Thermal power plant virtual controller data transmission optimization scheduling method based on 5G network

Abstract

The invention provides a thermal power plant virtual controller data transmission optimization scheduling based on a 5G network, and relates to the technical field of industrial control, a data mapping table is established through a virtual controller to determine key operation data, a long and short term memory neural network is adopted to calculate transmission characteristic parameters and construct a priority evaluation function to generate a transmission sequence, and the transmission sequence is optimized. A transmission period is divided into time slot units to establish a resource allocation table, path quality parameters are calculated by adopting an ant colony algorithm, a transmission path combination is optimally selected in combination with a dynamic graph neural network, and two-way transmission of key data and a control instruction is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, and in particular to a method for optimizing and scheduling data transmission of a virtual controller in a thermal power plant based on a 5G network. Background Technology

[0002] As industrial control systems develop towards intelligence, thermal power plant control systems are gradually adopting virtualization technology to reconstruct the traditional centralized control architecture. The introduction of 5G networks provides high-bandwidth and low-latency communication guarantees for virtual control systems.

[0003] However, in practical applications, existing technologies still have problems such as difficulty in accurately assessing data transmission needs, path selection methods often only consider a single performance indicator and cannot adapt to dynamic changes in network conditions, and lack of effective compensation mechanisms for transmission failures.

[0004] Therefore, a solution is urgently needed to address the problems existing in the current technology. Summary of the Invention

[0005] This invention provides a method for optimizing and scheduling data transmission of a virtual controller in a thermal power plant based on a 5G network, which can at least solve some of the problems existing in the prior art.

[0006] A first aspect of this invention provides a method for optimizing and scheduling data transmission of a virtual controller in a thermal power plant based on a 5G network, comprising:

[0007] The system acquires operational data from the field control unit of the thermal power plant, establishes a data mapping table and determines key operational data through the virtual controller, collects status parameters of the 5G network, establishes a network topology diagram, and plans multiple candidate transmission paths.

[0008] A long short-term memory neural network is used to calculate the transmission characteristic parameters of the key operational data, a transmission priority evaluation function is constructed based on the transmission characteristic parameters, and a priority transmission sequence is generated based on the transmission priority evaluation function.

[0009] The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit.

[0010] The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The path quality parameters are then matched with the priority transmission sequence. The optimal transmission path combination is selected and a transmission time slot is allocated to the optimal transmission path combination in the time slot resource allocation table.

[0011] According to the time slot resource allocation table, the key operating data is transmitted to the virtual controller through the optimal transmission path combination, and control commands are generated and sent to the field control unit through the reverse path in the optimal transmission path combination.

[0012] In one alternative implementation,

[0013] The virtual controller establishes a data mapping table and determines key operational data, collects 5G network status parameters, establishes a network topology diagram, and plans multiple candidate transmission paths, including:

[0014] The system acquires the operating data of the field control unit of the thermal power plant, establishes a data mapping table through the virtual controller, records data identifiers and data feature information, and uses a support vector classifier to classify the operating data based on the data feature information to determine key operating data.

[0015] Collect 5G network status parameters, calculate node availability score based on network node status in the status parameters, calculate link weight coefficient based on link status, and integrate the node availability score and the link weight coefficient to generate a network resource status matrix.

[0016] A network topology diagram is established by using the minimum tree generation algorithm in conjunction with the network resource state matrix. A path cost function is constructed based on the node availability score and the link weight coefficient. Multiple candidate transmission paths are planned in the network topology diagram according to the path cost function.

[0017] In one alternative implementation,

[0018] The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. A transmission priority evaluation function is constructed based on the transmission characteristic parameters. A priority transmission sequence is generated based on the transmission priority evaluation function, including:

[0019] The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. The long short-term memory neural network extracts features through a forget gate, an input gate, and an output gate. It performs unit state updates based on the gating combination of the current input and the historical state, and performs hidden layer state updates using the nonlinear transformation between the output gate and the unit state to obtain the transmission characteristic parameters of the key operational data. The transmission characteristic parameters include timeliness indicators, data value, historical transmission success rate, and bandwidth utilization rate.

[0020] A transmission priority evaluation function is constructed by minimizing the mean squared error between the predicted priority and the actual priority and by calculating the feature weight vector and the weight regularization term. The inner product of the feature weight vector and the transmission feature parameters is used as the transmission priority evaluation function.

[0021] The score result of the transmission priority evaluation function is calculated, and the key operating data is divided into multiple priority levels according to the score result of the transmission priority evaluation function. Within the priority level, a dynamic adjustment factor for waiting time and transmission failure count is set. The final priority score is obtained by multiplying the score result of the transmission priority evaluation function and the dynamic adjustment factor. The priority transmission sequence is generated according to the final priority score.

[0022] In one alternative implementation,

[0023] Calculate the score result of the transmission priority evaluation function, and divide the key operational data into multiple priority levels based on the score result of the transmission priority evaluation function, including:

[0024] Collect the correlation information between the key operational data, calculate the causal influence strength of the key operational data based on Bayesian network, and generate a causal strength matrix;

[0025] The causal strength matrix is ​​input into the transmission priority evaluation function, which includes a data criticality score item, a transmission timeliness score item, and a network resource consumption score item. Based on the causal strength matrix, the direct causal impact value and indirect causal impact value of the critical operational data are calculated.

[0026] A service quality indicator set is constructed, which includes throughput, latency, reliability, and jitter indicators. Based on the service quality indicator set, a multi-objective optimization function is constructed to obtain the weighted deviation between each indicator and the target value. The direct causal impact value, the indirect causal impact value, and the calculation result of the multi-objective optimization function are then weighted and combined with the data criticality score item, the transmission timeliness score item, and the network resource consumption score item to obtain the score result.

[0027] The scoring results are input into the priority level division model. The priority level division threshold is dynamically determined based on the causal intensity distribution in the scoring results. The key operation data is divided into three priority levels according to the priority level division threshold. Priority resource protection is implemented for the highest priority level, and resource reuse is implemented for the lowest priority level. Service quality compensation adjustment is achieved by dynamically adjusting the resource allocation ratio, monitoring service quality indicator deviations in real time, and adaptively compensating for bandwidth allocation.

[0028] In one alternative implementation,

[0029] The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit, including:

[0030] The data transmission period is obtained, and the transmission period is divided into multiple time slot units using the smallest time granularity of data transmission.

[0031] Create a time slot resource allocation table, and count and record the index number, occupancy identifier, remaining bandwidth and allocable duration of each time slot unit. The index number identifies the position of the time slot unit in the transmission cycle, the occupancy identifier represents the occupancy status of the time slot unit, and the remaining bandwidth and allocable duration represent the resource quantity of the time slot unit.

[0032] The availability status of each time slot unit is detected and updated in the time slot resource allocation table.

[0033] In one alternative implementation,

[0034] The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The matching degree between the path quality parameters and the priority transmission sequence is calculated, and the optimal transmission path combination is selected, including:

[0035] The link delay parameters are obtained by acquiring the transmission delay and processing delay between each hop node on the path. The bandwidth utilization rate is obtained by evaluating the ratio of the available bandwidth of the path to the total bandwidth of the link. A link reliability model is established based on the historical transmission success rate and link stability. The hop count cost is obtained by counting the number of relay nodes traversed by the path. The link delay parameters, the bandwidth utilization rate, the link reliability, and the hop count cost are constructed into a path quality evaluation index set.

[0036] A pheromone matrix is ​​created based on the path quality evaluation index set. A heuristic information matrix is ​​generated by performing weighted normalization calculation on the path quality evaluation index set. The pheromone matrix and the heuristic information matrix are used as initial parameters for path search.

[0037] By combining the pheromone matrix and the heuristic information matrix, the transition probability between nodes is calculated to generate candidate transmission paths. The path quality parameters are obtained by normalizing and weighting the path quality evaluation index set of the candidate transmission paths using the ant colony algorithm. The path quality parameters characterize the transmission performance of the candidate transmission paths.

