Complex business planning method based on data middle platform and multi-agent collaboration

Abstract

This invention relates to the field of business forecasting and management, specifically to a complex business planning method based on a data platform and multi-agent collaboration. First, it acquires logical sequence deviations and physical environment jitter in parallel, and constructs a multi-dimensional evaluation benchmark using modular arithmetic and a sliding window algorithm, effectively solving the problems of sequence number flipping misjudgment and single-delay fluctuation interference. Second, it introduces a thermodynamic dissipation model to calculate the residual potential energy of missing data, causing it to decay non-linearly with waiting time, and combines business priority and network stability to calculate the real-time utility value of new data. Finally, it constructs a dynamic energy barrier by comprehensively considering buffer backlog depth, and determines whether to break the sequence rules by calculating a jump decision factor. This invention can effectively eliminate head-of-line blocking, ensure the real-time performance of high-priority control commands, and improve the robustness and consistency of industrial distributed collaborative computing.

Description

Technical Field

[0001] This invention relates to the field of business forecasting and management, specifically to a complex business planning method based on a data platform and multi-agent collaboration. Background Technology

[0002] In real-world industrial automation control scenarios, edge computing systems need to process massive amounts of business data streams continuously over long periods. This data typically carries different types of business information, including conventional telemetry data such as voltage and temperature, which have relatively low real-time requirements, as well as control commands that are extremely sensitive to latency, such as fault trips and emergency power support. Due to the complex electromagnetic environment in industrial settings, data transmission inevitably experiences network jitter, packet loss, or out-of-order arrival. The system must respond to these anomalies within millisecond-level time windows to maintain the consistency of the entire network and the stability of control.

[0003] Existing technologies typically employ a single sequence number check or a fixed timeout mechanism to handle out-of-order data. However, in the aforementioned industrial control scenarios, this rigid decision-making logic has significant practical drawbacks: when network jitter causes data packets to become out of order, if the system mechanically blocks and waits based solely on sequence number order, it cannot recognize the urgency of the data content. This can lead to the system blocking subsequent, high-priority emergency control commands, such as emergency stop signals, for extended periods while waiting for an expired or low-value old data packet. This head-of-line blocking phenomenon not only wastes computing resources but also prevents critical control commands from being executed in a timely manner, potentially causing production accidents or even leading to the collapse of the control system. Summary of the Invention

[0004] To address the problem in existing technologies where critical control commands cannot be executed in a timely manner when the head of the queue is blocked, this invention proposes a complex service planning method based on a data platform and multi-agent collaboration. The method includes: responding to the receipt of a new data packet with a discontinuous sequence number, obtaining the transmission delay of the new data packet, and calculating a network environment stability coefficient based on the sliding window standard deviation of the transmission delay; obtaining the corresponding basic weight according to a preset service type identifier within the new data packet, and calculating the real-time utility value of the new data packet by combining the basic weight with the network environment stability coefficient; calculating the blocking waiting time corresponding to the desired sequence number, where the blocking waiting time is the time difference between the current moment and the arrival time of the previous valid data packet, and determining the residual value potential energy of the missing data using a dissipation model based on the blocking waiting time; obtaining the backlog depth of the current data packet buffer, and calculating a jump decision factor by combining the real-time utility value, the residual value potential energy, and the backlog depth; when the jump decision factor is greater than a preset jump threshold, forcibly updating the desired sequence number to match the sequence number of the new data packet, directly processing the new data packet, and marking the missing sequence number as invalid.

[0005] Unlike existing technologies that rely solely on fixed timeouts or simple sliding windows to handle out-of-order or lost data packets, this invention introduces a multi-dimensional dynamic decision-making mechanism. By comprehensively considering network environment stability, the basic weights of different service types, the waiting time for missing data, and the current backlog in the system buffer, it intelligently calculates a skip decision factor. This enables the system to make an optimal trade-off between waiting for missing data to ensure integrity and skipping missing data to ensure real-time performance. It effectively avoids system congestion or delays in critical control commands caused by rigidly waiting for non-critical data under network fluctuations or high load conditions, improving the robustness and timeliness of complex business planning in uncertain network environments.

