Resource optimization method and device using unmanned aerial vehicle to assist semantic communication

Abstract

The invention discloses a resource optimization method and device for assisting semantic communication by using an unmanned aerial vehicle, and the method comprises the steps: determining an original text, and generating corresponding semantic information according to the original text; recovering the semantic information, and generating a recovered text; according to the original text and the recovery text, a semantic similarity maximization problem is determined, semantic similarity is used for indicating semantic accuracy and semantic integrity between the original text and the recovery text, and the semantic similarity maximization problem comprises a resource allocation optimization problem and a semantic information transfer strategy optimization problem; creating an initial strategy, and updating parameters of the optimized initial strategy by using a standard gradient ascending method; determining the relationship between the importance distribution and the semantic similarity of the semantic information by utilizing the updated parameters; and according to the determined relationship, determining an optimal solution of the resource allocation optimization problem and an optimal solution of the semantic information transfer strategy optimization problem, so that the semantic similarity of the original text and the recovered text is maximized.

Description

Technical Field

[0001] The present application relates to the field of drone communication technology, and in particular to a resource optimization method and device using drone-assisted semantic communication. Background Art

[0002] At present, information and communication science is facing another important juncture. Communication technology guided by classical information theory has restricted the development of future communications. People's multi-dimensional and multi-channel communication needs will bring about the diversification of data modalities and task requirements, which will become challenges to the existing architecture. At present, these challenges need to be solved by combining perception and semantic communication.

[0003] In existing communication methods, people often use drones as intermediaries to connect semantic data transmission between users and base stations. That is, the semantic data transmission process is user (terminal device)-drone-base station. In other words, the terminal device encodes the original text data to generate corresponding semantic data and transmits the semantic data to the drone. The drone, as a transit device, transmits the received semantic data to the base station. The base station then decodes the semantic data and generates a restored text. In this process, the semantic similarity between the restored text generated by the base station and the original text is the key to measuring the quality of the semantic communication process, so it is necessary to increase the semantic similarity as much as possible (that is, maximize the semantic similarity). Among them, semantic similarity includes semantic accuracy and semantic completeness.

[0004] However, the existing optimization algorithms cannot solve the problem of maximizing semantic similarity, and thus cannot improve the quality of the semantic communication process.

[0005] Currently, no effective solution has been proposed for the technical problem that the traditional optimization algorithms in the prior art cannot solve the problem of maximizing semantic similarity and thus cannot improve the quality of the semantic communication process. Summary of the invention

[0006] The embodiments of the present disclosure provide a resource optimization method and device for using drone-assisted semantic communication, so as to at least solve the technical problem that the traditional optimization algorithm in the prior art cannot solve the problem of maximizing semantic similarity, thereby failing to improve the quality of the semantic communication process.

[0007] According to one aspect of an embodiment of the present disclosure, a resource optimization method for using drone-assisted semantic communication is provided, including: determining an original text and generating corresponding semantic information based on the original text; restoring the semantic information and generating a restored text; determining a semantic similarity maximization problem based on the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem; creating an initial strategy and updating the parameters of the optimized initial strategy using a standard gradient ascent method; determining the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters; and determining the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem based on the determined relationship, thereby maximizing the semantic similarity between the original text and the restored text.

[0008] According to another aspect of an embodiment of the present disclosure, a storage medium is further provided, the storage medium including a stored program, wherein when the program is running, a processor executes any one of the methods described above.

[0009] According to another aspect of the embodiments of the present disclosure, there is also provided a resource optimization device for using drone-assisted semantic communication, including: a semantic information generation module, used to determine the original text and generate corresponding semantic information based on the original text; a restored text generation module, used to restore the semantic information and generate a restored text; a maximization problem determination module, used to determine the semantic similarity maximization problem based on the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem; an updating module, used to create an initial strategy and update the parameters of the optimized initial strategy using a standard gradient ascent method; a relationship determination module, used to determine the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters; and a solving module, used to determine the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem based on the determined relationship, so as to maximize the semantic similarity between the original text and the restored text.

[0010] The present application discloses a resource optimization method using drone-assisted semantic communication. First, the processor determines the original text and generates corresponding semantic information based on the original text. Then, the processor restores the semantic information and generates a restored text. Further, the processor determines the semantic similarity maximization problem based on the original text and the restored text. After that, the processor creates an initial strategy and updates the pre-created initial strategy parameters using the standard gradient ascent method. Then, the processor determines the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters. Finally, the processor determines the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission optimization strategy based on the determined relationship, thereby maximizing the semantic similarity between the original text and the restored text.