[0038] The pheromone increment is calculated based on the path quality parameters, and the pheromone matrix is ​​updated by decaying in combination with the pheromone evaporation coefficient. The candidate transmission path generation and the path quality parameter calculation are repeated based on the updated pheromone matrix until a candidate transmission path that meets the requirements is found.

[0039] The path quality parameters of the candidate transmission paths obtained by the search are matched with the priority transmission sequence. Based on the preset matching degree threshold, a set of paths that meet the requirements of the priority transmission sequence is obtained. The set of paths is combined and optimized by combining dynamic graph neural network and dynamic programming method. The allocation scheme of the set of paths is adjusted with the path quality parameters and load balancing constraints as optimization objectives to obtain the optimal combination of transmission paths.

[0040] In one alternative implementation,

[0041] By combining dynamic graph neural networks and dynamic programming methods to optimize the path set, and adjusting the allocation scheme of the path set with the path quality parameters and load balancing constraints as optimization objectives, the optimal transmission path combination is obtained, including:

[0042] A dynamic graph neural network is constructed, which maps network nodes and links to graph nodes and graph edges, respectively. Node topology feature vectors and edge state feature vectors are extracted, and node local features are generated based on the edge convolution mechanism to obtain the network topology features of the path set. Node importance is calculated and hierarchical clustering is performed on the network topology features, and graph pooling is performed to obtain a hierarchical path set representation.

[0043] The latency, bandwidth, packet loss rate, jitter, and stability indicators of each path in the path set are obtained and normalized to obtain path quality parameters. The load difference between paths is calculated based on the current load rate of each path, and load balancing constraints are constructed. The hierarchical path set representation and load balancing constraints are constructed into a state space, and the allocation and adjustment strategy of the path set is constructed into an action space. A value function is constructed based on the path quality parameters and load balancing constraints.

[0044] By combining dynamic programming, state transitions are performed based on the value function and Bellman equation, the allocation scheme of the path set is dynamically adjusted, the network state update feature vector is monitored, the graph structure is adjusted in real time and path feature recalculation is triggered, and the optimal transmission path combination that satisfies the path quality parameters and load balancing constraints is obtained through iterative solution.

[0045] A second aspect of the present invention,

[0046] An electronic device is provided, comprising:

[0047] processor;

[0048] Memory used to store processor-executable instructions;

[0049] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0050] A third aspect of the present invention,

[0051] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0052] This invention integrates Long Short-Term Memory Neural Networks and Ant Colony Algorithms to achieve accurate extraction of data transmission features and dynamic optimization of transmission paths. A priority evaluation model based on multi-dimensional features, combined with a dynamic adjustment factor mechanism, ensures timely and reliable transmission of critical data. A dynamic graph neural network is introduced for path combination optimization, and network congestion is effectively avoided through load balancing constraints. A fine-grained time slot resource management mechanism significantly improves resource utilization efficiency, achieving efficient transmission of control data and continuous optimization of system performance, thus providing a reliable guarantee for the stable operation of the virtual control system of thermal power plants. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the data transmission optimization and scheduling method for a virtual controller in a thermal power plant based on a 5G network, according to an embodiment of the present invention.

[0054] Figure 2 This is a comparison chart of resource utilization under different load conditions for the data transmission optimization and scheduling method of the virtual controller of a thermal power plant based on a 5G network, according to an embodiment of the present invention.

[0055] Figure 3 This is a comparison chart of the multi-priority service support capabilities of the data transmission optimization and scheduling method for a thermal power plant virtual controller based on a 5G network, according to an embodiment of the present invention.

[0056] Figure 4 This diagram illustrates the path optimization effect of the data transmission optimization and scheduling method for a virtual controller in a thermal power plant based on a 5G network, according to an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0059] Figure 1This is a flowchart illustrating the data transmission optimization and scheduling method for a 5G network-based virtual controller in a thermal power plant, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0060] The system acquires operational data from the field control unit of the thermal power plant, establishes a data mapping table and determines key operational data through the virtual controller, collects status parameters of the 5G network, establishes a network topology diagram, and plans multiple candidate transmission paths.

[0061] A long short-term memory neural network is used to calculate the transmission characteristic parameters of the key operational data, a transmission priority evaluation function is constructed based on the transmission characteristic parameters, and a priority transmission sequence is generated based on the transmission priority evaluation function.

[0062] The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit.

[0063] The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The path quality parameters are then matched with the priority transmission sequence. The optimal transmission path combination is selected and a transmission time slot is allocated to the optimal transmission path combination in the time slot resource allocation table.

[0064] According to the time slot resource allocation table, the key operating data is transmitted to the virtual controller through the optimal transmission path combination, and control commands are generated and sent to the field control unit through the reverse path in the optimal transmission path combination.

[0065] In one alternative implementation,

[0066] The virtual controller establishes a data mapping table and determines key operational data, collects 5G network status parameters, establishes a network topology diagram, and plans multiple candidate transmission paths, including:

[0067] The system acquires the operating data of the field control unit of the thermal power plant, establishes a data mapping table through the virtual controller, records data identifiers and data feature information, and uses a support vector classifier to classify the operating data based on the data feature information to determine key operating data.

[0068] Collect 5G network status parameters, calculate node availability score based on network node status in the status parameters, calculate link weight coefficient based on link status, and integrate the node availability score and the link weight coefficient to generate a network resource status matrix.

[0069] A network topology diagram is established by using the minimum tree generation algorithm in conjunction with the network resource state matrix. A path cost function is constructed based on the node availability score and the link weight coefficient. Multiple candidate transmission paths are planned in the network topology diagram according to the path cost function.

[0070] Real-time acquisition of operational data from the field control units of thermal power plants. When establishing a data mapping table, a unique identifier is generated for each piece of operational data. The identifier includes information such as data type, acquisition time, and equipment number. Data feature information is extracted, the stability of the data generation cycle is calculated, the trend and fluctuation range of data values ​​are analyzed, and the data value level is determined based on the degree of impact of the data on the control process. All feature information is recorded in the data mapping table in a unified format.

[0071] Support vector classification is applied to the extracted data features, and the feature data is normalized to eliminate the influence of dimensions. A radial basis function kernel is selected to map the feature vectors to a high-dimensional feature space. In the high-dimensional feature space, an optimal classification hyperplane is constructed by maximizing the classification margin. The classification process uses a sequential minimum optimization algorithm to solve the dual problem, obtaining support vectors and a classification decision function. The decision function is used to classify the running data, and data with high classification priority are identified as key running data.

[0072] Network status parameters are collected and processed. Network node status data includes parameters such as processor utilization, memory usage, and task queue length. The collected node status parameters are then weighted and normalized, with weight coefficients determined based on the parameter's impact on node performance. A node availability score is calculated through weighted combination. Network link status parameters such as bandwidth utilization, latency, and packet loss rate are collected, and link weight coefficients are calculated using a fuzzy comprehensive evaluation method. The node availability score is then filled into the diagonal positions of the network resource status matrix, and the link weight coefficients are filled into the corresponding off-diagonal positions.

[0073] A network topology graph is constructed based on the network resource state matrix. A minimum spanning tree algorithm is employed, with network nodes as vertices, network links as edges, and link weights as edge weights. Starting from any node, the algorithm iteratively selects the edge with the minimum weight connecting the selected set of nodes to the unselected set of nodes, continuing until all nodes are connected. During the generation of the minimum spanning tree, suboptimal edges are recorded to prepare for planning multiple candidate paths later.

[0074] A path cost function is constructed, and candidate transmission paths are planned. The path cost function comprehensively considers the availability scores and link weight coefficients of all nodes on the path. For each node on the path, an influence coefficient is determined based on its position in the path, and the node cost is calculated by combining the node availability score. For each link on the path, the link cost is calculated by combining the link weight coefficient. The weighted sum of the node cost and the link cost constitutes the total path cost. In the network topology diagram, an improved shortest path algorithm is adopted, with path cost as the optimization objective, and multiple candidate transmission paths are planned using dynamic programming.