[0006] Furthermore, the specific method for calculating the jump decision factor is as follows:

[0007] ;

[0008] in This represents the real-time utility value; This represents the residual value potential energy; Indicates the depth of accumulation; This represents the preset minimum stable potential energy; This indicates the preset buffer coefficient.

[0009] Furthermore, the buffer coefficient is dynamically configured according to the system operating mode: when the system is configured in high reliability mode, the buffer coefficient ranges from 0.5 to 1.0; when the system is configured in low latency mode, the buffer coefficient ranges from 1.0 to 2.0.

[0010] Furthermore, the residual value potential energy is calculated using an exponential decay model or a linear approximation model based on the business tolerance half-life; the exponential decay model is as follows: The linear approximation model is: when hour, ,otherwise ;in As a preset basic information constant, The blocking wait time is... The preset business tolerance half-life.

[0011] Furthermore, the method for calculating the real-time utility value is as follows:

[0012] ;

[0013] in The basic weights; The network environment stability coefficient; This is a preset time delay reference constant; The transmission delay; This is the preset minimum value to prevent the elimination of zero.

[0014] This calculation method not only considers the importance of the service itself but also introduces network stability coefficients and transmission latency as penalty factors. This means that even high-priority service data will have its utility value reduced if it experiences excessive latency or network instability during transmission. This helps the system more objectively assess the actual utilization value of data at the current moment and prevents misleading control decisions due to blindly prioritizing low-quality, high-priority data.

[0015] Furthermore, the specific method for calculating the network environment stability coefficient is as follows:

[0016] ;

[0017] in This is the preset sensitivity coefficient; The standard deviation of the sliding window;

[0018] The standard deviation of the sliding window is calculated using an incremental variance update algorithm. When a data packet enters or leaves the sliding window, only the cumulative delay and the cumulative squared delay within the window are updated.

[0019] Furthermore, the basic weight is obtained from a priority mapping table pre-installed in the read-only storage area; the priority mapping table stores the mapping relationship between service type identifiers and basic weights, wherein the basic weight value set for security control services is greater than the basic weight value set for regular telemetry services.

[0020] Furthermore, when determining whether the new data packet is discontinuous with the expected sequence number, the difference between the sequence number of the new data packet and the expected sequence number is calculated; if the absolute value of the difference exceeds half of the preset sequence number range, the range is compensated for the smaller sequence number before the difference calculation is performed.

[0021] This invention resolves the logical ambiguity that may arise during cyclic counting of sequence numbers. In long-running systems, the sequence number resets after reaching its maximum value, and existing technologies are prone to calculation errors at this point. This invention, by determining the relationship between the difference and the range and performing compensation, ensures that the system's judgment of out-of-order orders and packet loss remains accurate even under the boundary condition of sequence number reversal, thus guaranteeing the logical continuity and stability of the business planning method during long-term continuous operation.

[0022] Furthermore, obtaining the backlog depth of the current data packet buffer includes: executing an atomic read instruction to directly access the memory counter that maintains the backlog depth, in order to prevent read-write conflicts under multi-threaded concurrent access.

[0023] In a multi-threaded concurrent environment, atomic instructions are used to directly access the memory counter, replacing the traditional heavyweight locking mechanism. This eliminates the thread blocking and context switching overhead that may occur when accessing the backlog depth during high concurrency, while preventing data inconsistency caused by read-write conflicts. This design significantly improves the concurrent throughput of the packet processing pipeline and ensures that the backlog depth data on which the jump decision factor calculation depends is always real-time and accurate.

[0024] Furthermore, after directly processing the new data packet, the method further includes: inputting the processed effective data stream into the distributed collaborative computing module, iteratively calculating the control parameters using the alternating direction multiplier method, and sending them to the actuator.

[0025] This architecture tightly integrates high-quality, efficient data streams cleaned at the front end with distributed collaborative computing at the back end. It utilizes the Alternating Direction Multiplier Method (ADMM) to decompose large-scale, complex collaborative optimization problems, enabling rapid iteration and convergence of control parameters among multiple agents while ensuring timely data input. This architecture not only reduces the computational burden on individual points but also ensures that the final instructions issued to the actuators are calculated based on intelligently selected optimal data, thus improving the accuracy and response speed of overall collaborative control.

