Optimization method and server therefor

Abstract

An optimization method and server thereof. The optimization method includes generating a constrained causal graph from observation data from a distribution unit, wherein a number of causal variables of the constrained causal graph and a causal structure of the constrained causal graph are determined together; performing a finite-domain representation planning using the constrained causal graph to generate action data on optimized radio unit parameters; and outputting the action data to the distribution unit. In order to avoid the inefficiency and high error rate of manually setting radio unit parameters, the optimization method and server thereof provided by the present application can infer the current state and plan the action to be taken to set the radio unit parameters in the best way, instead of manually setting the radio unit parameters.

Description

Technical Field

[0001] This invention relates to an optimization method and its server, and more particularly to an optimization method and its server that can improve the efficiency and accuracy of setting radio unit parameters. Background Technology

[0002] Different vendors may implement radio units and distribution units in different ways, making integration between radio units and distribution units from different vendors difficult. In 5G user scenarios, when connecting a radio unit from one vendor to a distribution unit from another vendor, the radio unit parameters listed in the Working Group 4 (WG4) interoperability testing (IOT) must be manually set to the distribution unit, a process that is very time-consuming and error-prone.

[0003] For example, when the radio unit and distribution unit are from the same vendor, their signal transmission and reception times are synchronized. However, for radio units and distribution units from different vendors, to support timing coordination between the control plane (C-plane) and user plane (U-plane), the Open Radio Access Network (O-RAN) interface specifies that control plane information must arrive at the radio unit some time earlier than the latest possible arrival time of the corresponding first user plane information (e.g., according to Tcp_adv_dl listed in Table 1). Since the interface between the radio unit and distribution unit is highly sensitive to latency, the radio unit parameters must be correctly configured to achieve tight synchronization between them.

[0004] Therefore, the existing methods for setting radio unit parameters need to be improved in terms of efficiency and accuracy.

[0005] Therefore, an optimization method and its server are needed to solve the above problems. Summary of the Invention

[0006] Therefore, the present invention mainly provides an optimization method and its server to improve the efficiency and accuracy of setting radio unit parameters.

[0007] The present invention discloses an optimization method comprising generating a constrained causal graph based on observation data from a distributed unit, wherein a number of a plurality of causal variables of the constrained causal graph and a causal structure of the constrained causal graph are determined together; performing a finite-domain representation programming using the constrained causal graph to generate action data with respect to optimized parameters of a plurality of radio units; and outputting the action data to the distributed unit.

[0008] The present invention discloses a server comprising a storage circuit for storing an instruction comprising generating a constrained causal graph based on observation data from a distributed unit, wherein a number of a plurality of causal variables of the constrained causal graph and a causal structure of the constrained causal graph are determined together; performing a finite-domain representation programming using the constrained causal graph to generate action data with respect to optimized parameters of a plurality of radio units; outputting the action data to the distributed unit; and a processing circuit coupled to the storage circuit for executing the instruction stored in the storage circuit.

[0009] To avoid the inefficiency and high error rate of manually setting radio unit parameters, the optimization method and server provided by this invention can infer the current state and plan the actions to be taken to set the radio unit parameters in the best way, instead of manually setting the radio unit parameters. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the communication system according to Embodiment 1 of the present invention.

[0011] Figure 2 This is a flowchart of the optimization method according to Embodiment 1 of the present invention.

[0012] Figure 3 This is a schematic diagram of the basic data and a partial cause-effect graph of Embodiment 1 of the present invention.

[0013] Figure 4 This is a schematic diagram of the state and local cause-effect graph of an embodiment of the present invention.

[0014] Figure 5 This is a schematic diagram of the planning tree in Embodiment 1 of the present invention.

[0015] Figure 6 This is a schematic diagram illustrating how a deep learning model is used to obtain a causal graph in an embodiment of the present invention.

[0016] Figure 7 This is a partial schematic diagram of the communication system according to Embodiment 1 of the present invention.

