A community property multi-source perception data cloud computing intelligent scheduling method

Abstract

The application discloses a community property multi-source sensing data cloud computing power intelligent scheduling method, and relates to the technical field of cloud computing resource scheduling.The application comprises the following steps: acquiring multi-source sensing data and device space task mapping, generating a current task according to abnormal data, and determining a candidate task based on a device action area, a property space unit and a task input-output relationship; determining an early calculation section and a real-time data dependence boundary according to a calculation step of the candidate task, an obtainable state of an input item and a corresponding space unit, and updating the boundary with a new output according to a current task; forming a heritable task set according to a calculation section output item and a candidate task input condition, determining a calculation section execution sequence according to avoidable repeated calculation amount and saving a pre-calculation state when cloud computing power is insufficient; and selecting a matching state closest to a boundary after the candidate task is triggered.The application can reduce the repeated calculation of subsequent tasks and the residual processing amount after real-time data arrives.

Description

Technical Field

[0001] This invention belongs to the field of cloud computing resource scheduling technology, and in particular relates to an intelligent scheduling method for cloud computing power of multi-source sensing data in community properties. Background Technology

[0002] With the development of smart communities, the Internet of Things, and cloud computing technologies, community property management systems are gradually integrating various sensing devices such as video surveillance, access control, elevators, fire protection, water immersion, parking, and environmental monitoring. These devices continuously generate sensing data with different data types and processing requirements. To meet the needs of community emergency identification, equipment status analysis, and coordinated response, it is necessary to utilize cloud computing resources to process and collaboratively analyze multi-source sensing data in a timely manner.

[0003] Currently, multi-source sensing data in community property management typically establishes corresponding cloud computing tasks based on data sources or business types. When a certain sensing data reaches a preset trigger condition, a corresponding computing task is created and cloud computing resources are allocated to it to complete the loading of running data, preparation of the computing environment, processing of sensing data, and output of results. Although different sensing tasks may be triggered consecutively by the same sudden event, and the data required for some calculation steps in subsequent computing tasks can be provided by the results of previous tasks, device configuration data, or existing sensing data, the existing scheduling method usually still treats each task as an independent computing unit. It only starts to execute as a whole after the subsequent sensing data is actually generated and forms a task, thus delaying the calculation steps that could have been completed in advance to be processed after the task is triggered. At the same time, if computing resources are reserved or allocated in advance for possible subsequent tasks, it is easy to generate invalid computing power occupation due to changes in the event development path, making it difficult to balance the response speed of subsequent tasks and the efficiency of cloud computing power utilization.

[0004] Existing cloud-based scheduling technology for multi-source sensing data in community properties cannot effectively utilize available computing power to complete the computation process of subsequent tasks that already have the conditions for execution based on the data dependency relationship between the sensing tasks that have already occurred and the candidate subsequent computation tasks. It also cannot effectively reduce the repeated computation and startup waiting after the subsequent tasks are triggered. Therefore, this invention proposes an intelligent cloud-based computing power scheduling method for multi-source sensing data in community properties. Summary of the Invention

[0005] The purpose of this invention is to provide a cloud computing power intelligent scheduling method for multi-source sensing data in community properties, which solves the problems of existing cloud scheduling technology for multi-source sensing data in community properties. It cannot make reasonable use of the available computing power to complete the calculation process of subsequent tasks that have the conditions for execution in advance based on the data dependency relationship between the sensing tasks that have occurred and the candidate subsequent calculation tasks, nor can it effectively reduce the problems of repeated calculations and start-up waiting after the subsequent tasks are triggered.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to a method for intelligent scheduling of cloud computing power for multi-source sensing data in community property management, comprising the following steps: S1. Acquire multi-source sensing data and device spatial task mapping, mapping and associating the device's operating area, property space unit, and computing task; generate the current task from abnormal data, and form an event-associated spatial unit set, referred to as the event set, with the abnormal device's operating area and the task output associated area; determine the task whose device's corresponding spatial unit coincides with the event set and whose at least one input item is provided by the current task output as a candidate task. S2. Obtain the sequential calculation steps and input items of the candidate task. All input items can be determined as the pre-calculated segments by the continuous steps provided by the acquired perception data, configuration data and the current task output. The first subsequent step that depends on the non-generated real-time perception data is determined as the real-time data dependency node and a boundary is formed. S3. When a new output is added to the current task, update the event set and the available status of node input items according to the associated region of the new output; when all input items of a node are available, include it in the pre-calculated segment, redetermine the node in the next step, and move the boundary to the next step. S4. Match the output items of the computation segment with the input items of the candidate tasks, and the set of input space units is included in the set of output space units of the computation segment to form an inheritable task set; determine the amount of computation that can avoid duplication by multiplying the computational power requirement of the computation segment by the number of tasks in the set minus one; when the available computing power in the cloud is insufficient to execute all computation segments, execute in descending order of computational power and save the pre-computation state, the set of inheritable tasks, the event set and the boundary. S5. When the real-time data in the event set meets the triggering conditions of the candidate task, it is determined as the target task. The state with the largest node step number is selected from the pre-calculated states of the target task in the inheritable task set. The real-time data is input into the corresponding node, so that the target task inherits the state and executes the remaining steps.