[0026] The technical effects of this invention are as follows:

[0027] This invention proposes an adaptive data stream processing mechanism based on a jump decision factor. Unlike traditional fixed-timeout retransmission strategies, this invention combines service weight, network stability, residual data value, and buffer backlog depth to construct a nonlinear model that dynamically decides whether to wait for out-of-order data or force a jump. This mechanism introduces a dissipation model of data value decaying over time and backlog relief logic to prevent congestion, effectively solving the problem of balancing data congestion and real-time performance in multi-agent collaboration under unstable networks while ensuring the timeliness of critical services. Attached Figure Description

[0028] Figure 1 This is a schematic flowchart illustrating a complex business planning method based on data platform and multi-agent collaboration in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram illustrating the tracking comparison of the actual received sequence number and the expected sequence number as the data packet index changes in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram illustrating the original transmission delay data and the jitter standard deviation curve after sliding window processing in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram illustrating the normalized curve of the network environment stability coefficient changing with the data stream processing process in an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram illustrating the decay curves of residual value potential energy under different half-lives based on the thermodynamic dissipation model in an embodiment of the present invention.

[0033] Figure 6 This is a schematic diagram illustrating the comparison between the exponential decay model for calculating residual value potential energy in this embodiment of the invention and the linear approximation implementation.

[0034] Figure 7 This is a schematic scatter plot illustrating the real-time utility values ​​of data packets of different service types in an embodiment of the present invention.

[0035] Figure 8 This is a line graph illustrating the dynamic changes in the backlog depth of the receive buffer and the jump trigger time in an embodiment of the present invention.

[0036] Figure 9 This is a scatter plot illustrating the relationship between the numerical distribution of the jump decision factor and the jump threshold in an embodiment of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention relates to a high-performance computing terminal operating at the edge of an industrial Internet of Things (IIoT) system. The terminal's physical architecture includes a core processing unit, a storage unit, a communication interface unit, and a high-precision clock unit.

[0039] The core processing unit, for example, employs an industrial-grade SoC based on a multi-core ARM Cortex-A53 or RISC-V architecture, with a clock speed of 1.2GHz and a floating-point unit (FPU) to support subsequent calculations. The storage unit includes DDR4 high-speed memory and eMMC non-volatile memory. Within the memory, the system has a capacity of [missing information]. A FIFO ring buffer is used to temporarily store out-of-order data packets; it connects via a CAN-FD bus or an industrial Ethernet interface. A distributed intelligent agent node (such as a servo driver or power monitoring terminal). The interface integrates a hardware timestamp unit with nanosecond precision; simultaneously, the controller internally maintains a sequence state machine that records the currently expected sequence number in real time. And monitor the number of backlogged packets in the buffer in real time, and record it as .

[0040] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0041] Example of a complex business planning method based on data platform and multi-agent collaboration:

[0042] like Figure 1 As shown, the complex business planning method based on data platform and multi-agent collaboration of the present invention includes:

[0043] S100 receives edge node service data streams in real time and decouples logical sequence deviations and physical environment jitter characteristics in parallel to construct a multi-dimensional evaluation benchmark, thereby solving the rigidity problem caused by single sequence number determination.

[0044] The controller receives raw business data streams from edge nodes in real time. To address the rigidity caused by relying on a single sequence number, the system first decomposes the input data, mapping it to the discreteness of the logical dimension and the environmental stability of the physical dimension.

[0045] S101. Parse the transport layer sequence number and use modular arithmetic logic to calculate the sequence deviation index between the current packet and the expected packet to quantify the dispersion of the data and thus accurately identify out-of-order data packets.

