A cloud platform-based teaching recording and broadcasting management method and system

Abstract

The application discloses a kind of teaching recording broadcast management method and system based on cloud platform, it is related to resource allocation technical field, the method of the present application includes collecting the full-link historical operation data of teaching recording broadcast task and the data of to-be-recorded teaching task, constructs historical resource consumption sequence based on full-link historical operation data;The teaching task duration in to-be-recorded teaching task data is time-sliced, and is endowed with tidal weight, and the historical resource consumption sequence is fitted by polynomial fitting function containing tidal weight, to generate load prediction function and channel gain matrix;The cooperative optimization scheduling strategy based on inner and outer layer nested architecture is executed, and the optimal scheduling scheme set is obtained;Dynamic decision locking based on service period characteristics, identify the campus service period to which the current physical time belongs, calculate the weighted Euclidean distance of scheme in optimal scheduling scheme set and virtual ideal scheme, obtain the final scheduling scheme and generate control instruction.

Description

Technical Field

[0001] This invention relates to the field of resource allocation technology, and in particular to a teaching recording and broadcasting management method and system based on a cloud platform. Background Technology

[0002] In the current digital teaching scenario, the teaching recording and broadcasting system undertakes the core task of transforming the on-site teaching process into digital resources. A large number of teachers need to conduct live teaching in the classroom environment and record courses simultaneously for subsequent resource sharing, online learning, and review. In order to break the information silos of traditional local deployment, the cloud-based teaching recording and broadcasting management model realizes centralized control and resource scheduling of large-scale concurrent recording and broadcasting tasks by moving computing, storage, and management functions to the cloud or edge servers, which greatly improves the production efficiency and management convenience of teaching resources.

[0003] In existing cloud platform recording and broadcasting management technologies, a cloud-edge collaborative architecture is typically used to solve the problem of remote control of terminal devices. Chinese patent CN113630463B discloses a smart classroom recording and broadcasting terminal device management system and method based on KubeEdge. It utilizes the collaboration between the cloud data center and the recording and broadcasting terminal, and realizes the management, command issuance and status feedback of the recording and broadcasting terminal nodes through Kubernetes cluster and KubeEdge architecture. Although the above-mentioned existing technologies have solved the connection and device management problems of recording and broadcasting terminals, they still have significant shortcomings in resource scheduling in multi-task concurrent scenarios. Existing technologies usually treat recording and broadcasting tasks as general computing instructions, failing to fully explore the special characteristics of teaching business in the time dimension, which can easily lead to insufficient or wasted resources. When scheduling tasks, existing technologies often consider computing resources and network transmission environment separately, lacking modeling and perception of wireless channel status, and failing to coordinate the optimization of the load capacity of computing nodes with the signal-to-interference-plus-noise ratio and transmission power of wireless links. In campus scenarios with complex wireless network environments or severe interference, this can easily lead to video stream transmission stuttering or excessive energy consumption. Summary of the Invention

[0004] The technical problem solved by this invention is that existing technologies usually treat recording and broadcasting tasks as general computing instructions, failing to fully explore the special characteristics of teaching business in the time dimension. When scheduling tasks, existing technologies often consider computing resources and network transmission environment separately, lacking modeling and perception of wireless channel status, and failing to coordinate the load capacity of computing nodes with the signal-to-interference-plus-noise ratio and transmission power of wireless links. In campus scenarios with complex wireless network environments or severe interference, this can easily lead to video stream transmission stuttering or excessive power consumption.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a cloud-based teaching recording and broadcasting management method, comprising the following steps: Step S1: Collect the full-link historical operation data of the teaching recording and broadcasting task and the teaching task data to be recorded and broadcasting, and construct the historical resource consumption sequence based on the full-link historical operation data; Step S2: Divide the duration of the teaching task in the data to be recorded into time slices and assign tidal weights. Fit the historical resource consumption sequence with a polynomial fitting function containing tidal weights to generate a load prediction function and a channel gain matrix. Step S3: Execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture on the randomly generated scheduling scheme to obtain the comprehensive fitness of the scheduling scheme and select the optimal scheduling scheme set; The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computing resources; in the inner continuous optimization process, calculating the comprehensive fitness of the scheduling scheme based on the channel gain matrix and combined with the convex optimization algorithm. Step S4: Dynamic decision locking based on business time period characteristics, identify the campus business time period to which the current physical time belongs, calculate the weighted Euclidean distance between the scheduling scheme in the optimal scheduling scheme set and the virtual ideal scheme, obtain the final scheduling scheme and generate control instructions.

[0006] Preferably, the end-to-end historical operational data includes: historical container operational metrics data, historical network transmission metrics data, and historical static data; The historical container operation metrics data include the CPU utilization and memory usage of the recording container; The historical network transmission index data includes the signal receiving power of the transmitting end in the recording classroom. The historical static data includes the rated transmission power of the recording classroom transmitter, the upper limit of the physical resource capacity of the heterogeneous computing nodes, and the ambient background noise power. The CPU utilization and memory usage at the relative sampling time are used to construct resource consumption values. The relative sampling time and resource consumption values ​​are paired to construct a historical resource consumption sequence. The data for the teaching tasks to be recorded includes the expected start time and expected course duration.

[0007] Preferably, in step S2, the process of generating the load prediction function specifically includes: The preset course duration of the teaching task to be recorded is divided into time segments, and the preset course duration is divided into rising period, steady period and ebb period. The rising period is defined as the first preset proportion from the start of the recording to the preset course duration, and a first tidal weight is assigned to the rising period. The stable period is defined as the second preset proportion from the end of the rising period to the preset course duration, and a second tidal weight is assigned to the stable period. The ebb tide period is the preset overtime tolerance period from the end of the stable period to the end of the preset course duration, and a third tide weight is assigned to the ebb tide period. The third tidal weight value is greater than the first tidal weight value and the second tidal weight value; Construct a polynomial fitting function containing unknown coefficients, construct a weighted residual sum of squares based on the polynomial fitting function and tidal weights, substitute the historical resource consumption sequence into the weighted residual sum of squares, and use the least squares method to solve for the unknown coefficients that minimize the weighted residual sum of squares. Substitute the unknown coefficients obtained from the solution into the polynomial fitting function to obtain the load prediction function.

