hours, while rural areas have a serious surplus of resources during non-peak hours. By predicting future communication traffic, implementing dynamic sleep control on communication base stations is the simplest, most intuitive and effective means to improve the efficiency of network resource utilization, reduce the energy consumption of communication base stations and promote green communication.

[0003] At present, my country has built the world's largest and most technologically advanced 5G communication network, which has significantly increased the data capacity and connection density of the communication network, laying a solid foundation for large-scale data collection and analysis. At the same time, the number of smartphones, smart watches, computers and other devices continues to grow, and each terminal device will continuously generate a large amount of data during use. These data contain rich spatiotemporal variation patterns of network traffic in users' daily activities, and can provide a data foundation with full coverage, strong dynamics and high reliability.

[0004] The base station sleep strategy aims to sleep as many base stations as possible on the basis of meeting the user's network traffic demand to reduce the base station energy consumption. However, network traffic is affected by many business environment factors such as festivals, weather, and traffic, and its change rules are complex and diverse, often changing with time and space. How to understand the spatial heterogeneity and dynamic changes of network traffic and make real-time and accurate predictions of network traffic is an important prerequisite for the implementation of the base station sleep strategy. The existing base station sleep methods based on traffic prediction are mainly divided into statistical theory-based methods and deep learning-based methods. The statistical theory-based method uses a linear parameterized mathematical model to fit the time dependence of historical network traffic data, which has the characteristics of simple calculation and interpretable model parameters. However, the actual network traffic exhibits complex nonlinear characteristics and spatial dependence, which makes it difficult for the inherent sample independence assumption of the statistical theory-based method to hold true, and it is impossible to fully and accurately fit the temporal and spatial variation law of network traffic. The deep learning-based method uses a deep neural network model to fit the complex temporal and spatial dependencies contained in large-scale historical network traffic data, which has the advantages of deep feature mining and strong fitting ability. These methods focus more on the overall temporal and spatial dependencies of network traffic, and pay less attention to local spatial heterogeneity. Even if some models take local spatial heterogeneity into account, their high modeling cost becomes the main limitation of their practical application. In addition, these methods ignore the dynamic nature of spatial heterogeneity and find it difficult to reveal the spatiotemporal dynamic laws of network traffic under different business environments at different times, which in turn affects the prediction efficiency and accuracy of user network traffic demand, leading to the problem of low accuracy in base station sleep control. Summary of the invention

[0005] The present application provides a base station sleep method based on traffic prediction, which can solve the problem of low accuracy in sleep control of the base station.

[0006] In a first aspect, an embodiment of the present application provides a base station sleep method based on traffic prediction, the base station sleep method comprising:

[0007] Obtain historical network traffic data and spatial environment data for each spatial unit in the target area; the historical network traffic data includes the network traffic of the spatial unit at multiple historical moments;

[0008] Perform feature extraction on each historical network traffic data to obtain the traffic feature vector of each spatial unit, and calculate the temporal similarity between every two spatial units based on all traffic feature vectors;

[0009] A spatiotemporal dependency graph is constructed based on all temporal similarities; multiple nodes in the spatiotemporal dependency graph correspond one-to-one to multiple spatial units, and the edge between two nodes is the similarity relationship between the corresponding two spatial units;

[0010] Obtain static environmental characteristics of the target area and dynamic environmental characteristics corresponding to each historical moment according to the spatial environmental data of all spatial units, and calculate model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics;

[0011] Based on the spatiotemporal dependency graph and model parameters, the traffic prediction model is used to predict the traffic in the target area and obtain the predicted network traffic of all spatial units;

[0012] An optimization objective function is constructed to describe the status of all base stations in the target area according to the predicted network traffic, and the optimization objective function is minimized to obtain a base station sleep control scheme for the target area; the base station sleep control scheme is used to describe the operating status of all base stations.

[0013] Optionally, feature extraction is performed on each historical network traffic data to obtain a traffic feature vector for each spatial unit, including:

[0014] For each historical network traffic data, Fourier transform is performed on the historical network traffic data to obtain a spectrum sequence, and the elements corresponding to the first multiple historical moments in the spectrum sequence are selected as key elements, and the traffic feature vector is constructed based on all key elements.

[0015] Optionally, perform Fourier transform on historical network traffic data to obtain a spectrum sequence, including:

[0016] By formula:

[0017]

[0018] Perform Fourier transform on the i-th historical network traffic data to obtain the i-th spectrum sequence S i (u);

[0019] Among them, x i (t) represents the network traffic at the tth historical moment in the i-th historical network traffic data, L-1 represents the number of the last historical moment, and j represents an imaginary unit;

[0020] Construct a traffic feature vector based on all key elements, including:

[0021] By formula:

[0022]

[0023] Calculate the i-th traffic feature vector F i ;

[0024] Among them, S i (u0) represents the 0th key element of the i-th spectrum sequence, represents the Kth ith spectral sequencef key elements, Indicates S i The real part of (u0), Indicates S i The imaginary part of (u0), express The real part of express The imaginary part of

[0025] The temporal similarity between every two spatial units is calculated based on all traffic feature vectors, including:

[0026] By formula:

[0027]

[0028]

[0029] Calculate the temporal similarity d(F i ,F g );

[0030] Among them, S g (u0) represents the 0th key element of the gth spectrum sequence, represents the Kth spectral sequence of the gth f key elements, Indicates S g The real part of (u0), express The imaginary part of , i,g∈(1,2,...,N), i≠g, N represents the number of spatial units.

[0031] Optionally, build a spatiotemporal dependency graph based on all temporal similarities, including:

[0032] For each spatial unit, all the time series similarities corresponding to the spatial unit are sorted from large to small, and the first multiple time series similarities in the sorting result are taken as the target time series similarities, and it is considered that there is a similar relationship between the spatial unit and the spatial unit corresponding to each target time series similarity;

[0033] Multiple spatial units are regarded as multiple nodes. If there is a similar relationship between the two spatial units corresponding to two nodes, an edge is generated between the two nodes to construct a spatiotemporal dependency graph.