[0017] Explanation of key component symbols:

[0018] 10, 70 Communication Systems

[0019] 10b Observational Data

[0020] 10c Motion Data

[0021] 10DU and 70DU distributed units

[0022] 10RU radio unit

[0023] 10SVR, 70SVR servers

[0024] 110R Causal Reasoning Module

[0025] 112B Bayesian Network Module

[0026] 112SCM Structural Causal Model Module

[0027] 120P Causal Programming Module

[0028] 122FDR Finite Field Representation Programming Module

[0029] 20 Optimization Methods

[0030] 30g basic data

[0031] 50 Planning Tree

[0032] 722c controller

[0033] 722p Planner

[0034] 722pm Strategy Customization Module

[0035] 722s Scheduler

[0036] 782p Application Module

[0037] 782s System Module

[0038] a. Actions 1-3, 11-13

[0039] CG, CG(Π) Cause-and-effect diagram

[0040] cv (i-1) ,cv i ,cv (j-1) ,cv j Causal variables CV1 to CVn

[0041] e0 Initial state

[0042] Effects of eff_a(u) and eff_a(v)

[0043] f (i-1) f i f(j-1) f j Observation function

[0044] pre_a(u) prerequisites

[0045] S200~S208 Steps

[0046] sg target state

[0047] u, v, e1~e3, e11~e13 states

[0048] v1~vm attributes

[0049] w (i-1) w i w (j-1) w j data Detailed Implementation

[0050] Figure 1 This is a schematic diagram of a communication system 10 according to an embodiment of the present invention. The communication system 10 may include a server 10SVR, a distributed unit (DU) 10DU, and a radio unit (RU) 10RU.

[0051] In one embodiment, the distribution unit 10DU and the radio unit 10RU may be from different vendors. When connecting a radio unit 10RU from one vendor to a distribution unit 10DU from another vendor, to avoid the inefficiency and high error rate of manually setting radio unit parameters, the server 10SVR can utilize algorithms (e.g., Figure 2 The optimization method 20) selects the most likely / optimized radio unit parameters, thereby optimally setting the radio unit parameters to the distribution unit 10DU.

[0052] Please refer to Figure 2 , Figure 2 This is a flowchart of an optimization method 20 according to an embodiment of the present invention. The optimization method 20 can be compiled into program code and executed by a processing circuit, and stored in a storage circuit. The optimization method 20 may include the following steps:

[0053] Step S200: Begin.

[0054] Step S202: Generate a constrained causal graph based on observation data 10b from distribution unit 10DU, wherein a number of multiple causal variables and a causal structure of the constrained causal graph are determined together.

[0055] Step S204: Use a constrained cause-effect graph to perform finite field representation planning to generate action data 10c for the optimized multiple radio unit parameters.

[0056] Step S206: Output motion data 10c to distribution unit 10DU.

[0057] Step S208: End.

[0058] exist Figure 1 The server 10SVR may include a causal reasoning module 110R and a causal planning module 120P. Figure 2 Steps S202 to S206 can be executed by the causal programming module 120P.

[0059] In step S202, the causal programming module 120P receives observation data 10b from the distribution unit 10DU. Observation data 10b may be related to the description of the system or network performance. In one embodiment, observation data 10b may be a log file of the distribution unit 10DU. In another embodiment, the causal programming module 120P may convert observation data 10b into grounding data.

[0060] In step S202, the causal programming module 110P can obtain a constrained causal graph from the basic data. The causal programming module 110P can select an optimized causal model based on maximum a posteriori (MAP) and point estimates. Accordingly, the causal variables (or the number of causal variables) and causal structure of the constrained causal graph are determined simultaneously, thus avoiding the problems caused by determining the causal variables first and then the causal structure.

[0061] For example, Figure 3 This is a schematic diagram of basic data 30g and a local causal graph CG according to Embodiment 1 of the present invention, where (a) and (b) respectively illustrate two possibilities for the basic data 30g and the causal graph CG. The causal graph CG can be used as a constrained causal graph in optimization method 20. Figure 3 The causal structure of causal graph CG can provide causal variables cv (i-1) ,cv i ,cv (j-1) ,cv j The relationship between them, and the observation function f (i-1) f if (j-1) f j Then the basic data of 30g can be used as w (i-1) w i w (j-1) w j Mapping to causal variable cv (i-1) ,cv i ,cv (j-1) ,cv j And providing causal variable cv (i-1) ,cv i ,cv (j-1) ,cv j Compared to the base data of 30g, the data w (i-1) w i w (j-1) w j The relationship between causal variables (cv). (i-1) ,cv i ,cv (j-1) ,cv j It is defined based on the user scenario between the distribution unit 10DU and the radio unit 10RU. In one embodiment, the causal variable cv (i-1) ,cv i ,cv (j-1) ,cv j Can be manually defined. i and j are positive integers.

