Disaster recovery data recovery method and system of disaster recovery system

Abstract

The invention provides a disaster recovery data recovery method and system of a disaster recovery system, and the method comprises the steps: adjusting a network bandwidth allocation proportion according to a priority, and if the target timeliness is higher than a preset recovery time threshold, increasing a real-time power grid alarm data bandwidth proportion, and generating a bandwidth allocation scheme; dynamically scheduling computing resource allocation according to a bandwidth allocation scheme, and if the key evaluation characteristic exceeds a set upper limit, increasing a computing resource share of a service key level, and generating a resource allocation decision; and after the priority updating scheme is obtained, monitoring a system load in concurrent request processing through a disaster recovery system, and if the system load exceeds a preset threshold value, triggering rebalance of network bandwidth allocation in combination with timeliness deviation to obtain a resource optimization result.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a disaster recovery data recovery method and system for a disaster recovery system. Background Art

[0002] With the improvement of the intelligence level of power grid, the role of data in power business has become increasingly prominent, especially when it comes to multiple key businesses such as safety verification, power generation planning, and market settlement. The timeliness and integrity of data directly affect the accuracy and economic benefits of system decisions. However, after a catastrophic event, the disaster recovery system needs to quickly respond to the recovery requests of multiple types of data with limited resources, which puts extremely high demands on the dynamic and priority scheduling of resource allocation. Research on how to optimize this process is not only related to the safety of power grid operation, but also to the fairness and efficiency of the power market. At present, the static priority allocation method or single-dimensional scheduling strategy commonly used in disaster recovery systems is incapable of dealing with concurrent requests from multiple businesses. These methods often only preset a fixed recovery order based on data type or business category, lack the ability to adapt to real-time scene changes, especially when resources are tight, and cannot effectively balance the needs of different businesses, resulting in some key businesses being damaged due to recovery delays. In particular, when real-time power grid alarm data and monthly electricity bill calculation data are requested to be restored at the same time, the existing solution is difficult to take into account the differentiated needs of the two, and the inefficiency of resource allocation is exposed. The core challenge in this area is how to achieve a dynamic balance between data timeliness and business criticality. Data timeliness requires the system to prioritize the recovery of data that is sensitive to real-time, such as power grid alarm information, while business criticality emphasizes avoiding major losses caused by critical system downtime, such as the accuracy of market settlement. The conflict between the two in concurrent scenarios complicates resource allocation decisions: differences in preset recovery time targets may lead to scheduling deviations, and the uncertainty of potential loss assessment further exacerbates the difficulty of priority determination. These unresolved technical factors directly give rise to the trade-off problem in dynamic resource allocation. Therefore, how to construct a two-dimensional priority matrix based on data timeliness and business criticality, and dynamically allocate network bandwidth and computing resources when concurrent requests for recovery of real-time power grid alarm data and monthly electricity bill calculation data are made, has become a key issue in optimizing the performance of disaster recovery systems. The solution to this problem requires the adaptive and efficient resource scheduling in changing scenarios to ensure full support for power business. Summary of the invention

[0003] The present invention provides a disaster recovery data recovery method for a disaster recovery system, which mainly includes:

[0004] Obtain the target timeliness and key assessment characteristics of concurrent requests, map the key assessment characteristics to business criticality levels, evaluate the gap between the current recovery status and the target timeliness, obtain the timeliness difference, combine the business criticality level and timeliness difference, build a priority matrix, and calculate the dynamic priority of different data recovery requests;

[0005] The network bandwidth allocation ratio is adjusted according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of real-time power grid alarm data is increased to generate a bandwidth allocation plan.

[0006] According to the bandwidth allocation plan, dynamically schedule computing resource allocation. If the key evaluation characteristics exceed the set upper limit, increase the computing resource share of the business criticality level and generate resource allocation decisions;

[0007] The actual recovery time of real-time grid alarm data and monthly electricity fee calculation data is extracted from resource allocation decisions, and compared with the target recovery time to obtain scheduling effect evaluation parameters, which include timeliness deviation and criticality satisfaction.

[0008] The scheduling effect evaluation parameter is used as the feedback signal, and the PID controller is introduced. The deviation between the actual scheduling effect and the preset target is used as the input error of the controller. The difference in data timeliness and the adjustment amount of the business criticality level weight are output, and the priority update plan is formed according to the adjusted weight combination.

[0009] After obtaining the priority update plan, the system load in concurrent request processing is monitored through the disaster recovery system. If the system load exceeds the preset threshold, the rebalancing of network bandwidth allocation is triggered in combination with the timeliness deviation to obtain resource optimization results.

[0010] The recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data is extracted from the resource optimization results. The stability of the dynamic scheduling framework is judged and the final dynamic scheduling framework is determined through the matching analysis between the criticality satisfaction and the business downtime loss assessment value.

[0011] Furthermore, the target timeliness and key evaluation characteristics of concurrent requests are obtained, the key evaluation characteristics are mapped to business criticality levels, the gap between the current recovery status and the target timeliness is evaluated, and the timeliness difference is obtained. In combination with the business criticality level and the timeliness difference, a priority matrix is ​​constructed to calculate the dynamic priority of different data recovery requests, including: constructing a timeliness target matrix based on the business data integrity score and resource occupancy rate in the concurrent data request, obtaining the recovery request baseline timeliness value for each concurrent data request from the timeliness target matrix, and judging the data recovery request level based on the baseline timeliness value. The real-time timeliness difference calculation is performed on the concurrent recovery requests based on the recovery request baseline timeliness value, the business requests are dynamically graded according to the timeliness difference value, and the timeliness difference matrix is ​​constructed based on the dynamic grading results. The timeliness difference matrix is ​​normalized, the business criticality baseline value is obtained from the data criticality quantitative indicator library, and the business criticality level matrix is ​​constructed based on the timeliness difference normalization result and the business criticality baseline value. Extract the criticality score of each request according to the business criticality level matrix, assign weights to the normalized results of timeliness difference, and build the initial priority matrix according to the timeliness difference value after weight assignment. Use the neural network model to optimize the initial priority matrix. The model input layer contains the business criticality score and the timeliness difference weight value. The hidden layer sets a three-layer structure for feature extraction, and the output layer generates a dynamic priority value matrix. Extract the priority score according to the dynamic priority value matrix, obtain the task allocation weight from the recovery task allocation rule library, calibrate the priority score according to the task allocation weight, and sort the calibrated priority values.

[0012] Further, key assessment characteristics are obtained, business criticality is graded according to the key assessment characteristics, business criticality is obtained, and current state information is extracted from the recovery state to obtain a state comparison benchmark value, and timeliness value is calculated to obtain timeliness benchmark data, including: extracting business impact degree data, compliance requirement index value, and core process dependency score in key assessment characteristics, calculating assessment characteristic weights through hierarchical analysis method, obtaining benchmark threshold intervals from the assessment benchmark rule library, and grading business criticality according to the weight calculation results to obtain criticality level values. A recovery time objective matrix is ​​constructed according to the business criticality level value, service level benchmark requirements are extracted from the service level agreement library, state monitoring is performed according to the processing deadline threshold, and the state benchmark score is calculated for the monitoring data. A recovery state prediction model is established through a random forest algorithm, the model input items include business criticality level values ​​and state benchmark scores, and the output items include state prediction values. A state comparison benchmark value is generated according to the prediction value and the actual monitoring data. A timeliness calculation rule is constructed according to the state comparison benchmark value, the target value is extracted from the recovery time objective matrix, constraint processing is performed according to the service level benchmark requirements, and data is filtered through the processing deadline threshold to obtain timeliness benchmark data. The difference analyzer is constructed based on the neural network model. The input layer contains the timeliness benchmark data and the state comparison benchmark value. The hidden layer performs feature mapping and nonlinear transformation. The output layer generates the timeliness difference result data. The timeliness difference result data is normalized, and the division criteria are extracted from the difference level division rule library. The normalized data is segmented according to the division criteria, and the timeliness difference level matrix is ​​generated for the segmentation results.