[0011] Since the application determines the semantic similarity maximization problem based on the original text and the restored text, the semantic similarity maximization problem can be further divided into the resource allocation optimization problem and the semantic information transmission strategy optimization problem. Among them, the resource allocation optimization problem is related to the uplink channel capacity allocated by the drone to the terminal device and the uplink channel capacity allocated by the drone to the base station; the semantic information transmission strategy optimization problem is related to the selection of semantic information.

[0012] Furthermore, since the present application creates an initial strategy, uses the standard gradient ascent method to update the parameters of the optimized initial strategy, and uses the updated parameters to determine the relationship between the importance distribution of semantic information and the semantic similarity, it is possible to determine the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem based on the determined relationship. Thus, the semantic similarity between the original text and the restored text can be maximized.

[0013] Thus, the technical effect of maximizing the semantic similarity between the original text and the restored text, thereby improving the quality of the semantic communication process, is achieved. This solves the technical problem that the traditional optimization algorithm in the prior art cannot solve the problem of maximizing the semantic similarity, thereby failing to improve the quality of the semantic communication process. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present application. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0015] Figure 1 is a hardware structure block diagram of a computing device for implementing the method according to Embodiment 1 of the present disclosure;

[0016] Figure 2 is a schematic diagram of a system model of semantic communication according to Embodiment 1 of the present disclosure;

[0017] Figure 3 is a flow chart of a resource optimization method using drone-assisted semantic communication according to the first aspect of Embodiment 1 of the present disclosure; and

[0018] Figure 4 It is a schematic diagram of a resource optimization device using drone-assisted semantic communication according to the first aspect of Example 2 of the present disclosure. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only embodiments of a part of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0021] Example 1

[0022] According to this embodiment, a method embodiment of resource optimization using drone-assisted semantic communication 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.

[0023] The method embodiment provided in this embodiment can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computing device for implementing a resource optimization method using drone-assisted semantic communication. Figure 1As shown, the computing device may include one or more processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, the transmission device, and the input / output interface are connected to the processor via a bus. In addition, it may also include: a display, a keyboard, and a cursor control device connected to the input / output interface. A person skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0024] It should be noted that the one or more processors 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. In addition, 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 computing device. As involved in the embodiments of the present disclosure, the data processing circuitry acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0025] The memory can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the resource optimization method using drone-assisted semantic communication in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the resource optimization method using drone-assisted semantic communication of the above-mentioned application is realized. The memory 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 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computing device 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 combinations thereof.

[0026] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computing device. In one example, the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet wirelessly.

[0027] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of the computing device.

[0028] It should be noted that, in some optional embodiments, the above Figure 1 The computing device shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computing devices described above.

[0029] Figure 2 is a schematic diagram of a system model of semantic communication according to this embodiment. Figure 2 As shown, the system model includes: N users, M drones and base stations. Among them, each of the N users is equipped with a corresponding terminal device.

[0030] User i can use the corresponding terminal device to read the original text L i Extract data and model the extracted data through knowledge graph to obtain the original text L i The corresponding semantic information G i In addition, user i can also use the corresponding terminal device to send semantic information G i Send to drone j.

[0031] Drone j is used to receive the semantic information G i Send to base station k.

[0032] Base station k is used to receive the semantic information G i Restore to generate the recovery text L i (a i ,G).

[0033] It should be noted that the terminal devices, drones j and base stations k in the system can all be applied to the hardware structure described above.

[0034] In the above operating environment, according to the first aspect of this embodiment, a resource optimization method using drone-assisted semantic communication is provided. The method comprises: Figure 2 The processor implementation shown in . Figure 3 A schematic diagram showing the process of the method is shown in FIG. Figure 3 As shown, the method includes:

[0035] S302: determining the original text, and generating corresponding semantic information according to the original text;

[0036] S304: restoring the semantic information and generating a restored text;

[0037] S306: determining a semantic similarity maximization problem according to the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem;

[0038] S308: Create an initial strategy, and use the standard gradient ascent method to update the parameters of the optimized initial strategy;

[0039] S310: Determine the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters; and

[0040] S312: According to the determined relationship, determine the optimal solution of the resource allocation optimization problem and the optimal solution of the semantic information transmission strategy optimization problem, so as to maximize the semantic similarity between the original text and the restored text. Figure 2 As shown, first, the processor in the terminal device determines the original text and generates corresponding semantic information according to the original text (S302). Specifically, the processor in the terminal device may receive the original text L sent by the user. i Then, for the original text L i Extract data and model the extracted data through knowledge graph to obtain semantic information G i Among them, semantic information in, And H i is G′ i The number of semantic triples in G. i The constraints should be met:

[0041]

[0042] Among them, c i (t) represents the semantic information G transmitted by drone j to user i i Uplink channel capacity. j(t) represents the number of semantic information G that UAV j assigns to base station k i Uplink channel capacity. Z(G′ i )R represents the transmitted semantic information G i The amount of data.

