Greenhouse climate monitoring method and system

Abstract

The invention discloses a greenhouse climate monitoring method and system, particularly relates to the technical field of Internet of Things, and aims to construct lightweight health self-inspection for distributed edge nodes for greenhouse climate monitoring, execute first heartbeat detection, perform first cross validation on adjacent edge nodes, and adjust the frequency of the first cross validation in real time, so as to improve the reliability of the greenhouse climate monitoring. Performing first node anomaly evaluation and classification on the edge nodes, performing second node anomaly evaluation on suspicious nodes by applying second cross validation, rechecking the suspicious nodes, marking the suspicious nodes separated from the distributed framework as abnormal nodes, performing intervention operation on the abnormal nodes, and detecting a network communication state if intervention fails. According to the method and the device, the contradiction between limited computing resources and efficient anomaly detection is effectively solved, the fault of the whole distributed framework caused by a single-point fault is avoided, and the fault tracing efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of the Internet of Things. More specifically, the present invention is a method and system for monitoring the climate in a greenhouse. Background Art

[0002] The environmental parameters in a greenhouse, such as temperature, humidity, carbon dioxide concentration, etc., usually do not change suddenly and violently, but slowly shift. As a result, it is difficult for the existing climate monitoring of greenhouses based on edge computing with a distributed architecture to detect in a timely manner whether some edge nodes fail. Moreover, the greenhouse adjustment strategy based on precise control tends to keep the climate parameters in the greenhouse at the edge position within the suitable range for crop growth. The slow shift of some climate parameters is difficult to detect in a timely manner. If complex predictive algorithms are directly adopted, it will inevitably occupy the limited computing resources of the edge nodes, affecting the main task of the edge nodes in processing climate data.

[0003] On the other hand, the climate monitoring of greenhouses based on a distributed architecture usually has redundant design to avoid the failure of climate monitoring in greenhouses caused by abnormal failures of some edge nodes. However, even if an abnormal node clearly sends a fault signal to notify the cloud server or other nodes to intervene, there is still a situation where the cloud server or other nodes fail to intervene, and it is difficult to determine the specific cause of the intervention failure.

[0004] To solve the above defects, a technical solution is proposed now. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for monitoring the climate in a greenhouse to solve the problems in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A method for monitoring the climate in a greenhouse, the specific steps include:

[0007] Construct a lightweight health self-check for the distributed edge nodes of the greenhouse climate monitoring, which is used to perform the first heartbeat detection, perform the first cross-verification on adjacent edge nodes, and adjust the frequency of the first cross-verification in real time in combination with the memory occupancy rate, historical failure rate and data stability of the edge nodes;

[0008] Perform the first node anomaly assessment on the edge nodes according to the results of the first heartbeat detection and the first cross-verification, and classify the edge nodes. The edge node types include suspicious nodes and non-suspicious nodes;

[0009] Perform the second node anomaly assessment on the suspicious nodes by using the second cross-verification, review the suspicious nodes, judge whether the suspicious nodes are detached from the distributed framework. If the suspicious nodes are detached from the distributed framework, mark the suspicious nodes detached from the distributed framework as abnormal nodes, and any adjacent edge node performs an intervention operation on the abnormal nodes;

[0010] If the intervention fails, the abnormal node sends a high-priority heartbeat signal to the adjacent edge node. When the high-priority heartbeat signal does not reach the adjacent edge node, the network communication status is detected. When the high-priority heartbeat signal reaches the adjacent edge node but there is no response, the computing resources of the adjacent edge node that receives the high-priority heartbeat signal are evaluated.

[0011] Determine the faulty path based on the network communication status and computing resource evaluation, and notify the staff.

[0012] Preferably, the construction method for constructing a lightweight health self-check for the first heartbeat detection is as follows:

[0013] Each edge node periodically sends a heartbeat signal to the cloud server to report the operating status of the edge node. The heartbeat signal includes the timestamp when the heartbeat signal is sent, the computing resource utilization rate of the edge node, the backlog status of the task queue, and the status of the monitoring sensor data. When it is detected that there is a delay in sending the heartbeat signal, insufficient computing resources of the edge node, backlog of the task queue, or invalid sensor data, it is determined that the heartbeat signal for the first heartbeat detection is abnormal; otherwise, it is determined that the heartbeat signal for the first heartbeat detection is not abnormal.

[0014] Preferably, the sending period of the heartbeat signal for the first heartbeat detection is adaptively adjusted according to the computing resource status of the edge node and historical heartbeat anomaly data. The adaptive adjustment method is as follows:

[0015] Define the initial cycle time R0 for the first heartbeat detection. Then the cycle time for the first heartbeat detection is Tlow = R0 * (1 + α1 * Rag + α2 * Ahr + α3 * And), where α1, α2, and α3 are the weight coefficients of the average CPU occupancy rate Rag, the abnormal heartbeat ratio Ahr in the past N cycles, and the average network delay And in the past N cycles, respectively, and α1, α2, and α3 are all positive numbers. Among them, the average CPU occupancy rate Rag is obtained through the greenhouse climate monitoring API, and the calculation expression for the abnormal heartbeat ratio Ahr in the past N cycles is The calculation expression for the average network delay And in the past N cycles is where ad i is the network delay value in the i-th cycle.