[0075] For example, when acquiring operational data from the field control unit, the following main steam pressure parameters were collected: real-time pressure value 16.8 MPa, steam temperature 545℃, and feedwater flow rate 1260 t / h. An identifier code PS_MS_001 (Main Steam Pressure_Main Steam System_Serial Number 001) was generated for the pressure data. Analysis extracted the following characteristic information: acquisition period 50 ms, data fluctuation range ±0.2 MPa, data change rate 0.15% / s, and process influence level one (highest level).

[0076] The pressure data features were input into a support vector classifier, and the feature values ​​were normalized: the 50ms acquisition period was normalized to 0.8, the fluctuation range of ±0.2MPa was normalized to 0.6, the rate of change of 0.15% / s was normalized to 0.75, and the first-level influence was normalized to 1.0. After classifier calculation, the score was 0.92, which is much higher than the critical data threshold of 0.75, and was therefore identified as critical operational data.

[0077] The status of three key nodes was obtained: Field control unit node A had a CPU utilization of 82% and a memory utilization of 65%, resulting in an availability score of 0.72; switch node B had a CPU utilization of 45% and a memory utilization of 38%, resulting in an availability score of 0.85; and virtual controller node C had a CPU utilization of 68% and a memory utilization of 58%, resulting in an availability score of 0.78. The inter-node link status was displayed as follows: AB link had a bandwidth utilization of 75%, a latency of 4.2ms, and a packet loss rate of 0.08%, resulting in a weighting coefficient of 0.76; BC link had a bandwidth utilization of 62%, a latency of 3.8ms, and a packet loss rate of 0.06%, resulting in a weighting coefficient of 0.82; and AC link had a bandwidth utilization of 70%, a latency of 4.5ms, and a packet loss rate of 0.07%, resulting in a weighting coefficient of 0.78.

[0078] In the network resource state matrix, the diagonal elements are the node availability scores [0.72, 0.85, 0.78], and the off-diagonal elements are the corresponding link weight coefficients. Through iterative calculation using the minimum spanning tree algorithm, the link with the highest weight (BC link, weight 0.82) is first selected, followed by the link with weight (AB link, weight 0.76), forming the complete network topology.

[0079] During the path planning phase, a node scoring weight coefficient of 0.6 and a link performance weight coefficient of 0.4 were set, resulting in two candidate paths: Path 1 (AC) with a total cost of 1.86 and Path 2 (ABC) with a total cost of 2.15. The direct path (AC) with the lower cost was selected as the primary channel. The system's switching thresholds were set to: bandwidth utilization of 85% and latency of 5ms. When the performance metrics of the primary path exceeded the thresholds, the system automatically switched to the backup path (ABC) to ensure reliable transmission of stress data.

[0080] In this embodiment, the radial basis kernel function is used to map features to a high-dimensional space. The optimal classification hyperplane is constructed by maximizing the classification interval, which realizes the accurate identification of key operating data and provides a basis for differentiated transmission. The minimum spanning tree algorithm is used to construct the network topology structure. By recording suboptimal edges, conditions are created for multi-path planning. The path cost function comprehensively considers the node position influence coefficient, availability score and link weight, making the path evaluation more comprehensive and providing a strong guarantee for the stable operation of the industrial control system.

[0081] In one alternative implementation,

[0082] The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. A transmission priority evaluation function is constructed based on the transmission characteristic parameters. A priority transmission sequence is generated based on the transmission priority evaluation function, including:

[0083] The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. The long short-term memory neural network extracts features through a forget gate, an input gate, and an output gate. It performs unit state updates based on the gating combination of the current input and the historical state, and performs hidden layer state updates using the nonlinear transformation between the output gate and the unit state to obtain the transmission characteristic parameters of the key operational data. The transmission characteristic parameters include timeliness indicators, data value, historical transmission success rate, and bandwidth utilization rate.

[0084] A transmission priority evaluation function is constructed by minimizing the mean squared error between the predicted priority and the actual priority and by calculating the feature weight vector and the weight regularization term. The inner product of the feature weight vector and the transmission feature parameters is used as the transmission priority evaluation function.

[0085] The score result of the transmission priority evaluation function is calculated, and the key operating data is divided into multiple priority levels according to the score result of the transmission priority evaluation function. Within the priority level, a dynamic adjustment factor for waiting time and transmission failure count is set. The final priority score is obtained by multiplying the score result of the transmission priority evaluation function and the dynamic adjustment factor. The priority transmission sequence is generated according to the final priority score.

[0086] A Long Short-Term Memory (LSTM) neural network is constructed to calculate transmission feature parameters. It comprises three layers: an input layer, an LSTM layer, and an output layer. The input layer receives historical transmission records of key operational data. In the LSTM layer, the data stream is processed through three gating structures: a forget gate, an input gate, and an output gate. The forget gate uses the sigmoid function to process the input data and determine the proportion of historical information to be forgotten; the input gate determines the information to be updated at the current moment; and the output gate controls the final information output level. During the unit state update process, the current input and historical state are combined through the gating structures to obtain candidate state values. The unit state is updated based on the output results of the forget gate and the input gate. The hidden layer state is updated through a tanh nonlinear transformation between the output gate and the unit state. After repeated iterative training, transmission feature parameters, including timeliness indicators, data value, historical transmission success rate, and bandwidth utilization, are extracted.

[0087] After extracting the feature parameters, a transmission priority evaluation function is constructed. Actual priority annotation information from historical transmission data is collected as the target value for supervised learning. The inner product of the feature weight vector and the transmission feature parameters is constructed as the basic form of the evaluation function. During the optimization of the evaluation function, a loss function is designed that includes the mean squared error of predicted and actual priorities, as well as an L2 regularization term for the weight vector. Stochastic gradient descent is used to optimize the loss function. The feature weight vector is continuously updated through multiple iterations until the loss function converges, resulting in the optimized feature weight vector. The inner product of this optimized feature weight vector and the transmission feature parameters is then used as the transmission priority evaluation function.

[0088] The priority transmission sequence generation process utilizes a pre-constructed transmission priority evaluation function to calculate a score for each key piece of operational data. Based on the score results, a reasonable priority division threshold is set to divide the data into different priority levels. Within each priority level, two dynamic adjustment factors are introduced: waiting time and the number of transmission failures. The waiting time adjustment factor gradually increases as the data waits in the transmission queue, while the number of transmission failures adjustment factor increases as the number of failed retransmissions increases. The original score obtained from the evaluation function is multiplied by these two dynamic adjustment factors to obtain the final score that takes into account the transmission history.

[0089] For example, in the boiler feedwater system of a thermal power plant, the system collects four key control parameters: feedwater flow rate, feedwater pressure, drum water level, and feedwater temperature. Through LSTM network analysis, the transmission characteristics of feedwater flow rate data are extracted as follows: 5ms timeliness requirement, 0.95 value score, 98% historical success rate, and 12% bandwidth occupancy; feedwater pressure data transmission characteristics are: 8ms timeliness requirement, 0.88 value score, 96% historical success rate, and 10% bandwidth occupancy; drum water level data transmission characteristics are: 10ms timeliness requirement, 0.85 value score, 97% historical success rate, and 8% bandwidth occupancy; and feedwater temperature data transmission characteristics are: 15ms timeliness requirement, 0.75 value score, 95% historical success rate, and 6% bandwidth occupancy.

[0090] After calculation using the priority evaluation function, the feedwater flow rate received an initial score of 0.92, feedwater pressure 0.85, drum water level 0.82, and feedwater temperature 0.73. Based on this, the system divided the data into three priority levels: feedwater flow rate (high priority), feedwater pressure and drum water level (medium priority), and feedwater temperature (low priority). During actual operation, due to a 100ms waiting time and two transmission failures for the feedwater pressure data, a waiting time adjustment factor of 1.2 and a failure count adjustment factor of 1.15 were assigned. Multiplying the original score of 0.85 for the feedwater pressure data by these two adjustment factors yielded a final priority score of 1.17, temporarily elevating its priority to the high priority level. The generated transmission order was: feedwater flow rate, feedwater pressure, drum water level, and feedwater temperature. The system continuously monitored the waiting time and transmission status of each data point during subsequent transmissions, dynamically adjusting the transmission priority accordingly.