[0062] In step S202, the causal programming module 110P can convert the basic data 30g of data w i Assigned to the observation function f i And the posterior probability P(f) of the causal structure of the causal graph CG. i Maximize C|w) to obtain the data from the base data 30g. i Obtain the causal structure and its causal variable cv. i Therefore, a Bayesian network can be used in conjunction with the observation function f. (i-1) f i f (j-1) f j This describes the inference of a causal model. It's worth noting that the causal variable cv... (i-1) ,cv i ,cv (j-1) ,cv j And the causal structure is obtained together, therefore the causal variable cv (i-1) ,cv i ,cv (j-1) ,cv jAnd causal structures can influence and constrain each other.

[0063] In one embodiment, the posterior probability P(f) i ,C|w i According to Bayesian rule, P(f) can satisfy the condition. i ,C|w i ,Int)∝P(f i ,C)P(w i |f i (,C,Int), where f i C can represent the observation function, and w can represent the causal structure. i This can represent a portion of the baseline data (30g), where Int can represent intervention. In one embodiment, the posterior probability P(f) i ,C|w i It can be proportional to P(f) i ,C)P(w i |f i C) or Where s t-1 This can represent the state at time point t-1, where T can represent the current time point, and γ can be 0.5 but is not limited to this. In one embodiment, P(w|f i C) can be In one embodiment, P(w) i，t |s t-1 C, f i ) can be or Where s i,t It can represent a causal variable cv i The state at time point t, N can represent all causal variables (e.g., causal variable cv). (i-1) ,cv i ,cv (j-1) ,cv j The total number of items, where N is a positive integer, R St This can represent the basic data 30g and the causal variable cv i state s i Compatible data w i The amount of data. In one embodiment, the minimum amount of data R can be utilized. St To select the causal variable cv i This allows for the inclusion of frequently used data within the 30g base data set (such as data w). i (Compared to less frequently used data) are divided into smaller parts.

[0064] As can be seen from the above, the Bayesian probability mechanism can combine causal variables cv i Number of causal variables, CV i State, causal structure, and causal variable cv i The observation function f i And related joint inferences are obtained to explain the underlying data 30g, thus generating the causal graph CG. Among them, the causal variable cv in the causal graph CG... (i-1) ,cv i ,cv (j-1) ,cv j (or causal variable cv) (i-1) ,cv i ,cv (j-1) ,cv j The number of elements and the causal structure C are determined simultaneously, based on which the causal programming module 110P can distinguish... Figure 3 The difference between (a) and (b).

[0065] like Figure 3 As shown, each causal variable (e.g., causal variable cv) i This will correspond to an observation function (e.g., observation function f). i In one embodiment, a causal semantic generative (CSG) model can be used to obtain the observation function (e.g., the observation function f). i This allows for the prediction of low-dimensional state attributes (such as causal variables cv) from high-dimensional environmental variables (e.g., basic data 30g). i The attributes of the state. Furthermore, causal semantic generation models can avoid misclassifying variation factors as causal variables (e.g., causal variable cv). i The cause is the semantic factor, which can be correctly identified as a causal variable (e.g., causal variable cv). i The cause of causal semantics. In one embodiment, the causal semantic generation model is primarily based on the causal invariance principle and involves variational Bayes.

[0066] In one embodiment, observe the function f i It can satisfy s i,t =f i (w i,t In one embodiment, observe the function f. iThis can be achieved using a multivariate Gaussian distribution, for example, one that satisfies... Alternatively, observe the function f i Related to Where z is the causal variable cv in the basic data 30g. i Data that did not contribute, average μ v Fixed as a zero vector, Σ can be parameterized through Cholesky decomposition, and for example, it can satisfy Σ = LL. T Matrix L can be a lower-triangular matrix with positive diagonals and, for example, can be parameterized to satisfy... matrix L zz It can be a smaller lower triangular matrix, matrix It can be any matrix. L zz It can be parameterized using the sum of the diagonal elements (as determined by the exponential map) and the lower triangular matrix (which does not have diagonal elements).