[0046] The processor parses the received data packets and extracts the transport layer sequence number. The system calculates the serial number and the expected serial number. Sequence bias index between:

[0047] ;

[0048] in This is the maximum range for the serial number. If... The packet is determined to be "out-of-order arrival" and is temporarily stored in a circular buffer. Modular operations are used to handle the natural overflow problem of finite-word-length sequence numbers. In long-term operation in industrial settings, sequence numbers... It will inevitably reach its maximum capacity. (For example, 65535, a 16-bit unsigned integer) and a flip occurs. Therefore, in the calculation When this happens, the system needs to execute loopback verification logic: if the absolute value of the directly calculated difference exceeds... If this occurs, it is determined that a cross-cycle flip has occurred, and compensation needs to be applied to the smaller sequence number. Then differential calculation is performed. This mechanism ensures that the system will not be misjudged as a serious out-of-order fault when the serial number naturally returns to zero during 24 / 7 uninterrupted operation.

[0049] like Figure 2 As shown in the figure, the relationship between the actual data packet sequence numbers received by the system and the expected sequence numbers maintained locally is illustrated over a period of time. The dashed line, rising in a stepped pattern, represents the sequence numbers the system expects to receive, while the solid line, fluctuating around this dashed line, represents the actual arriving sequence numbers. It can be seen that there is a clear out-of-order phenomenon in the actually arriving data packets, but the overall trend remains consistent.

[0050] S102. Extract timestamps to calculate transmission delay and use a sliding window statistical algorithm to obtain a normalized network environment stability coefficient as a weight parameter to prevent single delay fluctuations from misleading system decisions.

[0051] To prevent time delay fluctuations at a single moment from misleading decision-making, the system also introduces an environmental stability coefficient as a weight for subsequent utility evaluation.

[0052] The processor extracts data packets and generates timestamps. With current arrival time Calculate transmission delay At the same time, using a length of The sliding window statistics calculate the standard deviation of latency for the most recent 50 data packets. Considering the limited computing resources of edge computing terminals, to avoid performing a full traversal calculation every time the sliding window moves, this embodiment employs the Welford online algorithm or an incremental variance update strategy. Specifically, the controller memory maintains the latency of the current window and... and the sum of squared delays When new data packets enter the window and old data packets leave the window, only two addition and subtraction operations are needed to update these two statistics, which can then be updated using the formula. Fast solution. This approach reduces computational complexity from... Reduce to This ensures real-time performance in microsecond-level interrupt service routines.

[0053] like Figure 3 As shown in the figure, the transmission delay of data packets and the calculated jitter standard deviation vary with the data packet index. The thin, volatile line represents the instantaneous transmission delay of each data packet, while the smoother, thicker line represents the jitter standard deviation after processing using the sliding window algorithm. It can be seen that although the instantaneous delay has significant randomness and glitches, the jitter standard deviation curve obtained through window statistics can more stably reflect the fluctuation trend of the current network environment.

[0054] The processor calculates the network environment stability coefficient using the following formula normalization. :

[0055] ;

[0056] in The sensitivity coefficient can be set to 0.5 in this embodiment.

[0057] From the above formula, it can be seen that when network jitter... When it increases, It will approach 0; when the network is stable. Approaching 1.

[0058] like Figure 4 As shown in the figure, the normalized variation curve of the network environment stability coefficient with the packet processing process is illustrated. The boundary line of the filled area in the figure represents the calculated environment stability coefficient, and the horizontal dashed line represents the preset medium stability threshold. It can be observed from the figure that when network jitter intensifies, the coefficient drops rapidly and moves away from 1.0; while when network conditions improve, the coefficient gradually recovers. This figure verifies the effectiveness of the algorithm in mapping physical environment jitter to weights between 0 and 1, ensuring the system's dynamic perception of changes in network quality.

[0059] S200 monitors the blocking wait time in real time and calculates the residual value potential energy of missing data packets based on a nonlinear exponentially decaying thermodynamic dissipation model to quantify the decreasing waiting necessity over time.

[0060] For expected data packets that have not yet arrived and the system is waiting for This embodiment introduces a thermodynamic dissipation mechanism. The system assumes that the validity of missing data decreases non-linearly as the blocking time increases.

[0061] In one embodiment, the processor monitors in real time the time difference between the current moment and the arrival time of the previous valid packet, i.e., the blocking duration. The system calculates the residual value potential energy based on the following formula. :

[0062] ;

[0063] in This represents the basic information content constant, the normalization benchmark, which is preferably 100 in this embodiment; This indicates the service tolerance half-life. For 50Hz power frequency services, The preferred setting is 5ms (i.e., 1 / 4 cycle); for general industrial control, this value can be adjusted between 2ms and 20ms.