[0007] Furthermore, when establishing the device space task mapping, the community property monitoring area is divided into multiple property space units and space unit identifiers are configured; the device identifiers, installation locations, coverage boundaries, and corresponding calculation task identifiers of each sensing device are obtained; the coverage boundaries are projected onto the property space units, and the device operating area is determined based on the overlapping area between the coverage boundaries and the property space units; the correspondence between the device identifier, the device operating area, the space unit identifier, and the calculation task identifier is recorded.

[0008] Furthermore, in determining the set of event-related spatial units, the anomaly sensing device identifier is determined based on the anomaly sensing data, and the device spatial task mapping is queried to obtain the initial set of spatial units corresponding to the area of ​​action of the anomaly sensing device; the device identifier and spatial location data associated with the current task output data item are extracted, and the device spatial task mapping is queried to obtain the supplementary set of spatial units; the initial set of spatial units and the supplementary set of spatial units are merged to obtain the set of event-related spatial units.

[0009] Furthermore, the determination of candidate tasks includes the following: obtaining the set of spatial units and the set of input data item identifiers corresponding to the sensing devices associated with each computing task from the device space task mapping; the intersection of the computing spatial unit set and the event-associated spatial unit set; and determining the computing tasks whose intersection is not empty and whose set of input data item identifiers contains at least one current task output data item identifier as candidate tasks.

[0010] Furthermore, the determination of the pre-calculated segments and real-time data dependent nodes includes: writing the acquired sensing data, configuration data, and data item identifiers of the current task output into the data item availability table; for each candidate task, querying the input data item identifiers of each calculation step sequentially from the first calculation step according to the calculation step order; determining the consecutive calculation steps where all input data item identifiers exist in the data item availability table as pre-calculated segments; and determining the first calculation step after the pre-calculated segment that contains real-time sensing data item identifiers not written into the data item availability table as a real-time data dependent node.

[0011] Furthermore, the shifting of the real-time data dependency boundary includes: when the current task generates a new output data item, the identifier of the new output data item is written into the data item availability table, and the event association spatial unit set is updated according to the device action area associated with the new output data item; when all input data item identifiers of the current real-time data dependency node exist in the data item availability table, the node is included in the advance calculation segment; starting from the next calculation step of the node, the input data item identifiers are queried sequentially, and the continuous calculation steps where all input data item identifiers exist in the data item availability table are continued to be included in the advance calculation segment, and the first subsequent calculation step containing a real-time sensing data item identifier not written into the data item availability table is determined as the new real-time data dependency node, and the real-time data dependency boundary is updated.

[0012] Furthermore, when forming an inheritable task set, output description data is generated for each pre-computed segment. The output description data includes output data item identifiers, output data types, and output space unit sets. Input condition data for each candidate task is obtained. Input condition data includes input data item identifiers, input data types, and input space unit sets. Candidate tasks whose output data item identifiers are the same as their input data item identifiers, whose output data types are the same as their input data types, and whose input space unit sets are contained within their output space unit sets are added to the inheritable task set of the corresponding pre-computed segment.

[0013] Furthermore, a mapping version identifier is configured for the device space task mapping; the pre-calculated state is associated with the current task identifier, the set of inheritable tasks, the set of event-associated spatial units, the mapping version identifier, the real-time data dependent node step identifier, and the state generation time; after determining the target task, the pre-calculated states that contain the target task in the set of inheritable tasks, that contain the set of spatial units corresponding to the target task in the set of event-associated spatial units, and that have the same mapping version identifier as the mapping version identifier used when the target task is triggered are selected, and the target task trigger time is later than the state generation time and the time interval between the two is not greater than the preset state validity duration. The pre-calculated state with the largest real-time data dependent node step number is selected from the pre-calculated states obtained from the selection.