[0013] Further, the network bandwidth allocation ratio is adjusted according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of the real-time power grid alarm data is increased to generate a bandwidth allocation plan, including: constructing a bandwidth reference matrix according to the network bandwidth capacity and the network congestion level, extracting the priority value from the priority classification value database, calculating the alarm priority weight according to the amount of real-time power grid alarm data, and obtaining the initial bandwidth allocation ratio by a weighted calculation method. Read the preset recovery time threshold from the recovery time record library to obtain the current data transmission rate. If the target timeliness is higher than the preset recovery time threshold, increase the proportion of real-time power grid alarm data in the initial bandwidth allocation ratio to obtain the bandwidth adjustment coefficient. Perform a proportional amplification calculation on the initial bandwidth allocation ratio according to the bandwidth adjustment coefficient, obtain the maximum adjustment interval from the bandwidth adjustment amplitude rule library, constrain the amplification result according to the interval range, and obtain the alarm data bandwidth ratio value. A bandwidth allocation optimization model is constructed by a neural network algorithm. The input layer contains the alarm data bandwidth ratio value and the network congestion level data, the hidden layer performs feature extraction and nonlinear transformation, and the output layer generates an optimized bandwidth allocation plan. The optimized bandwidth allocation scheme is resource mapped, the physical channel parameters are extracted from the network bandwidth capacity data, the channels are grouped according to the alarm data transmission priority, and a complete bandwidth allocation scheme is generated for the grouping results. A real-time monitoring indicator matrix is ​​constructed based on the complete bandwidth allocation scheme, the real-time alarm frequency is obtained from the power grid alarm data collection terminal, and the alarm data bandwidth ratio is dynamically adjusted according to the frequency change trend to obtain the adaptive bandwidth allocation result.

[0014] Furthermore, according to the bandwidth allocation scheme, the computing resource allocation is dynamically scheduled. If the key evaluation characteristics exceed the set upper limit, the computing resource share of the business criticality level is increased to generate a resource allocation decision, including: constructing a resource benchmark matrix according to the bandwidth demand ratio and the number of computing resources, extracting the priority values ​​of each business process from the business priority database, and calculating the starting value of resource allocation according to the current resource usage to obtain the computing resource allocation benchmark value. Read the upper limit values ​​of the three key evaluation characteristics of business impact, compliance requirements, and process dependency from the performance threshold library to obtain the current resource occupancy index. If any key evaluation characteristic exceeds the set upper limit, increase the corresponding business process resource quota to obtain the resource quota adjustment value. Update the computing resource allocation benchmark value according to the resource quota adjustment value, obtain the minimum guaranteed resource quota from the scheduling rule library, calculate the resource demand of each business process according to the real-time resource load data, and obtain the dynamic resource scheduling benchmark table. The deep reinforcement learning algorithm is used to train the resource allocation agent. The input items include the dynamic resource scheduling benchmark table and the real-time performance indicators. The reward function drives the agent to optimize the allocation strategy under resource constraints and outputs the resource allocation optimization plan. The resource allocation optimization plan is grouped according to the business process, the resource status data is read from the computing node pool, and the business process is mapped according to the node processing capacity to obtain a specific node allocation plan. According to the node allocation plan, a resource scheduling instruction set is constructed, the resource guarantee level is obtained from the business criticality level table, the resource preemption priority is set according to the guarantee level, and the resource allocation decision is generated for the priority sequence.

[0015] Furthermore, the actual recovery time of real-time power grid alarm data and monthly electricity bill calculation data is extracted from the resource allocation decision, and compared with the target recovery time to obtain the scheduling effect evaluation parameters, which include timeliness deviation and critical satisfaction, including: extracting power grid alarm processing records and electricity bill calculation task records from the resource allocation decision, obtaining the real-time alarm recovery time and the monthly electricity bill calculation completion time according to the timestamp, and constructing a time benchmark matrix based on the historical target time record to obtain the actual recovery progress data of the two types of businesses. The actual recovery progress data is standardized, the standard alarm processing time and the standard electricity bill calculation cycle are obtained from the business timeliness benchmark library, and the actual progress deviation is calculated based on the standard time to obtain the initial value of the timeliness deviation. The real-time alarm data stream is processed by a convolutional neural network. The input layer receives the alarm processing rate and the alarm backlog, the hidden layer extracts the timeliness features, and the output layer generates the alarm processing timeliness evaluation value. The monthly electricity bill calculation data is analyzed by a time series algorithm, the calculation time distribution is extracted from the billing cycle record, and the processing weight is calculated based on the business priority matrix to obtain the electricity bill calculation timeliness evaluation value. A comprehensive evaluation matrix is ​​constructed based on the timeliness evaluation value of alarm processing and the timeliness evaluation value of electricity fee calculation. The business level data is obtained from the key indicator library. The satisfaction benchmark value is calculated based on the business level to obtain the key satisfaction data. The initial value of the timeliness deviation is weighted, the timeliness scoring standard is obtained from the evaluation rule library, and the deviation value is graded according to the scoring standard to obtain the timeliness deviation result. Based on the key satisfaction data and the timeliness deviation result, the scheduling effect evaluation parameters are generated, the standard interval of the evaluation parameters is extracted from the parameter mapping table, and the evaluation parameters are classified according to the interval range to obtain the scheduling effect evaluation result.

[0016] Furthermore, the scheduling effect evaluation parameter is used as a feedback signal, and a PID controller is introduced. The deviation between the actual scheduling effect and the preset target is used as the input error of the controller, and the adjustment amount of the data timeliness difference and the business criticality level weight is output, and a priority update scheme is formed according to the adjusted weight combination, including: extracting the timeliness deviation value and the criticality satisfaction from the scheduling effect evaluation parameter to construct a feedback signal, calculating the control error according to the preset scheduling target value, and constructing the PID controller input matrix according to the error value to obtain the control error sequence. The proportional control item is calculated according to the control error sequence, the proportional coefficient is obtained from the controller parameter library, and the error is linearly amplified according to the proportional coefficient to obtain the proportional adjustment amount. The control error sequence is time-integrated, the integral time constant is extracted from the controller parameter library, and the long-term error accumulation value is calculated according to the integral result to obtain the integral adjustment amount. The error change rate is calculated by the time difference method, the differential time constant is read from the controller parameter library, and the error development trend is predicted according to the change rate to obtain the differential adjustment amount. The proportional adjustment amount, the integral adjustment amount and the differential adjustment amount are input into the combiner, and the weighted combination is performed according to the control weight parameter, and the total timeliness difference adjustment amount is calculated according to the combination result. The weight mapping relationship is established through a deep neural network. The input layer receives the total amount of timeliness difference adjustment, the hidden layer extracts the adjustment features, and the output layer generates the business criticality weight adjustment value. The total amount of timeliness difference adjustment and the business criticality weight adjustment value are normalized, and the combination coefficient is obtained from the priority calculation rule library. The priority update scheme is generated based on the combination coefficient. A new round of scheduling parameters is constructed based on the priority update scheme, and the scheduling effect evaluation results are obtained from the evaluation feedback signal. The PID controller parameters are updated based on the evaluation results to complete the closed-loop control cycle.

[0017] Furthermore, after obtaining the priority update scheme, the system load in the concurrent request processing is monitored through the disaster recovery system. If the system load exceeds the preset threshold, the rebalancing of the network bandwidth allocation is triggered in combination with the timeliness deviation to obtain the resource optimization result, including: obtaining three load indicators of processor occupancy, memory usage, and network throughput from the disaster recovery monitoring point, classifying the concurrent request data according to the priority update value, and constructing a load state matrix based on the resource occupancy data to obtain real-time load monitoring data. The load monitoring data is processed by the neural network predictor, the input layer receives the three load indicators and the number of concurrent requests, the hidden layer extracts the load characteristics, and the output layer generates the future load prediction value. The three preset values ​​of the processor threshold, memory threshold, and network threshold are read from the threshold rule library to obtain the current load prediction value and the timeliness deviation data. If any load prediction value exceeds the corresponding threshold, the bandwidth allocation adjustment signal is triggered. The rebalancing rule matrix is ​​constructed according to the bandwidth allocation adjustment signal, the available bandwidth data is obtained from the network resource pool, and the business bandwidth demand is calculated based on the timeliness deviation to obtain the initial bandwidth allocation plan. The initial bandwidth allocation plan is processed by a deep reinforcement learning agent. The input items include business priority and resource occupancy data. The reward function is designed based on load balancing and the bandwidth allocation optimization plan is output. Resource mapping is performed on the bandwidth allocation optimization plan, link status data is extracted from the network topology library, bandwidth is allocated according to the link load, and a specific link configuration plan is obtained. Bandwidth resource reallocation is performed according to the link configuration plan, and the monitoring threshold is obtained from the business performance indicator library. The reallocation result is verified based on the monitoring data to obtain the resource optimization result data.