[0034] Optionally, the spatial environment data includes static environment data of the spatial unit and dynamic environment data of multiple historical moments;

[0035] According to the spatial environment data of all spatial units, the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment are obtained, including:

[0036] Extract features from the static environment data of all spatial units to obtain the static environment features of the target area;

[0037] The dynamic environment data of all spatial units at each historical moment are extracted to obtain the dynamic environment features corresponding to the target area at each historical moment.

[0038] Optionally, the model parameters of the traffic prediction model are calculated based on the static environment characteristics and all dynamic environment characteristics, including:

[0039] Acquire the geographic meta-knowledge of the target area at each historical moment based on static environmental features and all dynamic environmental features;

[0040] For each historical moment, model parameters of the traffic prediction model at the historical moment are generated based on the geographic meta-knowledge corresponding to the historical moment.

[0041] Optionally, the geographic meta-knowledge of the target area at each historical moment is obtained based on the static environmental characteristics and the dynamic environmental characteristics, including:

[0042] By formula:

[0043]

[0044] Calculate the geographical meta-knowledge MK at the tth historical moment t ;

[0045] in, represents the convolution filter, SF represents the static environment feature, DF t represents the dynamic environment characteristics at the t-th historical moment, || represents the feature connection operator, and Reshape() represents the reorganization function;

[0046] The model parameters of the traffic prediction model at historical moments are generated based on geographic meta-knowledge, including:

[0047] By formula:

[0048] M t =Reshape -1 (MK t )M shared

[0049] Calculate the model parameter M at the tth historical moment t ;

[0050] Among them, M shared represents the shared parameter matrix, Wr,i ,W z,i ,W h,i ,b r,i ,b z,i ,b h,i These are all parameters in the traffic prediction model.

[0051] Optionally, based on the spatiotemporal dependency graph and model parameters, a traffic prediction model is used to predict traffic in the target area to obtain predicted network traffic for all spatial units, including:

[0052] Based on all historical network traffic data and all dynamic environment data, obtain the potential time-dependent characteristics of the target area at each historical moment;

[0053] Calculate the spatiotemporal dependency features of each historical moment using the spatiotemporal dependency graph and the potential temporal dependency features of each historical moment;

[0054] For each historical moment, the model parameters and all spatiotemporal dependency features of the historical moment are used to extract the dynamic heterogeneous spatiotemporal dependency features of the historical moment;

[0055] Traffic prediction is performed based on the dynamic heterogeneous spatiotemporal dependency characteristics of the last historical moment to obtain the predicted network traffic of all spatial units.

[0056] Optionally, based on all historical network traffic data and all dynamic environment data, obtain the potential time-dependent characteristics of the target area at each historical moment, including:

[0057] By formula:

[0058]

[0059] Calculate the potential time-dependent features at the t-th historical moment

[0060] in, represents the potential time-dependent characteristics of the ith spatial unit at the tth historical moment, N represents the number of spatial units, and X t represents the network traffic of all spatial units at the tth historical moment, DE t represents the dynamic environment data of all spatial units at the t-th historical moment, GRU() represents the gated recurrent unit, represents the potential time-dependent characteristics at the t-1th historical moment. When t = 0, is the preset initial value;

[0061] The spatiotemporal dependency features of each historical moment are calculated using the spatiotemporal dependency graph and the potential temporal dependency features of each historical moment, including:

[0062] By formula:

[0063]

[0064] Calculate the spatiotemporal dependency characteristics of the t-th historical moment

[0065] in, represents the similarity weight vector between the i-th spatial unit and the j-th spatial unit at the t-th historical moment, represents the potential time-dependent characteristics of the jth spatial unit at the tth historical moment, W g and b g Indicates the parameters, represents the attention score vector between the i-th spatial unit and the j-th spatial unit at the t-th historical moment, N(i) represents the number of the spatial units corresponding to all nodes adjacent to the node corresponding to the i-th spatial unit in the spatiotemporal dependency graph, represents the spatiotemporal dependency characteristics of the i-th spatial unit at the t-th historical moment, LeakyReLU represents the activation function, x represents the input data of the activation function, and α represents the hyperparameter;

[0066] Using the model parameters and all spatiotemporal dependency features of historical moments, the dynamic heterogeneous spatiotemporal dependency features of historical moments are extracted, including:

[0067] By formula:

[0068]

[0069] Compute the dynamic heterogeneous spatiotemporal dependency characteristics at the t-th historical moment

[0070] in, represents the reset gate output corresponding to the i-th spatial unit at the t-th historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the ith spatial unit at the tth historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the ith spatial unit at the t-1th historical moment, represents the update gate output corresponding to the ith spatial unit at the tth historical moment, represents the candidate hidden state of the i-th spatial unit at the t-th historical moment;

[0071] Traffic prediction is performed based on the dynamic heterogeneous spatiotemporal dependency characteristics of the last historical moment to obtain the predicted network traffic of all spatial units, including:

[0072] By formula:

[0073]

[0074] Calculate the predicted network traffic for all spatial units

[0075] in, Represents the predicted dynamic environmental data of all spatial units, MK t represents geographic meta-knowledge, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the last historical moment, W p and b p All represent parameters.

[0076] Optionally, the optimization objective function is:

[0077]

[0078] Where E represents the coverage energy consumption matrix, e m,i represents the energy consumption when the mth base station serves the ith spatial unit, M represents the number of base stations in the target area, N represents the number of spatial units, S t′ represents the base station state matrix at the future time, Represents the state of the mth base station. When , it means that the mth base station is in working state at the future time. When , it means that the mth base station is in sleep state at the future time, E 0 represents the fixed energy consumption matrix of the base station, represents the fixed energy consumption of the mth base station, E s represents the base station state switching energy consumption matrix, represents the energy consumption of the mth base station switching state, S t′-1 represents the base station state matrix at the previous moment in the future, F represents the coverage traffic matrix, f m,i It represents the traffic load that the m-th base station can provide when serving the i-th spatial unit.