[0008] Preferably, in step S2, the process of constructing the channel gain matrix specifically includes: Construct a local spatial coordinate system on campus to obtain the spatial coordinates of all recording classroom transmitters and all heterogeneous computing node receivers. The difference between the rated transmit power and the signal receive power at the recording classroom transmitter is used as the historical signal path loss value. The Euclidean distance between the transmitter and receiver and the historical signal path loss value between the transmitter and receiver are used as signal path loss sample pairs. The path loss sample pairs are fitted using the logarithmic distance path loss formula and linear regression is performed. The slope of the logarithmic distance path loss formula is used as the signal path loss exponent. The path loss exponent is substituted back into the logarithmic distance path loss formula to obtain the spatial signal attenuation function. Traverse the set of transmitting points and the set of receiving points, calculate the Euclidean distance between any transmitting end and any heterogeneous computing node, substitute the Euclidean distance into the spatial signal attenuation function, and calculate the channel gain value. Construct a channel gain matrix, the elements of which are the channel gain values ​​between the transmitter and the corresponding heterogeneous computing nodes.

[0009] Preferably, step S3 includes the following sub-steps: Step S301: Obtain the teaching task to be recorded and the available heterogeneous computing nodes, initialize the genetic algorithm with a preset population size, encode each chromosome as a scheduling scheme, and randomly generate several scheduling schemes to form a candidate scheme set; The scheduling scheme involves allocating the teaching tasks to be recorded to available heterogeneous computing nodes.

[0010] Preferably, step S3 further includes the following sub-steps: Step S302: Traverse all candidate schemes in the candidate scheme set and calculate the peak load of each heterogeneous computing node in each candidate scheme. The peak load is the sum of the resource requirements of the sending end of all teaching tasks to be recorded and broadcast allocated to the heterogeneous computing node. If the peak load of any of the heterogeneous computing nodes is greater than or equal to the physical resource limit of the heterogeneous computing node, then the corresponding candidate solution is given a first penalty value.

[0011] Preferably, step S3 further includes the following sub-steps: Step S303: For the transmitting end of each teaching task to be recorded in the candidate schemes, calculate the signal-to-interference-plus-noise ratio (SINR) of the transmitting end. The mathematical expression for the SINR is: ; in, For the first Signal-to-interference-to-noise ratio for each task For the sending end The channel gain of the heterogeneous computing nodes to which it is allocated. For the sending end The transmission power, For the sending end For the sending end Interference channel gain of the receiving node For the sending end The transmission power, Environmental noise power; For the signal-to-interference-plus-noise ratio (SINR) of the transmitting end of any teaching task to be recorded, a constraint formula is constructed, wherein the SINR is greater than or equal to a preset signal quality threshold. The minimum power vector that satisfies the constraint formula is solved using a convex optimization algorithm. If there are negative values ​​in the minimum power vector, then a second penalty value is assigned to the corresponding candidate scheme; Step S304: Calculate the comprehensive fitness value of each scheduling scheme in the candidate scheme set. The comprehensive fitness value is calculated by: calculating the weighted sum of the transmission energy consumption and the computation energy consumption of the scheduling scheme, and adding the weighted result, the first penalty value and the second penalty value as the comprehensive fitness value. Step S305: Sort the candidate schemes in ascending order according to the comprehensive fitness value, select the candidate schemes corresponding to the top several comprehensive fitness values ​​as the parent chromosomes, perform gene crossover and site mutation on the parent chromosomes to generate the next generation population, and obtain the optimal scheduling scheme set after iterative optimization.

[0012] Preferably, in step S4, obtaining the final scheduling scheme specifically includes: The physical time is divided into peak business periods and off-peak business periods. The current physical time is obtained, and the campus business time period to which the current physical time belongs is identified. The campus business time period includes peak business periods and off-peak business periods. The campus business time period is divided based on a preset time period. A virtual scheduling ideal scheme is defined, in which the normalized transmission energy consumption and normalized computation energy consumption are both 0. Determine the decision weights for transmission energy consumption and computing energy consumption based on the campus business time period to which the current physical time belongs; For each scheduling scheme in the optimal scheduling scheme set, calculate the weighted Euclidean distance between the normalized transmission energy consumption and the normalized computational energy consumption of the virtual ideal scheduling scheme. The scheduling scheme with the minimum weighted Euclidean distance is selected as the final scheduling scheme.

[0013] Preferably, in step S4, generating control commands specifically includes: Based on the final scheduling scheme, node allocation instructions and power control instructions are generated for each teaching task to be recorded and broadcast, and then sent to the heterogeneous computing nodes and the sending end of the teaching task to be recorded and broadcast for execution.

[0014] A cloud-based teaching recording and broadcasting management system includes: an acquisition module, a prediction module, an optimization module, and a management module; The acquisition module is used to collect the full-link historical operation data of teaching recording and broadcasting tasks and the data of teaching tasks to be recorded and broadcasting, and to construct a historical resource consumption sequence based on the full-link historical operation data. The prediction module is used to divide the duration of the teaching tasks in the data to be recorded into time slices and assign tidal weights. It then uses a polynomial fitting function containing tidal weights to fit the historical resource consumption sequence and generate a load prediction function and a channel gain matrix. The optimization module is used to execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture to obtain the optimal scheduling scheme set. The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computing resources; in the inner continuous optimization process, solving for the minimum transmit power based on the channel gain matrix and combined with a convex optimization algorithm. The management module is used for dynamic decision locking based on business time period characteristics, identifying the campus business time period to which the current physical time belongs, calculating the weighted Euclidean distance between the optimal scheduling scheme set and the virtual ideal scheme, obtaining the final scheduling scheme and generating control commands.