[0018] Furthermore, if it is detected that the timeliness deviation exceeds the preset threshold, the trigger mechanism is activated, the current state of the network bandwidth is obtained, and the allocation adjustment algorithm is used to obtain a preliminary bandwidth allocation plan. For the allocation result in the preliminary plan, it is judged whether the allocation balance is achieved through resource status evaluation. If not, it is iteratively optimized through the adjustment process to obtain the updated value of the optimization result, and the timeliness change is predicted. The resource optimization plan is determined according to the predicted timeliness, including: obtaining the timeliness deviation value and reading the bandwidth trigger threshold. If the timeliness deviation value exceeds the trigger threshold, the current bandwidth occupancy data is obtained from the bandwidth status collection point, and the bandwidth allocation reference matrix is ​​constructed based on the resource load data to obtain the initial state of the bandwidth resource. The initial state of the bandwidth resource is processed by the allocation adjustment algorithm, the input layer receives the service priority data and the bandwidth occupancy data, the middle layer performs feature extraction and priority mapping, and the output layer generates a preliminary bandwidth allocation plan. According to the preliminary bandwidth allocation plan, a balance evaluation matrix is ​​constructed, and the balance threshold data is obtained from the resource evaluation library. The current allocation balance is calculated based on the proportion of each service bandwidth, and it is judged whether the allocation plan has reached a balanced state. If the allocation balance is lower than the threshold, the iterative optimization mechanism is started, and the iterative optimization goal is set to increase the bandwidth share of low-balance services. The adjustment coefficient is generated based on the optimization goal to obtain the optimized iteration benchmark value. The optimization iteration benchmark value is processed by the deep reinforcement learning algorithm, and the reward function is set to the degree of improvement in balance. When the balance improvement in three consecutive rounds of iterations is less than the threshold, the iteration is stopped and the updated optimization value is output. The updated optimization value is predicted for timeliness, and the timeliness change model parameters are read from the prediction model library. The prediction model is trained based on historical data to generate timeliness prediction results. A resource calibration matrix is ​​constructed based on the timeliness prediction results, and the calibration parameters are obtained from the resource optimization rule library. The updated optimization value is corrected based on the calibration parameters to obtain a resource optimization solution.

[0019] Furthermore, the recovery completion time of the real-time power grid alarm data and the monthly electricity bill calculation data is extracted from the resource optimization results, and the stability of the dynamic scheduling framework is judged by the matching analysis of the criticality satisfaction and the business downtime loss assessment value, and the final dynamic scheduling framework is determined, including: extracting the completion time records of the real-time power grid alarm data and the monthly electricity bill calculation data from the resource optimization results, constructing a recovery progress matrix according to the completion time, calculating the recovery time deviation according to the preset recovery time threshold, and obtaining the business recovery completion data. The business recovery completion data is processed by a neural network, the input layer receives the recovery progress and time deviation data, the hidden layer extracts the recovery features, and the output layer generates the initial value of criticality satisfaction. The downtime duration is calculated according to the business interruption record, the unit time loss benchmark value is obtained from the loss assessment library, and the business downtime loss assessment value is calculated according to the downtime duration and the loss benchmark value. The matching degree of the initial value of criticality satisfaction and the downtime loss assessment value is calculated, the evaluation standard is obtained from the matching rule library, and the business matching coefficient is calculated according to the evaluation standard to obtain the matching degree evaluation result. The stability evaluator is constructed through a deep learning model. The input layer contains the matching evaluation results and historical stability data. The hidden layer learns the stability features. The output layer generates the stability prediction value of the dynamic scheduling framework. The scheduling parameters are optimized according to the stability prediction value. The parameter configuration template is extracted from the scheduling rule library. The parameters are calibrated according to the framework stability prediction value to obtain the scheduling framework parameter scheme. The final scheduling framework matrix is ​​constructed based on the scheduling framework parameter scheme. The scheduling scenario data is obtained from the business scenario library. The framework applicability is verified based on the scenario data to determine the final dynamic scheduling framework.

[0020] The present invention provides a disaster recovery data recovery system of a disaster recovery system, which mainly includes:

[0021] The priority evaluation module is used to obtain the target timeliness and key evaluation characteristics of concurrent requests, map the key evaluation characteristics to the business criticality level, evaluate the gap between the current recovery status and the target timeliness, obtain the timeliness difference, combine the business criticality level and timeliness difference, build a priority matrix, and calculate the dynamic priority of different data recovery requests;

[0022] The bandwidth allocation module is used to adjust the network bandwidth allocation ratio according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of real-time power grid alarm data is increased to generate a bandwidth allocation plan;

[0023] The resource scheduling module is used to dynamically schedule computing resource allocation according to the bandwidth allocation plan. If the key evaluation characteristics exceed the set upper limit, the computing resource share of the business criticality level is increased to generate resource allocation decisions;

[0024] The effect evaluation module is used to extract the actual recovery time of real-time power grid alarm data and monthly electricity fee calculation data from the resource allocation decision, compare it with the target recovery time, and obtain the scheduling effect evaluation parameters. The scheduling effect evaluation parameters include timeliness deviation and criticality satisfaction;

[0025] The weight adjustment module is used to use the scheduling effect evaluation parameter as a feedback signal, introduce a PID controller, use the deviation between the actual scheduling effect and the preset target as the input error of the controller, output the adjustment amount of the data timeliness difference and the business criticality level weight, and form a priority update plan based on the adjusted weight combination;

[0026] The load monitoring module is used to monitor the system load in concurrent request processing through the disaster recovery system after obtaining the priority update plan. If the system load exceeds the preset threshold, the rebalancing of network bandwidth allocation is triggered in combination with the timeliness deviation to obtain resource optimization results;

[0027] The stability analysis module is used to extract the recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data from the resource optimization results, and to judge the stability of the dynamic scheduling framework and determine the final dynamic scheduling framework through matching analysis between criticality satisfaction and business downtime loss assessment value.

[0028] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0029] The present invention discloses a disaster recovery data recovery method for a disaster recovery system. Aiming at the disaster recovery needs of different business data such as real-time power grid alarm data and monthly electricity fee calculation data in the power system, the present invention constructs a priority matrix by evaluating the target timeliness and key characteristics of concurrent requests, and dynamically adjusts the network bandwidth and computing resource allocation. The method introduces a PID controller to continuously optimize the priority strategy according to the deviation between the actual scheduling effect and the preset target. At the same time, the present invention monitors the system load and triggers resource rebalancing when the threshold is exceeded. The stability of the scheduling framework is judged by analyzing the matching degree of criticality satisfaction and the business machine loss assessment value. The method can effectively balance the recovery needs of different business data, improve disaster recovery efficiency, reduce the risk of business interruption, and provide strong support for the reliable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a flowchart of a method for recovering data from a disaster recovery system.

[0031] Figure 2 The present invention is a structural diagram of a disaster recovery data recovery system of a disaster recovery system. DETAILED DESCRIPTION

[0032] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0033] like Figure 1 In this embodiment, a disaster recovery data recovery method of a disaster recovery system may specifically include:

[0034] S101. After the disaster recovery function is started, the target timeliness and key evaluation characteristics of concurrent requests are obtained, and the key evaluation characteristics are mapped to the business criticality level. The timeliness difference is obtained by evaluating the gap between the current recovery status and the target timeliness. A priority matrix is ​​constructed in combination with the business criticality level and the timeliness difference to calculate the dynamic priority of different data recovery requests.

[0035] S1011. Extract indicators including business data integrity and resource occupancy rate from concurrent data requests, construct a timeliness target matrix, obtain the baseline timeliness value of the recovery request for each request from the matrix, judge the level of the data recovery request according to the baseline timeliness value, and calculate the timeliness difference value according to the real-time recovery progress to generate a timeliness difference matrix. In an embodiment of the present invention, the timeliness target matrix reflects the recovery time targets of different business data. For example, power trading data may be set to 10 minutes, while historical data is 60 minutes. By normalizing the timeliness difference value and combining it with the business criticality baseline value in the data criticality quantitative indicator library, a business criticality level matrix is ​​constructed.

[0036] S1012. Extract the criticality score of each request from the business criticality level matrix, assign weights to the normalized results of the time difference, generate an initial priority matrix, and optimize the matrix using a neural network model, wherein the model input layer receives the criticality score and the time difference weight value, performs feature extraction through three hidden layers, and the output layer generates a dynamic priority numerical matrix. In an embodiment of the present invention, the number of neurons in the hidden layer of the neural network model is 128, 64, and 32, respectively, and residual connections are used to improve performance. Priority scores are extracted based on the dynamic priority numerical matrix, task allocation weights are obtained from the recovery task allocation rule library, priority scores are calibrated and sorted, and an initial resource allocation plan is formed.

[0037] In an embodiment of the present invention, for the scenario of power transaction data recovery, concurrent requests for the day's transaction data and historical transaction data are obtained, where the resource occupancy rate of the day's transaction data is 85%, the data integrity is 95%, and the benchmark timeliness value is 10 minutes. If the actual recovery takes 15 minutes, the timeliness difference value is 5 minutes, which is 0.75 after normalization. The business criticality benchmark value extracted from the quantitative indicator library is 0.9, and the criticality score is calculated as 0.95 based on the transaction amount and customer level, which is multiplied by the timeliness difference value to obtain the initial priority value of 0.7125. After optimization, the neural network model outputs a dynamic priority value of 0.57, indicating that the request has a higher priority in resource allocation.