[0079] In a second aspect, an embodiment of the present application provides a base station sleep device based on traffic prediction, including:

[0080] The first acquisition module acquires the historical network traffic data and spatial environment data of each spatial unit in the target area; the historical network traffic data includes the network traffic of the spatial unit at multiple historical moments;

[0081] The feature extraction module extracts features from each historical network traffic data to obtain the traffic feature vector of each spatial unit, and calculates the temporal similarity between every two spatial units based on all traffic feature vectors;

[0082] The construction module constructs a spatiotemporal dependency graph based on all temporal similarities; multiple nodes in the spatiotemporal dependency graph correspond one-to-one to multiple spatial units, and the edge between two nodes is the similarity relationship between the corresponding two spatial units;

[0083] The second acquisition module acquires the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment according to the spatial environmental data of all spatial units, and calculates the model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics;

[0084] The traffic prediction module uses the traffic prediction model to predict the traffic flow in the target area based on the spatiotemporal dependency graph and model parameters, and obtains the predicted network traffic flow of all spatial units;

[0085] The minimization solution module constructs an optimization objective function for describing the status of all base stations in the target area according to the predicted network traffic, and minimizes the optimization objective function to obtain the base station sleep control plan in the target area; the base station sleep control plan is used to describe the operating status of all base stations.

[0086] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned base station sleep method based on traffic prediction when executing the above-mentioned computer program.

[0087] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned base station sleep method based on traffic prediction is implemented.

[0088] The above solution of the present application has the following beneficial effects:

[0089] In an embodiment of the present application, historical network traffic data and spatial environment data of each spatial unit in the target area are obtained, and then feature extraction is performed on each historical network traffic data to obtain a traffic feature vector for each spatial unit, and the temporal similarity between every two spatial units is calculated based on all traffic feature vectors, and then a spatiotemporal dependency graph is constructed based on all temporal similarities, and then the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment are obtained according to the spatial environment data of all spatial units, and the model parameters of the traffic prediction model are calculated based on the static environmental characteristics and all dynamic environmental characteristics, and then based on the spatiotemporal dependency graph and the model parameters, the traffic prediction model is used to predict the traffic in the target area to obtain the predicted network traffic of all spatial units, and finally, an optimization objective function for describing the status of all base stations in the target area is constructed based on the predicted network traffic, and the optimization objective function is minimized to obtain a base station sleep control scheme for the target area. Among them, traffic prediction is performed based on spatial environment data and historical network traffic data, the characteristics of network traffic in both spatial and temporal dimensions are analyzed, the influence of space and time on network traffic is characterized, and the accuracy of traffic prediction is effectively improved. The optimization objective function of the base station is constructed based on the highly accurate predicted network traffic, so that the optimization objective function can reasonably and accurately describe the state of the base station at future moments. The base station is put into sleep mode based on a reasonable and accurate optimization objective function, which effectively improves the accuracy of sleep control of the base station.

[0090] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0092] Figure 1 A flowchart of a base station sleep method based on traffic prediction provided in an embodiment of the present application;

[0093] Figure 2 A schematic diagram of a specific flow chart of a base station sleep method based on traffic prediction provided in an embodiment of the present application;

[0094] Figure 3 A schematic diagram of the structure of a base station sleep device based on traffic prediction provided in one embodiment of the present application;

[0095] Figure 4 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0096] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0097] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0098] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0099] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0100] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0101] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0102] In response to the problem of low accuracy of existing sleep control of base stations, an embodiment of the present application provides a base station sleep method based on traffic prediction. The base station sleep method obtains historical network traffic data and spatial environment data of each spatial unit in the target area, and then extracts features from each historical network traffic data to obtain a traffic feature vector for each spatial unit, and calculates the time series similarity between every two spatial units based on all traffic feature vectors, and then constructs a spatiotemporal dependency graph based on all time series similarities, and then obtains the static environment characteristics of the target area and the dynamic environment characteristics corresponding to each historical moment based on the spatial environment data of all spatial units, and calculates the model parameters of the traffic prediction model based on the static environment characteristics and all dynamic environment characteristics, and then uses the traffic prediction model based on the spatiotemporal dependency graph and the model parameters to predict the traffic in the target area, and obtains the predicted network traffic of all spatial units, and finally constructs an optimization objective function for describing the status of all base stations in the target area based on the predicted network traffic, and minimizes the optimization objective function to obtain a base station sleep control scheme for the target area. Among them, traffic prediction is performed based on spatial environment data and historical network traffic data, the characteristics of network traffic in both spatial and temporal dimensions are analyzed, the influence of space and time on network traffic is characterized, and the accuracy of traffic prediction is effectively improved. The optimization objective function of the base station is constructed based on the highly accurate predicted network traffic, so that the optimization objective function can reasonably and accurately describe the state of the base station at future moments. The base station is put into sleep mode based on a reasonable and accurate optimization objective function, which effectively improves the accuracy of sleep control of the base station.

[0103] Next, an exemplary description is given of the base station sleep method based on traffic prediction provided in this application.

[0104] like Figure 1 As shown, the base station sleep method based on traffic prediction provided by the present application includes the following steps:

[0105] Step 11, obtaining historical network traffic data and spatial environment data of each spatial unit in the target area.

[0106] The above-mentioned historical network traffic data includes the network traffic of the spatial unit at multiple historical moments (such as incoming calls, outgoing calls, text message reception, text message transmission, etc.). The above-mentioned target area is the area where the base station needs to be put under sleep control, and the spatial unit is the sub-area divided by the target area. For example, the target area can be divided into grids, and the area corresponding to each grid is a spatial unit. The spatial environment data includes the static environment data of the spatial unit (such as the location of the spatial unit, different road densities, building coverage ratio, number of buildings, impervious surface area, night light intensity, number of points of interest, intensity of social activities, number of base stations, etc.) and dynamic environment data at multiple historical moments (such as temperature, relative humidity, total radiation, net radiation, atmospheric pressure, precipitation intensity, precipitation coverage, precipitation type, week corresponding to the historical moment, hour corresponding to the historical moment, etc.).