[0014] Furthermore, for the same candidate task, the pre-calculated state includes spatial state segments divided according to property space units. Each spatial state segment is associated with a spatial unit identifier, pre-calculated output, and real-time data dependency boundary. The pre-calculated output of each spatial state segment is generated from configuration data, current task output, and the perception data of the corresponding property space unit, and does not depend on the perception data of other property space units. After the event-related spatial unit set is updated, the intersection of the event-related spatial unit sets before and after the update is determined as the state-preserving spatial unit, the spatial units added to the updated set relative to the updated set are determined as the state-supplementing spatial units, and the spatial units reduced from the updated set relative to the updated set are determined as the state-invalidating spatial units. The spatial state fragments corresponding to the retained state spatial units are made invalid, and the spatial state fragments corresponding to the invalid state spatial units are made invalid. For the state supplement spatial units, the pre-calculated segments are determined and executed according to the order of the candidate tasks to obtain the supplementary state fragments. The retained spatial state fragments and supplementary state fragments are combined according to the spatial unit identifier to form the updated pre-calculated state and record the real-time data dependency boundaries corresponding to each spatial state fragment. After the candidate task is determined as the target task, the remaining calculation steps are executed from the real-time data dependency nodes corresponding to each spatial state fragment.

[0015] Furthermore, the formation of the updated pre-computation state includes: obtaining the step numbers of the real-time data dependency nodes corresponding to each reserved spatial state fragment and supplementary state fragment of the same candidate task, determining the real-time data dependency node corresponding to the smallest step number as the shared dependency node; and combining the pre-computation outputs formed by each spatial state fragment before the shared dependency node into a shared pre-computation state. In the spatial state segment where the step number of the real-time data dependency node is greater than the step number of the shared dependency node, the pre-computed output formed from the shared dependency node to the corresponding real-time data dependency node is saved as the spatial depth state. The spatial depth state is associated with the spatial unit identifier and the corresponding real-time data dependency boundary. After the target task is triggered, the shared pre-computation state is inherited. The target task processing spatial unit is matched with the spatial depth state. The spatial unit that is successfully matched executes the remaining calculation steps from the real-time data dependency node of the corresponding spatial depth state, and the spatial unit that is not matched with the spatial depth state executes the remaining calculation steps from the shared dependency node.

[0016] The present invention has the following beneficial effects: This invention establishes a correspondence between the operating area of ​​sensing devices, property space units, and computing tasks through device space task mapping, and forms a set of event-related spatial units based on the operating area of ​​abnormal devices and the output of the current task; candidate tasks are simultaneously constrained by spatial overlap and input-output relationships, so that the cloud only processes tasks that are related to the current property event by both device and space, reducing the participation of irrelevant tasks in the calculation and the mismatch of data in different property areas.

[0017] This invention determines the pre-calculated segments and real-time data dependency boundaries based on the calculation steps of candidate tasks, the available states of input data, and corresponding spatial units, and dynamically moves the output boundaries as the current task is added. When cloud computing power is insufficient, the execution order is determined according to the computing power requirements of the calculation segment and the number of inheritable tasks, so that multiple subsequent tasks inherit the same pre-calculated state, reducing the repeated execution of the same calculation steps and shortening the processing time of the target task after the arrival of real-time data.

[0018] When the set of event-related spatial units changes, this invention divides the pre-computation state into spatial state segments, retains the segments still within the event's influence range, performs supplementary calculations on newly added spatial units, and invalidates segments that move out of the range, thus avoiding the overall abandonment of existing calculation results. Given the different dependency boundaries of each spatial state segment, a shared pre-computation state is formed using the foremost dependency node, and calculation results exceeding this node are saved as spatial depth states, allowing the target task to continue execution according to the existing calculation progress of each spatial unit, reducing state mismatches and invalid recalculations. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the overall process of intelligent cloud computing power scheduling in this invention.

[0021] Figure 2 This is a flowchart illustrating the pre-calculated segment determination and boundary shifting process in this invention.

[0022] Figure 3 This is a flowchart of the spatial state fragment update and hierarchical inheritance process in this invention. Detailed Implementation

[0023] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0024] Example 1

[0025] See Figure 1 This invention relates to a method for intelligent scheduling of cloud computing power for multi-source sensing data in community property management, comprising the following steps: This embodiment takes a fire in a community underground garage as an example. A data gateway is set up in the community, and smoke sensors, temperature sensors, access control devices, elevator controllers and smoke exhaust devices communicate with the data gateway. After receiving the sensing data, the data gateway sends the device identification, collection time and data value to the cloud server. The cloud server stores the device space task mapping, the calculation steps of each calculation task and the input and output relationships of the steps, and performs candidate task determination, pre-calculation and pre-calculation state inheritance.

[0026] S1. Acquire multi-source sensing data and device spatial task mapping, mapping and associating the device's operating area, property space unit, and computing task; generate the current task from abnormal data, and form an event-associated spatial unit set, referred to as the event set, with the abnormal device's operating area and the task output associated area; determine the task whose device's corresponding spatial unit coincides with the event set and whose at least one input item is provided by the current task output as a candidate task. When establishing the device space task mapping, the community property monitoring area is divided into multiple property space units and space unit identifiers are configured; the device identifiers, installation locations, coverage boundaries, and corresponding calculation task identifiers of each sensing device are obtained; the coverage boundaries are projected onto the property space units, and the device operating area is determined based on the overlapping area between the coverage boundaries and the property space units; the correspondence between the device identifier, device operating area, space unit identifier, and calculation task identifier is recorded.