[0091] In this embodiment, by introducing the optimization process of mean squared error loss term and regularization term, the evaluation results can not only reflect historical transmission experience, but also have strong generalization ability. This avoids the rigidity problem of traditional fixed priority schemes and provides a scientific basis for flexible adjustment of data transmission priority. The dynamic priority mechanism ensures the timely transmission of high-priority data, while also providing transmission opportunities for other data, thus improving the overall transmission efficiency. By dynamically increasing the priority, the retransmission process is accelerated, and the average latency of data transmission is reduced.

[0092] In one alternative implementation,

[0093] Calculate the score result of the transmission priority evaluation function, and divide the key operational data into multiple priority levels based on the score result of the transmission priority evaluation function, including:

[0094] Collect the correlation information between the key operational data, calculate the causal influence strength of the key operational data based on Bayesian network, and generate a causal strength matrix;

[0095] The causal strength matrix is ​​input into the transmission priority evaluation function, which includes a data criticality score item, a transmission timeliness score item, and a network resource consumption score item. Based on the causal strength matrix, the direct causal impact value and indirect causal impact value of the critical operational data are calculated.

[0096] A service quality indicator set is constructed, which includes throughput, latency, reliability, and jitter indicators. Based on the service quality indicator set, a multi-objective optimization function is constructed to obtain the weighted deviation between each indicator and the target value. The direct causal impact value, the indirect causal impact value, and the calculation result of the multi-objective optimization function are then weighted and combined with the data criticality score item, the transmission timeliness score item, and the network resource consumption score item to obtain the score result.

[0097] The scoring results are input into the priority level division model. The priority level division threshold is dynamically determined based on the causal intensity distribution in the scoring results. The key operation data is divided into three priority levels according to the priority level division threshold. Priority resource protection is implemented for the highest priority level, and resource reuse is implemented for the lowest priority level. Service quality compensation adjustment is achieved by dynamically adjusting the resource allocation ratio, monitoring service quality indicator deviations in real time, and adaptively compensating for bandwidth allocation.

[0098] Bayesian networks were used for data association analysis. During data collection, the temporal variation characteristics of key operational data were recorded, including the frequency, magnitude, trend, and synchronicity with other data. When constructing the Bayesian network, noise interference was eliminated through data preprocessing, and a structure learning algorithm was used to determine the network topology. Directed edges between network nodes were determined through conditional independence tests, and the direction of the edges was determined based on temporal information and expert knowledge. In the parameter learning phase, the maximum likelihood estimation method was used, combined with the expectation-maximization algorithm to iteratively optimize the parameter values ​​in the conditional probability table. When calculating the causal influence strength, the conditional probability values ​​of direct connections and the indirect influence transmission effect were comprehensively considered to obtain a complete causal strength matrix.

[0099] In constructing the transmission priority evaluation function, a data criticality scoring item is designed. A scoring model is established by analyzing factors such as the impact of data on the control process, the decay characteristics of data value over time, and the stability of the data update cycle. The transmission timeliness scoring item is based on the data's lifecycle characteristics, combined with transmission delay tolerance and real-time requirements, to construct a quantitative timeliness index. The network resource consumption scoring item is implemented by establishing a resource usage prediction model, considering factors such as bandwidth requirements, processing load, and storage overhead.

[0100] When calculating the direct causal impact value, the direct connection strength in the causal strength matrix is ​​normalized. The calculation of the indirect causal impact value needs to consider the impact propagation along multi-hop paths. A dynamic programming method is used to calculate the path with the maximum impact, and a decay function is used to handle the weakening of the impact caused by multi-stage propagation. The combination of impact value and scoring items adopts an adaptive weighting scheme, with the weight coefficients dynamically adjusted according to the network state.

[0101] The service quality indicator set is constructed using a layered architecture, establishing composite and integrated indicators on top of four basic indicators: throughput, latency, reliability, and jitter. The multi-objective optimization function employs goal programming, calculating the deviation between actual and target values ​​by setting expected values ​​and tolerance ranges for different indicators. A penalty term is introduced during the optimization process to weight deviations exceeding the tolerance range.

[0102] The priority hierarchy model employs an adaptive threshold mechanism, determining the hierarchy threshold by analyzing the probability distribution characteristics of the scoring results and combining this with a clustering method based on kernel density estimation. This model can automatically adjust the threshold according to dynamic changes in the scoring distribution, maintaining a reasonable allocation ratio for each priority level. In the resource protection mechanism, a token bucket algorithm is used to control resource usage at the highest priority level, ensuring the transmission performance of critical data. The resource reuse strategy improves the resource utilization efficiency of the lowest priority level through a dynamic time slot allocation method.

[0103] Service quality compensation adjustments employ a closed-loop control mechanism, continuously monitoring deviations in various service quality indicators. When an anomaly is detected, the severity and scope of the anomaly are assessed, and intervention is implemented through methods such as adjusting resource allocation ratios and bandwidth compensation allocation. The compensation process adopts a gradual adjustment strategy to avoid system turbulence caused by drastic changes. Simultaneously, a compensation effectiveness evaluation mechanism is established to assess the effectiveness of compensation actions in real time and dynamically optimize the compensation strategy based on the evaluation results.

[0104] For example, taking the steam turbine control system of a thermal power plant as an example, key operating data such as speed, inlet steam pressure, exhaust steam pressure, and vibration are involved. Bayesian network analysis reveals that the causal strength between speed and inlet steam pressure is 0.85, and the causal strength between speed and vibration is 0.75; the causal strength between inlet steam pressure and exhaust steam pressure is 0.65; and the causal strength between vibration and other parameters is less than 0.3. Based on this, a causal strength matrix is ​​generated.

[0105] In the priority assessment, the engine speed data had the highest direct causal impact value of 0.85 and indirect causal impact value of 0.54. Service quality indicators showed that current throughput reached 95% of the target value, latency exceeded the target value by 15%, reliability reached 98% of the target value, and jitter was within the target range. After weighted calculation, the engine speed data received a comprehensive score of 0.92, inlet pressure was 0.85, vibration was 0.78, and exhaust pressure was 0.72.

[0106] Based on the distribution characteristics of the scoring results, 0.9 was set as the high-priority threshold, and 0.75 as the medium-priority threshold. Finally, the engine speed data was classified as high-priority, with 20% of its bandwidth reserved; inlet steam pressure and vibration data were classified as medium-priority; and exhaust steam pressure was classified as low-priority, using a dynamic multiplexing mechanism. When the transmission delay of the inlet steam pressure data was detected to be continuously exceeding the standard, its bandwidth allocation was dynamically adjusted by 15% to bring the delay index back to the target range.

[0107] In this embodiment, the causal reasoning-based analysis method breaks through the limitation of traditional techniques that only consider a single data feature, making priority evaluation more in line with the actual operating characteristics of the control system. The adaptive weighting mechanism introduced in the evaluation function enables the evaluation results to be dynamically adjusted with changes in the network environment, enhancing the system's environmental adaptability.

[0108] In existing technologies, the determination of data transmission priority in power plant control systems mainly relies on static configuration and simple numerical comparison, ignoring the correlation between data and failing to accurately reflect the actual importance of data in the control process. The use of fixed evaluation indicators and weights lacks dynamic adjustment capabilities and is difficult to adapt to changes in network status and transmission requirements. In terms of service quality assurance, existing technologies mostly adopt independent compensation methods, failing to achieve multi-indicator synergistic optimization, resulting in low resource utilization efficiency.

[0109] The dynamic priority adjustment mechanism in this embodiment effectively prevents the transmission starvation phenomenon of low-priority data, ensuring the fairness of the overall transmission. The multi-level priority division and resource protection mechanism provides a stable transmission channel for high-priority data, significantly reducing transmission latency and jitter. The introduction of resource reuse strategy improves network resource utilization efficiency and reduces transmission congestion, providing a new solution for improving the transmission performance of industrial control systems.