[0107] In some embodiments of the present application, historical network traffic data can be obtained by accessing a system that monitors the network traffic of the target area (such as an open API interface for communication data provided by the three major communication operators), and spatial environment data can be obtained by accessing a website that discloses relevant data of the target area.

[0108] Step 12: extract features from each historical network traffic data to obtain a traffic feature vector for each spatial unit, and calculate the temporal similarity between every two spatial units based on all traffic feature vectors.

[0109] In some embodiments of the present application, the above steps of extracting features from each historical network traffic data, obtaining a traffic feature vector for each spatial unit, and calculating the temporal similarity between every two spatial units based on all traffic feature vectors are specifically as follows:

[0110] In the first step, for each historical network traffic data, Fourier transform is performed on the historical network traffic data to obtain a spectrum sequence, and the elements corresponding to the first multiple historical moments in the spectrum sequence are selected as key elements, and the traffic feature vector is constructed based on all key elements.

[0111] Specifically, through the formula:

[0112]

[0113] Perform Fourier transform on the i-th historical network traffic data to obtain the i-th spectrum sequence S i (u).

[0114] Among them, x i (t) represents the network traffic at the tth historical moment in the i-th historical network traffic data, L-1 represents the number of the last historical moment, and j represents an imaginary unit.

[0115] By formula:

[0116]

[0117] Calculate the i-th traffic feature vector F i .

[0118] Among them, S i (u0) represents the 0th key element of the i-th spectrum sequence, represents the Kth ith spectral sequence f key elements, Indicates S i The real part of (u0), Indicates S i The imaginary part of (u0), express The real part of express The imaginary part of .

[0119] In the second step, the temporal similarity between every two spatial units is calculated based on all traffic feature vectors.

[0120] Specifically, through the formula:

[0121]

[0122] Calculate the temporal similarity d(F i ,F g ).

[0123] Among them, S g (u0) represents the 0th key element of the gth spectrum sequence, represents the Kth spectral sequence of the gth f key elements, Indicates S g The real part of (u0), express The imaginary part of , i,g∈(1,2,...,N), i≠g, N represents the number of spatial units.

[0124] It should be noted that the number of key elements is a preset number.

[0125] Step 13: construct a spatiotemporal dependency graph based on all temporal similarities.

[0126] The multiple nodes in the above-mentioned space-time dependency graph correspond one-to-one to the multiple space units, and the edge between two nodes is the similarity relationship between the corresponding two space units.

[0127] In some embodiments of the present application, the above-mentioned construction of the spatiotemporal dependency graph based on all temporal similarities is specifically as follows:

[0128] In the first step, for each spatial unit, all the time series similarities corresponding to the spatial unit are sorted from large to small, and the first multiple time series similarities in the sorting result are used as the target time series similarities, and it is considered that there is a similar relationship between the spatial unit and the spatial unit corresponding to each target time series similarity.

[0129] Exemplarily, for the first spatial unit, the temporal similarities between it and the second spatial unit, the third spatial unit, the fourth spatial unit, and the fifth spatial unit are 1.4, 2.5, 0.2, and 1.3, respectively, where the number of target temporal similarities is preset to 3. It is considered that the spatial units corresponding to 2.5, 1.4, and 1.3 (the third spatial unit, the second spatial unit, and the fifth spatial unit) have a similar relationship with the first spatial unit.

[0130] In the second step, multiple spatial units are used as multiple nodes. If there is a similar relationship between the two spatial units corresponding to two nodes, an edge is generated between the two nodes to construct a spatiotemporal dependency graph.

[0131] It is worth mentioning that by calculating the temporal similarity between spatial units based on historical network traffic data, it is possible to spatially analyze the status of historical network traffic in the target area and analyze the correlation between network traffic between spatial units.

[0132] Step 14, obtaining the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment according to the spatial environmental data of all spatial units, and calculating the model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics.

[0133] In some embodiments of the present application, the steps of obtaining the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment according to the spatial environmental data of all spatial units, and calculating the model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics are specifically as follows:

[0134] In the first step, the static environment data of all spatial units are extracted to obtain the static environment features of the target area.

[0135] For example, two fully connected layers connected in sequence can be used to extract features from the static environment data of all spatial units to obtain the static environment features of the target area, which can be expressed as:

[0136]

[0137] Among them, SF represents the static environment characteristics, and They all represent learnable parameters, σ() represents the activation function, x represents the input of the activation function, and SE represents the static environment data of all spatial units.

[0138] The second step is to extract features from the dynamic environment data of all spatial units at each historical moment to obtain the dynamic environment features corresponding to the target area at each historical moment.

[0139] For example, the gated recurrent units and fully connected layers connected in sequence can be used to extract features from all dynamic environment data of all spatial units to obtain the dynamic environment features corresponding to the target area at each historical moment, expressed as:

[0140]

[0141] Among them, DF t represents the dynamic environment characteristics at the t-th historical moment, Denotes DE t The output hidden state, DE t represents the dynamic environment data of all spatial units at the t-th historical moment, Denotes DE t-1 The output hidden state, DE t-1 represents the dynamic environment data of all spatial units at the t-1th historical moment, σ() represents the activation function, GRU() represents the calculation of the gated recurrent unit, and the expression is:

[0142]

[0143] Among them, R t Represents the reset gate, Z t represents the update gate, represents the candidate hidden state, X t represents the input data of the gated recurrent unit (i.e., the dynamic environment data of all spatial units at the t-th historical moment), W xr , W hr , W xz , W hz , W xh , W hh , b r , b z , b h These are parameters in the gated neural unit.

[0144] The third step is to obtain the geographic meta-knowledge of the target area at each historical moment based on static and dynamic environmental characteristics.

[0145] Specifically, through the formula:

[0146]

[0147] Calculate the geographical meta-knowledge MK at the tth historical moment t .

[0148] in, represents the convolution filter, SF represents the static environment feature, DF t represents the dynamic environment characteristics at the t-th historical moment, || represents the feature connection operator, and Reshape() represents the reorganization function.

[0149] The fourth step is to generate the model parameters of the traffic prediction model at each historical moment based on the geographic meta-knowledge corresponding to the historical moment.