[0015] The beneficial effects of this invention are as follows: This invention uses tidal weights to fit time-sliced ​​teaching task durations, generating a load prediction function that accurately predicts the dynamic characteristics of resource consumption at different course stages. In particular, it enhances the fitting sensitivity to resource demands at the end of a course and during extended periods. By constructing a channel gain matrix and combining it with a collaborative optimization scheduling strategy based on an inner and outer nested architecture, it achieves collaborative optimization of computing resources and wireless transmission resources. In the outer discrete optimization, the load prediction function is called to perform hard constraint verification of computing resources, ensuring that the task load allocated to heterogeneous computing nodes does not exceed their physical upper limit. In the inner continuous optimization, the minimum transmit power is solved based on the channel gain matrix and using a convex optimization algorithm, effectively reducing transmit power and transmission energy consumption while ensuring that the video stream signal-to-interference-plus-noise ratio meets the quality threshold. Attached Figure Description

[0016] Figure 1 The present invention provides a flowchart of a cloud-based teaching recording and broadcasting management method according to an embodiment of the present invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example 1, referring to Figure 1 This provides a cloud-based teaching recording and broadcasting management method, which includes the following steps: Step S1, collect the full-link historical operation data of teaching recording and broadcasting tasks and the teaching task data to be recorded and broadcasting, and construct a historical resource consumption sequence based on the full-link historical operation data; Step S2: Divide the duration of the teaching task in the data to be recorded into time slices and assign tidal weights. Fit the historical resource consumption sequence with a polynomial fitting function containing tidal weights to generate a load prediction function and a channel gain matrix. Step S3: Execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture on the randomly generated scheduling scheme to obtain the comprehensive fitness of the scheduling scheme and select the optimal scheduling scheme set; The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computing resources; in the inner continuous optimization process, calculating the comprehensive fitness of the scheduling scheme based on the channel gain matrix and combined with the convex optimization algorithm. Step S4: Dynamic decision locking based on business time period characteristics, identify the campus business time period to which the current physical time belongs, calculate the weighted Euclidean distance between the scheduling scheme in the optimal scheduling scheme set and the virtual ideal scheme, obtain the final scheduling scheme and generate control instructions.

[0019] This invention collects historical operational data across the entire link and constructs a historical resource consumption sequence. By combining time-slicing and tidal weighting of teaching task durations, it generates a load prediction function and a channel gain matrix, enabling the prediction of resource requirements for recording and broadcasting tasks. Through the execution of a collaborative optimization scheduling strategy based on an inner and outer nested architecture, the load prediction function is called in the outer discrete optimization layer to perform hard constraint verification of computing resources. In the inner continuous optimization layer, the minimum transmit power is solved based on the channel gain matrix and combined with a convex optimization algorithm, achieving joint optimization of computing resource allocation and wireless transmission power control. Through dynamic decision locking based on service time period characteristics, the final scheduling scheme that best matches the current campus service time period characteristics is selected from the set of optimal scheduling schemes, and control commands are generated, realizing closed-loop intelligent management from data prediction, collaborative optimization to dynamic decision-making.

[0020] The end-to-end historical operational data includes: historical container operation metrics data, historical network transmission metrics data, and historical static data; Historical container performance metrics include CPU utilization and memory usage of the recording container; Historical network transmission metrics data include the signal reception power of the transmitting end in the recording classroom; Historical static data includes the rated transmission power of the recording classroom transmitter, the upper limit of the physical resource capacity of heterogeneous computing nodes, and the ambient background noise power; The CPU utilization and memory usage at the relative sampling time are used to construct resource consumption values. The relative sampling time and resource consumption values ​​are paired to construct a historical resource consumption sequence. The data for the teaching tasks to be recorded includes the expected start time and expected course duration.

[0021] In one specific embodiment of the present invention, the end-to-end historical operation data comes from container-side data, network-side data and spatial static data, and is aggregated to the cloud platform database via the network. The time period for the full-link historical operation data is from the start of the recording (start of teaching) to the end of the recording (end of teaching). The operation data is collected according to the sampling frequency with the start time of the recording as 0. The sampling timestamp of the operation data includes the relative sampling time relative to 0 and the physical timestamp. The relative sampling times and resource consumption values ​​within the time period from the start to the end of the recording are paired and sorted by time to form discrete time-series data, which serves as the historical resource consumption sequence. It records the resource usage of the recording container when it performed tasks in the past; The relative timestamp is an offset relative to the course start time, which eliminates the time differences between classes on different dates and extracts the common features of the course. The resource consumption value includes two dimensions: CPU utilization and memory usage. After normalizing the CPU utilization and memory usage, a weighted sum is performed to obtain the resource consumption value. In this embodiment, the weight of CPU utilization is 0.7 and the weight of memory usage is 0.3. The ambient background noise power is directly read from the underlying driver interface (OAM interface of 5G base station) by the monitoring probe of the cloud platform, and converted into a linear power value in watts.

[0022] Step S2, the process of generating the load prediction function specifically includes: The pre-set course duration of the teaching task to be recorded is divided into time segments, and the pre-set course duration is divided into rising period, steady period and ebb period. The rising phase is defined as the first preset proportion from the start of the recording to the preset course duration, and the rising phase is assigned a first tidal weight. The stable period is defined as the second preset proportion from the end of the rising period to the preset course duration, and a second tidal weight is assigned to the stable period. The ebb tide period is the preset overtime tolerance period from the end of the steady period to the end of the preset course duration, and a third tide weight is assigned to the ebb tide period. The third tidal weight value is greater than the first tidal weight value and the second tidal weight value; it is used to enhance the fitting sensitivity to the end of the course and the overtime period during the fitting process. Construct a polynomial fitting function containing unknown coefficients, construct a weighted residual sum of squares based on the polynomial fitting function and tidal weights, substitute the historical resource consumption sequence into the weighted residual sum of squares, and use the least squares method to solve for the unknown coefficients that minimize the weighted residual sum of squares. Substitute the unknown coefficients obtained from the solution into the polynomial fitting function to obtain the load prediction function.