[0110] Figure 2 This diagram compares the resource utilization of the 5G-based virtual controller data transmission optimization scheduling method for thermal power plants under different load conditions, illustrating the resource utilization of various priority allocation schemes under different network load conditions. The horizontal axis represents the network load percentage, and the vertical axis represents the resource utilization rate. The proposed scheme (square markers) achieves a resource utilization rate of 78.5% under light load (30%), 85.3% under medium load (50%), 89.7% under heavy load (70%), and maintains a high efficiency of 87.2% under extreme load (90%). In contrast, the static three-level priority scheme (circle markers) has a utilization rate of only 62.3% under light load. Although it increases with increasing load, it drops sharply to 58.6% under extreme load, showing significant resource waste. The priority scheme based on queue management (cross markers) performs well under medium load (76.8%), but drops to 69.1% under extreme load conditions. Of particular note is the excellent stability demonstrated by this technical solution in load fluctuation scenarios (75% to 85% range in the figure), with resource utilization maintained at approximately 88.5% and fluctuations not exceeding 2%. This is attributed to the introduction of a network resource consumption scoring term in the multi-objective optimization function and the adaptive threshold mechanism in the priority hierarchy partitioning model. In actual thermal power plant turbine control system testing, this technical solution maintained an average resource utilization rate of 89.3% during typical operating condition changes (such as rapid load increases and decreases), which is 21.7 percentage points higher than the traditional solution.

[0111] In one alternative implementation,

[0112] The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit, including:

[0113] The data transmission period is obtained, and the transmission period is divided into multiple time slot units by using the minimum time granularity of data transmission.

[0114] Create a time slot resource allocation table, and count and record the index number, occupancy identifier, remaining bandwidth and allocable duration of each time slot unit. The index number identifies the position of the time slot unit in the transmission cycle, the occupancy identifier represents the occupancy status of the time slot unit, and the remaining bandwidth and allocable duration represent the resource quantity of the time slot unit.

[0115] The availability status of each time slot unit is detected and updated in the time slot resource allocation table.

[0116] The transmission cycle is acquired and divided by analyzing the acquisition cycle, processing latency, and transmission requirements of all key operational data in the control system to determine a complete data transmission cycle. After determining the transmission cycle, an appropriate time granularity needs to be selected for cycle division. The selection of time granularity is based on the minimum processing unit time for data transmission; the minimum time granularity should be less than the shortest time required for all data processing and transmission operations. The entire transmission cycle is then uniformly divided according to the selected time granularity to obtain a series of continuous time slot units.

[0117] A time slot resource allocation table is created, with a complete attribute record for each time slot unit, including four core attributes: an index number to identify the relative position of the time slot unit in the transmission cycle, using a consecutive integer sequence; an occupancy identifier to indicate the current usage status of the time slot unit, including idle, reserved, and occupied states; remaining bandwidth to indicate the network bandwidth resources that can be allocated in the current time slot; and allocable duration to indicate the duration during which the time slot can be used for new transmission tasks. These attribute information records are recorded in a standardized format for easy subsequent querying and updating.

[0118] After creating the time slot resource allocation table, a real-time monitoring mechanism is established to monitor the availability status of each time slot unit. Status monitoring includes multiple dimensions: detecting the number of transmission tasks currently executing in the time slot unit and their resource usage; monitoring the current network bandwidth usage, actual transmission rate, and bandwidth fluctuations; and assessing whether the remaining resources in the time slot unit can meet new transmission demands. The monitoring results need to be updated in the time slot resource allocation table in a timely manner to ensure that the information in the allocation table remains synchronized with the actual operating status.

[0119] The status update employs a real-time update mechanism. Once a status change is detected, an update operation is immediately triggered. This includes recalculating the remaining bandwidth based on the bandwidth occupied by currently executing transmission tasks, determining the actual available bandwidth resources; updating the allocatable duration, taking into account the time already occupied in the current time slot and reserved safety margins; and adjusting the occupancy flag to reflect the current actual usage status of the time slot unit. This dynamic update mechanism ensures the accuracy and real-time nature of resource allocation.

[0120] For example, in the steam turbine control system of a thermal power plant, the minimum data processing time is found to be 5 milliseconds, so 5 milliseconds is chosen as the time granularity. Assuming the control cycle is 1 second, this 1-second transmission cycle is divided into 200 time slot units.

[0121] In the time slot resource allocation table, attribute records are recorded for each time slot unit. For example, the record for a certain time slot unit is: index number 45, indicating that this is the 45th time slot in the transmission cycle; occupancy identifier is "partially occupied", indicating that the time slot has been partially allocated but there are still resources remaining; remaining bandwidth shows that there is still some available bandwidth; allocable duration shows that a certain amount of transmission time can still be allocated in this time slot.

[0122] During the status detection process, the resource usage of the time slot unit is continuously monitored. When a new transmission task is detected to be starting, the occupancy flag of the time slot unit is updated to "occupied," and the remaining bandwidth and allocable duration are reduced accordingly. When the transmission task is completed, the status is updated again, and the released resources are added back to the available resource pool to ensure that the information in the resource allocation table always reflects the actual resource usage.

[0123] In this embodiment, fine-grained management of time resources improves the accuracy of time resource utilization and provides a more flexible scheduling space for subsequent resource allocation. By recording attributes such as index number, occupancy identifier, remaining bandwidth, and allocable duration, a comprehensive description of the resource status of each time slot unit is achieved. The time slot resource allocation table can accurately reflect the current resource usage, avoiding conflicts and waste in the resource allocation process. Through refined time slot division, comprehensive status recording, and real-time monitoring and updates, efficient management of transmission resources is achieved, providing strong support for the stable operation of control system data transmission.

[0124] In one alternative implementation,

[0125] The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The matching degree between the path quality parameters and the priority transmission sequence is calculated, and the optimal transmission path combination is selected, including:

[0126] The link delay parameters are obtained by acquiring the transmission delay and processing delay between each hop node on the path. The bandwidth utilization rate is obtained by evaluating the ratio of the available bandwidth of the path to the total bandwidth of the link. A link reliability model is established based on the historical transmission success rate and link stability. The hop count cost is obtained by counting the number of relay nodes traversed by the path. The link delay parameters, the bandwidth utilization rate, the link reliability, and the hop count cost are constructed into a path quality evaluation index set.

[0127] A pheromone matrix is ​​created based on the path quality evaluation index set. A heuristic information matrix is ​​generated by performing weighted normalization calculation on the path quality evaluation index set. The pheromone matrix and the heuristic information matrix are used as initial parameters for path search.

[0128] By combining the pheromone matrix and the heuristic information matrix, the transition probability between nodes is calculated to generate candidate transmission paths. The path quality parameters are obtained by normalizing and weighting the path quality evaluation index set of the candidate transmission paths using the ant colony algorithm. The path quality parameters characterize the transmission performance of the candidate transmission paths.

[0129] The pheromone increment is calculated based on the path quality parameters, and the pheromone matrix is ​​updated by decaying in combination with the pheromone evaporation coefficient. The candidate transmission path generation and the path quality parameter calculation are repeated based on the updated pheromone matrix until a candidate transmission path that meets the requirements is found.

[0130] The path quality parameters of the candidate transmission paths obtained by the search are matched with the priority transmission sequence. Based on the preset matching degree threshold, a set of paths that meet the requirements of the priority transmission sequence is obtained. The set of paths is combined and optimized by combining dynamic graph neural network and dynamic programming method. The allocation scheme of the set of paths is adjusted with the path quality parameters and load balancing constraints as optimization objectives to obtain the optimal combination of transmission paths.