[0038] S1013. According to the calibrated priority value, the available resource status is extracted from the network bandwidth and computing resource pool, the initial resource allocation plan is generated in combination with the task allocation weight, and the feasibility of the plan is verified through real-time monitoring. In this embodiment of the present invention, if the CPU occupancy rate of a recovery node is 75%, and the corresponding weight coefficient is 0.8, the priority value is adjusted to ensure that key businesses obtain more resource support. In this way, the embodiment of the present invention realizes differentiated scheduling based on timeliness and criticality, laying the foundation for resource optimization.

[0039] In the embodiment of the present invention, by constructing a priority matrix and calculating dynamic priorities, resources can be reasonably allocated according to the different requirements of real-time power grid alarm data and monthly electricity bill calculation data. Compared with the traditional static scheduling method, this method significantly improves the efficiency of resource utilization, reduces business losses caused by insufficient timeliness or unmet criticality, and provides adaptive support for the disaster recovery system. Subsequent steps will further optimize the resource allocation plan to ensure the efficient recovery of power services.

[0040] In the embodiment of the present invention, the disaster recovery system ensures that the recovery requirements of various business data in the power system are met by dynamically scheduling network bandwidth and computing resources. The following steps implement adaptive adjustment of resource allocation based on priority for concurrent requests such as real-time power grid alarm data and monthly electricity bill calculation data.

[0041] S102. During the operation of the disaster recovery system, the network bandwidth allocation ratio is adjusted according to the dynamic priority. If it is detected that the target timeliness exceeds the preset recovery time threshold, the bandwidth ratio of the real-time power grid alarm data is increased to generate a bandwidth allocation plan.

[0042] In practical applications, the current bandwidth capacity and network congestion data are first obtained from the network status monitoring module, and a bandwidth benchmark matrix is ​​constructed to reflect resource availability. Subsequently, the priority weights of real-time power grid alarm data are extracted from the priority classification value database, and the initial bandwidth allocation ratio is generated through weighted calculation. Next, the preset recovery time threshold is read from the recovery time record library and compared with the current data transmission rate. If the target timeliness is higher than the threshold, the bandwidth adjustment coefficient is calculated to increase the proportion of alarm data in the initial ratio, and the allocation scheme is further optimized through the neural network algorithm, and finally an adaptive bandwidth allocation result is formed.

[0043] S1021. Construct a bandwidth reference matrix according to the network bandwidth capacity and the network congestion level, obtain the alarm data priority weight from the priority classification value database, and determine the initial bandwidth allocation ratio by weighted calculation method in combination with the real-time power grid alarm data volume, and extract the preset recovery time threshold from the recovery time record library and compare it with the current transmission rate to generate a bandwidth adjustment coefficient. In an embodiment of the present invention, the bandwidth reference matrix is ​​generated by collecting the throughput and delay data of the network interface. For example, in a scenario where the bandwidth capacity is 1000Mbps and the congestion level is 65%, the matrix records the available bandwidth as 350Mbps. The priority weight is graded according to the urgency of the alarm data. The first-level alarm weight can be set to 0.9 and the ordinary alarm to 0.3. When weighted calculation, it is assumed that the alarm data volume accounts for 15% of the total traffic, and the initial allocation ratio is 20%, that is, 200Mbps. If the preset threshold is 5 seconds, and the current transmission rate is only 30Mbps, which is lower than the demand, the adjustment coefficient is calculated to be 1.8 to ensure that the emergency alarm data is supported by more bandwidth.

[0044] S1022, use the bandwidth adjustment coefficient to perform proportional amplification calculation on the priority weight of the alarm data, obtain the maximum adjustment interval from the bandwidth adjustment amplitude rule library to constrain the amplification result, generate the alarm data bandwidth ratio value, and construct a bandwidth allocation optimization model through a neural network algorithm, wherein the input layer contains the alarm data bandwidth ratio value and network congestion degree data, the hidden layer performs feature extraction and nonlinear transformation, and the output layer generates an optimized bandwidth allocation scheme, and then performs physical resource mapping on the scheme and groups channels to form a complete allocation result. In an embodiment of the present invention, after proportional amplification, the alarm data bandwidth is increased from 200Mbps to 360Mbps, and the rule library limits the maximum adjustment interval to 400Mbps to ensure that the result is controllable. The neural network model adopts a three-layer hidden layer structure, with 64, 32 and 16 neurons respectively. The input data is processed by an activation function such as ReLU, optimized in combination with the congestion degree, and finally outputs a bandwidth allocation scheme. For example, among 8 physical channels, 3 are allocated for emergency alarms, and the bandwidth share is increased to 45%, that is, 450Mbps. This design can be dynamically adjusted according to the network status to avoid waste or insufficiency of resources.

[0045] For the power grid operation monitoring scenario, taking a 330 kV substation as an example, its bandwidth capacity is 1000Mbps, the current network congestion level is 65%, and the first-level alarm such as transformer oil temperature overlimit requires a transmission rate of no less than 50Mbps. Under normal conditions, the alarm data is initially allocated 200Mbps, but when the oil temperature overlimit alarm is triggered, the transmission rate drops to 30Mbps, and the target timeliness is 8 seconds, which is higher than the 5-second threshold. At this time, the system increases the bandwidth to 360Mbps based on the adjustment factor 1.8, and allocates it to 450Mbps after neural network optimization. Real-time monitoring shows that the data volume during the high alarm period reaches 2MB / s, and drops to 0.5MB / s during the low period. The system dynamically adjusts the proportion of emergency channel groups accordingly, reaching 60% during peak hours and 25% during low hours, ensuring that the delay is controlled within 3 seconds.

[0046] In the embodiment of the present invention, through the combination of bandwidth reference matrix and neural network algorithm, the system can adjust the bandwidth allocation ratio in real time according to the trend of alarm frequency changes. For example, when multiple substations generate alarms at the same time, frequency data is obtained from the real-time alarm data acquisition terminal, a monitoring indicator matrix is ​​constructed, and the channel grouping strategy is dynamically updated. This mechanism not only improves the bandwidth utilization rate to 85%, but also avoids transmission congestion of high-priority alarms, providing efficient support for disaster recovery.

[0047] In the embodiment of the present invention, the disaster recovery system ensures that the recovery requirements of various business data in the power system are efficiently met by dynamically scheduling network bandwidth and computing resources. The following steps further optimize the allocation process of computing resources based on the bandwidth allocation scheme for concurrent requests such as real-time power grid alarm data and monthly electricity bill calculation data.

[0048] S103. When the disaster recovery system is running, computing resource allocation is dynamically scheduled according to the bandwidth allocation plan. If it is detected that the key evaluation characteristic exceeds the set upper limit, the computing resource share corresponding to the business criticality level is increased to generate a resource allocation decision.

[0049] In an embodiment of the present invention, a resource benchmark matrix is ​​constructed based on the bandwidth demand ratio and the total amount of computing resources, the priority value of each business process is extracted from the business priority database, and the initial allocation benchmark value is calculated in combination with the current resource usage. Subsequently, the upper limit value of the key evaluation characteristic is obtained from the performance threshold library and compared with the real-time resource occupancy index. If it exceeds the upper limit, the resource quota of the corresponding business is adjusted. The resource allocation agent is trained through a deep reinforcement learning algorithm, and an optimization plan is generated by combining the adjusted quota and real-time load data, which is finally mapped to the computing node to form a resource allocation decision.

[0050] S1031. Construct a resource benchmark matrix based on the bandwidth demand ratio and the number of computing resources, extract the priority value of each business process from the business priority database, and calculate the resource allocation starting value in combination with the current resource usage to obtain the computing resource allocation benchmark value, then read the upper limit of the business impact, compliance requirements and process dependency from the performance threshold library, and compare it with the current resource occupancy index. If any characteristic exceeds the upper limit, increase the resource quota of the corresponding business process to generate a resource quota adjustment value. In the power trading system, for example, the bandwidth demand of the transaction settlement business accounts for 35% of the total bandwidth, the total computing resources are 100 CPU cores and 320GB of memory, and the initial allocation benchmark value is calculated as 30 CPU cores based on the priority of 0.8. If the daily transaction amount reaches 5 billion yuan, exceeding the upper limit of 4 billion yuan, the system automatically increases the resource quota by 20% and adjusts it to 36 CPU cores. This adjustment mechanism ensures that key businesses receive sufficient support under high load.

[0051] For power trading scenarios, three key evaluation characteristics are extracted from real-time monitoring data: business impact, compliance requirements, and process dependency. For example, the settlement business processing time is 85 minutes, close to the compliance limit of 90 minutes, the core process depends on 5 related modules, and the resource occupancy rate shows 45% CPU utilization and 75% memory utilization. When any indicator exceeds the standard, the system calculates the adjustment value based on the minimum guaranteed quota of the dispatch rule base (such as 25 core CPUs and 80GB memory) to ensure that resource allocation meets basic needs and has room for expansion.