[0043] The semantic information G extracted by the processor in the terminal device i The amount of data will be much smaller than the original text L i This is because the semantic information G i The two-token relationship between entity pairs in the original text L can be reduced i Redundant context in .

[0044] Then, user i uses the processor in the terminal device to convert the determined semantic information G i The processor in drone j then transmits the semantic information G i Transmitted to base station k.

[0045] Finally, the base station k recovers the semantic information and generates a recovered text (S304). Specifically, the processor in the base station k recovers the semantic information G received by the text generation model. i In the above example, we recover the coherent multi-sentence text. The recovered text is And among them, w′ i,m Represents the restored sentence text.

[0046] Then, the processor in the drone k determines the semantic similarity maximization problem based on the original text and the restored text (S306). The semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text. The semantic accuracy between the original text and the restored text is:

[0047]

[0048] Among them, σ(L′ i (a i (t),G′ i ),w′ i,m ) represents the restored text L′ i (α i (t),G′ i ), the restored sentence text w′ i,m The number of occurrences; σ(L i ,w′ i,m ) indicates that in the original text L i In the example, the restored sentence text w′ i,m Number of occurrences.

[0049] The semantic integrity between the original text and the restored text is:

[0050]

[0051] Among them, σ(L′ i (a i (t),G′ i ),w′ i,m ) represents the restored text L′ i (α i (t),G′ i ), the restored sentence text w′ i,m The number of occurrences, σ(L i ,w′ i,m ) indicates that in the original text L i In the example, the restored sentence text w′ i,m Number of occurrences.

[0052] Based on the above formula 2 and formula 3, the text L′ is restored i (α i (t),G′ i ) The semantic similarity can be expressed as:

[0053]

[0054] Among them, φ∈(0,1) is a parameter used to adjust the contribution of semantic accuracy and semantic completeness to semantic similarity. That is, φ is used to balance the two. Increasing values ​​will increase the impact of semantic accuracy on semantic similarity. That is, when φ increases, semantic accuracy becomes more important for semantic similarity. θ i is an additional penalty for short texts, which can be expressed as follows:

[0055]

[0056] Among them, when the restored text is greater than or equal to the original text, it means that the restored text is relatively complete, and θ i =1; when the recovered text is smaller than the original text, it means that the recovered text is incomplete.

[0057] The goal of this solution is to maximize the semantic similarity between the original text and the restored text while meeting the transmission delay requirements. This maximization problem includes the resource allocation optimization problem and the semantic information transmission strategy optimization problem. The semantic similarity maximization problem can be expressed by the following formula:

[0058]

[0059] sta i,q (t)∈{0,1},i∈U,q∈Q,t∈T (Formula 6b)

[0060]

[0061]

[0062]

[0063] ||q(t+1)-q(t)|| 2 ≤(ν max Δ T ) 2 (Formula 6f)

[0064]

[0065] Among them, formula 6b, formula 6c and formula 6d ensure that each user i can only occupy one uplink orthogonal resource block, and each uplink orthogonal resource block can only be allocated to one user i for semantic information transmission. 6e is the delay requirement for semantic information transmission. Formula 6f represents the constraint that the movement of drone j in adjacent time slots does not exceed the maximum movement distance. Formula 6g represents the distance constraint between two drones. It can be seen from formula 6a that the semantic similarity depends on the selected semantic information subset G′ i and uplink orthogonal resource block allocation a i Among them, a i represents the association variable between the user and the subchannel. If the user is associated with a subchannel, then a i =1; if the user is not associated with the subchannel, then a i = 0. In addition, the operation steps for determining the problem of maximizing the semantic similarity will be described in detail later, so they will not be repeated here.

[0066] However, since formula 6a is non-convex and depends on the text generation model used to restore the text, formula 6a cannot be solved by traditional optimization algorithms. To solve this problem, this application will use a reinforcement learning algorithm with an attention network to calculate the semantic similarity between the original text and the restored text.