[0023] In a specific embodiment of the present invention, although ordinary polynomial regression can fit the curve, it is not sensitive enough to the abrupt change point (the moment the bell rings to signal the end of get out of class), and is prone to oversmoothing, which leads to an underestimation of the peak value or an overslow prediction of the release of resources after class. Therefore, it is necessary to introduce a tidal factor to correct the load prediction. In this embodiment, the preset course duration of the teaching course is obtained. (The preset course duration data comes from the teaching plan; in this example, it is 45 minutes.) The teaching course is divided into three periods: the rising period, the stable period, and the ebb period. like ,but Located in the upward phase, among which, The rising period proportionality coefficient is taken in this embodiment. ~ (That is, the first 10%-20% of the course), this period corresponds to the teacher turning on the equipment, debugging the courseware, students connecting one after another, and the start of the streaming. like ,but Located in a stable period, among which This embodiment uses the threshold for the end of the plateau period as an example. ~ (That is, when the course is 80%-90% complete), the teaching process is stable during this period, and the changes in the screen are mostly PPT switching or blackboard writing; like ,but Located during low tide, among which, The preset tolerance window for overtime is 15 minutes in this embodiment. This period is the stage near the end of get out of class and the stage where overtime occurs. This is a high-risk area for resource scheduling. If the course is misjudged as over and resources are taken back in advance, the recorded content of the overtime part will be lost. like If so, recording will be forcibly stopped; Selecting a third-order polynomial model As a benchmark function; Solving using the least squares method yields the weighted sum of squared residuals. Minimum coefficient vector The mathematical expression for the weighted sum of squared residuals is: ; in, This represents the total fitting error. According to The tidal weights allocated to the time period. As a historical reality, , , and The coefficients to be solved are... Relative time; After obtaining the coefficient vector, the coefficients in the coefficient vector are substituted into the third-order polynomial model to obtain the load prediction function that can characterize the changes in resource consumption of the recorded teaching course at different course stages.

[0024] This invention divides the preset course duration of the teaching task to be recorded into a rising period, a stable period, and a receding period (including a period of tolerance for overtime), and assigns different tidal weights to different periods. In particular, it gives the receding period (overtime stage) the largest third weight value. By fitting the historical resource consumption sequence with a polynomial fitting function that includes tidal weights, the resulting load prediction function can more accurately capture the resource consumption characteristics of different stages of the course. It significantly enhances the fitting sensitivity to sudden changes in resource demand at the end of the course and during overtime, effectively avoiding the problems of insufficient sensitivity to sudden change points, easy oversmoothing, and underestimation of peak values ​​or slow prediction of resource release that ordinary polynomial regression can solve. Thus, it can more accurately predict the resource demand during high-risk periods and prevent the loss of recorded content due to insufficient resource estimation.

[0025] Step S2, the process of constructing the channel gain matrix specifically includes: Construct a local spatial coordinate system on campus to obtain the spatial coordinates of all recording classroom transmitters and all heterogeneous computing node receivers. The difference between the rated transmit power and the signal receive power at the recording classroom transmitter is used as the historical signal path loss value. The Euclidean distance between the transmitter and receiver and the historical signal path loss value between the transmitter and receiver are used as signal path loss sample pairs. The path loss sample pairs are fitted using the logarithmic distance path loss formula and solved by linear regression. The slope of the logarithmic distance path loss formula is used as the signal path loss exponent. The path loss exponent is substituted back into the logarithmic distance path loss formula to obtain the spatial signal attenuation function. Traverse the set of transmitting points and the set of receiving points, calculate the Euclidean distance between any transmitting end and any heterogeneous computing node, substitute the Euclidean distance into the spatial signal attenuation function, and calculate the channel gain value. Construct a channel gain matrix, where each element represents the channel gain value between the transmitter and the corresponding heterogeneous computing node.

[0026] In a specific embodiment of the present invention, a campus local spatial coordinate system is constructed, with the southwest corner of the campus as the origin, and a spatial rectangular coordinate system covering the entire campus is constructed. The coordinates of the recording classroom transmitter (classroom) and the heterogeneous computing node receiver (back-end processing server distributed in the campus network) are obtained. The signal receiving power is read from the log of the heterogeneous computing node receiver, and the difference between the signal receiving power and the rated transmission power of the recording classroom transmitter is used as the historical signal path loss value. Based on the constructed campus local coordinate system, the coordinates of the historical recording classroom transmitter and the historical heterogeneous node computing end are obtained and the Euclidean distance is calculated. The historical signal path loss value is extracted, and the historical signal path loss value and the corresponding Euclidean distance constitute a historical sample. Signals travel through the air and weaken with distance; therefore, using all historical samples, a method containing unknowns is employed. The standard logarithmic distance path loss formula (SLM) is used to quantify the signal attenuation process. The formula for the logarithmic distance path loss model is as follows: ; in, The distance is Path loss value at time, Reference distance Signal path loss value at that time, is the signal path loss exponent, and d is the actual distance between the transmitter and receiver; The Euclidean distance in each historical sample is used as... Historical signal path loss value as ,by As the independent variable, As a dependent variable; Linear regression is performed using the least squares method to find the fitted slope ( This refers to the path loss index of the signal in the current campus environment. Substituting into the standard logarithmic distance path loss formula, we obtain the definite logarithmic distance path loss formula. The preset reference distance (1m in this embodiment; the application scenario of this embodiment is a campus LAN and an indoor recording classroom, which is a typical microcell communication environment, indicating that within 1 meter of the transmitting antenna, the signal propagation follows the free space propagation model and is not affected by wall reflection and multipath effect). The logarithmic distance path loss formula is transformed into a linear formula by taking the negative and antilogarithm, yielding the channel gain function. The mathematical expression of the channel gain function is: ; in, For channel gain, Let be the Euclidean distance between the sending and receiving ends. The signal path loss index; It is a constant coefficient related to the reference distance and frequency, and its physical meaning is the free-space path gain at the reference distance. The calculation formula is: ,in At the speed of light, This is the wireless signal carrier frequency (network configuration parameter, which can be read directly). This is the set reference distance; Calculate the Euclidean distance between any transmitter and any heterogeneous computing node. Substitute the Euclidean distance into the channel gain function to calculate the channel gain value between the transmitter and the heterogeneous computing node. Construct a channel gain matrix through a preset mapping relationship. The preset mapping relationship specifically includes: All transmitters and heterogeneous computing nodes are numbered. The element values ​​of the channel gain matrix correspond to the channel gain values ​​between the transmitter and the heterogeneous computing node. The element in the first row and first column is the channel gain value between transmitter No. 1 and heterogeneous computing node No. A, and so on.

[0027] This invention constructs a campus local spatial coordinate system. Based on the spatial coordinates of the transmitter and receiver and the historical difference between the received signal power and the rated transmit power, it uses the logarithmic distance path loss formula to perform linear regression on the path loss sample pairs to obtain the signal path loss exponent. Substituting this exponent back into the formula, it obtains the spatial signal attenuation function. Then, it calculates the channel gain value between any transmitter and heterogeneous computing nodes and constructs a channel gain matrix. The obtained channel gain matrix can accurately reflect the wireless channel quality between the transmitter of each recording classroom and the receiver of each heterogeneous computing node. This provides key channel environment parameters for solving the minimum transmit power and reducing transmission energy consumption in subsequent collaborative optimization, which helps to reduce signal interference and improve transmission stability.