[0064] From the above formula, we can see that the residual value potential energy It's about time. It is a monotonically decreasing function.

[0065] when hour, At this point, the value of waiting for this data is greatest. Exceed At that time, the exponent term The potential energy drops sharply to 37% of its original value. This model prevents the system from idling for meaningless historical data.

[0066] like Figure 5 As shown in the figure, the residual value potential of missing data packets decays with varying blocking duration under different service tolerance half-life parameter settings. The figure includes multiple curves with different slopes, corresponding to different half-life settings (e.g., 2ms, 5ms, 10ms, etc.). It can be intuitively seen that as the blocking duration increases, the value potential of all curves exhibits a non-linear decreasing trend. The shorter the half-life, the steeper the curve's descent, indicating a lower system tolerance for latency in this type of service. This figure vividly illustrates how the thermodynamic dissipation model quantifies the loss of data value over time.

[0067] In another embodiment, to cover low-computing-power scenarios, such as FPGA hardware, the following calculation method can also be used:

[0068] like Then perform linear operations. ;otherwise This embodiment achieves a similar attenuation effect through shifting and subtraction, and thus also falls within the protection scope of this invention.

[0069] like Figure 6 As shown in the figure, the accurate exponential decay model is compared with the linear approximation model under low computing power scenarios. The smoothly descending solid line in the figure represents the theoretical exponential decay trajectory, while the dashed line, which descends linearly and is truncated at a specific time point, represents the simplified linear approximation trajectory. Both maintain a high degree of fit within a specific time window. This figure demonstrates that, under limited hardware resources, using linear operations instead of exponential operations can still effectively simulate the dissipation characteristics of data value.

[0070] S300: Based on the business emergency flag, query the preset priority mapping table and combine it with the network environment stability coefficient to calculate the real-time utility value of the new data in order to clearly quantify the business gain brought by directly processing the data.

[0071] For the new data packets that arrive out of order The system assesses the direct processing value of the data packet through the physical channel. First, the processor reads the urgent flag bit of the data packet. The basic weights are obtained through a pre-configured business priority mapping table (Lookup Table). .

[0072] Specifically, the service priority mapping table is stored in the controller's read-only memory (Flash) and contains key-value pairs of 'service type ID' and 'basic weight'. This embodiment defines the following typical industrial scenario mapping relationship:

[0073] (1) Type ID 0x01 (Regular Telemetry): Corresponding weight This type of data includes slowly varying parameters such as voltage and temperature; the loss of a few packets has a negligible impact on the system.

[0074] (2) Type ID 0x02 (State Synchronization): Corresponding weight This type of data involves the coordinated action of multiple agents and requires close attention.

[0075] (3) Type ID 0x03 (Stability Control): Corresponding weight This type of data includes fault trips and emergency power support commands, and is of the highest priority.

[0076] Through this hierarchical mapping, the system can ensure that under the same network jitter... The utility value calculated from the stability control message. It is much higher than that of conventional telemetry messages.

[0077] Example: When the flag is "heartbeat packet", When the flag is set to "Emergency Stop", Then, combined with the results obtained in step S102... Calculate real-time utility value :

[0078] ;

[0079] in This is the time delay reference constant, which can be set to 10ms; To prevent the value from being zero or a minimum.

[0080] If network jitter is severe or the current new data itself has a large latency, even The value is very high, the final It will also be limited to a smaller value.

[0081] like Figure 7 As shown in the figure, the distribution of real-time utility values ​​for different types of service data packets in a mixed service flow scenario is illustrated. The scatter points in the figure represent individual data packets, and their vertical axes represent the calculated real-time utility values. It is evident that data points representing "routine telemetry" are mainly concentrated in the bottom low-utility region, while data points representing "emergency shutdown" or "stability control" are distributed in the higher utility range. This figure verifies that the proposed method can effectively stratify data flows of different importance along the value dimension based on service urgency flags and network environment.