[0067] Thus, to solve the optimization problem in Formula 6a, the UAV as an intelligent agent first creates an initial strategy π θ . And using the initial strategy π θ Sampling semantic communication actions to generate action sets The drone then defines the expected reward corresponding to the set of actions and maximize the expected reward (Right now, ), thereby generating an optimized initial policy. Then, the optimized initial policy π can be updated using the standard gradient ascent method θThe parameter θ (S308) is calculated as follows:

[0068]

[0069] Among them, α is the learning rate, is the gradient of the parameter θ.

[0070] Further, the drone determines the relationship between the importance distribution of the semantic information and the semantic similarity using the updated strategy parameters (S308), wherein the semantic information corresponds to the original text. The above content will be described in detail later, so it will not be repeated here.

[0071] Finally, the drone determines the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transfer strategy optimization problem based on the determined relationship, thereby maximizing the semantic similarity between the original text and the restored text (S310).

[0072] As described in the background technology, in existing communication methods, people often use drones as intermediaries to connect semantic data transmission between users and base stations. That is, the semantic data transmission process is user (terminal device)-drone-base station. In other words, the terminal device encodes the original text data to generate corresponding semantic data, and transmits the semantic data to the drone. The drone acts as a transit device to transmit the received semantic data to the base station. The base station then decodes the semantic data and generates a restored text. In this process, the semantic similarity between the restored text generated by the base station and the original text is the key to measuring the quality of the semantic communication process, so it is necessary to increase the semantic similarity as much as possible (that is, maximize the semantic similarity). Among them, semantic similarity includes semantic accuracy and semantic completeness.

[0073] However, the existing optimization algorithms cannot solve the problem of maximizing semantic similarity, and thus cannot improve the quality of the semantic communication process.

[0074] Since the application determines the semantic similarity maximization problem based on the original text and the restored text, the semantic similarity maximization problem can be further divided into the resource allocation optimization problem and the semantic information transmission strategy optimization problem. Among them, the resource allocation optimization problem is related to the uplink channel capacity allocated by the drone to the terminal device and the uplink channel capacity allocated by the drone to the base station; the semantic information transmission strategy optimization problem is related to the selection of semantic information.

[0075] Furthermore, since the present application creates an initial strategy, uses the standard gradient ascent method to update the parameters of the optimized initial strategy, and uses the updated parameters to determine the relationship between the importance distribution of semantic information and the semantic similarity, it is possible to determine the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem based on the determined relationship. Thus, the semantic similarity between the original text and the restored text can be maximized.

[0076] Thus, the technical effect of maximizing the semantic similarity between the original text and the restored text, thereby improving the quality of the semantic communication process, is achieved. This solves the technical problem that the traditional optimization algorithm in the prior art cannot solve the problem of maximizing the semantic similarity, thereby failing to improve the quality of the semantic communication process.

[0077] Optionally, the operation of determining the original text and generating corresponding semantic information based on the original text includes: extracting data from the original text using a terminal device, and modeling the extracted data using a knowledge graph to obtain semantic information composed of multiple semantic triples. Further optionally, the operation of restoring the semantic information and generating a restored text includes: restoring the semantic information using a text generation model pre-set by a base station, and generating a restored text.

[0078] Specifically, first, the processor builds a semantic communication model for semantic information transmission based on the terminal device, the drone and the base station. The semantic communication model includes a system model, a perception model, a queuing model and a semantic model.

[0079] Creation of system model: The system model includes N users, M drones, cellular base stations, and edge servers, such as Figure 2 The system model is a multi-time slot system, and each time slot is represented by t.

[0080] Creation of perception model: the perception duration of the i-th user in the perception model at the t-th time slot It is expressed as:

[0081]

[0082] in, represents the perceived duration of user i in the tth time slot, Qi(t) represents the amount of data that can be processed in the future, and o i Represents the rate at which sensory data is generated.

[0083] The energy consumed by user i for data perception is It can be expressed by the following formula:

[0084]

[0085] in, represents the energy consumed by user i for data perception, Qi(t) represents the amount of data that can be processed in the future, and e i Indicates the perceived consumption per bit.

[0086] Creation of queuing model: The queuing model considers a multi-task model. The amount of data Qi(t) that can be processed in the future will be cached at user i and then migrated to drone j. The cache queue state at the user end is expressed as:

[0087]

[0088] Among them, Q 1 (t) represents the data that the first user terminal has not yet processed, ..., Q N (t) represents the data that the Nth user terminal has not had time to process.