[0052] S1032, using a deep reinforcement learning algorithm to train a resource allocation agent, by inputting a dynamic resource scheduling benchmark table and real-time performance indicators, using a reward function to drive the agent to optimize the allocation strategy under resource constraints, outputting a resource allocation optimization plan, and reading resource status data from the computing node pool according to the plan, mapping the business process to generate a specific node allocation plan based on the node processing capacity, and then setting the resource preemption priority and forming a resource allocation decision. In an embodiment of the present invention, the deep reinforcement learning algorithm uses resource utilization and task completion time as reward targets, and inputs the adjusted 36-core CPU quota and real-time load data. Through multiple rounds of iterative learning, it is optimized to the 45-core CPU upper limit. The computing node pool includes 8 high-performance nodes (each node has 16-core CPUs and 64GB of memory) and 12 standard nodes. The settlement business is mapped to 3 high-performance nodes, and each node reserves 15% of resources to cope with sudden demands. According to the business criticality level table, the settlement business is classified as a critical level and has preemption authority to ensure priority allocation when competing for resources.

[0053] In actual operation, when the peak of clearing arrives, the single-core CPU utilization rate rises to 85%. The system identifies high-priority demands through deep reinforcement learning algorithms, reclaims resources for non-critical businesses, increases the CPU allocation for settlement business to 45 cores, and increases the memory to 120GB. The resource scheduling instruction set generates a specific allocation plan based on the node status, such as using all three high-performance nodes for settlement business, while dynamically adjusting the resource ratio of other businesses. Real-time monitoring shows that the settlement business processing time has dropped from 85 minutes to 65 minutes, the CPU utilization rate has stabilized at 80%, and the memory usage rate has been controlled within 70%.

[0054] In an embodiment of the present invention, dynamic scheduling of computing resources is achieved through the synergy of the resource benchmark matrix and the deep reinforcement learning algorithm. In particular, in the scenario where multiple key businesses are processed in parallel, the system can quickly respond to changes in resource requirements based on business priorities and real-time performance indicators. For example, when transaction settlement and real-time transaction processing are running at the same time, the system prioritizes the 45-core CPU requirements of the settlement business while reserving basic operating resources for other businesses. This resource allocation strategy based on criticality assessment not only improves the recovery efficiency of core businesses, but also maintains the overall stability of the system when resources are tight, providing a reliable foundation for subsequent optimization.

[0055] In the embodiment of the present invention, the disaster recovery system ensures that the recovery process of real-time grid alarm data and monthly electricity fee calculation data in the power system meets the timeliness and criticality requirements by evaluating the resource allocation effect. The following steps are based on the resource allocation decision, analyze the deviation between the actual recovery time and the target, and generate the scheduling effect evaluation parameters.

[0056] S104. Extract the actual recovery time of the real-time power grid alarm data and the monthly electricity fee calculation data from the resource allocation decision, and compare them with the target recovery time to generate scheduling effect evaluation parameters, including timeliness deviation and criticality satisfaction.

[0057] First, the alarm processing records and electricity fee calculation task records are obtained from the resource allocation decision database, and the actual recovery time of the two types of services is calculated by timestamp. Subsequently, the time reference matrix is ​​constructed in combination with the historical target time records, and the actual recovery progress is standardized to generate the initial matrix of progress deviation. Next, the alarm data stream is analyzed using a convolutional neural network, and the timeliness features are extracted to generate an evaluation value. At the same time, the time consumption distribution of electricity fee calculation is processed through a time series algorithm to calculate its timeliness evaluation value. Combined with the business criticality level and the preset evaluation rules, the comprehensive scheduling effect evaluation parameters are generated.

[0058] In the power grid dispatching business, taking a power supply bureau as an example, the actual processing time of the transformer temperature over-limit alarm is 15 minutes, the target time is 10 minutes, and the monthly electricity fee calculation task actually takes 6 hours, the target is 4 hours. The time benchmark matrix shows that the alarm processing deviation is 5 minutes and the electricity fee calculation deviation is 2 hours. The standard time is obtained from the business timeliness benchmark library, the alarm processing is 12 minutes, the electricity fee calculation is 5 hours, and the alarm deviation after standardization is 0.25 and the electricity fee deviation is 0.4. This standardization process ensures the comparability of different businesses by normalizing the time scale, laying the foundation for subsequent analysis.

[0059] S1041. Extract the power grid alarm processing records and electricity fee calculation task records from the resource allocation decision, calculate the real-time alarm recovery time and the monthly electricity fee calculation completion time according to the timestamp, and build a time reference matrix in combination with the historical target time record to obtain the actual recovery progress data. Then, standardize the progress data and calculate the initial value of the timeliness deviation according to the standard time, and then process the alarm data stream through the convolutional neural network to generate a timeliness evaluation value. In an embodiment of the present invention, the convolutional neural network receives an alarm processing rate of 3 per minute and a backlog of 15 as input, and uses a three-layer convolution structure, each layer contains 32, 16, and 8 convolution kernels, respectively. The ReLU activation function is used to extract time features, and the output alarm timeliness evaluation value is 0.75, indicating that the processing efficiency is relatively high. The standardization process maps the alarm deviation of 0.25 to the 0-1 interval, which is convenient for comparison with other services.

[0060] S1042. Use a time series algorithm to analyze the time consumption distribution of monthly electricity bill calculation data, extract calculation weights from billing cycle records and generate electricity bill calculation timeliness evaluation values, then build a comprehensive evaluation matrix based on the timeliness evaluation values ​​of alarms and electricity bills, calculate the satisfaction benchmark value and generate critical satisfaction data in combination with the business level data of the key indicator library, and finally classify the timeliness deviation and critical satisfaction according to the preset evaluation rules to generate scheduling effect evaluation parameters. In the time series analysis, the time consumption of electricity bill calculation shows periodic fluctuations, with a peak from the 1st to the 3rd of each month. The weight is set to 0.8, and the calculated timeliness evaluation value is 0.6. The comprehensive evaluation matrix integrates the evaluation values ​​of alarms and electricity bills. In terms of critical satisfaction, the alarm business is the first-level priority, with a benchmark value of 0.9 and an actual value of 0.85; the electricity bill is the second-level priority, with a benchmark value of 0.7 and an actual value of 0.65. The evaluation rules grade the deviation into 0-0.2 as excellent, 0.2-0.4 as good, 0.4-0.6 as fair, and 0.6 and above as needing improvement. The final warning is good and the electricity fee is fair.

[0061] For scenarios with multiple concurrent services, the combination of convolutional neural networks and time series algorithms enables accurate evaluation of recovery effects. For example, the high timeliness of alarm processing benefits from priority resource allocation, while the large deviation in electricity fee calculation reflects the disadvantage of secondary priority services in resource competition. The evaluation parameters show that the comprehensive score of alarms is 0.8 and the electricity fee is 0.62, indicating that the system performs well in ensuring high-priority services. This evaluation mechanism provides data support for the adjustment of resource scheduling strategies by quantifying timeliness and criticality deviations, and can effectively identify bottlenecks and optimize resource allocation efficiency, especially when the load fluctuates.

[0062] In the embodiment of the present invention, the disaster recovery system optimizes resource scheduling through a closed-loop feedback mechanism to ensure the recovery efficiency of real-time grid alarm data and monthly electricity fee calculation data in the power system. The following steps are based on the scheduling effect evaluation parameters and use the PID controller to achieve dynamic update of the priority.

[0063] S105. The scheduling effect evaluation parameters are input into the PID controller as feedback signals, the control error is calculated by the deviation between the actual scheduling effect and the preset target, the adjustment amount of the timeliness difference and the business criticality level weight is output, and a priority update scheme is generated in combination with a deep neural network to optimize the next round of scheduling parameters.

[0064] Extract the timeliness deviation and criticality satisfaction from the scheduling effect evaluation parameters, build a feedback signal and compare it with the preset target value to generate a control error sequence. Then, through the proportional, integral and differential operations of the PID controller, calculate the total amount of timeliness difference adjustment, and use the deep neural network to map the business criticality weight adjustment value, and generate an update plan based on the priority calculation rules. This closed-loop control mechanism can continuously optimize the scheduling strategy according to the actual operating status.