[0028] Step S3 includes the following sub-steps: Step S301: Obtain the teaching task to be recorded and the available heterogeneous computing nodes, initialize the genetic algorithm with a preset population size (the population size is 50 in this embodiment), encode each chromosome as a scheduling scheme, and randomly generate several scheduling schemes to form a candidate scheme set; The scheduling scheme involves allocating the teaching tasks to be recorded to available heterogeneous computing nodes.

[0029] This invention uses a genetic algorithm with a preset population size to encode each scheduling scheme as a chromosome and randomly generate a set of candidate schemes. This provides a basic solution space search framework for subsequent collaborative optimization scheduling strategies. By adopting the heuristic global optimization method of genetic algorithm, it can effectively solve complex combinatorial optimization problems such as multi-task resource allocation. Through iterative evolution, it approaches the globally optimal or near-optimal scheduling scheme, providing an effective solution for finding low-energy-consumption scheduling schemes under the premise of satisfying resource constraints and signal quality constraints.

[0030] Step S3 also includes the following sub-steps: Step S302: Traverse all candidate schemes in the candidate scheme set and calculate the peak load of each heterogeneous computing node in each candidate scheme. The peak load is the sum of the resource requirements of the sending end of all teaching tasks to be recorded and broadcast allocated to the heterogeneous computing node. If the peak load of any heterogeneous computing node is greater than or equal to the physical resource limit of the heterogeneous computing node, the corresponding candidate scheme is given a first penalty value (109 in this embodiment).

[0031] This invention traverses the candidate solution set and calculates the peak load of each heterogeneous computing node. When the peak load of any node reaches its physical resource limit, a first penalty value is assigned to the corresponding candidate solution. This mechanism effectively performs hard constraint verification of computing resources in the outer discrete optimization process, ensuring that the scheduling scheme generated during the optimization process will not cause any computing node to overload, guaranteeing the feasibility of computing resource allocation, and preventing system crashes or serious degradation of service quality due to resource overruns.

[0032] Step S3 also includes the following sub-steps: Step S303: For the transmitting end of each teaching task to be recorded in the candidate schemes, calculate the signal-to-interference-plus-noise ratio (SINR) of the transmitting end. The mathematical expression for the SINR is: ; in, For the first Signal-to-interference-to-noise ratio for each task For the sending end The channel gain of the heterogeneous computing nodes to which it is allocated. For the sending end The transmission power, For the sending end For the sending end Interference channel gain of the receiving node For the sending end The transmission power, Environmental noise power; molecular Indicates the sending end Useful signal strength sent to heterogeneous computing nodes, denominator Represents all others The sum of interference generated by each transmitter; For the signal-to-interference-plus-noise ratio (SINR) of the transmitting end of any teaching task to be recorded, a constraint formula is constructed, wherein the SINR is greater than or equal to a preset signal quality threshold. The minimum power vector that satisfies the constraint formula is solved using a convex optimization algorithm. If there is a negative value in the minimum power vector, a second penalty value is assigned to the corresponding candidate scheme (106 in this embodiment). Step S304: Calculate the comprehensive fitness value of each scheduling scheme in the candidate scheme set. The comprehensive fitness value is calculated by: calculating the weighted sum of the transmission energy consumption and the computation energy consumption of the scheduling scheme, and adding the weighted result, the first penalty value and the second penalty value as the comprehensive fitness value. Step S305: Sort the candidate schemes in ascending order according to the size of the comprehensive fitness value, select the candidate schemes corresponding to the top several comprehensive fitness values ​​as the parent chromosomes, perform gene crossover and site mutation on the parent chromosomes to generate the next generation population, and obtain the optimal scheduling scheme set after iterative optimization. In one specific embodiment of the present invention, the teaching task to be recorded is obtained. The classroom where the recorded teaching task is located and the available heterogeneous computing node receiving end. ; There are within the teaching area One concurrent recording and broadcasting task and Heterogeneous computing nodes; The resource requirements of the teaching task to be recorded are calculated using the load prediction function, and the channel gain value between the classroom (task sending end) where the teaching task to be recorded is located and the available heterogeneous computing node receiving end is extracted from the channel gain matrix. Initialize a population size of (In this embodiment, 50) genetic algorithms are used, with each chromosome encoded as a sequence of length... sequence, the first in the sequence The value of the gene locus represents the _th _ ... The computing node number assigned to each recording task (range of values) Genetic algorithms randomly generate a set of candidate solutions through encoding; For example, the recording tasks for classroom 1 are assigned to the receiver of heterogeneous computing node A, and the recording tasks for classrooms 2 and 3 are assigned to the receiver of heterogeneous computing node B. for One teaching task to be recorded, corresponding to The constraint formula for the signal-to-interference-plus-noise ratio (SINR) is used to ensure that the video stream runs smoothly without interruption. The constraint formulas for each signal-to-interference-plus-noise ratio (SIR) must all be greater than or equal to a preset signal quality threshold. ( ); Get The inequalities are solved using the convex optimization algorithm, the Lagrange multiplier method. The equations are transformed into matrix operations, and the mathematical expressions for these matrix operations are as follows: ; in, yes The identity matrix, yes A diagonal matrix, where the diagonal elements are the signal quality thresholds. , yes The normalized channel gain matrix, elements in Describes the sending end of the teaching task to be recorded. For the sending end of recorded teaching tasks The degree of interference; yes dimensional noise vector, This represents the basic base power required to overcome background noise. Each element The calculation formula is: ; in, Sending end for recorded teaching tasks Basic base power, For signal quality threshold, For the sending end The channel gain to the heterogeneous computing nodes to which it is allocated; Solve the equation Obtain the minimum power vector ; The weighted sum of transmission energy consumption and computational energy consumption of the computational scheduling scheme is calculated. The weight coefficients of the weighted sum are determined based on the campus business period to which the current physical time belongs. The weighted result, the first penalty value and the second penalty value are added together to obtain the comprehensive fitness value. To ensure smooth video streaming in a wireless environment (meeting signal quality thresholds), the transmitting end (camera and encoder) in the recording studio must consume power to transmit wireless signals. Greater environmental interference requires higher transmission power and thus higher energy consumption. The mathematical expression for the transmission energy consumption of the scheduling scheme is: ; in, The preset course duration, This represents the total number of teaching tasks to be recorded and broadcast. For steps The One element, Representing the The instantaneous transmission power required by a classroom to overcome interference and maintain connection; the unit of instantaneous transmission power is watt. When heterogeneous computing nodes (servers) process video transcoding, encapsulation, and storage, their CPUs and memory operate under high load. Server energy consumption is positively correlated with load, and the mathematical expression for computing energy consumption is: ; in, This represents the total number of heterogeneous computing nodes. In the current scheduling scheme, it is allocated to heterogeneous computing nodes. The collection of teaching tasks to be recorded and broadcast. For the first The load prediction function value for each task. The energy conversion factor is expressed in joules per unit load. It is a constant and its value ranges from 200 to 400. Based on the overall fitness value, the candidate scheme with low energy consumption is used as the parent scheme. Gene crossover and site mutation are performed on the preserved parent chromosomes to generate the next generation population. Repeat steps S301 to S305 until the overall fitness reaches the preset convergence condition (10⁻⁴ in this embodiment). The final output population is the optimal scheduling scheme set.