[0089] Creation of semantic model: Using w i,n Represents words, symbols, or punctuation marks in text data, so it is called w i,n is a token. The text data that user i needs to transmit to base station k consists of a series of tokens, as shown below:

[0090]

[0091] Where V is the vocabulary, N i is the original text L i The number of symbols in .

[0092] Then, the processor in the terminal device processes the original text L i Extract data and model the extracted data through knowledge graph to obtain semantic information G i The amount of data extracted from the semantic information will be much smaller than the original text L i This is because the semantic information G i The two-token relationship between entity pairs in the original text L can be reduced i Redundant context in .

[0093] Furthermore, a transmission model is created. First, the coordinates of user i, drone j, and base station k are expressed in a three-dimensional coordinate system. The coordinates of user i are q i =[q i,x ,q i,y ], the coordinates of UAV j are And the flight height H of drone j remains unchanged. The coordinates of base station k are q 0 =[x 0 ,y 0 ,H 0 ]. Among them, H0 represents the height of base station k relative to user i. The channel gain between the i-th user and the j-th drone is:

[0094]

[0095] The channel gain between the jth UAV and base station k is:

[0096]

[0097] Among them, α = 2, the distance between the i-th user and the j-th drone is d i,j (t), the distance d between the jth UAV and base station k m,0 (t). β 0 is the channel gain per unit distance.

[0098] As mentioned above, drone j will use OFDMA technology to allocate Q uplink orthogonal resource blocks to user i. Assuming that each user i can only occupy one uplink orthogonal resource block, and each uplink orthogonal resource block can only be allocated to one user. Then user i transmits semantic information G i The uplink channel capacity is:

[0099]

[0100] Where W represents the bandwidth of each uplink orthogonal resource block. i,j (t) indicates that user i transfers semantic information G i The transmission power transmitted to UAV j. q Represents the resource block P allocated to the user i,j (t) interference. 0 is the noise power spectral density. And it is assumed that the delay for each user i to transmit data to drone j is limited to T.

[0101] Semantic information G transmitted by drone j to base station k i The uplink channel capacity is:

[0102]

[0103] Among them, P j,0 (t) indicates that drone j sends semantic information G i The transmission power transmitted to base station k. j,0 (t) represents the channel gain between UAV j and base station k.

[0104] Finally, base station k uses a preset text generation model to restore semantic information and generate restored text.

[0105] Optionally, the operation of determining the problem of maximizing semantic similarity based on the original text and the restored text includes: determining the semantic accuracy between the original text and the restored text; determining the semantic integrity between the original text and the restored text; and determining the problem of maximizing semantic similarity based on the semantic accuracy and the semantic integrity. Further optionally, the operation of determining the problem of maximizing semantic similarity based on the semantic accuracy and the semantic integrity includes: determining the semantic similarity between the original text and the restored text based on the semantic accuracy and the semantic integrity; determining the constraint conditions corresponding to the semantic similarity; and maximizing the semantic similarity when the constraint conditions are met, and determining the problem of maximizing semantic similarity.

[0106] Specifically, the semantic accuracy between the original text and the restored text is:

[0107]

[0108] Among them, σ(L′ i (a i (t),G′ i ),w′ i,m ) represents the restored text L′ i (α i (t),G′ i ), the restored sentence text w′ i,m The number of occurrences; σ(L i ,w′ i,m ) indicates that in the original text L i In the example, the restored sentence text w′ i,m Number of occurrences.

[0109] The semantic integrity between the original text and the restored text is:

[0110]

[0111] Among them, σ(L′ i (a i (t),G′ i ),w′ i,m ) represents the restored text L′ i (α i (t),G′ i ), the restored sentence text w′ i,m The number of occurrences, σ(L i ,w′ i,m ) indicates that in the original text L i In the example, the recovered sentence text w′ i,m Number of occurrences.

[0112] Based on the above formula 2 and formula 3, the text L′ is restored i(α i (t),G′ i ) The semantic similarity can be expressed as:

[0113]

[0114] Among them, φ∈(0,1) is a parameter used to adjust the contribution of semantic accuracy and semantic completeness to semantic similarity. That is, φ is used to balance the two. Increasing values ​​will increase the impact of semantic accuracy on semantic similarity. That is, when φ increases, semantic accuracy becomes more important for semantic similarity. θ i is an additional penalty for short texts, which can be expressed as follows:

[0115]

[0116] Among them, when the restored text is greater than or equal to the original text, it means that the restored text is relatively complete, and θ i =1; when the recovered text is smaller than the original text, it means that the recovered text is incomplete.