[0150] Specifically, through the formula:

[0151] M t =Reshape -1 (MK t )M shared

[0152] Calculate the model parameter M at the tth historical moment t .

[0153] Among them, M shared represents the shared parameter matrix, W r,i ,W z,i ,W h,i ,b r,i ,b z,i ,b h,i These are all parameters in the traffic prediction model.

[0154] It is worth mentioning that obtaining static environmental characteristics and dynamic environmental characteristics can describe the environmental information of the target area in terms of static environment and dynamic environment respectively, and calculating model parameters based on static environmental characteristics and dynamic environmental characteristics makes the calculation of model parameters contain information of static environment and dynamic environment, thereby improving the practicality and accuracy of model parameters.

[0155] Step 15: Based on the spatiotemporal dependency graph and model parameters, the traffic prediction model is used to predict the traffic in the target area to obtain the predicted network traffic of all spatial units.

[0156] The above-mentioned predicted network traffic is the network traffic of the spatial unit at a future time (such as incoming phone calls, outgoing phone calls, received text messages, sent text messages, etc.).

[0157] In some embodiments of the present application, the above-mentioned steps of performing traffic prediction on the target area using the traffic prediction model based on the spatiotemporal dependency graph and the model parameters to obtain the predicted network traffic of all spatial units are specifically as follows:

[0158] In the first step, based on all historical network traffic data and all dynamic environment data, the potential time-dependent characteristics of the target area at each historical moment are obtained.

[0159] Specifically, through the formula:

[0160]

[0161] Calculate the potential time-dependent features at the t-th historical moment

[0162] in, represents the potential time-dependent characteristics of the ith spatial unit at the tth historical moment, N represents the number of spatial units, and X t represents the network traffic of all spatial units at the tth historical moment, DE t represents the dynamic environment data of all spatial units at the t-th historical moment, GRU() represents the gated recurrent unit, represents the potential time-dependent characteristics at the t-1th historical moment. When t = 0, The default initial value.

[0163] In the second step, the spatiotemporal dependency features of each historical moment are calculated using the spatiotemporal dependency graph and the latent temporal dependency features of each historical moment.

[0164] Specifically, through the formula:

[0165]

[0166]

[0167] Calculate the spatiotemporal dependency characteristics of the t-th historical moment

[0168] in, represents the similarity weight vector between the i-th spatial unit and the j-th spatial unit at the t-th historical moment, represents the potential time-dependent characteristics of the jth spatial unit at the tth historical moment, W g and b g Indicates the parameters, represents the attention score vector between the i-th spatial unit and the j-th spatial unit at the t-th historical moment, N(i) represents the number of the spatial units corresponding to all nodes adjacent to the node corresponding to the i-th spatial unit in the spatiotemporal dependency graph, represents the spatiotemporal dependency characteristics of the i-th spatial unit at the t-th historical moment, LeakyReLU represents the activation function, x represents the input data of the activation function, and α represents the hyperparameter.

[0169] The third step is to extract the dynamic heterogeneous spatiotemporal dependency features of each historical moment using the model parameters and all spatiotemporal dependency features of the historical moment.

[0170] Specifically, through the formula:

[0171]

[0172] Compute the dynamic heterogeneous spatiotemporal dependency characteristics at the t-th historical moment

[0173] in, represents the reset gate output corresponding to the i-th spatial unit at the t-th historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the ith spatial unit at the tth historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the ith spatial unit at the t-1th historical moment, represents the update gate output corresponding to the ith spatial unit at the tth historical moment, represents the candidate hidden state of the ith spatial unit at the tth historical moment. When t = 0, The default initial value.

[0174] In the fourth step, traffic prediction is performed based on the dynamic heterogeneous spatiotemporal dependency characteristics of the last historical moment to obtain the predicted network traffic of all spatial units.

[0175] Specifically, through the formula:

[0176]

[0177] Calculate the predicted network traffic for all spatial units

[0178] in, Represents the predicted dynamic environmental data of all spatial units, MK t represents geographic meta-knowledge, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the last historical moment, W p and b p All represent parameters.

[0179] It can be understood that all the above calculation formulas (the above formula for calculating potential time dependency features, the formula for calculating spatiotemporal dependency features, the formula for calculating dynamic heterogeneous spatiotemporal dependency features, and the formula for calculating predicted network traffic) are expressions of the prediction model. If the current time is 9 o'clock, the multiple historical moments can be the previous moments including the current time, such as 6 o'clock, 7 o'clock, 8 o'clock and 9 o'clock, and the future moment is the moment after the current moment, such as 10 o'clock.

[0180] Exemplarily, before performing the traffic prediction in the fourth step, the above-mentioned formula for predicting network traffic can be used to predict traffic for the dynamic heterogeneous spatiotemporal dependency characteristics of each historical moment, and the predicted network traffic at the next moment of each historical moment can be obtained to form training data. Then, a loss function is constructed based on the training data, and the prediction model is trained based on the loss function. If the calculated loss function value is less than the preset value, the prediction model is used to perform the above-mentioned fourth step to obtain the predicted network traffic required for this step; if the value of the loss function is greater than or equal to the preset value, the back propagation algorithm is used to pass the value of the loss function to each learnable parameter of the prediction model, and the learnable parameters of the prediction model are updated based on the Adam optimizer, and the step of obtaining the potential time-dependent characteristics of the target area at each historical moment based on all historical network traffic data and all dynamic environmental data is returned until the value of the loss function is less than the preset value. The loss function is:

[0181]

[0182] in, represents the value of the loss function, represents the predicted network traffic of all spatial units at the i-th historical moment, X' i represents the network traffic of all spatial units at the i-th historical moment, N T represents the number of historical moments used as training samples, represents the predicted dynamic environmental data of all spatial units at the i-th historical moment, DE' i represents the dynamic environment data of all spatial units at the i-th historical moment, and λ represents an adjustable hyperparameter.