[0131] Transmission performance parameters between each hop node of the path are obtained through network probing. Delay is measured for each hop, including the transmission delay of data packets on the link and the processing delay at the node. These delay parameters are combined to obtain a complete link delay characteristic. By continuously monitoring the actual transmission rate of the link and combining it with the theoretical bandwidth capacity of the link, the utilization status of bandwidth resources is calculated. Historical transmission data is collected, and the transmission success rate is analyzed. A reliability assessment model is established based on the physical characteristics and operational status of the link. The number of relay nodes traversed by the path is counted, and routing costs are evaluated. After standardization, a complete set of path quality assessment indicators is formed.

[0132] Based on the path quality assessment index set, an initial pheromone matrix is ​​created. The elements in the matrix represent the path selection tendency, and the initial values ​​are set to the same small positive numbers. The indices in the path quality assessment index set are weighted and normalized to generate a heuristic information matrix. The matrix reflects the prior knowledge of path selection and guides the path search direction in the initial stage.

[0133] The transition probabilities between nodes are calculated using a pheromone matrix and a heuristic information matrix. The calculation of transition probabilities comprehensively considers the influence of pheromone concentration and heuristic information to determine the selection probability of the next-hop node. Path searching is performed based on the transition probabilities to generate multiple candidate transmission paths. For each candidate path, its complete path quality parameters are calculated, which comprehensively reflect the transmission performance of the path.

[0134] The pheromone increment is calculated based on the quality parameters of the candidate paths, and the magnitude of the increment is positively correlated with the path quality. Considering the natural volatility of pheromones, a volatility coefficient is introduced to decay and update the pheromone matrix. The updated pheromone matrix is ​​used to re-search for paths and evaluate quality, and the search results are continuously optimized through multiple iterations until a candidate path that meets the transmission requirements is found.

[0135] The candidate paths obtained from the search are matched with priority transmission sequences to calculate the matching degree, and the path can be evaluated to determine whether it can meet the transmission requirements of data with different priorities. A set of qualified paths is selected based on a preset matching degree threshold. A dynamic graph neural network is used to establish a dynamic representation of the network topology, and dynamic programming is used to optimize the path set. During the optimization process, path quality parameters and load balancing constraints are considered simultaneously, and the optimal combination of transmission paths is obtained by adjusting the path allocation scheme.

[0136] For example, in an industrial control network, there is a data transmission requirement from source node S to target node D. The link parameters in the network are obtained: the transmission latency from node A to B is 2ms, the processing latency is 1ms, the bandwidth utilization is 60%, and the link reliability is 95%; the path passes through 3 relay nodes. Similarly, the parameters of other links are obtained to construct a complete set of evaluation metrics.

[0137] When creating the initial pheromone matrix, the pheromone value for all links is set to 0.1. After weighted normalization, a heuristic pheromone matrix reflecting link transmission performance is obtained. During path search, when an ant is at node A, it selects the next hop node based on the transition probability. Assuming path SABCD is selected, calculate the quality parameters of this path.

[0138] After multiple rounds of iterative search, several candidate paths are obtained. These candidate paths are matched with priority transmission sequences to select a subset of paths that meet the requirements for high-priority data transmission. Using dynamic graph neural networks and dynamic programming methods, a set of optimal transmission path combinations is obtained, which satisfies both transmission performance requirements and achieves a balanced distribution of network load.

[0139] In this embodiment, the multi-dimensional evaluation method provides a more accurate path quality metric, enabling path selection decisions to be based on more complete performance information and improving the accuracy of path selection. The pheromone matrix guides the search direction by recording historical search experience, while the heuristic information matrix provides real-time guidance based on the current network state. This dual guidance mechanism accelerates the convergence speed of the optimal path and reduces the computational overhead of invalid searches. The dynamic adjustment of pheromones ensures that the search process can respond promptly to changes in network state, avoiding getting trapped in local optima and improving the global optimality of path search results. The matching degree-based filtering mechanism ensures that data transmission needs of different priorities can obtain corresponding path support, improving the quality assurance capability of transmission services and significantly enhancing the performance and reliability of data transmission, providing strong support for the efficient operation of industrial control networks.

[0140] Figure 3 This is a comparison chart of the multi-priority service support capabilities of the 5G-based virtual controller data transmission optimization scheduling method for thermal power plants, presented in this embodiment of the invention. A grayscale heatmap visually illustrates the comparison of different algorithms' support capabilities for various priority services. Lower grayscale values ​​represent better performance, while higher values ​​represent stronger support capabilities. The data shows that for the highest priority service, critical control commands, this technical solution achieved the best score of 92.5, significantly higher than the traffic engineering algorithm (80.2), the hierarchical service algorithm (86.2), and the multi-constraint routing algorithm (74.1), with an average improvement rate of 12.3%. For real-time video streaming services, this technical solution scored 88.3, ​​9.2% higher than the other three algorithms on average. The traffic engineering algorithm performed better (82.5), while the hierarchical service algorithm only scored 76.3. For voice services, this technical solution and the hierarchical service algorithm performed similarly (86.2 vs. 84.2), but still significantly better than the other two algorithms. In terms of interactive data service support, this technical solution scored 82.4 points, which is 7.5% higher than the average score of other algorithms. It is worth noting that in low-priority services such as background transmission, the multi-constraint routing algorithm performed best (80.5 points), slightly higher than this technical solution's 78.5 points. This is the only indicator where this technical solution did not achieve the highest score, showing a difference of -1.3%. In terms of average support capability, this technical solution achieved 85.6 points, higher than the hierarchical service algorithm (79.1 points), the traffic engineering algorithm (77.5 points), and the multi-constraint routing algorithm (76.3 points), with an average improvement of 9.6%.

[0141] Figure 3 The grayscale differences in the data intuitively demonstrate the comprehensive advantages of this technical solution in supporting various priority services, especially in high-priority services where it has the best performance (lowest grayscale), thanks to the matching degree calculation based on path quality parameters and priority transmission sequences, and the path combination optimization mechanism.

[0142] In one alternative implementation,

[0143] By combining dynamic graph neural networks and dynamic programming methods to optimize the path set, and adjusting the allocation scheme of the path set with the path quality parameters and load balancing constraints as optimization objectives, the optimal transmission path combination is obtained, including:

[0144] A dynamic graph neural network is constructed, which maps network nodes and links to graph nodes and graph edges, respectively. Node topology feature vectors and edge state feature vectors are extracted, and node local features are generated based on the edge convolution mechanism to obtain the network topology features of the path set. Node importance is calculated and hierarchical clustering is performed on the network topology features, and graph pooling is performed to obtain a hierarchical path set representation.

[0145] The latency, bandwidth, packet loss rate, jitter, and stability indicators of each path in the path set are obtained and normalized to obtain path quality parameters. The load difference between paths is calculated based on the current load rate of each path, and load balancing constraints are constructed. The hierarchical path set representation and load balancing constraints are constructed into a state space, and the allocation and adjustment strategy of the path set is constructed into an action space. A value function is constructed based on the path quality parameters and load balancing constraints.

[0146] By combining dynamic programming, state transitions are performed based on the value function and Bellman equation, the allocation scheme of the path set is dynamically adjusted, the network state update feature vector is monitored, the graph structure is adjusted in real time and path feature recalculation is triggered, and the optimal transmission path combination that satisfies the path quality parameters and load balancing constraints is obtained through iterative solution.

[0147] In the construction phase of a dynamic graph neural network, the physical network needs to be mapped to a graph structure representation. This process maps each physical node in the network to a vertex in the graph, and the communication links between nodes to edges. For each node, its complete topological feature vector is extracted, containing key information such as the node's degree, centrality, and connections to other nodes. For each link, its state feature vector is extracted, covering dynamic operating parameters such as bandwidth capacity, current utilization, and transmission latency. Based on feature extraction, an edge convolution mechanism is used to process the local connectivity relationships of nodes. Through aggregation operations on neighborhood information, feature representations that characterize the local network properties of nodes are generated.