[0117] Further, determine the constraints corresponding to the semantic similarity. Specifically, referring to the above, this embodiment assumes that each user i can only occupy one uplink orthogonal resource block, and each uplink orthogonal resource block can only be allocated to one user i for semantic information transmission, such as the above formula 6b, formula 6c and formula 6d. The delay of semantic information transmission should be less than the given transmission delay threshold T, such as the above formula 6e. The movement of the drone in adjacent time slots does not exceed the maximum movement distance, such as the above formula 6f. The distance between the two drones is greater than the given distance threshold, such as the above formula 6g.

[0118] Finally, under the condition of satisfying the above constraints, the semantic similarity is maximized, thereby determining the semantic similarity maximization problem.

[0119] Optionally, the operation of creating an initial strategy and updating the parameters of the optimized initial strategy using a standard gradient ascent method includes: creating an initial strategy; sampling actions related to semantic communication according to the initial strategy and generating an action set; defining an expected reward for the action set and maximizing the expected reward to determine an optimized initial strategy, wherein the expected reward corresponds to semantic similarity; and updating the parameters of the optimized initial strategy using a standard gradient ascent method.

[0120] Specifically, in order to solve the problem of maximizing semantic similarity, the drone first creates an initial strategy π θ , where θ is the parameter of the initial strategy. And the semantic communication actions are sampled to obtain the action set D. Among them, the action set To evaluate the strategy of maximizing semantic similarity, the expected reward of the action in D is defined as:

[0121]

[0122] Among them, s represents the action state, a d represents the action set, π θ Represents the optimized initial strategy, and the reward for action state s to select action a is R. In addition, this formula is equivalent to the problem of maximizing semantic similarity.

[0123] Optimized initial strategy π θ The goal is to maximize the semantic similarity between the original text and the restored text. That is,

[0124] Then, the optimized initial policy π can be updated using the standard gradient ascent method θ The parameter θ in the formula is:

[0125]

[0126] in, represents the learning rate, represents the gradient of the parameter θ.

[0127] By iterating the above update steps, the optimized initial strategy π θ The parameter θ in can find the relationship between the importance distribution of all semantic information and the semantic similarity. Thus, the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem can be obtained, thereby maximizing the semantic similarity between the original text and the restored text.

[0128] Optionally, the operation of determining the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters includes: predetermining the importance distribution of semantic information; and determining the relationship between the importance distribution of semantic information and the semantic similarity based on the importance distribution of semantic information. Further optionally, the operation of predetermining the importance distribution of semantic information includes: constructing an importance vector, wherein the importance vector is used to represent the importance distribution of semantic triples in the semantic information; marking the sentence text in the original text and marking the semantic triples; determining a first importance for indicating the correlation between the semantic triple and the marked sentence text; determining a second importance for indicating the correlation between the semantic triple and the original text; determining the importance vector based on the first importance and the second importance; and determining the importance distribution of semantic information based on the importance vector.

[0129] Specifically, first, create an importance vector. The importance vector f i (Gi ) is used to represent the importance distribution of semantic triplets in semantic information.

[0130] Then, the base station processes the original text The sentence text in w i,n Marking and semantic information Semantic triples in Then, define the The vector is Used to mark w i,n The vector is in, Representing semantic triples The sentence text itself. x Represents the dimension of each token vector.

[0131] Semantic triples With the original text L i Mark w i,n The correlation between them can be expressed as:

[0132]

[0133] in, is a semantic triple Using the trained parameters, the attention network can calculate (W tri x i,b g ) T (W tok x i,n ). And among them, (W tri x i,b g ) T (W tok x i,n ) represents a semantic triple Medium Mark With the original text L i Mark w i,n The correlation between them.

[0134] Defining semantic triples With the original text L i The importance of the correlation between can be expressed by the following formula:

[0135]

[0136] Semantic information G i The importance distribution of can be expressed by the following formula:

[0137]

[0138] in,

[0139] Therefore, the importance evaluation is performed using a policy gradient-based reinforcement learning algorithm, which enables the base station to analyze the relationship between the importance distribution and the measure of semantic similarity, so as to achieve the purpose of optimizing semantic information selection (i.e., semantic information transmission strategy optimization) and resource block allocation (i.e., resource allocation optimization).