[0183] It is worth mentioning that traffic prediction is performed based on spatial environment data and historical network traffic data, and the characteristics of network traffic in both spatial and temporal dimensions are analyzed to characterize the impact of space and time on network traffic, which effectively improves the accuracy of traffic prediction.

[0184] Step 16, constructing an optimization objective function for describing the status of all base stations in the target area according to the predicted network traffic, and minimizing the optimization objective function to obtain a base station sleep control solution for the target area.

[0185] The above base station sleep control scheme is used to describe the operating status of all base stations. For example, at a future moment, the first base station is in a sleep state, and the second base station is in a working state.

[0186] Specifically, the above optimization objective function is:

[0187]

[0188] Where E represents the coverage energy consumption matrix, e m,i represents the energy consumption when the mth base station serves the ith spatial unit, M represents the number of base stations in the target area, N represents the number of spatial units, S t′ represents the base station state matrix at the future time, Represents the state of the mth base station. When , it means that the mth base station is in working state at the future time. When , it means that the mth base station is in sleep state at the future time, E 0 represents the fixed energy consumption matrix of the base station, represents the fixed energy consumption of the mth base station, E s represents the base station state switching energy consumption matrix, represents the energy consumption of the mth base station switching state, S t′-1 represents the base station state matrix at the previous moment in the future, F represents the coverage traffic matrix, f m,i It represents the traffic load that the m-th base station can provide when serving the i-th spatial unit.

[0189] For example, a genetic algorithm can be used to minimize the optimization objective function to obtain a base station sleep control scheme for the target area. After the base station sleep control scheme is obtained, at a future time, the states of all base stations in the target area are controlled according to the base station sleep control scheme.

[0190] It is worth mentioning that traffic prediction is performed based on spatial environmental data and historical network traffic data, and the characteristics of network traffic in both spatial and temporal dimensions are analyzed. The impact of space and time on network traffic is characterized, which effectively improves the accuracy of traffic prediction. The optimization objective function of the base station is constructed based on the highly accurate predicted network traffic, so that the optimization objective function can reasonably and accurately describe the state of the base station at future moments. The base station is put into sleep mode based on a reasonable and accurate optimization objective function, which effectively improves the accuracy of sleep control of the base station.

[0191] The base station sleep method based on traffic prediction in the present application applies meta-learning technology to the prediction model, characterizes the impact of the environment in time and space on network traffic, solves the dynamic spatial heterogeneity problem of network traffic modeling, and realizes efficient and high-precision prediction of user network traffic demand, which can meet the dynamic and intelligent needs of base station sleep strategy.

[0192] The method of the present application is illustrated below with reference to a specific example.

[0193] like Figure 2 As shown in FIG. 1 , the historical network traffic data and the historical dynamic environment data are input into the spatiotemporal learning unit (i.e., the prediction model mentioned above) composed of the gated neural unit (GRU, Gate Recurrent Unit), the graph attention network (GAT, Graph Attention Networks) and the meta-gated neural unit (MetaGRU, Meta Gate Recurrent Unit) connected in sequence. The data corresponding to each historical moment is input into the moment learning unit in sequence. At the same time, in the meta-learning module, the static environment data is input into two multi-layer perceptrons (MLP, Multi-Layer Perceptron) connected in sequence, and the historical dynamic environment data is input into the GRU and MLP connected in sequence, and both outputs are input into the convolution ConV. Combined with the shared parameter matrix M shared Calculate and obtain model parameters, input the model parameters into the spatiotemporal learning model, so that the spatiotemporal learning model outputs predicted network traffic, and obtains the future base station sleep plan based on the predicted network traffic. For multiple moments after the future moment, the predicted traffic of the next moment is obtained by inputting the future network traffic data and the future dynamic environment data into the spatiotemporal learning unit for step-by-step prediction, and the base station sleep is controlled based on the predicted traffic. represents the potential time-dependent features at the t-th historical moment, Represents the dynamic heterogeneous spatiotemporal dependency characteristics at the t-th historical moment.

[0194] In an example of the present application, network traffic data of a city center (including 1,600 spatial units, the time range is from November 3, 2013 to December 5, 2013, and the time interval is 1 hour) and spatial environment data (including weather data, precipitation data, date data, spatial location, frequency domain characteristics, road network data, building data, artificial impervious surface data, night light data, point of interest data, social media data and communication base station data) are used as real data sets of the embodiment to illustrate the implementation process of the present application.

[0195] The method of the present application is used to predict the traffic of the above-mentioned real data set, and the mean absolute error (MAE) and the root mean square error (RMSE) are selected as the evaluation indicators of the prediction results. The compared models include: Historical Average Model (HA), Gated Recurrent Unit (GRU), Convolutional Long Short-Term Memory Network (ConvLSTM), Temporal Graph Convolutional Network (T-GCN), Spatio-Temporal Graph Convolutional Network (STGNN), Spatio-Temporal Cross-Domain Network (STCNet). The accuracy comparison results of the single-step prediction task are shown in Table 1.

[0196]

[0197] Table 1

[0198] Among them, Call-in refers to the number of incoming calls, Call-out refers to the number of outgoing calls, SMS-in refers to the number of text messages received, and SMS-out refers to the number of text messages sent.

[0199] The accuracy comparison results of the five-step prediction task are shown in Table 2.

[0200]

[0201]

[0202] Table 2

[0203] It can be seen that the accuracy of the prediction results obtained by the method of this application is greater than that of other methods, whether it is a single-step prediction or a five-step prediction. This method can be applied to the actual base station sleep control scenario to predict the most accurate future user network traffic demand, thereby serving the base station dynamic sleep control plan and the green and low-carbon development needs of the information and communication industry.

[0204] The following is an exemplary description of the base station sleep device based on traffic prediction provided in the present application.