[0067] In another embodiment, a deep learning model can be used to obtain the observation function (e.g., the observation function f). i This allows for the prediction of low-dimensional state attributes (such as causal variables cv) from high-dimensional environmental variables (e.g., basic data 30g). i (attributes of the state), however, deep learning models may mistakenly classify changing factors as causal variables (cv) in addition to semantic factors. i The reason.

[0068] like Figure 1 As shown, the causal programming module 120P may include a finite domain representation (FDR) programming module 122FDR.

[0069] In step S204, the finite domain representation planning module 122FDR can dynamically perform finite domain representation planning using a constrained causal graph (e.g., causal graph CG). In finite domain representation planning, preconditions define the conditions (states) under which an action can be performed. Actions can change states and produce new states (i.e., cause effects), ultimately achieving a goal state (e.g., goal state sg). The solution of the finite domain representation planning is a plan in the search space from the initial state (e.g., initial state e0) to the goal state. In one embodiment, the solution of the finite domain representation planning can be a directed graph.

[0070] In one embodiment, the finite field representation of the planning task can be Π = (V, A, c, I, G), where V is a finite set of state variables, A is a finite set of actions, c is the loss function, I is the initial state, and G is the target state. For example, Figure 4 This is a schematic diagram of states u and v and a local causal graph CG(Π) according to an embodiment of the present invention, where (a) and (b) illustrate two possibilities of the causal graph CG(Π). If state u is not equal to state v and there exists an action a satisfying a∈A such that "there exists a∈A such that both the prerequisite pre_a(u) and the effect eff_a(v) are defined" or "there exists a∈A such that both the effect eff_a(u) and the effect eff_a(v) are defined", then the causal graph CG(Π) representing the planning task Π in a finite field can be a directed graph with nodes v and paths (arc), and there can be a path (u,v) between states u and v. Here, the prerequisite pre_a(u) refers to the state where node u is a prerequisite given action a, and the effect eff_a(v) refers to the state where node v is an effect given action a. In other words, in the present invention, the causal graph representing the planning task Π in a finite field is a constrained causal graph. Constrained cause-effect graphs can be optimized according to user scenarios to eliminate other possibilities in the search space, thereby improving efficiency, reducing computational load, and lowering power consumption.

[0071] In one embodiment, in step S204, after constructing the constrained causal graph using the algorithm, the constrained causal graph can be converted into a domain file (i.e., a description of the system) of the Planning Domain Description Library (PDDL) for finite domain representation planning by the finite domain representation planning module 122FDR. In other words, the domain file of the present invention does not need to be manually defined.

[0072] In one embodiment, a domain file can be used to describe actions, and a problem file can be used to describe the initial state and the target state. In one embodiment, the content of the domain file may include the following:

[0073] precond_1->action_1->effect_1

[0074] precond_2->action_2->effect_2 ...

[0076] precond_n->action_n->effect_n

[0077] Here, states precond_1 to precond_n can serve as prerequisites, action_1 to action_n can be actions, and states effect_1 to effect_n can be effects, where n is a positive integer. After executing action_i, state precond_i can enter state effect_i. In one embodiment, precond_i->action_i->effect_i can correspond to a portion of a causal graph (e.g., causal graph CG or CG(Π)), for example, in... Figure 4 (a) The causal graph CG(Π) is drawn such that the state u as the cause corresponds to the precondition precond_i, and the state v as the effect corresponds to the effect_i. In other words, the causal graph can represent the structure of states and actions.

[0078] In one embodiment, effect_1 may be a cause of state precond_3, and effect_3 may be a cause of state precond_7. Accordingly, the solution of the finite field representation programming may at least include a sequence of actions action_1, action_3, and action_7, but not, for example, action_2.