[0027] Taking the first basement level of Building 3 as an example, it is divided into the East Garage Area P01, Elevator Lobby P02, Evacuation Passage P03, and West Garage Area P04. Each property space unit is represented by the area boundary in the building plan. After the installation location and coverage boundary of the sensing device are converted to the same building coordinate system, they are judged to coincide with the area boundary of the property space unit. The smoke sensor corresponds to the garage area covered by its detection range, the access control device corresponds to the evacuation area on both sides of the door, and the elevator controller corresponds to the area where the elevator lobby and elevator shaft are located.

[0028] In the equipment space task mapping, in addition to recording the correspondence between sensing devices and property space units, the sensing devices required for smoke exhaust linkage tasks, elevator return and descent tasks, and evacuation access control tasks are also recorded. When the installation location, detection range, or task configuration of the equipment changes, the corresponding mapping records are updated, and a new mapping version identifier is configured for the updated mapping.

[0029] In determining the set of spatial units associated with an event, the anomaly sensing device identifier is determined based on the anomaly sensing data, and the device spatial task mapping is queried to obtain the initial set of spatial units corresponding to the area of ​​action of the anomaly sensing device; the device identifier and spatial location data associated with the current task output data item are extracted, and the device spatial task mapping is queried to obtain the supplementary set of spatial units; the initial set of spatial units and the supplementary set of spatial units are merged to obtain the set of spatial units associated with the event.

[0030] The determination of candidate tasks includes the following: obtaining the set of spatial units and the set of input data item identifiers corresponding to the sensing devices associated with each computing task from the device space task mapping; the intersection of the computing spatial unit set and the event-associated spatial unit set; and determining the computing tasks whose intersection is not empty and whose set of input data item identifiers contains at least one current task output data item identifier as candidate tasks.

[0031] When the smoke sensor in P01 of the East Zone of the garage generates abnormal data, the cloud server generates a fire assessment task as the current task. The current task outputs the event type, event level, and event impact space. Each output data item carries the source device identifier or space unit identifier. The smoke exhaust linkage task requires the event type and event impact space as input, the elevator descent task requires the event type as input, and the evacuation access control task requires the event level and event impact space as input. Therefore, the above tasks can use the data items output by the current task. The input items being provided by the current task output means that the candidate task has at least one input data item that can be provided by the current task output, not that all of its input data items have been generated.

[0032] When the initial abnormal device is located in the east area of ​​the garage, P01, the initial spatial unit set includes P01. After the current task determines that the elevator lobby P02 is affected based on the subsequent data from the temperature sensor, P02 is added to the supplementary spatial unit set, and the event-related spatial unit set is updated accordingly to include P01 and P02.

[0033] S2. Obtain the sequential calculation steps and input items of the candidate task. All input items can be determined as the pre-calculated segments by the continuous steps provided by the acquired perception data, configuration data and the current task output. The first subsequent step that depends on the non-generated real-time perception data is determined as the real-time data dependency node and a boundary is formed. See Figure 2 The determination of the pre-calculated segments and real-time data dependent nodes includes: writing the acquired sensing data, configuration data, and data item identifiers of the current task output into the data item availability table; for each candidate task, querying the input data item identifiers of each calculation step in sequence, starting from the first calculation step; determining the consecutive calculation steps where all input data item identifiers exist in the data item availability table as pre-calculated segments; and determining the first calculation step after the pre-calculated segment that contains real-time sensing data item identifiers not written into the data item availability table as a real-time data dependent node.

[0034] Taking the smoke exhaust linkage task as an example, its calculation process includes the following steps: determining the space unit to be processed based on the event type and building smoke exhaust configuration; querying the smoke exhaust equipment and ventilation path based on the space unit to be processed; reading the real-time operating status of the smoke exhaust equipment and the smoke concentration of the corresponding area; and generating the smoke exhaust control sequence based on the reading results.

[0035] After the current task outputs the event type and the space affected by the event, the corresponding data item identifier, along with the data item identifier of the building smoke exhaust configuration, is written into the data item availability table. The data required for the first two calculation steps are already available, and their processing space is located within the event-related space unit set, so the first two calculation steps are executed first. The real-time operating status of the smoke exhaust equipment required for the third calculation step has not yet been uploaded, so the third calculation step is determined as the real-time data dependency node, and the execution results of the first two calculation steps are saved as the pre-calculation status.

[0036] The data item availability table is used to record the data items that can be read by the current computation step; for received sensing data, currently valid configuration data, and output data generated by the current task, their corresponding data item identifiers are marked as available.