[0140] In addition, it is worth noting that the policy gradient-based reinforcement learning algorithm consists of six parts: 1) agent, 2) environment, 3) action, 4) state, 5) strategy, and 6) reward. The specific process is as follows:

[0141] Input: The original text L of each user i i and transmission delay threshold T.

[0142] Initialization: Randomly generate the initial strategy parameters θ, the interference I of each channel q , task learning rate δ, number of iterations E

[0143] 1: Calculate the importance distribution f(G according to the above importance distribution formula i )

[0144] 2: Fori=1→Edo

[0145] 3: Using the initial strategy π θ The action trajectory of mobile drone j α = [α 1 ,...,α U ]

[0146] 4: Update the parameters θ of the initial policy πθ using the standard gradient ascent method (iteratively run step 4)

[0147] 5: endfor

[0148] Therefore, according to the first aspect of this embodiment, the semantic similarity between the original text and the restored text can be maximized, thereby achieving the technical effect of improving the quality of the semantic communication process.

[0149] In addition, reference Figure 1 As shown, according to the second aspect of this embodiment, a storage medium is provided, wherein the storage medium includes a stored program, wherein when the program is run, a processor executes any one of the above methods.

[0150] Therefore, according to this embodiment, the semantic similarity between the original text and the restored text can be maximized, thereby achieving the technical effect of improving the quality of the semantic communication process.

[0151] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0152] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0153] Example 2

[0154] Figure 4 The resource optimization device 400 using drone-assisted semantic communication according to the first aspect of this embodiment is shown, and the device 400 corresponds to the method according to the first aspect of embodiment 1. Figure 4 As shown, the device 400 includes: a semantic information generation module 410, which is used to determine the original text and generate corresponding semantic information based on the original text; a restored text generation module 420, which is used to restore the semantic information and generate a restored text; a maximization problem determination module 430, which is used to determine the semantic similarity maximization problem based on the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem; an updating module 440, which is used to create an initial strategy and update the parameters of the optimized initial strategy using a standard gradient ascent method; a relationship determination module 450, which is used to determine the relationship between the importance distribution of semantic information and the semantic similarity using the updated parameters; and a solution module 460, which is used to determine the optimal solution to the resource allocation optimization problem and the optimal solution to the semantic information transmission strategy optimization problem based on the determined relationship, so as to maximize the semantic similarity between the original text and the restored text.

[0155] Optionally, the semantic information generation module 410 includes: a semantic information generation sub-module, which is used to extract data from the original text using a terminal device, and model the extracted data through a knowledge graph to obtain semantic information composed of multiple semantic triples.

[0156] Optionally, the recovery text generation module 420 includes: the recovery text generation module 420 is used to use the text generation model preset by the base station to restore semantic information and generate recovery text.

[0157] Optionally, the maximization problem determination module 430 includes: a semantic accuracy determination module, used to determine the semantic accuracy between the original text and the restored text; a semantic integrity determination module, used to determine the semantic integrity between the original text and the restored text; and a maximization problem determination submodule, which determines the semantic similarity maximization problem based on the semantic accuracy and semantic integrity.

[0158] Optionally, the maximization problem determination submodule includes: a semantic similarity determination module, used to determine the semantic similarity between the original text and the restored text based on semantic accuracy and semantic completeness; a constraint determination module, used to determine the constraint corresponding to the semantic similarity; and a first determination module, used to maximize the semantic similarity while satisfying the constraint, and determine the semantic similarity maximization problem.

[0159] Optionally, the update module 440 includes: a creation module for creating an initial strategy; a generation module for sampling actions related to semantic communication according to the initial strategy and generating an action set; an expected reward definition module for defining the expected reward of the action set and maximizing the expected reward to determine the optimized initial strategy, wherein the expected reward corresponds to the semantic similarity; and an update submodule for updating the parameters of the optimized initial strategy using a standard gradient ascent method.

[0160] Optionally, the relationship determination module 450 includes: a second determination module for predetermining the importance distribution of the semantic information; and a second determination module for determining the relationship between the importance distribution of the semantic information and the semantic similarity according to the importance distribution of the semantic information.

[0161] Optionally, the second determination module includes: a construction module for constructing an importance vector, wherein the importance vector is used to represent the importance distribution of semantic triples in the semantic information; a marking module for marking the sentence text in the original text and marking the semantic triples; a third determination module for determining a first importance for indicating the correlation between the semantic triple and the marked sentence text; a fourth determination module for determining a second importance for indicating the correlation between the semantic triple and the original text; a fifth determination module for determining the importance vector based on the first importance and the second importance; and a sixth determination module for determining the importance distribution of the semantic information based on the importance vector.