[0033] This invention calculates the signal-to-interference-plus-noise ratio (SIR) of each teaching task to be recorded and broadcast, and uses a convex optimization algorithm to find the minimum power vector that satisfies this constraint, with the SIR being greater than or equal to a preset signal quality threshold. This inner-layer continuous optimization process finds the minimum transmission power required to maintain the wireless connection while ensuring that the video stream is not interrupted. At the same time, if there is a negative value in the minimum power vector, it means that the constraint has no solution or the solution is infeasible, so a second penalty value is assigned to further filter out infeasible solutions. Combining the weighted sum of the calculation of transmission energy consumption and computation energy consumption, and incorporating the first and second penalty values ​​into the comprehensive fitness value, the optimal scheduling scheme set is finally obtained through iterative optimization using a genetic algorithm. This set of schemes is the one with the best comprehensive energy consumption under the dual hard constraints of computational resource capacity and wireless signal quality, realizing deep collaborative optimization of computational and transmission resources and maximizing overall energy efficiency.

[0034] Step S4, obtaining the final scheduling scheme specifically includes: The physical time is divided into peak business periods and off-peak business periods. The current physical time is obtained, and the campus business time period to which the current physical time belongs is identified. The campus business time period includes peak business periods and off-peak business periods. The campus business time period is divided based on a preset time period. A virtual scheduling ideal scheme is defined, in which the normalized transmission energy consumption and normalized computation energy consumption are both 0. Determine the decision weights for transmission energy consumption and computing energy consumption based on the campus business time period to which the current physical time belongs; For each scheduling scheme in the optimal scheduling scheme set, calculate the weighted Euclidean distance between the normalized transmission energy consumption and the normalized computational energy consumption of the virtual ideal scheduling scheme. The scheduling scheme with the minimum weighted Euclidean distance is selected as the final scheduling scheme.

[0035] In one specific embodiment of the present invention, the periods from 08:00 to 12:00 and 14:00 to 18:00 on weekdays are divided into peak business periods, and the remaining time periods are divided into off-peak business periods. The division of campus business time periods is based on the school's working time periods, with working time periods as peak business periods and rest time periods as off-peak business periods. For each scheduling scheme in the optimal scheduling scheme set, the transmission energy consumption and computing energy consumption are normalized. A two-dimensional coordinate system is established, with the origin (0, 0) as the zero transmission energy consumption and zero computation energy consumption of the virtual ideal scheme. The transmission energy consumption and computation energy consumption of the normalized scheduling scheme are mapped to coordinates in the two-dimensional coordinate system. The Euclidean weighted distance to the origin is calculated, and a selection is made. The minimum solution is taken as the final scheduling solution. The mathematical expression for the Euclidean weighted distance is: ; in, Euclidean weighted distance, , These are the weighting coefficients. To normalize transmission energy consumption, To normalize the energy consumption calculation; Transmission power consumption is usually related to transmission power and channel interference. The higher the value, the greater the power the base station transmits to ensure signal quality, or the more severe the environmental interference (the worse the stability). Computation power consumption is related to server CPU load. The higher the value, the more electricity is consumed. If the current time is during peak business hours, then The value is 0.8. The value is 0.2, increasing the transmission weight means that any solution with high transmission energy consumption will have a higher transmission weight over distance. They will all grow bigger and thus be eliminated; If the current time is during a business off-peak period, then The value is 0.2. The value is 0.8, which increases the calculation weight.

[0036] This invention divides physical time into peak and off-peak periods. Based on the campus business period to which the current time belongs, it dynamically determines the decision weights of transmission and computing energy consumption in the final scheme selection. It also calculates the weighted Euclidean distance between each scheme in the optimal scheduling scheme set and the virtual ideal scheme with normalized energy consumption of 0, and selects the scheme with the smallest distance as the final scheduling scheme. This dynamic decision-making mechanism based on business period characteristics can flexibly adjust the optimization objectives according to the actual business load status. During peak periods, it focuses on selecting schemes with low transmission costs and good channels to alleviate network pressure, while during off-peak periods, it focuses more on saving computing resources. This further improves the adaptability and flexibility of resource scheduling, ensuring that the scheduling decision always best matches the actual business needs of the campus, thereby achieving optimal resource utilization in different time periods.

[0037] In step S4, generating control commands specifically includes: Based on the final scheduling scheme, node allocation instructions and power control instructions are generated for each teaching task to be recorded and broadcast, and then sent to the heterogeneous computing nodes and the sending end of the teaching task to be recorded and broadcast for execution.

[0038] This invention generates specific node allocation instructions and power control instructions for each recorded teaching task based on the final selected scheduling scheme, and sends them to heterogeneous computing nodes and the sending end of the recorded teaching task for execution. This ensures that the optimal scheduling scheme obtained through complex collaborative optimization and dynamic decision-making can be accurately and timely implemented in specific computing resource allocation and wireless device power control, so that the prediction, optimization and decision-making strategies in the early stage can really play a role, ensuring that the teaching recording task is executed smoothly according to the optimal scheme, and ultimately achieving the goal of improving the quality of recording service and the overall energy efficiency of the system.