[0037] S3. When a new output is added to the current task, update the event set and the available status of node input items according to the associated region of the new output; when all input items of a node are available, include it in the pre-calculated segment, redetermine the node in the next step, and move the boundary to the next step. The shifting of the real-time data dependency boundary includes: when the current task generates a new output data item, the identifier of the new output data item is written into the data item availability table, and the event association spatial unit set is updated according to the device action area associated with the new output data item; when all input data item identifiers of the current real-time data dependency node exist in the data item availability table, the node is included in the advance calculation segment; starting from the next calculation step of the node, the input data item identifiers are queried sequentially, and the continuous calculation steps where all input data item identifiers exist in the data item availability table are continued to be included in the advance calculation segment, and the first subsequent calculation step containing a real-time sensing data item identifier that has not been written into the data item availability table is determined as the new real-time data dependency node, and the real-time data dependency boundary is updated.

[0038] After the smoke exhaust equipment uploads its operating status, the cloud server writes the data item identifier of that operating status into the data item availability table. If its corresponding spatial unit has already entered the updated event-associated spatial unit set, the input conditions for the third calculation step are met, the cloud server executes this step, and continues to check the next calculation step. If there is still no real-time sensing data to be generated in the next calculation step, this step is treated as a new real-time data dependency node. If the input data for the next calculation step is available, execution continues until a new real-time data dependency node is encountered.

[0039] The update of the event-related spatial unit set and the update of the data item availability table reflect the scope of event influence and the data generation status, respectively. The update of the event-related spatial unit set reflects the current scope of event influence, and the update of the data item availability table reflects the generation status of the input data of the calculation step. The real-time data dependency boundary is determined based on the available status of the input data items of the candidate task calculation step. The event-related spatial unit set is used for candidate task screening, pre-computation state space matching, and spatial state fragment update.

[0040] S4. Match the output items of the computation segment with the input items of the candidate tasks, and the set of input space units is included in the set of output space units of the computation segment to form an inheritable task set; determine the amount of computation that can avoid duplication by multiplying the computational power requirement of the computation segment by the number of tasks in the set minus one; when the available computing power in the cloud is insufficient to execute all computation segments, execute in descending order of computational power and save the pre-computation state, the set of inheritable tasks, the event set and the boundary. When forming an inheritable task set, output description data is generated for each pre-computed segment. The output description data includes the output data item identifier, the output data type, and the output space unit set. The input condition data of each candidate task is obtained. The input condition data includes the input data item identifier, the input data type, and the input space unit set. Candidate tasks whose output data item identifier is the same as the input data item identifier, whose output data type is the same as the input data type, and whose input space unit set is contained in the output space unit set are added to the inheritable task set of the corresponding pre-computed segment.

[0041] The cloud server writes the output data item identifier, data type, and output space unit set of the pre-calculated segment into the output description data; when the input data item identifier and data type of the candidate task are consistent with the output description data, and its input space unit set does not exceed the output space unit set, the candidate task is added to the corresponding inheritable task set.

[0042] The computing power requirement of a computing segment can be determined by the processor time occupied by each execution of the segment. If there is no execution record, the estimated processor time recorded in the task configuration is used. The cloud server combines the computing power requirement of the computing segment with the number of candidate tasks that can inherit the computing segment to determine the amount of redundant computation that can be reduced after the computing segment is executed in advance. When the available computing power is insufficient, computing segments that can reduce more redundant computation are executed first.

[0043] The pre-computation state stores the output data of completed computation steps, the last completed step, and the next real-time data dependency node, and associates them with the corresponding set of inheritable tasks, the set of event-associated spatial units, the mapping version identifier, and the state generation time.

[0044] Specifically, let the computational power requirement of a pre-calculated segment be T, and its corresponding inheritable task set contain N candidate tasks. Without pre-calculated state inheritance, each of the N candidate tasks would need to execute the computation segment N times. With pre-calculated state inheritance, only one pre-calculated segment needs to be executed, and the remaining N-1 candidate tasks can directly inherit the pre-calculated output of the computation segment. Therefore, the amount of redundant computation that can be avoided for a computation segment is determined as follows:

[0045] The available computing power in the cloud is insufficient to execute all pre-compute segments, which means that the computing resources that can be allocated to pre-compute tasks in the current scheduling cycle are lower than the total computing power requirement of all pre-compute segments to be executed.

[0046] S5. When the real-time data in the event set meets the triggering conditions of the candidate task, it is determined as the target task. The state with the largest node step number is selected from the pre-calculated states of the target task in the inheritable task set. The real-time data is input into the corresponding node, so that the target task inherits the state and executes the remaining steps.