[0148] To better understand the overall characteristics of the network, the system conducts in-depth analysis of the extracted network topology features. By comprehensively considering multiple factors such as the node's location, connectivity, and current transmission load, the system calculates the importance weight of each node, performs hierarchical clustering, groups nodes with similar features in the network, and compresses fine-grained node feature information into hierarchical path set representations through graph pooling.

[0149] An optimization framework is constructed, collecting complete performance metrics for each available path in the path set. These metrics include multiple dimensions such as end-to-end transmission latency, available bandwidth capacity, packet loss rate, transmission jitter, and link stability. To ensure comparability across different dimensions, all raw metrics are normalized, mapping them to standardized numerical ranges to obtain standardized path quality parameters. The current load on each path is continuously monitored, and network load balancing constraints are established by calculating load differences between paths.

[0150] The obtained hierarchical path set representation and load balancing constraints are combined to construct the system's state space, which fully describes all possible operating states of the system. Possible path allocation adjustment strategies, such as dynamic adjustment of bandwidth allocation ratios and switching to backup paths, are constructed as the system's action space. Based on standardized path quality parameters and load balancing constraints, a value function is constructed to evaluate the merits of different combinations of states and actions.

[0151] The solution is obtained using dynamic programming. Based on the constructed value function and Bellman equation, the system calculates the value changes during state transitions and continuously seeks the strategy combination that maximizes the value function by adjusting the allocation scheme of the path set. During this process, the network's operating state is continuously monitored. When a significant change in the network state is detected, the feature vectors of nodes and edges are updated in a timely manner, and the graph structure is dynamically adjusted, triggering a recalculation of path features. Through multiple rounds of iterative optimization, an optimal transmission path combination scheme that satisfies both path quality requirements and ensures load balancing is obtained.

[0152] For example, in an industrial control network environment, there are multiple available transmission paths between source node S and target node D. A dynamic graph representation is constructed, where the feature vector of node A records its topological features such as degree 3 and betweenness centrality 0.4, and the feature vector of the edge connecting nodes A and B contains operating parameters such as current bandwidth utilization of 60% and average latency of 3ms. Through edge convolution processing, the system generates a local feature representation containing complete information about node A and its neighboring nodes B, C, and D. After node importance calculation and hierarchical clustering, the network nodes are divided into two layers: the core layer and the access layer. Performance metrics for the path set show that path P1 has a latency of 10ms, bandwidth utilization of 70%, and a packet loss rate of 0.1%; the load distribution of each path in the current network is 65%, 45%, and 55%, with a maximum load difference of 20%. The constructed state space includes the current path allocation scheme and network load status, while the action space covers operation options such as bandwidth allocation ratio adjustment and backup path switching. During the dynamic programming solution process, when a sudden increase in the load of node B is detected, its feature vector is immediately updated and the path is recalculated to obtain a set of optimal path combinations that can balance transmission performance and network load.

[0153] In this embodiment, local features of nodes are extracted through edge convolution mechanism to deeply mine the topological information contained in the network structure, providing a more accurate decision basis for path optimization. It not only retains the key topological information of nodes, but also reflects the dynamic operating status of links, providing more comprehensive decision information for path optimization. By monitoring the network status in real time and dynamically adjusting the allocation scheme, it can quickly respond to network changes, maintain continuous optimization of resource allocation, and significantly improve the utilization efficiency of network resources. Through real-time monitoring and dynamic adjustment of path load, the system achieves a balanced distribution of network load and improves the overall carrying capacity of the network.

[0154] In existing technologies, network path optimization mainly relies on static routing algorithms and simple load balancing strategies, which cannot accurately capture the dynamic changes in network topology, nor can they deeply understand the complex relationships between nodes. Path selection methods often consider certain performance indicators alone and lack the ability to comprehensively analyze multi-dimensional network characteristics. Existing load balancing mechanisms usually adopt fixed allocation strategies and cannot dynamically adjust resource allocation schemes according to network status, resulting in low network resource utilization efficiency.

[0155] This embodiment establishes an intelligent network path optimization scheme through innovative technologies such as dynamic graph neural network modeling, multidimensional feature extraction, and dynamic programming optimization. It has stronger network perception capabilities, higher optimization accuracy, and better environmental adaptability, providing effective technical support for improving network transmission performance.

[0156] Figure 4 This diagram illustrates the path optimization effect of a 5G-based virtual controller data transmission optimization and scheduling method for thermal power plants, as described in this embodiment of the invention. It showcases the extraction and optimization effect of dynamic graph neural networks on network topology features, and uses a node connection graph to visually present the path selection strategies of different algorithms. The graph contains 20 nodes (labeled as S, N1 to N18, and D) and 31 links, forming a typical enterprise WAN topology. The size of a node directly reflects its importance and weight in the network, while the thickness of the link represents the corresponding bandwidth capacity and utilization.

[0157] The calculation results of node importance in this technical solution show that nodes N7, N12, N15, and N18 have the highest importance weights, at 82%, 78%, 84%, and 79%, respectively, which matches the actual network traffic monitoring results by 92.3%. In comparison, the key nodes identified by the reinforcement learning-based path selection algorithm have a match rate of 78.5%, while the OSPF algorithm based on the shortest path only achieves 63.7%.

[0158] Regarding link characteristics, the link load rate between nodes N4 and N9 is 68%, with a latency of 2.8ms; the link load rate between N9 and N13 is 61%, with a latency of 3.2ms. These key link characteristics are accurately captured by this technical solution and incorporated into the path optimization decision-making process.

[0159] In selecting the transmission path from node S to node D, this technical solution employs a combination strategy of three parallel paths:

[0160] First path (solid line): S→N1→N4→N9→N13→N16→D, total latency 18.2ms, average link utilization 62.4%; Second path (solid line): S→N2→N5→N9→N14→N17→D, total latency 24.7ms, average link utilization 58.1%; Third path (solid line): S→N2→N5→N10→N14→N18→D, total latency 25.3ms, average link utilization 55.8%.

[0161] In contrast, the reinforcement learning-based path selection algorithm (dashed line) selected only two paths:

[0162] S→N1→N4→N9→N13→N16→D, total latency 25.3ms, average link utilization 73.8%.

[0163] S→N2→N5→N10→N15→N18→D, total latency 29.5ms, average link utilization 68.7%.

[0164] The OSPF algorithm based on the shortest path (point-to-line) selects only a single path S→N1→N4→N9→N13→N16→D, with a total latency of 32.1ms and a link utilization rate as high as 87.6%.

[0165] This technical solution compresses the original complex topology into a three-layer structure through path hierarchical clustering and graph pooling operations, reducing computational complexity by 63.5% while preserving 91.2% of the path feature information. It is particularly noteworthy that although the OSPF algorithm based on the shortest path selects the path with the shortest hop count, it suffers from high congestion and the highest latency because it does not consider the actual load on the links. This technical solution, by comprehensively considering multiple dimensions such as latency, bandwidth, and load balancing, achieves globally optimal path combination selection, significantly improving network transmission efficiency.

[0166] As clearly observed in the diagram, the path combination selected by this technical solution (solid line) avoids high-load nodes such as N15 and effectively balances network load through multi-path traffic distribution. When network traffic fluctuates, this technical solution can complete path recalculation and load distribution adjustment within 208ms, which is 37.5% faster than reinforcement learning algorithms and 42.3% shorter than OSPF's routing convergence time, demonstrating excellent adaptability and optimization performance.

[0167] A second aspect of the present invention,

[0168] An electronic device is provided, comprising:

[0169] processor;

[0170] Memory used to store processor-executable instructions;

[0171] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0172] A third aspect of the present invention,

[0173] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0174] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing and scheduling data transmission of a virtual controller in a thermal power plant based on a 5G network, characterized in that, include: The system acquires operational data from the field control unit of the thermal power plant, establishes a data mapping table and determines key operational data through the virtual controller, collects status parameters of the 5G network, establishes a network topology diagram, and plans multiple candidate transmission paths. A long short-term memory neural network is used to calculate the transmission characteristic parameters of the key operational data, a transmission priority evaluation function is constructed based on the transmission characteristic parameters, and a priority transmission sequence is generated based on the transmission priority evaluation function. The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit. The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The path quality parameters are then matched with the priority transmission sequence. The optimal transmission path combination is selected and a transmission time slot is allocated to the optimal transmission path combination in the time slot resource allocation table. According to the time slot resource allocation table, the key operating data is transmitted to the virtual controller through the optimal transmission path combination, and control commands are generated and sent to the field control unit through the reverse path in the optimal transmission path combination.