[0039] A cloud-based teaching recording and broadcasting management system includes: an acquisition module, a prediction module, an optimization module, and a management module; The data acquisition module is used to collect the full-link historical operation data of teaching recording and broadcasting tasks and the data of teaching tasks to be recorded and broadcasting, and to construct a historical resource consumption sequence based on the full-link historical operation data. The prediction module is used to time-slice the duration of teaching tasks in the data to be recorded and broadcast, assign tidal weights, and fit the historical resource consumption sequence using a polynomial fitting function that includes tidal weights to generate a load prediction function and a channel gain matrix. The optimization module is used to execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture to obtain the optimal scheduling scheme set. The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computing resources; in the inner continuous optimization process, solving for the minimum transmit power based on the channel gain matrix and combined with the convex optimization algorithm. The management module is used for dynamic decision locking based on business time period characteristics. It identifies the campus business time period to which the current physical time belongs, calculates the weighted Euclidean distance between the optimal scheduling scheme set scheme and the virtual ideal scheme. The weight used in the weighted Euclidean distance is consistent with the weight used in step S3 to calculate the comprehensive fitness value, obtains the final scheduling scheme and generates control instructions.

[0040] The cloud-based teaching recording and broadcasting management method and system provided by this invention can accurately construct historical resource consumption sequences and perform time-slicing fitting of teaching task duration based on tidal weights to generate a load prediction function. This enables accurate prediction of the dynamic characteristics of resource consumption at different course stages, especially enhancing the fitting sensitivity of resource demand at the end of the course and during extended class periods. This effectively avoids the loss of recorded content due to insufficient resource estimation or the waste of resources due to overestimation, significantly improving the foresight and accuracy of resource scheduling. By constructing a channel gain matrix and combining it with a collaborative optimization scheduling strategy based on an inner and outer nested architecture, this invention achieves collaborative optimization of computing resources and wireless transmission resources. In the outer discrete optimization, a load prediction function is invoked to perform hard constraint verification of computing resources, ensuring that the task load allocated to heterogeneous computing nodes does not exceed their physical upper limit. In the inner continuous optimization, the minimum transmit power is solved based on the channel gain matrix and using a convex optimization algorithm, effectively reducing transmit power and transmission energy consumption while ensuring that the signal-to-interference-plus-noise ratio of the video stream meets the quality threshold. This collaborative optimization mechanism can balance computing energy consumption and transmission energy consumption, avoid local resource bottlenecks or excessive energy consumption, and thus achieve overall system energy efficiency optimization while ensuring the quality of multi-channel concurrent recording and broadcasting services. This invention features a dynamic decision-making locking function based on business time period characteristics. It can dynamically adjust the decision weights of transmission and computing energy consumption according to the campus business time period (such as peak or off-peak hours) at the current physical time. During peak hours, the system prioritizes reducing transmission energy consumption, favoring solutions with good channel quality and low transmission costs. During off-peak hours, it prioritizes reducing computing energy consumption, severely penalizing high-energy-consuming solutions. This dynamic adaptive scheduling strategy allows the system to flexibly adjust its optimization objectives based on actual business load changes, further improving resource utilization flexibility and overall operational economy.

[0041] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0042] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A cloud-based teaching recording and broadcasting management method, characterized in that, Includes the following steps: Step S1: Collect the full-link historical operation data of the teaching recording and broadcasting task and the teaching task data to be recorded and broadcasting, and construct the historical resource consumption sequence based on the full-link historical operation data; Step S2: Divide the duration of the teaching task in the data to be recorded into time slices and assign tidal weights. Fit the historical resource consumption sequence with a polynomial fitting function containing tidal weights to generate a load prediction function and a channel gain matrix. Step S3: Execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture on the randomly generated scheduling schemes, obtain the comprehensive fitness of the scheduling schemes, and select the optimal scheduling scheme set. The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computing resources; and in the inner continuous optimization process, calculating the comprehensive fitness of the scheduling scheme based on the channel gain matrix and combined with the convex optimization algorithm. Step S4: Dynamic decision locking based on business time period characteristics, identify the campus business time period to which the current physical time belongs, calculate the weighted Euclidean distance between the scheduling scheme in the optimal scheduling scheme set and the virtual ideal scheme, obtain the final scheduling scheme and generate control instructions.

2. The teaching recording and broadcasting management method based on a cloud platform as described in claim 1, characterized in that, The full-link historical operation data includes: historical container operation index data, historical network transmission index data, and historical static data; The historical container operation metrics data include the CPU utilization and memory usage of the recording container; The historical network transmission index data includes the signal receiving power of the transmitting end in the recording classroom. The historical static data includes the rated transmission power of the recording classroom transmitter, the upper limit of the physical resource capacity of the heterogeneous computing nodes, and the ambient background noise power. The CPU utilization and memory usage at the relative sampling time are used to construct resource consumption values. The relative sampling time and resource consumption values ​​are paired to construct a historical resource consumption sequence. The data for the teaching tasks to be recorded includes the expected start time and expected course duration.

3. The teaching recording and broadcasting management method based on a cloud platform as described in claim 2, characterized in that, In step S2, the process of generating the load prediction function specifically includes: The preset course duration of the teaching task to be recorded is divided into time segments, and the preset course duration is divided into rising period, steady period and ebb period. The rising period is defined as the first preset proportion from the start of the recording to the preset course duration, and a first tidal weight is assigned to the rising period. The stable period is defined as the second preset proportion from the end of the rising period to the preset course duration, and a second tidal weight is assigned to the stable period. The ebb tide period is the preset overtime tolerance period from the end of the stable period to the end of the preset course duration, and a third tide weight is assigned to the ebb tide period. The third tidal weight value is greater than the first tidal weight value and the second tidal weight value; Construct a polynomial fitting function containing unknown coefficients, construct a weighted residual sum of squares based on the polynomial fitting function and tidal weights, substitute the historical resource consumption sequence into the weighted residual sum of squares, and use the least squares method to solve for the unknown coefficients that minimize the weighted residual sum of squares. Substitute the unknown coefficients obtained from the solution into the polynomial fitting function to obtain the load prediction function.