[0047] Configure a mapping version identifier for the device space task mapping. Associate the pre-calculated state with the current task identifier, the set of inheritable tasks, the set of event-associated spatial units, the mapping version identifier, the real-time data dependency node step identifier, and the state generation time. After determining the target task, filter the pre-calculated states where the set of inheritable tasks contains the target task, the set of spatial units corresponding to the target task is contained in the set of event-associated spatial units, and the mapping version identifier is the same as the mapping version identifier used when the target task is triggered. Ensure that the target task trigger time is later than the state generation time and the time interval between the two is not greater than the preset state validity duration. Select the pre-calculated state with the largest real-time data dependency node step number from the filtered pre-calculated states.

[0048] When any real-time sensing data of the smoke exhaust equipment status, elevator position, or access control status meets the triggering condition of the corresponding candidate task, the corresponding candidate task is determined as the target task. The cloud server filters the pre-calculated states based on the target task identifier, processing space, and current mapping version, and excludes states that have exceeded the validity period. When multiple pre-calculated states are obtained, the state with the latest real-time data dependency node is used, and the newly received real-time sensing data is input into that node to continue executing the calculation steps that have not yet been completed.

[0049] Example 2

[0050] See Figure 3In the above embodiments, as the current task continues to generate new outputs, the set of event-related spatial units may expand or shrink, causing the pre-calculated state and the updated event impact range to no longer be completely consistent. The pre-calculated results of some spatial units can still be used, while some spatial units need to be supplemented with calculations. At the same time, different spatial units are affected by the timing of the generation of sensing data and the availability of data, resulting in differences in the number of calculation steps that can be executed in advance, leading to different real-time data dependency boundary positions corresponding to each spatial state. If the original pre-calculated state is directly discarded, it will result in a waste of completed calculation results and cloud computing power. If it is directly inherited as a whole, it may cause a mismatch between the spatial range and the calculation progress. In order to solve this problem, in this embodiment, for the same candidate task, the pre-calculated state includes spatial state segments divided according to property spatial units. Each spatial state segment is associated with a spatial unit identifier, pre-calculated output, and real-time data dependency boundary. The pre-calculated output of each spatial state segment is generated by configuration data, current task output, and sensing data of the corresponding property spatial unit, and does not depend on the sensing data of other property spatial units. When forming spatial state segments, each segment uses the same configuration data and current task output, and reads the perception data of the corresponding property space unit respectively; a spatial state segment can be retained independently if its pre-calculated output does not need to read the perception data of other property space units; calculation steps that require the simultaneous use of perception data of multiple property space units are not included in a single spatial state segment, but are executed as calculation steps by all spatial state segments.

[0051] The spatial state segment records the spatial unit identifier, the completed calculation steps, the corresponding pre-calculation output, and the next real-time data dependency node. When updating the pre-calculation state, the cloud server finds the corresponding segment based on the spatial unit identifier and does not directly splice the perceived values ​​of different spatial units.

[0052] After the event-related spatial unit set is updated, the intersection of the event-related spatial unit sets before and after the update is determined as the state-preserving spatial unit, the spatial units added to the updated set relative to the updated set are determined as the state-supplementing spatial units, and the spatial units reduced from the updated set relative to the updated set are determined as the state-invalidating spatial units. The spatial state fragments corresponding to the retained state spatial units are made invalid, and the spatial state fragments corresponding to the invalid state spatial units are made invalid. For the state supplement spatial units, the pre-calculated segments are determined and executed according to the order of the candidate tasks to obtain the supplementary state fragments. The retained spatial state fragments and supplementary state fragments are combined according to the spatial unit identifier to form the updated pre-calculated state and record the real-time data dependency boundaries corresponding to each spatial state fragment. After the candidate task is determined as the target task, the remaining calculation steps are executed from the real-time data dependency nodes corresponding to each spatial state fragment.

[0053] For example, the set of event-related spatial units before the update included P01 in the east area of ​​the garage, P02 in the elevator lobby, and P03 in the evacuation passage. The set after the update includes P01, P02, and P04 in the west area of ​​the garage. The spatial state fragments corresponding to P01 and P02 are retained, while the spatial state fragment corresponding to P03 is invalidated. For the newly added P04, the parts that can be completed in advance are executed according to the calculation steps of the candidate tasks, and a supplementary state fragment corresponding to P04 is generated.

[0054] The combined updated pre-computation state contains the correspondence between spatial unit identifiers and spatial state fragments. When the target task processes a spatial unit, it reads the corresponding spatial state fragment based on the spatial unit identifier and continues execution from the real-time data dependency node recorded in that fragment.