[0205] like Figure 3 As shown, the embodiment of the present application provides a base station sleep device based on traffic prediction, and the base station sleep device 300 includes:

[0206] The first acquisition module 301 acquires historical network traffic data and spatial environment data of each spatial unit in the target area; the historical network traffic data includes the network traffic of the spatial unit at multiple historical moments;

[0207] The feature extraction module 302 extracts features from each historical network traffic data to obtain a traffic feature vector for each spatial unit, and calculates the temporal similarity between every two spatial units based on all traffic feature vectors;

[0208] A construction module 303 constructs a spatiotemporal dependency graph based on all temporal similarities; a plurality of nodes in the spatiotemporal dependency graph correspond one-to-one to a plurality of spatial units, and an edge between two nodes is a similarity relationship between the corresponding two spatial units;

[0209] The second acquisition module 304 acquires the static environment characteristics of the target area and the dynamic environment characteristics corresponding to each historical moment according to the spatial environment data of all spatial units, and calculates the model parameters of the traffic prediction model based on the static environment characteristics and all dynamic environment characteristics;

[0210] The traffic prediction module 305 uses the traffic prediction model to perform traffic prediction on the target area based on the spatiotemporal dependency graph and model parameters to obtain the predicted network traffic of all spatial units;

[0211] The minimization solution module 306 constructs an optimization objective function for describing the status of all base stations in the target area according to the predicted network traffic, and minimizes the optimization objective function to obtain a base station sleep control scheme for the target area; the base station sleep control scheme is used to describe the operating status of all base stations.

[0212] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0213] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0214] like Figure 4 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0215] Specifically, when the processor D100 executes the computer program D102, it obtains the historical network traffic data and spatial environment data of each spatial unit in the target area, and then extracts features from each historical network traffic data to obtain the traffic feature vector of each spatial unit, and calculates the time series similarity between every two spatial units based on all traffic feature vectors, and then builds a spatiotemporal dependency graph based on all time series similarities, and then obtains the static environmental characteristics of the target area and the dynamic environmental characteristics corresponding to each historical moment based on the spatial environment data of all spatial units, and calculates the model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics, and then based on the spatiotemporal dependency graph and the model parameters, uses the traffic prediction model to predict the traffic in the target area, and obtains the predicted network traffic of all spatial units, and finally constructs an optimization objective function for describing the status of all base stations in the target area based on the predicted network traffic, and minimizes the optimization objective function to obtain a base station sleep control scheme for the target area. Among them, traffic prediction is performed based on spatial environment data and historical network traffic data, the characteristics of network traffic in both spatial and temporal dimensions are analyzed, the influence of space and time on network traffic is characterized, and the accuracy of traffic prediction is effectively improved. The optimization objective function of the base station is constructed based on the highly accurate predicted network traffic, so that the optimization objective function can reasonably and accurately describe the state of the base station at future moments. The base station is put into sleep mode based on a reasonable and accurate optimization objective function, which effectively improves the accuracy of sleep control of the base station.

[0216] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0217] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0218] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0219] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0220] 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 present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code to the base station dormancy method device / terminal device based on traffic prediction. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0221] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0222] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0223] The above is 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 described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A base station sleep method based on traffic prediction, characterized in that: include: Obtain historical network traffic data and spatial environment data for each spatial unit in the target area; The historical network traffic data includes the network traffic of the spatial unit at multiple historical moments; Performing feature extraction on each of the historical network traffic data to obtain a traffic feature vector of each of the spatial units, and calculating the temporal similarity between every two spatial units based on all the traffic feature vectors; Build a spatiotemporal dependency graph based on all temporal similarities; The multiple nodes in the spatiotemporal dependency graph correspond one-to-one to the multiple spatial units, and the edge between two nodes is the similarity relationship between the corresponding two spatial units; Obtaining static environmental characteristics of the target area and dynamic environmental characteristics corresponding to each historical moment according to the spatial environmental data of all spatial units, and calculating model parameters of the traffic prediction model based on the static environmental characteristics and all dynamic environmental characteristics; Based on the spatiotemporal dependency graph and the model parameters, using the traffic prediction model to perform traffic prediction on the target area to obtain predicted network traffic of all spatial units; An optimization objective function is constructed based on the predicted network traffic to describe the status of all base stations in the target area, and the optimization objective function is minimized to obtain a base station sleep control scheme for the target area; the base station sleep control scheme is used to describe the operating status of all base stations.

2. The base station sleep method according to claim 1, characterized in that: The feature extraction of each of the historical network traffic data to obtain the traffic feature vector of each of the spatial units includes: For each of the historical network traffic data, Fourier transform is performed on the historical network traffic data to obtain a spectrum sequence, and elements corresponding to the first multiple historical moments in the spectrum sequence are selected as key elements, and a traffic feature vector is constructed based on all the key elements.

3. The base station sleep method according to claim 2, characterized in that: The performing Fourier transform on the historical network traffic data to obtain a spectrum sequence includes: By formula: Perform Fourier transform on the i-th historical network traffic data to obtain the i-th spectrum sequence S i (u); Among them, x i (T) represents the network traffic at the Tth historical moment in the i-th historical network traffic data, L-1 represents the number of the last historical moment, and j represents an imaginary unit; The construction of the traffic feature vector based on all key elements includes: By formula: Calculate the i-th traffic feature vector F i ; Among them, S i (u0) represents the 0th key element of the i-th spectrum sequence, represents the Kth frequency spectrum sequence of the i-th i key elements, Indicates S i The real part of (u0), Indicates S i The imaginary part of (u0), express The real part of express The imaginary part of The calculating of the time series similarity between every two spatial units based on all traffic feature vectors includes: By formula: Calculate the temporal similarity d(F i , F g ); Among them, S g (u0) represents the 0th key element of the gth spectrum sequence, represents the Kth g-th spectrum sequence f key elements, Indicates S g The real part of (u0), express The imaginary part of , i, g∈(1, 2, ..., N), i≠g, N represents the number of spatial units.

4. The base station sleep method according to claim 3, characterized in that: The construction of a spatiotemporal dependency graph based on all temporal similarities includes: For each spatial unit, all the time series similarities corresponding to the spatial unit are sorted from large to small, the first multiple time series similarities in the sorting result are all used as the target time series similarities, and it is considered that there is a similar relationship between the spatial unit and the spatial unit corresponding to each of the target time series similarities; Multiple spatial units are regarded as multiple nodes. If there is a similar relationship between the two spatial units corresponding to two nodes, an edge is generated between the two nodes to construct a spatiotemporal dependency graph.