[0065] S1051. Extract the timeliness deviation value and criticality satisfaction from the dispatch effect evaluation parameters to construct a feedback signal, calculate the control error according to the preset dispatch target value and generate a control error sequence, then obtain the proportional coefficient, integral time constant and differential time constant from the controller parameter library, and obtain the proportional adjustment amount, integral adjustment amount and differential adjustment amount through three operations respectively, and weightedly combine them into the total timeliness difference adjustment amount. In a power supply bureau scenario, the timeliness deviation is 0.3, the target is 0.1, the criticality satisfaction is 0.7, the target is 0.9, and the errors are 0.2 and 0.2 respectively. The proportional coefficient is set to 1.5, and the error is amplified to 0.3; the integral time constant is 0.1, the cumulative 10-cycle deviation is 2.8, and the adjustment amount is 0.28; the differential time constant is 0.2, the error change rate is 0.05, and the adjustment amount is 0.01. After weighted combination, the total adjustment amount is 0.354, reflecting the comprehensive correction needs of the deviation.

[0066] In power grid dispatching, the core of the PID controller is to achieve fast response and long-term stability through three adjustments. The proportional term provides immediate correction, the integral term eliminates cumulative errors, and the differential term predicts the trend of changes. For example, when the alarm processing delay increases, the proportional term quickly increases resource allocation, the integral term ensures that the deviation does not continue to deviate from the target through historical data, and the differential term avoids over-adjustment. This design allows the system to maintain balance under dynamic loads.

[0067] S1052. Receive the total amount of timeliness difference adjustment through a deep neural network, extract characteristic parameters and establish a weight mapping model to generate a business critical weight adjustment value, then normalize the total amount of adjustment and the weight adjustment value, generate a priority update scheme based on the combination coefficient of the priority calculation rule base, and build a new round of scheduling parameters based on the update scheme to complete closed-loop control. In an embodiment of the present invention, the deep neural network includes three hidden layers, with the number of neurons being 64, 32, and 16 respectively. The ReLU function is used to process the input 0.354, and the output high-priority business weight is increased by 0.15, and the low priority is reduced by 0.1. After normalization, they are 0.7 and 0.3 respectively, and the combination coefficients of 0.6 and 0.4 are calculated to increase the priority by 20%. After a new round of scheduling, the timeliness deviation is reduced to 0.15, and the satisfaction is increased to 0.85. The controller parameters are adjusted to a proportional coefficient of 1.3 and an integral time constant of 0.12 to optimize the control performance.

[0068] For multi-business scenarios, deep neural networks use nonlinear mapping to ensure that weight adjustments adapt to the characteristics of different businesses. For example, real-time alarm data has a higher weight increase due to its high timeliness requirements, while electricity bill calculations are relatively conservative. This differentiated strategy improves the pertinence of resource allocation. The closed-loop feedback mechanism gradually approaches the target through multiple rounds of iterations, especially when the load fluctuates, and can quickly adjust the priority to ensure the efficient recovery of key businesses.

[0069] In the embodiment of the present invention, the disaster recovery system ensures that the recovery requirements of various business data in the power system are met through real-time monitoring and dynamic optimization. The following steps are based on the priority update scheme to further adjust resource allocation to cope with system load changes.

[0070] S106. After obtaining the priority update plan, the system load in the concurrent request processing is monitored. If the system load exceeds the preset threshold, the rebalancing of the network bandwidth allocation is triggered in combination with the timeliness deviation. The bandwidth allocation optimization plan is generated through the neural network predictor and the deep reinforcement learning agent, and resource reallocation is performed to obtain resource optimization results. At the same time, an iterative optimization mechanism is used to further adjust the bandwidth allocation when the timeliness deviation exceeds the standard.

[0071] Load indicators such as processor occupancy, memory usage, and network throughput are collected from monitoring points, a load state matrix is ​​constructed, and load prediction values ​​are generated through a neural network predictor. If the prediction value exceeds the threshold, a bandwidth adjustment signal is triggered, and a deep reinforcement learning agent is used to optimize bandwidth allocation based on business priority and resource occupancy data, mapped to specific links for redistribution. In addition, when the timeliness deviation exceeds the threshold, the final resource optimization solution is generated through allocation adjustment algorithm and iterative optimization.

[0072] In the power grid disaster recovery scenario, taking a substation as an example, the current processor occupancy rate is 75%, the memory usage accounts for 80% of the total, that is, 16GB, and the network throughput of the total bandwidth of 1000Mbps reaches 800Mbps. The neural network predictor receives these indicators and adopts a three-layer structure with 64, 32, and 16 neurons in each layer. It extracts features through the ReLU activation function and predicts that the processor occupancy rate will rise to 85%, the memory usage will increase to 18GB, and the throughput will reach 900Mbps in the next 30 minutes. The preset thresholds are 80%, 16GB, and 850Mbps. The predicted value exceeds the standard and the timeliness deviation is 0.3, triggering bandwidth rebalancing. The initial bandwidth allocation is 300Mbps for emergency alarm, 500Mbps for electricity bill calculation, and 200Mbps for daily query. After optimization, it is adjusted to 400Mbps, 400Mbps, and 200Mbps to ensure that key businesses are prioritized.

[0073] S1061. Obtain processor occupancy, memory usage and network throughput indicators from the disaster recovery monitoring point and construct a load state matrix. Process the load data through a neural network predictor to generate future load forecast values. Then, determine whether to trigger a bandwidth allocation adjustment signal based on a preset threshold and timeliness deviation, and use a rebalancing rule matrix and a deep reinforcement learning agent to generate a bandwidth allocation optimization scheme to perform link resource reallocation. In an embodiment of the present invention, the load state matrix integrates concurrent request classification data, with emergency alarms accounting for 20%, electricity bill calculations accounting for 50%, and daily inquiries accounting for 30%. After the predictor outputs an over-threshold result, a rebalancing rule matrix is ​​constructed. Combined with the available bandwidth data and business demand, the deep reinforcement learning agent uses load balancing as a reward function to adjust the bandwidth to 400Mbps for emergency alarms, 400Mbps for electricity bill calculations, and 200Mbps for daily inquiries. Link mapping is based on the status of four trunk links. Emergency alarms give priority to link 3 with a load rate of 55%, and the response time is reduced from 2 seconds to 1.2 seconds, meeting the 1.5-second requirement.

[0074] S1062. When it is detected that the timeliness deviation exceeds the preset threshold, the current occupancy data is obtained from the bandwidth status collection point and a bandwidth allocation benchmark matrix is ​​constructed. A preliminary plan is generated through the allocation adjustment algorithm and the balance is evaluated. If it does not meet the standard, it is iteratively optimized to a balanced state through the deep reinforcement learning algorithm, and then the timeliness change is predicted and calibrated to generate the final resource optimization plan. Taking the substation as an example, the timeliness deviation of 0.35 exceeds the 0.3 threshold. Among the total bandwidth of 1000Mbps, monitoring data accounts for 400Mbps, fault alarms account for 300Mbps, and metering data accounts for 200Mbps. The allocation adjustment algorithm is adjusted to 350Mbps, 350Mbps, and 200Mbps according to the priority of 0.4, 0.35, and 0.25. The balance evaluation shows that the difference in proportion exceeds 10%. After iterative optimization, it is adjusted to 300Mbps, 300Mbps, and 300Mbps, and the standard deviation is reduced to 0.05. The timeliness prediction shows that the deviation has dropped to 0.25. The final solution is 300Mbps monitoring, 350Mbps fault alarm including 50Mbps elastic bandwidth, and 250Mbps metering, with a resource utilization rate of 90%.

[0075] For high-load scenarios with multiple services running concurrently, flexible resource allocation is achieved through prediction and optimization mechanisms. For example, when a link is interrupted, load changes trigger immediate adjustments, and fault alarm traffic is switched to low-load links to ensure continuity. This dynamic management method significantly reduces congestion, stabilizes service response time, and provides efficient support for power grid operation.

[0076] In the embodiment of the present invention, the disaster recovery system evaluates the stability of the dynamic scheduling framework by analyzing the resource optimization results to ensure that the recovery requirements of the real-time grid alarm data and the monthly electricity fee calculation data in the power system are met. The following steps determine the final scheduling framework based on the recovery completion time and critical matching degree.

[0077] S107. Extract the recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data from the resource optimization results, analyze the matching degree of criticality satisfaction and business downtime loss assessment value through neural network and deep learning model, judge the stability of the dynamic scheduling framework and generate the final scheduling parameter plan.

[0078] The completion time records of alarm data and electricity cost calculation data are obtained from the resource optimization results, and the recovery progress matrix is ​​constructed and the business recovery completion degree is calculated. Subsequently, the initial value of critical satisfaction is extracted through the neural network, and the matching coefficient is calculated in combination with the downtime loss assessment value. The deep learning model further analyzes the matching results, predicts the stability of the scheduling framework, and finally optimizes the parameters and determines the scheduling framework. This method ensures that the scheduling strategy takes into account both timeliness and stability through multi-level analysis.