[0079] For example, Figure 5 This is a schematic diagram of a planning tree 50 according to Embodiment 1 of the present invention. When the planner of the finite domain representation planning module 122FDR reads the domain file, the planner can generate a planning tree 50 corresponding to the domain file (or a constrained cause-effect graph). The path of the planning tree 50 (i.e. Figure 5The arrows represent actions, and nodes represent states, each state being represented by a set of state variables. Paths can show the dependencies between states. For action act1, the initial state e0 can be used as a prerequisite, and state e1 can be used as an effect. For action act11, the effect of action act1 (i.e., state e1) can be used as a prerequisite, and state e11 can be used as an effect. The programming tree 50 can serve as the search space. The solution to the finite field representation programming module 122FDR can be an ordered sequence of actions in the programming tree 50 that begins with the initial state e0 and ends with the target state sg. For example, it can be determined that the branch for action act1 is not a solution to the finite field representation programming module 122FDR, while the branches for actions act2 and act3 are solutions to the finite field representation programming module 122FDR, but this is not the only possibility.

[0080] In one embodiment, certain actions may be reversible, such that a past state (e.g., state e11) can be reached from a present state (e.g., state e1). In one embodiment, a loop may be formed between two adjacent states (e.g., states e1 and e11). In one embodiment, certain states (e.g., state e11) may be nondeterministic states.

[0081] In one embodiment, the planner may use a search algorithm to find a solution to the finite field representation planning module 122FDR, such as best-first search, iterative deepening search, hill-climbing search, or greedy best-first search.

[0082] In step S204, the causal reasoning module 110R can provide an initial state e0 to the causal programming module 120P. The causal programming module 110P can use the initial state e0 as a starting point and use the constrained causal graph to perform finite domain representation programming.

[0083] like Figure 1As shown, the causal inference module 110R may include a structural causal model (SCM) module 112SCM or a Bayesian network module 112B. In step S204, the causal inference module 110R may input a (generic) causal graph into the structural causal model module 112SCM or the Bayesian network module 112B to output an initial state e0. In other words, the causal inference module 110R may verify or process the (generic) causal graph based on the structural causal model or the Bayesian network, then predict / infer the current state, and use the current state as the initial state e0 of the causal programming module 110P. The generic causal graph is an unconstrained causal graph, but in one embodiment it may also be a constrained causal graph.

[0084] In one embodiment, the causal reasoning module 110R may only include one of the structural causal model module 112SCM and the Bayesian network module 112B while removing the other.

[0085] In one embodiment, similar to the causal graph CG, the causal inference module 110R can obtain a (ordinary) causal graph based on the observation data 10b of the distribution unit 10DU (or the data transformed from the observation data 10b).

[0086] In one embodiment, the causal reasoning module 110R can obtain a (normal) causal graph based on the observation data 10b of the distribution unit 10DU (or data transformed from the observation data 10b) using a deep learning model. For example, Figure 6 This is a schematic diagram illustrating how a (normal) causal graph is obtained using a deep learning model, as described in an embodiment of the present invention. The (normal) causal graph includes causal variables CV1 to CVn, where n is a positive integer. Figure 6 The causal variable CV1 can be formed as a feature vector by the attributes v1 to vm of the state, where m is a positive integer. Causal variable CV1 can be mapped to the input layer of the deep learning model, while causal variables CV2 to CVn-1 are the outputs of the hidden layers HD2 to HDn-1 of the deep learning model after passing through the activation function. Causal variable CVn can be the predicted state, and it can be mapped to the output layer of the deep learning model. Figure 6 It can be seen that, Figure 6 The deep learning model presented above corresponds to... Figure 6 The cause-and-effect diagram is shown below.

[0087] As described above, the algorithm of the 10SVR server software can be divided into a causal inference phase and a causal programming phase. The causal inference phase can infer the initial state using high-dimensional environmental variables; the causal programming phase can perform planning based on the initial state, thus taking the current state into account. Furthermore, the causal programming phase can plan the most probable / optimal radio unit parameters.

[0088] In step S206, the causal reasoning module 110R can output the action data 10c to the distribution unit 10DU, thereby setting the most probable / optimized radio unit parameters to the distribution unit 10DU. When the radio unit parameters are set to the distribution unit 10DU, different distribution units 10DUs and radio units 10RU from different suppliers can be integrated.

[0089] In one embodiment, action data 10c may contain a solution to the finite field representation programming or be associated with a solution to the finite field representation programming. For example, action data 10c may contain all actions in the solution to the finite field representation programming. In another embodiment, action data 10c may contain the most probable / optimized radio unit parameters or be associated with the most probable / optimized radio unit parameters.