[0082] S400 integrates the utility of new data with the residual potential energy of old data and introduces a buffer backlog deep feedback mechanism to calculate the jump decision factor in order to construct a dynamic energy barrier and thus decide whether to break the sequence rules.

[0083] The controller calculates the dynamically changing jump decision factor based on reference values ​​of real-time utility value and residual value potential energy. This is used to determine whether strict sequence rules have been broken, as shown below:

[0084] ;

[0085] in This represents the real-time utility value output in step S300; This represents the residual value potential energy of the old data output in step S200; This represents the number of buffer backlog packets directly read from the memory counter (it is an integer, and its range is 1). ); Represents the minimum stable potential energy, preferably taking ; This represents the buffer coefficient, which determines the system's sensitivity to congestion. In actual deployment, it can be dynamically configured according to the network topology.

[0086] In high reliability mode, set At this point, the weight of the subtrahend term is relatively small, and the system tends to wait conservatively, trying to restore the complete sequence as much as possible, and only jumping when there is extreme congestion;

[0087] In low latency mode, set If even a small amount of pressure buildup occurs in the buffer zone at this point, It will rise rapidly, prompting the system to aggressively discard old data to ensure that the latest control commands can be executed immediately.

[0088] It should be noted that, due to It is a critical resource shared by the 'Receive Interrupt Service Routine' and the 'Main Processing Thread'. To prevent data inconsistency caused by read-write conflicts, the processor uses atomic operation instructions, such as the LDREX / STREX instructions under the ARM architecture, to directly read the counter value in the memory address, or briefly disables interrupts at the moment of reading, to ensure that the obtained backlog depth can truly reflect the current congestion status.

[0089] like Figure 8 As shown in the figure, the dynamic change of the backlog in the receive buffer over time and the jump trigger point are illustrated. The jagged line in the figure represents the real-time backlog depth of the buffer, which fluctuates continuously due to enqueue and dequeue operations. The dots at the peaks of the line indicate that a sequence jump operation was triggered at that moment. It can be seen that when the buffer backlog depth reaches a local high point, it is often accompanied by a jump event, which reflects the algorithm's regulatory role in eliminating congestion.

[0090] The above formula uses The numerator increases quadratically as the real-time utility value of new data increases. Therefore, for extremely high-priority tasks, the system will very likely ignore residual value potential. The obstruction directly triggers a jump, thereby eliminating the blockage at the head of the queue;

[0091] When the buffer is free, for example hour, , The subtrahend term is large. If the price is suppressed, wait for the old data and do not jump to a new level easily.

[0092] When the buffer is severely overloaded, for example hour, , The subtrahend term approaches 0. No longer suppressed, the system is under greater pressure and no longer waits for excellent sequences to be generated. Instead, it tends to jump aggressively to clear the backlog and avoid deadlock.

[0093] To ensure the program's success during runtime in the above embodiments, the processor checks the denominator term before calculation. .like Injunction This ensures the jump occurs immediately.

[0094] S500, in response to the jump decision factor exceeding the threshold, forcibly corrects the expected sequence state machine and drives distributed collaborative computation based on the cleaned effective data stream to output consistency control instructions.

[0095] The processor will calculate the jump decision factor With preset jump threshold In comparison, in this embodiment, the jump threshold The preferred setting is 5.0. When When this occurs, a jump is triggered, at which point the controller forcibly sends the desired sequence number. Updated to It marks the missing sequence number as "permanently invalid" and directly processes the current new data packet, mapping its payload to the network state vector table in memory.

[0096] If no jump is triggered, the data packet remains in the buffer, and the system continues to perform the loop monitoring in step S200 until old data arrives or a timeout occurs. Decay until it is sufficient to trigger a jump.

[0097] like Figure 9 As shown in the figure, this is a time-series analysis of the calculated jump decision factors. The scatter points in the figure represent the decision factor values ​​calculated at each time step, and the horizontal dashed line represents the preset jump threshold. The decision factor values ​​of the vast majority of data points are below the threshold, indicating that the system continues to wait; while the few scatter points that are significantly higher than the threshold correspond to the moments when the system decides to break the sequence rules and execute a jump.