[0079] In a scenario of a regional power supply bureau, the resource optimization results show that the real-time grid alarm data recovery time is 8 minutes, the target is 5 minutes, and the monthly electricity bill calculation task completion time is 4 hours, the target is 6 hours. The recovery progress matrix calculation completion degrees are 0.6 and 0.8 respectively. The neural network processes these data, using a three-layer structure, with 64 neurons in the hidden layer. The features are extracted through the ReLU function, and the output alarm criticality satisfaction is 0.7 and the electricity bill calculation is 0.85. The alarm service is down for 15 minutes, the unit loss is 1,000 yuan / minute, and the total loss is 15,000 yuan; the electricity bill calculation is uninterrupted and the loss is 0. The matching coefficient calculation shows that the alarm is 0.65, which is lower than the 0.8 threshold, and the electricity bill is 0.9, indicating that the alarm guarantee is insufficient.

[0080] S1071. Extract the completion time records of the power grid alarm data and electricity fee calculation data from the resource optimization results and construct a recovery progress matrix to calculate the business recovery completion degree. Generate a critical satisfaction initial value by processing the completion data through a neural network. Then calculate the business downtime loss assessment value based on the downtime duration and the loss benchmark value and perform a matching analysis with the satisfaction to obtain an assessment result. In an embodiment of the present invention, the recovery progress matrix is ​​based on time deviation, with an alarm deviation of 3 minutes and an electricity fee of 2 hours in advance. The neural network inputs these data and outputs satisfaction to reflect the business response capability. The downtime loss is calculated through historical interruption records, and the matching analysis quantifies the degree of fit between satisfaction and loss. A low alarm match indicates that resource allocation needs to be adjusted.

[0081] S1072. Receive the matching evaluation results through the deep learning model and combine the historical stability data to learn the stability characteristics to generate the scheduling framework stability prediction value, then optimize the scheduling parameters according to the prediction value and build the final scheduling framework matrix, and verify the applicability of the framework through business scenarios to determine the final dynamic scheduling framework. The deep learning model contains three hidden layers, with 128, 64, and 32 neurons, input alarm matching coefficient 0.65 and electricity fee 0.9, and output stability index 0.75. After parameter optimization, the alarm resource reservation is increased to 30%, the priority weight is increased to 0.85, and the scheduling cycle is shortened to 30 seconds. The final framework layered design ensures rapid response to high-priority tasks.

[0082] For power grid business, the optimized framework reduces the processing time to 6 minutes when alarms are high, and maintains stable operation through queue optimization during the monthly settlement peak period. This mechanism improves resource utilization and business continuity through data-driven stability evaluation and parameter adjustment.

[0083] like Figure 2 The present invention provides a disaster recovery data recovery system of a disaster recovery system, which mainly includes:

[0084] The priority evaluation module is used to obtain the target timeliness and key evaluation characteristics of concurrent requests, map the key evaluation characteristics to the business criticality level, evaluate the gap between the current recovery status and the target timeliness, obtain the timeliness difference, combine the business criticality level and timeliness difference, build a priority matrix, and calculate the dynamic priority of different data recovery requests;

[0085] The bandwidth allocation module is used to adjust the network bandwidth allocation ratio according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of real-time power grid alarm data is increased to generate a bandwidth allocation plan;

[0086] The resource scheduling module is used to dynamically schedule computing resource allocation according to the bandwidth allocation plan. If the key evaluation characteristics exceed the set upper limit, the computing resource share of the business criticality level is increased to generate resource allocation decisions;

[0087] The effect evaluation module is used to extract the actual recovery time of real-time power grid alarm data and monthly electricity fee calculation data from the resource allocation decision, compare it with the target recovery time, and obtain the scheduling effect evaluation parameters. The scheduling effect evaluation parameters include timeliness deviation and criticality satisfaction;

[0088] The weight adjustment module is used to use the scheduling effect evaluation parameter as a feedback signal, introduce a PID controller, use the deviation between the actual scheduling effect and the preset target as the input error of the controller, output the adjustment amount of the data timeliness difference and the business criticality level weight, and form a priority update plan based on the adjusted weight combination;

[0089] The load monitoring module is used to monitor the system load in concurrent request processing through the disaster recovery system after obtaining the priority update plan. If the system load exceeds the preset threshold, the rebalancing of network bandwidth allocation is triggered in combination with the timeliness deviation to obtain resource optimization results;

[0090] The stability analysis module is used to extract the recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data from the resource optimization results, and to judge the stability of the dynamic scheduling framework and determine the final dynamic scheduling framework through matching analysis between criticality satisfaction and business downtime loss assessment value.

[0091] The above embodiment is only one of the preferred implementation modes of the present invention and should not be used to limit the protection scope of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. A disaster recovery data recovery method for a disaster recovery system, characterized in that: The method comprises: Obtain the target timeliness and key assessment characteristics of concurrent requests, map the key assessment characteristics to business criticality levels, evaluate the gap between the current recovery status and the target timeliness, obtain the timeliness difference, combine the business criticality level and timeliness difference, build a priority matrix, and calculate the dynamic priority of different data recovery requests; The network bandwidth allocation ratio is adjusted according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of real-time power grid alarm data is increased to generate a bandwidth allocation plan; According to the bandwidth allocation plan, dynamically schedule computing resource allocation. If the key evaluation characteristics exceed the set upper limit, increase the computing resource share of the business criticality level and generate resource allocation decisions; The actual recovery time of real-time grid alarm data and monthly electricity fee calculation data is extracted from resource allocation decisions, and compared with the target recovery time to obtain scheduling effect evaluation parameters, which include timeliness deviation and criticality satisfaction. The scheduling effect evaluation parameter is used as the feedback signal, and the PID controller is introduced. The deviation between the actual scheduling effect and the preset target is used as the input error of the controller. The difference in data timeliness and the adjustment amount of the business criticality level weight are output, and the priority update plan is formed according to the adjusted weight combination. After obtaining the priority update plan, the system load in concurrent request processing is monitored through the disaster recovery system. If the system load exceeds the preset threshold, the rebalancing of network bandwidth allocation is triggered in combination with the timeliness deviation to obtain resource optimization results. The recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data is extracted from the resource optimization results. The stability of the dynamic scheduling framework is judged and the final dynamic scheduling framework is determined through the matching analysis between the criticality satisfaction and the business downtime loss assessment value.

2. The method according to claim 1, characterized in that The target timeliness and key evaluation characteristics of concurrent requests are obtained, the key evaluation characteristics are mapped to business criticality levels, the gap between the current recovery status and the target timeliness is evaluated, and the timeliness difference is obtained. In combination with the business criticality level and the timeliness difference, a priority matrix is ​​constructed to calculate the dynamic priority of different data recovery requests, including: Acquire concurrent data requests with business data integrity indicators and resource occupancy rate indicators, wherein the concurrent data requests are used to construct a timeliness target matrix; Calculate the aging difference value according to the aging target matrix, and use the aging difference value to construct the aging difference matrix; Acquiring a business criticality benchmark value from a data criticality quantitative index library, and using the business criticality benchmark value and a normalized result of the timeliness difference matrix to construct a business criticality level matrix; A criticality score is extracted from the business criticality level matrix, and the criticality score and the timeliness difference weight value are input into a neural network model. The neural network model generates a dynamic priority value matrix through a three-layer hidden layer structure.

3. The method according to claim 2, characterized in that Also includes: Obtain key assessment characteristics, classify business criticality according to key assessment characteristics, obtain business criticality level, extract current state information from recovery state, obtain state comparison benchmark value, calculate timeliness value, and obtain timeliness benchmark data, including: A hierarchical analysis is performed on the business impact data and the compliance requirement index values ​​to obtain a criticality value; the criticality value is used to construct a recovery time objective matrix, and status monitoring is performed through the processing deadline threshold in the recovery time objective matrix, and a status benchmark score is calculated for the status monitoring data; a random forest prediction model is established based on the status benchmark score, and the random forest prediction model receives the criticality value and the status benchmark score to generate a prediction value, and the prediction value is compared with the status monitoring data to obtain a status comparison benchmark value; a timeliness calculation rule is constructed based on the status comparison benchmark value, and the target value is extracted from the recovery time objective matrix, and constraint processing is performed according to the service level benchmark requirements. Data is filtered through the processing deadline threshold to obtain timeliness benchmark data.

4. The method according to claim 1, characterized in that The network bandwidth allocation ratio is adjusted according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of the real-time power grid alarm data is increased to generate a bandwidth allocation plan, including: Constructing a bandwidth reference matrix according to the network bandwidth capacity and the network congestion degree, and obtaining the alarm data priority weight from the priority classification value database through the bandwidth reference matrix; Obtaining the priority weight of the alarm data, reading a preset recovery time threshold from a recovery time record library, and obtaining a bandwidth adjustment coefficient if the data transmission rate is higher than the preset recovery time threshold; The bandwidth adjustment coefficient is used to perform proportional amplification calculation on the priority weight of the alarm data, and a bandwidth allocation optimization model is constructed through a neural network algorithm to obtain a bandwidth allocation solution.