[0055] The process of updating the pre-computation state includes: obtaining the step numbers of the real-time data dependency nodes corresponding to each reserved spatial state fragment and supplementary state fragment of the same candidate task, and determining the real-time data dependency node corresponding to the smallest step number as the shared dependency node; and combining the pre-computation outputs formed by each spatial state fragment before the shared dependency node into a shared pre-computation state. In the spatial state segment where the step number of the real-time data dependency node is greater than the step number of the shared dependency node, the pre-computed output formed from the shared dependency node to the corresponding real-time data dependency node is saved as the spatial depth state. The spatial depth state is associated with the spatial unit identifier and the corresponding real-time data dependency boundary. After the target task is triggered, the shared pre-computation state is inherited. The target task processing spatial unit is matched with the spatial depth state. The spatial unit that is successfully matched executes the remaining calculation steps from the real-time data dependency node of the corresponding spatial depth state, and the spatial unit that is not matched with the spatial depth state executes the remaining calculation steps from the shared dependency node.

[0056] In one execution of this embodiment, the real-time data dependency node corresponding to P01 is the fourth calculation step, the real-time data dependency node corresponding to P02 is the fifth calculation step, and the real-time data dependency node corresponding to P04 is the third calculation step. The cloud server determines the third calculation step as the shared dependency node and forms a shared pre-calculation state by combining the pre-calculation outputs of each spatial state fragment formed before the third calculation step.

[0057] The output generated by P01 after the third calculation step and before the fourth calculation step is saved as the spatial depth state of P01; the output generated by P02 after the third calculation step and before the fifth calculation step is saved as the spatial depth state of P02; P04 has not yet executed the third calculation step and does not generate a spatial depth state; after the target task is triggered, the shared pre-computation state is loaded first; when processing P01 and P02, the corresponding spatial depth states are read respectively, and execution continues from their respective real-time data dependency nodes; when processing P04, execution continues from the shared dependency node.

[0058] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling of cloud computing power for multi-source sensing data in community property management, characterized in that, Includes the following steps: S1. Acquire multi-source sensing data and device spatial task mapping, wherein the mapping is associated with the device's operating area, property space unit, and computing task; generate the current task from abnormal data, and form an event-associated spatial unit set, referred to as the event set, with the abnormal device's operating area and the task output associated area; determine the task whose device's corresponding spatial unit coincides with the event set and whose at least one input item is provided by the current task output as a candidate task. S2. Obtain the sequential calculation steps and input items of the candidate task. All input items can be determined as the pre-calculated segments by the continuous steps provided by the acquired perception data, configuration data and the current task output. The first subsequent step that depends on the non-generated real-time perception data is determined as the real-time data dependency node and a boundary is formed. S3. When a new output is added to the current task, the status can be obtained by updating the event set and node input items according to the associated region of the new output. When all input items for a node are available, it is included in the pre-calculated segment, and the node is redefined in the next step, thus shifting the boundary to the next step. S4. Match the output items of the computation segment with the input items of the candidate tasks, and form an inheritable task set by including the input space unit set of the output space unit set of the computation segment. The amount of computation that can avoid duplication is determined by multiplying the computation power requirement of the computation segment by the number of tasks in the set minus one. When the available computing power in the cloud is insufficient to execute all computation segments, execute in descending order of the computation amount and save the pre-computation state, inheritable task set, event set and boundary. S5. When the real-time data in the event set meets the triggering conditions of the candidate task, it is determined as the target task. The state with the largest node step number is selected from the pre-calculated states of the target task in the inheritable task set. The real-time data is input into the corresponding node, so that the target task inherits the state and executes the remaining steps.

2. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 1, characterized in that, When establishing the device space task mapping, the community property monitoring area is divided into multiple property space units and space unit identifiers are configured; the device identifier, installation location, coverage boundary and corresponding calculation task identifier of each sensing device are obtained; the coverage boundary is projected onto the property space unit, the device action area is determined according to the overlapping area of ​​the coverage boundary and the property space unit, and the correspondence between the device identifier, device action area, space unit identifier and calculation task identifier is recorded.

3. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 2, characterized in that, In determining the set of event-related spatial units, the anomaly sensing device identifier is determined based on the anomaly sensing data, and the device spatial task mapping is queried to obtain the initial set of spatial units corresponding to the area of ​​action of the anomaly sensing device; the device identifier and spatial location data associated with the current task output data item are extracted, and the device spatial task mapping is queried to obtain the supplementary set of spatial units; the initial set of spatial units and the supplementary set of spatial units are merged to obtain the set of event-related spatial units.

4. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 3, characterized in that, The determination of the candidate tasks includes the following: Obtain the set of spatial units and the set of input data item identifiers corresponding to the sensing devices associated with each computing task from the device space task mapping; calculate the intersection of the set of spatial units and the set of event-associated spatial units; determine the computing tasks whose intersection is not empty and whose set of input data item identifiers contains at least one current task output data item identifier as candidate tasks.

5. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 1, characterized in that, The determination of the pre-calculated segments and real-time data-dependent nodes includes: writing the acquired sensing data, configuration data, and data item identifiers of the current task output into the data item availability table; for each candidate task, querying the input data item identifiers of each calculation step sequentially from the first calculation step according to the calculation step order; determining the consecutive calculation steps where all input data item identifiers exist in the data item availability table as pre-calculated segments; and determining the first calculation step after the pre-calculated segments that contains real-time sensing data item identifiers not written into the data item availability table as real-time data-dependent nodes.

6. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 5, characterized in that, The shifting of the real-time data dependency boundary includes: when the current task generates a new output data item, the identifier of the new output data item is written into the data item availability table, and the event association spatial unit set is updated according to the device action area associated with the new output data item; when all the input data item identifiers of the current real-time data dependency node exist in the data item availability table, the node is included in the advance calculation segment; starting from the next calculation step of the node, the input data item identifiers are queried sequentially, and the continuous calculation steps where all the input data item identifiers exist in the data item availability table are continued to be included in the advance calculation segment, and the first subsequent calculation step containing a real-time sensing data item identifier that has not been written into the data item availability table is determined as the new real-time data dependency node, and the real-time data dependency boundary is updated.

7. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 1, characterized in that, When forming the inheritable task set, output description data is generated for each pre-computed segment. The output description data includes an output data item identifier, an output data type, and an output space unit set. Input condition data for each candidate task is obtained. The input condition data includes an input data item identifier, an input data type, and an input space unit set. Candidate tasks whose output data item identifier is the same as the input data item identifier, whose output data type is the same as the input data type, and whose input space unit set is contained in the output space unit set are added to the inheritable task set of the corresponding pre-computed segment.

8. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 1, characterized in that, Configure a mapping version identifier for the device space task mapping. The pre-calculated state is associated with the current task identifier, the set of inheritable tasks, the set of event-associated spatial units, the mapping version identifier, the real-time data dependent node step identifier, and the state generation time. After determining the target task, filter the pre-calculated states where the set of inheritable tasks contains the target task, the set of spatial units corresponding to the target task is contained in the set of event-associated spatial units, and the mapping version identifier is the same as the mapping version identifier used when the target task is triggered. Ensure that the target task trigger time is later than the state generation time and the time interval between the two is not greater than the preset state validity duration. Select the pre-calculated state with the largest real-time data dependent node step number from the filtered pre-calculated states.

9. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 1, characterized in that, For the same candidate task, the pre-calculated state includes spatial state segments divided according to property space units. Each spatial state segment is associated with a spatial unit identifier, pre-calculated output, and real-time data dependency boundary. The pre-calculated output of each spatial state segment is generated by configuration data, current task output, and perception data of the corresponding property space unit, and does not depend on the perception data of other property space units. After the event-related spatial unit set is updated, the intersection of the event-related spatial unit sets before and after the update is determined as the state-preserving spatial unit, the spatial units added to the updated set relative to the updated set are determined as the state-supplementing spatial units, and the spatial units reduced from the updated set relative to the updated set are determined as the state-invalidating spatial units. The spatial state fragments corresponding to the retained state spatial units are made invalid, and the spatial state fragments corresponding to the invalid state spatial units are made invalid. For the state supplement spatial units, the pre-calculated segments are determined and executed according to the order of the candidate tasks to obtain supplementary state fragments. The retained spatial state fragments and supplementary state fragments are combined according to the spatial unit identifier to form an updated pre-calculated state and record the real-time data dependency boundaries corresponding to each spatial state fragment. After the candidate task is determined as the target task, the remaining calculation steps are executed from the real-time data dependency nodes corresponding to each spatial state fragment.

10. The intelligent scheduling method for cloud computing power of multi-source sensing data in community property management according to claim 9, characterized in that, The formation of the updated pre-computation state includes: obtaining the step sequence number of the real-time data dependency node corresponding to each reserved spatial state segment and supplementary state segment of the same candidate task, determining the real-time data dependency node corresponding to the smallest step sequence number as the shared dependency node; and combining the pre-computation outputs formed by each spatial state segment before the shared dependency node into a shared pre-computation state. In the spatial state segment where the step number of the real-time data dependency node is greater than the step number of the shared dependency node, the pre-calculated output formed from the shared dependency node to the corresponding real-time data dependency node is saved as the spatial depth state. The spatial depth state is associated with the spatial unit identifier and the corresponding real-time data dependency boundary. After the target task is triggered, the shared pre-computation state is inherited. The target task processing spatial unit is matched with the spatial depth state. The spatial unit that is successfully matched executes the remaining calculation steps from the real-time data dependency node of the corresponding spatial depth state, and the spatial unit that is not matched with the spatial depth state executes the remaining calculation steps from the shared dependency node.

Images