4. The teaching recording and broadcasting management method based on a cloud platform as described in claim 3, characterized in that, In step S2, the process of constructing the channel gain matrix specifically includes: Construct a local spatial coordinate system on campus to obtain the spatial coordinates of all recording classroom transmitters and all heterogeneous computing node receivers. The difference between the rated transmit power and the signal receive power at the recording classroom transmitter is used as the historical signal path loss value. The Euclidean distance between the transmitter and receiver and the historical signal path loss value between the transmitter and receiver are used as signal path loss sample pairs. The path loss sample pairs are fitted using the logarithmic distance path loss formula and linear regression is performed. The slope of the logarithmic distance path loss formula is used as the signal path loss exponent. The path loss exponent is substituted back into the logarithmic distance path loss formula to obtain the spatial signal attenuation function. Traverse the set of transmitting points and the set of receiving points, calculate the Euclidean distance between any transmitting end and any heterogeneous computing node, substitute the Euclidean distance into the spatial signal attenuation function, and calculate the channel gain value. Construct a channel gain matrix, the elements of which are the channel gain values ​​between the transmitter and the corresponding heterogeneous computing nodes.

5. The teaching recording and broadcasting management method based on a cloud platform as described in claim 4, characterized in that, Step S3 includes the following sub-steps: Step S301: Obtain the teaching task to be recorded and the available heterogeneous computing nodes, initialize the genetic algorithm with a preset population size, encode each chromosome as a scheduling scheme, and randomly generate several scheduling schemes to form a candidate scheme set; The scheduling scheme involves allocating the teaching tasks to be recorded to available heterogeneous computing nodes.

6. The teaching recording and broadcasting management method based on a cloud platform as described in claim 5, characterized in that, Step S3 further includes the following sub-steps: Step S302: Traverse all candidate schemes in the candidate scheme set and calculate the peak load of each heterogeneous computing node in each candidate scheme. The peak load is the sum of the resource requirements of the sending end of all teaching tasks to be recorded and broadcast allocated to the heterogeneous computing node. If the peak load of any of the heterogeneous computing nodes is greater than or equal to the physical resource limit of the heterogeneous computing node, then the corresponding candidate solution is given a first penalty value.

7. The teaching recording and broadcasting management method based on a cloud platform as described in claim 6, characterized in that, Step S3 further includes the following sub-steps: Step S303: For the transmitting end of each teaching task to be recorded in the candidate schemes, calculate the signal-to-interference-plus-noise ratio (SINR) of the transmitting end. The mathematical expression for the SINR is: ； in, Let be the signal-to-interference-plus-noise ratio (SIR) for the i-th task. Let i be the channel gain from the transmitter i to its assigned heterogeneous computing nodes. Let i be the transmit power. Let j be the interference channel gain of the receiver node i relative to the transmitter j. Let j be the transmit power of the transmitter. Environmental noise power; For the signal-to-interference-plus-noise ratio (SINR) of the transmitting end of any teaching task to be recorded, a constraint formula is constructed, wherein the SINR is greater than or equal to a preset signal quality threshold. The minimum power vector that satisfies the constraint formula is solved using a convex optimization algorithm. If there are negative values ​​in the minimum power vector, then a second penalty value is assigned to the corresponding candidate scheme; Step S304: Calculate the comprehensive fitness value of each scheduling scheme in the candidate scheme set. The comprehensive fitness value is calculated by: calculating the weighted sum of the transmission energy consumption and the computation energy consumption of the scheduling scheme, and adding the weighted result, the first penalty value and the second penalty value as the comprehensive fitness value. Step S305: Sort the candidate schemes in ascending order according to the comprehensive fitness value, select the candidate schemes corresponding to the top several comprehensive fitness values ​​as the parent chromosomes, perform gene crossover and site mutation on the parent chromosomes to generate the next generation population, and obtain the optimal scheduling scheme set after iterative optimization.

8. The teaching recording and broadcasting management method based on a cloud platform as described in claim 7, characterized in that, In step S4, obtaining the final scheduling scheme specifically includes: The physical time is divided into peak business periods and off-peak business periods. The current physical time is obtained, and the campus business time period to which the current physical time belongs is identified. The campus business time period includes peak business periods and off-peak business periods. The campus business time period is divided based on a preset time period. A virtual scheduling ideal scheme is defined, in which the normalized transmission energy consumption and normalized computation energy consumption are both 0. Determine the decision weights for transmission energy consumption and computing energy consumption based on the campus business time period to which the current physical time belongs; For each scheduling scheme in the optimal scheduling scheme set, calculate the weighted Euclidean distance between the normalized transmission energy consumption and the normalized computational energy consumption of the virtual ideal scheduling scheme. The scheduling scheme with the minimum weighted Euclidean distance is selected as the final scheduling scheme.

9. The teaching recording and broadcasting management method based on a cloud platform as described in claim 8, characterized in that, In step S4, generating control commands specifically includes: Based on the final scheduling scheme, node allocation instructions and power control instructions are generated for each teaching task to be recorded and broadcast, and then sent to the heterogeneous computing nodes and the sending end of the teaching task to be recorded and broadcast for execution.

10. A cloud-based teaching recording and broadcasting management system, characterized in that, include: Data acquisition module, prediction module, optimization module, and management module; The acquisition module is used to collect the full-link historical operation data of teaching recording and broadcasting tasks and the data of teaching tasks to be recorded and broadcasting, and to construct a historical resource consumption sequence based on the full-link historical operation data. The prediction module is used to divide the duration of the teaching tasks in the data to be recorded into time slices and assign tidal weights. It then uses a polynomial fitting function containing tidal weights to fit the historical resource consumption sequence and generate a load prediction function and a channel gain matrix. The optimization module is used to execute a collaborative optimization scheduling strategy based on an inner and outer nested architecture to obtain the optimal scheduling scheme set; The collaborative optimization scheduling strategy specifically includes: in the outer discrete optimization process, calling the load prediction function to perform hard constraint verification of computational resources; and in the inner continuous optimization process, solving for the minimum transmit power based on the channel gain matrix and combining it with a convex optimization algorithm. The management module is used for dynamic decision locking based on business time period characteristics, identifying the campus business time period to which the current physical time belongs, calculating the weighted Euclidean distance between the optimal scheduling scheme set and the virtual ideal scheme, obtaining the final scheduling scheme and generating control commands.

Images