5. The method according to claim 1, characterized in that The method dynamically schedules computing resource allocation according to the bandwidth allocation scheme, and if the key evaluation characteristic exceeds the set upper limit, increases the computing resource share of the business criticality level, and generates a resource allocation decision, including: Build a resource benchmark matrix based on the bandwidth demand ratio, obtain the business process priority value from the business priority database, and obtain the computing resource allocation benchmark value; Reading the upper limit value of the business impact degree and the resource occupancy index from the performance threshold library, if the resource occupancy index exceeds the upper limit value of the business impact degree, increasing the corresponding business process resource quota to obtain the resource quota adjustment value; The deep reinforcement learning algorithm is used to train a resource allocation agent, and a resource allocation optimization scheme is obtained through the resource quota adjustment value and the computing resource allocation benchmark value; The resource status data of the computing node pool is read according to the resource allocation optimization scheme, resource mapping is performed on the business process according to the node processing capacity, and a resource allocation decision is generated.

6. The method according to claim 1, characterized in that The actual recovery time of the real-time power grid alarm data and the monthly electricity fee calculation data is extracted from the resource allocation decision, and compared with the target recovery time to obtain the scheduling effect evaluation parameters, which include timeliness deviation and criticality satisfaction, including: Acquire alarm processing records and electricity fee calculation task records from a resource allocation decision database, and obtain real-time alarm recovery duration data and monthly electricity fee calculation completion duration data according to the alarm processing records and the electricity fee calculation task records; Performing standardization processing on the real-time alarm recovery time data and the monthly electricity fee calculation completion time data to obtain a progress deviation initial matrix; Receiving the progress deviation initial matrix through a convolutional neural network, and the output layer of the convolutional neural network generates an alarm processing timeliness evaluation parameter; A scheduling effect evaluation result is generated according to the alarm processing timeliness evaluation parameter and preset evaluation rules, wherein the preset evaluation rules include a timeliness scoring standard and a criticality satisfaction benchmark value.

7. The method according to claim 1, characterized in that The scheduling effect evaluation parameter is used as a feedback signal, a PID controller is introduced, the deviation between the actual scheduling effect and the preset target is used as the input error of the controller, the adjustment amount of the output data timeliness difference and the business criticality level weight is output, and a priority update scheme is formed according to the adjusted weight combination, including: Obtaining a timeliness deviation threshold from a preset scheduling target value, analyzing the scheduling effect evaluation parameters according to the timeliness deviation threshold to obtain a control error sequence; Obtaining a proportional coefficient, an integral time constant, and a differential time constant from a controller parameter library, and using the proportional coefficient, the integral time constant, and the differential time constant to operate the control error sequence to obtain a total amount of timeliness difference adjustment; A deep neural network is used to obtain characteristic parameters of the total amount of timeliness difference adjustment, and a weight mapping model is established according to the characteristic parameters to obtain a business criticality weight adjustment value; A normalization operation is performed on the total amount of timeliness difference adjustment and the business criticality weight adjustment value, and a priority update scheme is generated according to the combination coefficient in the priority calculation rule base.

8. The method according to claim 1, characterized in that After obtaining the priority update scheme, the system load in the concurrent request processing is monitored through the disaster recovery system. If the system load exceeds the preset threshold, the rebalancing of the network bandwidth allocation is triggered in combination with the timeliness deviation to obtain the resource optimization result, including: Obtaining a load index from a monitoring point, and inputting the load index into a neural network predictor, wherein the neural network predictor extracts load features through a hidden layer to obtain a load prediction value; If the load prediction value exceeds a preset threshold, a bandwidth allocation adjustment signal is triggered, and the bandwidth allocation adjustment signal is used to construct a rebalancing rule matrix; According to the rebalancing rule matrix, the deep reinforcement learning agent generates a bandwidth allocation optimization plan according to the service priority and resource occupancy data, and the bandwidth allocation optimization plan is used to perform link resource reallocation; It also includes: if it is detected that the timeliness deviation exceeds a preset threshold, the trigger mechanism is activated, the current state of the network bandwidth is obtained, and a preliminary bandwidth allocation plan is obtained by using an allocation adjustment algorithm. For the allocation result in the preliminary plan, it is determined whether the allocation balance is achieved through resource status evaluation. If not, it is optimized iteratively through the adjustment process to obtain an updated value of the optimization result, and the timeliness change is predicted. The resource optimization plan is determined according to the predicted timeliness, which specifically includes: Obtaining a timeliness deviation value and a bandwidth trigger threshold, if the timeliness deviation value exceeds the bandwidth trigger threshold, obtaining bandwidth occupancy data from a bandwidth status collection point, and constructing a bandwidth allocation reference matrix according to the bandwidth occupancy data to obtain an initial state of bandwidth resources; Processing the initial state of the bandwidth resources by an allocation adjustment algorithm, generating a preliminary bandwidth allocation plan based on feature extraction and priority mapping; Construct a balance evaluation matrix according to the preliminary bandwidth allocation plan, obtain balance threshold data from the resource evaluation library, and calculate the current allocation balance according to the bandwidth proportion of each service; If the current allocation balance is lower than the balance threshold, the iterative optimization mechanism is started, the deep reinforcement learning algorithm is used to process the optimized iteration benchmark value, the iteration termination condition is determined according to the balance improvement, and the updated optimization value is output; The calibration parameters are obtained from the resource optimization rule base, and the updated optimization values ​​are corrected according to the calibration parameters to obtain the resource optimization solution.

9. The method according to claim 1, characterized in that: The recovery completion time of the real-time grid alarm data and the monthly electricity fee calculation data is extracted from the resource optimization results, and the stability of the dynamic scheduling framework is judged through the matching analysis of the critical satisfaction and the business downtime loss assessment value, and the final dynamic scheduling framework is determined, including: Obtaining completion time records of power grid alarm data and electricity fee calculation data in the resource optimization result, constructing a recovery progress matrix according to the completion time records, and obtaining service recovery completion degree data; According to the business recovery completion data, the data is input into a neural network model, and the neural network model extracts features from the business recovery completion data to obtain an initial value of criticality satisfaction; The initial value of the criticality satisfaction is used to calculate the matching degree with the loss benchmark value in the downtime loss assessment library, and the business matching coefficient is determined according to the evaluation criteria in the matching rule library to obtain the matching degree evaluation result; The matching evaluation result is input into a deep learning model, and the deep learning model performs stability feature learning on the matching evaluation result to obtain a scheduling framework parameter solution; The final scheduling framework matrix is ​​constructed based on the scheduling framework parameter scheme, the scheduling scenario data is obtained from the business scenario library, the framework applicability is verified based on the scenario data, and the final dynamic scheduling framework is determined.

10. A disaster recovery data recovery system of a disaster recovery system, characterized in that: The system comprises: The priority evaluation module is used to obtain the target timeliness and key evaluation characteristics of concurrent requests, map the key evaluation characteristics to the business criticality level, evaluate the gap between the current recovery status and the target timeliness, obtain the timeliness difference, combine the business criticality level and timeliness difference, build a priority matrix, and calculate the dynamic priority of different data recovery requests; The bandwidth allocation module is used to adjust the network bandwidth allocation ratio according to the priority. If the target timeliness is higher than the preset recovery time threshold, the bandwidth ratio of real-time power grid alarm data is increased to generate a bandwidth allocation plan; The resource scheduling module is used to dynamically schedule computing resource allocation according to the bandwidth allocation plan. If the key evaluation characteristics exceed the set upper limit, the computing resource share of the business criticality level is increased to generate resource allocation decisions; The effect evaluation module is used to extract the actual recovery time of real-time power grid alarm data and monthly electricity fee calculation data from the resource allocation decision, compare it with the target recovery time, and obtain the scheduling effect evaluation parameters. The scheduling effect evaluation parameters include timeliness deviation and criticality satisfaction; The weight adjustment module is used to use the scheduling effect evaluation parameter as a feedback signal, introduce a PID controller, use the deviation between the actual scheduling effect and the preset target as the input error of the controller, output the adjustment amount of the data timeliness difference and the business criticality level weight, and form a priority update plan based on the adjusted weight combination; The load monitoring module is used to monitor the system load in concurrent request processing through the disaster recovery system after obtaining the priority update plan. If the system load exceeds the preset threshold, the rebalancing of network bandwidth allocation is triggered in combination with the timeliness deviation to obtain resource optimization results; The stability analysis module is used to extract the recovery completion time of real-time power grid alarm data and monthly electricity bill calculation data from the resource optimization results, and to judge the stability of the dynamic scheduling framework and determine the final dynamic scheduling framework through matching analysis between criticality satisfaction and business downtime loss assessment value.

Images