[0090] In one embodiment, after setting the most likely / optimized radio unit parameters to the distribution unit 10DU in step S206, the radio unit parameters can be fine-tuned manually again, so that the radio unit parameters set to the distribution unit 10DU are optimized after testing, and not just optimized by algorithm judgment. When the radio unit parameters are set to the distribution unit 10DU, distribution units 10DUs and radio units 10RUs from different suppliers can be integrated.

[0091] Figure 7 This is a partial schematic diagram of a communication system 70 according to an embodiment of the present invention. The communication system 70 may include a server 70SVR and a distribution unit 70DU.

[0092] The distribution unit 70DU may include an application module 782p and a system module 782s. The application module 782p may output event signals to the system module 782s and receive commands from the system module 782s. The system module 782s may output observation data 10b to the server 70SVR and receive action data 10c from the server 70SVR.

[0093] Server 70SVR may include a controller 722c, a scheduler 722s, a planner 722p, and a policy maker module 722pm. Controller 722c can convert received observation data 10b into Extensible Markup Language (XML) format and output an execution status in XML format to scheduler 722s. Scheduler 722s can convert variable-length XML format into a fixed-length format and output a reactivity signal (e.g., basic data 30g) to planner 722p.

[0094] Planner 722p can output an unscheduled plan regarding the content (what) to scheduler 722s based on the initial state e0, the objective signal, the response signal from scheduler 722s, and the system description from strategy customization module 722pm. Scheduler 722s can then output a scheduling plan regarding the time (when) and method (how) to controller 722c.

[0095] In one embodiment, after setting the most likely radio unit parameters to the distribution unit 10DU in step S206 (or after manually fine-tuning the radio unit parameters again), the planner 722p can determine whether replanning is needed based on the response signal. If the planner 722p determines that the radio unit parameters need to be adjusted again, the server 70SVR can output the adjusted action data 10c to the distribution unit 70DU to reset the radio unit parameters of the distribution unit 70DU.

[0096] In one embodiment, the interface between the radio unit and the distribution unit may be an eCPRI interface, but is not limited thereto; for example, it may also be a CPRI interface.

[0097] In one embodiment, server 10SVR may be an Oracle Access Manager (OAM) server.

[0098] In one embodiment, the storage circuit is used to store image data or instructions. The storage circuit may be a Subscriber Identity Module (SIM), Read-Only Memory (ROM), Flash Memory or Random-Access Memory (RAM), CD-ROM / DVD-ROM / BD-ROM, Magnetic Tape, Hard Disk, Optical Data Storage Device, Non-volatile Storage Device, Non-transitory Computer-readable Medium, and is not limited thereto.

[0099] In one embodiment, the processing circuitry is used to execute instructions, which may be a central processing unit (CPU), a microprocessor, or an application-specific integrated circuit (ASIC), but is not limited thereto.

[0100] Table 1 lists some of the interoperability testing (IOT) profiles. Table 1 shows which radio unit parameters can be included, which sections of the Open Radio Access Network (O-RAN) specification these items relate to, and the possible settings for these items; however, radio unit parameters are not limited to these. According to Table 1, radio unit parameters may be subject to constraints.

[0101] (Table 1)

[0102]

[0103]

[0104] In summary, to avoid the inefficiency and high error rate of manually setting radio unit parameters, this invention can infer the current state and plan the actions to be taken to set the radio unit parameters in the best way, instead of manually setting the radio unit parameters.

[0105] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made within the scope of the claims of the present invention should be included within the scope of the present invention.

Claims

1. An optimization method, the optimization method comprising: A constrained causal graph is generated from observation data from a distribution unit, wherein a number of multiple causal variables and a causal structure of the constrained causal graph are determined together. The constrained causal graph is generated by maximizing multiple posterior probabilities of the underlying data corresponding to the observation data, assigning them to multiple observation functions, and the causal structure of the constrained causal graph. One of these multiple posterior probabilities is proportional to... , where s t-1 f represents the state at time point t-1, where T represents the current time point, and f i C represents the observation function, and w represents the causal structure. i，t γ represents a data point from the multiple data points at a time point t that corresponds to the observed function, where γ is a real number; The constrained cause-effect graph is used to perform a finite-domain representation programming to generate action data regarding the optimized parameters of multiple radio units; and Output the action data to the distribution unit.