[0098] Whether the data arrives normally or after skip processing, it is sent to the downstream ADMM distributed collaborative computing module. Based on the cleaned latest state slice, this module iteratively calculates control parameters such as power allocation coefficients or torque commands, and sends them to the actuators via an optocoupler-isolated interface.

Claims

1. A complex business planning method based on a data middle platform and multi-agent collaboration, characterized in that, The method includes: in response to receiving a new data packet with a non-continuous sequence number, obtaining the transmission delay of the new data packet, and calculating a network environment stability coefficient based on the sliding window standard deviation of the transmission delay; The corresponding basic weight is obtained based on the preset service type identifier in the new data packet, and the real-time utility value of the new data packet is calculated by combining the basic weight with the network environment stability coefficient. Calculate the blocking wait time corresponding to the expected sequence number, where the blocking wait time is the time difference between the current time and the arrival time of the previous valid data packet, and use a dissipation model based on the blocking wait time to determine the residual value potential energy of the missing data. Obtain the backlog depth of the current data packet buffer, and calculate the jump decision factor by combining the real-time utility value, the residual value potential energy, and the backlog depth; When the jump decision factor is greater than the preset jump threshold, the expected sequence number is forcibly updated to match the sequence number of the new data packet, the new data packet is processed directly, and the missing sequence number in the middle is marked as invalid. 2.The complex business planning method based on data middle platform and multi-agent collaboration according to claim 1, wherein, The specific method for calculating the jump decision factor is as follows: ； wherein represents the real-time utility value; represents the residual value potential; represents the backorder depth; represents a preset minimum stable potential; represents a preset buffer coefficient. 3.The complex business planning method based on data platform and multi-agent collaboration according to claim 2, characterized in that, The buffer coefficient is dynamically configured according to the system operating mode: When the system is configured in high reliability mode, the buffer coefficient ranges from 0.5 to 1.

0. When the system is configured in low latency mode, the buffer coefficient ranges from 1.0 to 2.

0. 4.The method of claim 2, wherein, The residual value potential is calculated using an exponential decay model or a linear approximation model based on the business tolerance half-life. The exponential decay model is: ; The linear approximation model is: when , , otherwise ; wherein is a preset basic information amount constant, is the blocking waiting time length, is a preset service tolerance half-life.

5. The complex business planning method based on data platform and multi-agent collaboration according to claim 2, characterized in that, The specific method for calculating the real-time utility value is as follows: ； in The basic weights; The network environment stability coefficient; This is a preset time delay reference constant; The transmission delay; This is the preset minimum value to prevent the elimination of zero.

6. The complex business planning method based on data platform and multi-agent collaboration according to claim 5, characterized in that, The specific method for calculating the network environment stability coefficient is as follows: ； in This is the preset sensitivity coefficient; The standard deviation of the sliding window; The standard deviation of the sliding window is calculated using an incremental variance update algorithm. When a data packet enters or leaves the sliding window, only the cumulative delay and the cumulative squared delay within the window are updated.

7. The complex business planning method based on data platform and multi-agent collaboration according to claim 5, characterized in that, The basic weights are obtained from a priority mapping table pre-placed in the read-only storage area; The priority mapping table stores the mapping relationship between service type identifiers and basic weights, wherein the basic weight value set for stability control services is greater than the basic weight value set for regular telemetry services.

8. The complex business planning method based on data platform and multi-agent collaboration according to claim 1, characterized in that, When determining whether the new data packet is discontinuous with the expected sequence number, the difference between the sequence number of the new data packet and the expected sequence number is calculated; If the absolute value of the difference exceeds half of the preset serial number range, the range is compensated for the smaller serial number before the difference calculation is performed.

9. The complex business planning method based on data platform and multi-agent collaboration according to claim 1, characterized in that, Get the current backlog depth of the packet buffer, including: An atomic read instruction is executed to directly access the memory counter that maintains the backlog depth, in order to prevent read-write conflicts under multi-threaded concurrent access.

10. The complex business planning method based on data platform and multi-agent collaboration according to claim 1, characterized in that, After directly processing the new data packet, the process also includes: The processed effective data stream is input into the distributed collaborative computing module, which uses the alternating direction multiplier method to iteratively calculate the control parameters and then sends them to the actuator.

Images