2. The method according to claim 1, characterized in that, The virtual controller establishes a data mapping table and determines key operational data, collects 5G network status parameters, establishes a network topology diagram, and plans multiple candidate transmission paths, including: The system acquires the operating data of the field control unit of the thermal power plant, establishes a data mapping table through the virtual controller, records data identifiers and data feature information, and uses a support vector classifier to classify the operating data based on the data feature information to determine key operating data. Collect 5G network status parameters, calculate node availability score based on network node status in the status parameters, calculate link weight coefficient based on link status, and integrate the node availability score and the link weight coefficient to generate a network resource status matrix. A network topology diagram is established by using the minimum tree generation algorithm in conjunction with the network resource state matrix. A path cost function is constructed based on the node availability score and the link weight coefficient. Multiple candidate transmission paths are planned in the network topology diagram according to the path cost function.

3. The method according to claim 1, characterized in that, The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. A transmission priority evaluation function is constructed based on the transmission characteristic parameters. A priority transmission sequence is generated based on the transmission priority evaluation function, including: The transmission characteristic parameters of the key operational data are calculated using a long short-term memory neural network. The long short-term memory neural network extracts features through a forget gate, an input gate, and an output gate. It performs unit state updates based on the gating combination of the current input and the historical state, and performs hidden layer state updates using the nonlinear transformation between the output gate and the unit state to obtain the transmission characteristic parameters of the key operational data. The transmission characteristic parameters include timeliness indicators, data value, historical transmission success rate, and bandwidth utilization rate. A transmission priority evaluation function is constructed by minimizing the mean squared error between the predicted priority and the actual priority and by calculating the feature weight vector and the weight regularization term. The inner product of the feature weight vector and the transmission feature parameters is used as the transmission priority evaluation function. The score result of the transmission priority evaluation function is calculated, and the key operating data is divided into multiple priority levels according to the score result of the transmission priority evaluation function. Within the priority level, a dynamic adjustment factor for waiting time and transmission failure count is set. The final priority score is obtained by multiplying the score result of the transmission priority evaluation function and the dynamic adjustment factor. The priority transmission sequence is generated according to the final priority score.

4. The method according to claim 1, characterized in that, Calculate the score result of the transmission priority evaluation function, and divide the key operational data into multiple priority levels based on the score result of the transmission priority evaluation function, including: Collect the correlation information between the key operational data, calculate the causal influence strength of the key operational data based on Bayesian network, and generate a causal strength matrix; The causal strength matrix is ​​input into the transmission priority evaluation function, which includes a data criticality score item, a transmission timeliness score item, and a network resource consumption score item. Based on the causal strength matrix, the direct causal impact value and indirect causal impact value of the critical operational data are calculated. A service quality indicator set is constructed, which includes throughput, latency, reliability, and jitter indicators. Based on the service quality indicator set, a multi-objective optimization function is constructed to obtain the weighted deviation between each indicator and the target value. The direct causal impact value, the indirect causal impact value, and the calculation result of the multi-objective optimization function are then weighted and combined with the data criticality score item, the transmission timeliness score item, and the network resource consumption score item to obtain the score result. The scoring results are input into the priority level division model. The priority level division threshold is dynamically determined based on the causal intensity distribution in the scoring results. The key operation data is divided into three priority levels according to the priority level division threshold. Priority resource protection is implemented for the highest priority level, and resource reuse is implemented for the lowest priority level. Service quality compensation adjustment is achieved by dynamically adjusting the resource allocation ratio, monitoring service quality indicator deviations in real time, and adaptively compensating for bandwidth allocation.

5. The method according to claim 1, characterized in that, The data transmission cycle is divided into multiple time slot units, and a time slot resource allocation table is established. The time slot resource allocation table records the availability status of each time slot unit, including: The data transmission period is obtained, and the transmission period is divided into multiple time slot units by using the minimum time granularity of data transmission. Create a time slot resource allocation table, and count and record the index number, occupancy identifier, remaining bandwidth and allocable duration of each time slot unit. The index number identifies the position of the time slot unit in the transmission cycle, the occupancy identifier represents the occupancy status of the time slot unit, and the remaining bandwidth and allocable duration represent the resource quantity of the time slot unit. The availability status of each time slot unit is detected and updated in the time slot resource allocation table.

6. The method according to claim 1, characterized in that, The ant colony algorithm is used to calculate the path quality parameters of the candidate transmission paths. The matching degree between the path quality parameters and the priority transmission sequence is calculated, and the optimal transmission path combination is selected, including: The link delay parameters are obtained by acquiring the transmission delay and processing delay between each hop node on the path. The bandwidth utilization rate is obtained by evaluating the ratio of the available bandwidth of the path to the total bandwidth of the link. A link reliability model is established based on the historical transmission success rate and link stability. The hop count cost is obtained by counting the number of relay nodes traversed by the path. The link delay parameters, the bandwidth utilization rate, the link reliability, and the hop count cost are constructed into a path quality evaluation index set. A pheromone matrix is ​​created based on the path quality evaluation index set. A heuristic information matrix is ​​generated by performing weighted normalization calculation on the path quality evaluation index set. The pheromone matrix and the heuristic information matrix are used as initial parameters for path search. By combining the pheromone matrix and the heuristic information matrix, the transition probability between nodes is calculated to generate candidate transmission paths. The path quality parameters are obtained by normalizing and weighting the path quality evaluation index set of the candidate transmission paths using the ant colony algorithm. The path quality parameters characterize the transmission performance of the candidate transmission paths. The pheromone increment is calculated based on the path quality parameters, and the pheromone matrix is ​​updated by decaying in combination with the pheromone evaporation coefficient. The candidate transmission path generation and the path quality parameter calculation are repeated based on the updated pheromone matrix until a candidate transmission path that meets the requirements is found. The path quality parameters of the candidate transmission paths obtained by the search are matched with the priority transmission sequence. Based on the preset matching degree threshold, a set of paths that meet the requirements of the priority transmission sequence is obtained. The set of paths is combined and optimized by combining dynamic graph neural network and dynamic programming method. The allocation scheme of the set of paths is adjusted with the path quality parameters and load balancing constraints as optimization objectives to obtain the optimal combination of transmission paths.

7. The method according to claim 6, characterized in that, By combining dynamic graph neural networks and dynamic programming methods to optimize the path set, and adjusting the allocation scheme of the path set with the path quality parameters and load balancing constraints as optimization objectives, the optimal transmission path combination is obtained, including: A dynamic graph neural network is constructed, which maps network nodes and links to graph nodes and graph edges, respectively. Node topology feature vectors and edge state feature vectors are extracted, and node local features are generated based on the edge convolution mechanism to obtain the network topology features of the path set. Node importance is calculated and hierarchical clustering is performed on the network topology features, and graph pooling is performed to obtain a hierarchical path set representation. The latency, bandwidth, packet loss rate, jitter, and stability indicators of each path in the path set are obtained and normalized to obtain path quality parameters. The load difference between paths is calculated based on the current load rate of each path, and load balancing constraints are constructed. The hierarchical path set representation and load balancing constraints are constructed into a state space, and the allocation and adjustment strategy of the path set is constructed into an action space. A value function is constructed based on the path quality parameters and load balancing constraints. By combining dynamic programming, state transitions are performed based on the value function and Bellman equation, the allocation scheme of the path set is dynamically adjusted, the network state update feature vector is monitored, the graph structure is adjusted in real time and path feature recalculation is triggered, and the optimal transmission path combination that satisfies the path quality parameters and load balancing constraints is obtained through iterative solution.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images