2. The optimization method as described in claim 1, wherein, The steps for generating the constrained causal graph based on the observation data from the distribution unit include: Transform the observation data into a base data set; and The constrained causal graph is generated from the underlying data based on maximum a posteriori and point estimation.

3. The optimization method as described in claim 2, wherein, The steps for generating the constrained causal graph from the underlying data based on maximum a posteriori and point estimation include: Multiple observation functions are used to map multiple data points from the underlying data to the multiple causal variables in the constrained causal graph.

4. The optimization method as described in claim 3, wherein, These multiple observation functions are obtained based on a causal semantic generation model.

5. The optimization method as described in claim 2, wherein, The observation data is first converted into an Extensible Markup Language (EXPLAIN) format and then converted back into the underlying data.

6. The optimization method as described in claim 1, wherein, The observation data is a record file for this distribution unit.

7. The optimization method as described in claim 1, wherein, The steps for performing the finite field representation programming using this constrained cause-effect graph include: The constrained cause-effect graph is converted into a domain file of a planning domain description library for the finite domain representation planning.

8. The optimization method as described in claim 7, wherein, In this constrained cause-effect graph, a cause corresponds to a prerequisite for an action in the domain file, and an effect corresponding to that cause corresponds to an effect of that action in the domain file.

9. The optimization method as described in claim 1, wherein, The steps for performing the finite field representation programming using this constrained cause-effect graph include: The finite field representation programming is performed using an initial state and the constrained cause-effect graph.

10. The optimization method as described in claim 9, wherein, The initial state is generated from another causal graph based on a structural causal model or a Bayesian network.

11. A server comprising: A storage circuit for storing an instruction, the instruction comprising: A constrained causal graph is generated from observation data from a distribution unit, wherein a number of multiple causal variables and a causal structure of the constrained causal graph are determined together. The constrained causal graph is generated by maximizing multiple posterior probabilities of the underlying data corresponding to the observation data, assigning them to multiple observation functions, and the causal structure of the constrained causal graph. One of these multiple posterior probabilities is proportional to... , where s t-1 f represents the state at time point t-1, where T represents the current time point, and f i C represents the observation function, and w represents the causal structure. i，t γ represents a data point from the multiple data points at a time point t that corresponds to the observed function, where γ is a real number; The constrained cause-effect graph is used to perform a finite-domain representation programming to generate action data regarding the optimized parameters of multiple radio units; and The motion data is output to the distribution unit; and A processing circuit coupled to the storage circuit is used to execute the instruction stored in the storage circuit.

12. The server as claimed in claim 11, wherein, The instructions for generating the constrained causal graph based on the observation data from the distribution unit include: Transform the observation data into a base data set; and The constrained causal graph is generated from the underlying data based on maximum a posteriori and point estimation.

13. The server as claimed in claim 12, wherein, The instructions for generating the constrained cause-effect graph from the base data based on maximum a posteriori and point estimation include: Multiple observation functions are used to map multiple data points from the underlying data to the multiple causal variables in the constrained causal graph.

14. The server as claimed in claim 13, wherein, These multiple observation functions are obtained based on a causal semantic generation model.

15. The server as claimed in claim 12, wherein, The observation data is first converted into an Extensible Markup Language (EXPLAIN) format and then converted back into the underlying data.

16. The server as claimed in claim 11, wherein, The observation data is a record file for this distribution unit.

17. The server as claimed in claim 11, wherein, The instructions for using this constrained cause-effect graph to perform the finite field representation programming include: The constrained cause-effect graph is converted into a domain file of a planning domain description library for the finite domain representation planning.

18. The server as claimed in claim 17, wherein, In this constrained cause-effect graph, a cause corresponds to a prerequisite for an action in the domain file, and an effect corresponding to that cause corresponds to an effect of that action in the domain file.

19. The server as claimed in claim 11, wherein, The instructions for using this constrained cause-effect graph to perform the finite field representation programming include: The finite field representation programming is performed using an initial state and the constrained cause-effect graph.

20. The server as claimed in claim 19, wherein, The initial state is generated from another causal graph based on a structural causal model or a Bayesian network.

Images