5. The base station sleep method according to claim 1, characterized in that: The spatial environment data includes static environment data of the spatial unit and dynamic environment data of multiple historical moments; The step of acquiring the static environmental features of the target area and the dynamic environmental features corresponding to each historical moment according to the spatial environmental data of all spatial units includes: Extracting features from the static environment data of all spatial units to obtain static environment features of the target area; Feature extraction is performed on the dynamic environment data of all spatial units at each historical moment to obtain the dynamic environment features corresponding to the target area at each historical moment.

6. The base station sleep method according to claim 1, characterized in that: The model parameters of the flow prediction model are calculated based on the static environmental features and all dynamic environmental features, including: Acquire geographic meta-knowledge of the target area at each historical moment according to the static environmental features and all dynamic environmental features; For each historical moment respectively, model parameters of the traffic prediction model at the historical moment are generated based on the geographic meta-knowledge corresponding to the historical moment.

7. The base station sleep method according to claim 6, characterized in that: The acquiring of geographic meta-knowledge of the target area at each historical moment according to the static environmental features and the dynamic environmental features includes: By formula: Calculate the geographical meta-knowledge MK at the tth historical moment t ; in, represents the convolution filter, SF represents the static environment feature, DF t represents the dynamic environment characteristics at the t-th historical moment, || represents the feature connection operator, and Reshape() represents the reorganization function; The generating of the model parameters of the traffic prediction model at the historical moment based on the geographic meta-knowledge corresponding to the historical moment includes: By formula: M t =Reshape -1 (MK t )M shared Calculate the model parameter M at the tth historical moment t ; Among them, M shared represents the shared parameter matrix, W r，i , W z，i , W h，i , b r，i , b z，i , b h，i All of them are parameters in the traffic prediction model.

8. The base station sleep method according to claim 7, characterized in that: The method of performing traffic prediction on the target area based on the spatiotemporal dependency graph and the model parameters using the traffic prediction model to obtain predicted network traffic of all spatial units includes: Based on all historical network traffic data and all dynamic environment data, obtaining potential time-dependent characteristics of the target area at each historical moment; Calculating the spatiotemporal dependency feature of each historical moment using the spatiotemporal dependency graph and the potential time dependency feature of each historical moment; For each historical moment, respectively, the model parameters and all spatiotemporal dependency features of the historical moment are used to extract the dynamic heterogeneous spatiotemporal dependency features of the historical moment; Traffic prediction is performed based on the dynamic heterogeneous spatiotemporal dependency characteristics of the last historical moment to obtain the predicted network traffic of all spatial units.

9. The base station sleep method according to claim 8, characterized in that: The step of obtaining the potential time-dependent characteristics of the target area at each historical moment based on all historical network traffic data and all dynamic environment data includes: By formula: Calculate the potential time-dependent features at the Tth historical moment in, represents the potential time-dependent characteristics of the ith spatial unit at the Tth historical moment, N represents the number of spatial units, and X t represents the network traffic of all spatial units at the Tth historical moment, DE t represents the dynamic environment data of all spatial units at the Tth historical moment, GRU() represents the gated recurrent unit, represents the potential time-dependent characteristics of the T-1th historical moment. When T = 0, is the preset initial value; The calculating of the spatiotemporal dependency features of each historical moment by using the spatiotemporal dependency graph and the potential time dependency features of each historical moment comprises: By formula: Calculate the spatiotemporal dependency characteristics of the Tth historical moment in, represents the similarity weight vector between the i-th spatial unit and the j-th spatial unit at the T-th historical moment, represents the potential time-dependent characteristics of the j-th spatial unit at the T-th historical moment, W g and b g Indicates the parameters, represents the attention score vector between the i-th spatial unit and the j-th spatial unit at the t-th historical moment, N(i) represents the number of the spatial units corresponding to all nodes adjacent to the node corresponding to the i-th spatial unit in the spatiotemporal dependency graph, represents the spatiotemporal dependency characteristics of the i-th spatial unit at the T-th historical moment, LeakyReLU represents the activation function, x represents the input data of the activation function, and α represents the hyperparameter; The extracting the dynamic heterogeneous spatiotemporal dependency features of the historical moments by using the model parameters and all spatiotemporal dependency features of the historical moments includes: By formula: Compute the dynamic heterogeneous spatiotemporal dependency characteristics at the Tth historical moment in, represents the reset gate output corresponding to the i-th spatial unit at the T-th historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the i-th spatial unit at the T-th historical moment, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the i-th spatial unit at the T-1th historical moment, represents the update gate output corresponding to the i-th spatial unit at the T-th historical moment, represents the candidate hidden state of the i-th spatial unit at the T-th historical moment; The traffic prediction is performed based on the dynamic heterogeneous spatiotemporal dependency characteristics of the last historical moment to obtain the predicted network traffic of all spatial units, including: By formula: Calculate the predicted network traffic for all spatial units in, Represents the predicted dynamic environmental data of all spatial units, MK t represents geographic meta-knowledge, represents the dynamic heterogeneous spatiotemporal dependence characteristics of the last historical moment, W p and b p All represent parameters.

10. The base station sleep method according to claim 1, characterized in that: The optimization objective function is: Where E represents the coverage energy consumption matrix, e mi represents the energy consumption when the mth base station serves the ith spatial unit, M represents the number of base stations in the target area, N represents the number of spatial units, S t′ represents the base station state matrix at the future time, Represents the state of the mth base station. When , it means that the mth base station is in working state at the future time. When , it means that the mth base station is in sleep state at the future time, E 0 represents the fixed energy consumption matrix of the base station, represents the fixed energy consumption of the mth base station, E s represents the base station state switching energy consumption matrix, represents the energy consumption of the mth base station switching state, S t′-1 represents the base station state matrix at the previous moment in the future, F represents the coverage traffic matrix, f m，i It represents the traffic load that the m-th base station can provide when serving the i-th spatial unit.

Images