[0162] Therefore, according to this embodiment, the semantic similarity between the original text and the restored text can be maximized, thereby achieving the technical effect of improving the quality of the semantic communication process.

[0163] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0164] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0165] 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 schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, 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.

[0166] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0167] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0168] 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 invention, in essence, or the part that contributes to the prior art, 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 a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0169] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A resource optimization method using drone-assisted semantic communication, characterized in that: include: Determining an original text, and generating corresponding semantic information according to the original text; Restoring the semantic information and generating a restored text; Determine a semantic similarity maximization problem according to the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem; Create an initial strategy and use the standard gradient ascent method to update the parameters of the optimized initial strategy; Determining the relationship between the importance distribution of the semantic information and the semantic similarity using the updated parameters; as well as According to the determined relationship, an optimal solution to the resource allocation optimization problem and an optimal solution to the semantic information transfer strategy optimization problem are determined, so as to maximize the semantic similarity between the original text and the restored text.

2. The method according to claim 1, characterized in that The operation of determining the original text and generating corresponding semantic information according to the original text includes: The original text is extracted using a terminal device, and the extracted data is modeled through a knowledge graph to obtain semantic information consisting of multiple semantic triples.

3. The method according to claim 1, characterized in that The operation of restoring the semantic information and generating a restored text includes: The semantic information is restored and the restored text is generated by using a text generation model preset by the base station.

4. The method according to claim 1, characterized in that: Determining an operation of maximizing the semantic similarity problem according to the original text and the restored text includes: determining semantic accuracy between the original text and the restored text; determining semantic integrity between the original text and the restored text; and Based on the semantic accuracy and the semantic completeness, the semantic similarity maximization problem is determined.

5. The method according to claim 3, characterized in that: Based on the semantic accuracy and the semantic completeness, determining the operation of maximizing the semantic similarity problem includes: Determining the semantic similarity between the original text and the restored text based on the semantic accuracy and the semantic completeness; determining constraints corresponding to the semantic similarity; and Under the condition that the constraint condition is satisfied, the semantic similarity is maximized, and the semantic similarity maximization problem is determined.

6. The method according to claim 1, characterized in that The operation of creating an initial strategy and updating the parameters of the optimized initial strategy using the standard gradient ascent method includes: Create an initial strategy; Sampling actions related to semantic communication according to the initial strategy and generating an action set; defining an expected reward for the action set and maximizing the expected reward to thereby determine an optimized initial strategy, wherein the expected reward corresponds to the semantic similarity; and The parameters of the optimized initial policy are updated using the standard gradient ascent method.

7. The method according to claim 6, characterized in that The operation of determining the relationship between the importance distribution of the semantic information and the semantic similarity using the updated parameters includes: predetermining the importance distribution of the semantic information; and According to the importance distribution of the semantic information, a relationship between the importance distribution of the semantic information and the semantic similarity is determined.

8. The method according to claim 7, characterized in that The operation of predetermining the importance distribution of the semantic information includes: Constructing an importance vector, wherein the importance vector is used to represent the importance distribution of the semantic triples in the semantic information; Marking the sentence text in the original text and marking the semantic triples; determining a first importance indicating a relevance between the semantic triple and the labeled sentence text; determining a second importance indicating a relevance between the semantic triple and the original text; Determining the importance vector according to the first importance and the second importance; Based on the importance vector, an importance distribution of the semantic information is determined.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the processor executes the method according to any one of claims 1 to 8.

10. A resource optimization device using drone-assisted semantic communication, characterized in that: include: A semantic information generation module, used to determine the original text and generate corresponding semantic information according to the original text; A recovery text generation module, used to recover the semantic information and generate a recovery text; A maximization problem determination module, used to determine a semantic similarity maximization problem based on the original text and the restored text, wherein the semantic similarity is used to indicate the semantic accuracy and semantic completeness between the original text and the restored text, and the semantic similarity maximization problem includes a resource allocation optimization problem and a semantic information transmission strategy optimization problem; The update module is used to create the initial policy and update the parameters of the optimized initial policy using the standard gradient ascent method; A relationship determination module, used to determine the relationship between the importance distribution of the semantic information and the semantic similarity using the updated parameters; as well as A solution module is used to determine the optimal solution of the resource allocation optimization problem and the optimal solution of the semantic information transmission strategy optimization problem according to the determined relationship, so as to maximize the semantic similarity between the original text and the restored text.

Images