[0016] Preferably, the method for performing the first cross-validation on adjacent edge nodes is as follows:

[0017] Each edge node periodically sends its own sensor data to the adjacent edge nodes, and calibrates the sensor data as Sep k, where k is the number of sensor data types, and k = {1, 2, 3…, p}, p is the total number of sensor data types, and p is a positive integer. Calculate the data deviation of adjacent edge nodes at the same time point. The calculation formula is In the formula, u and v are the numbers of adjacent edge nodes. Set the deviation threshold of different sensor data types as Sth k , if is greater than or equal to Sth k , then the sensor data of this edge node is abnormal relative to its adjacent edge node, and mark this edge node as a trembling node. If an edge node is a trembling node relative to its M adjacent nodes, then update and mark this edge node as a disabled node.

[0018] Preferably, the frequency of the first cross-validation is adjusted in real time in combination with the memory occupancy rate, historical failure rate, and data stability of the edge node:

[0019] An adaptive adjustment strategy is adopted to adjust the cross-validation frequency. The calculation formula of the cross-validation frequency is Crv = β1*Mel + β2*Hfr + β3*Dst. In the formula, β1, β2, and β3 are the weight coefficients of the memory occupancy rate Mel, historical failure rate Hfr, and data stability Dst respectively, and β1, β2, and β3 are all positive numbers. Among them, the memory occupancy rate Mel is obtained through the system resource monitoring of the edge node. The calculation formula of the historical failure rate Hfr is Tw is the historical time window. The sensor data is collected H times periodically within the historical time window Tw. The calculation formula of the data stability Dst is In the formula, X g is the sensor data of the gth one, is the mean value of H times of sensor data. The calculation formula is

[0020] Preferably, the logic for performing the first node anomaly assessment and classifying edge nodes is:

[0021] When it is determined that the heartbeat signal of the lightweight health self-check is abnormal and the edge node is marked as a disabled node, record the disabled node and transmit a disabled signal to the cloud server and other edge nodes, then record the disabled node as a suspicious node;

[0022] When it is determined that the heartbeat signal of the lightweight health self-check does not appear abnormal and the edge node is marked as a disabled node, record the disabled node as a non-suspicious node;

[0023] When it is determined that the heartbeat signal of the lightweight health self-check is abnormal and the edge node is not marked as a disabled node, record the edge node as a non-suspicious node;

[0024] When it is determined that the heartbeat signal of the lightweight health self-check is abnormal and the edge node is marked as a disabled node, record the edge node as a non-suspicious node.

[0025] Preferably, use the second cross-validation to perform a second node anomaly assessment on the suspicious node. The method for rechecking the suspicious node is as follows:

[0026] Use the method of random sampling based on dynamic weights to recheck the suspicious node, and label the suspicious node as S e , calculate the suspicious node S e The weights of all adjacent edge nodes {A1, A2, …, A d}, and the calculation expression is W s =γ1*Dst + γ2*Mel + γ3*Fvc. In the formula, W s is the weight of all adjacent edge nodes {A1, A2, …, A e} of the suspicious node S d}, Dst is the data stability of the adjacent node A s in the previous historical time window, Mel is the memory occupancy rate of the adjacent node A s , Fvc is the consistency of the first cross-validation of the adjacent node A s in the previous time. Among them, d is the total number of all adjacent edge nodes, s is the serial number of the adjacent edge node, and s = {1, 2, 3 …, d}, d is a positive number. Adopt the weighted random sampling method, and randomly select ran adjacent edge nodes for cross-validation according to the distribution of the weight W s of the suspicious node S e . The value of ran is a positive integer and ran is less than or equal to d. Mark the selected ran adjacent edge nodes as control nodes. Each control node independently measures the data of the suspicious node S e and compares it with its own data. Set the deviation threshold as Devth. Record the number of control nodes whose deviation from the suspicious node S e exceeds the deviation threshold Devth as Bec. Record the number of control nodes whose deviation from the suspicious node S e does not exceed the deviation threshold Devth as Nec. Then calculate the recheck index as Set the recheck threshold as Reth. If the recheck index Rei is greater than or equal to the recheck threshold Reth, it is determined that the suspicious node breaks away from the distributed framework. If the recheck index Rei is less than the recheck threshold Reth, it is determined that the suspicious node does not break away from the distributed framework;

[0027] If a suspicious node detaches from the distributed framework, any adjacent edge node will perform an intervention operation on the suspicious node that has detached from the distributed framework; if the suspicious node does not detach from the distributed framework, no intervention operation will be performed.

[0028] Preferably, adjacent node A s The method for obtaining the first cross-validation consistency last time is as follows:

[0029] Collect the sensor data of adjacent node A s in the first cross-validation and calibrate it as data representing the k-th sensor data type, and there are p sensor data types in total;

[0030] Collect the average data of the node set A nor recorded as non-suspicious nodes in the first node anomaly assessment and calibrate it as where in the formula, A f is an edge node in the non-suspicious node set A nor and represents the mean value of the data of all edge nodes in the non-suspicious node set A nor on the sensor data type k;

[0031] Calculate the deviation between adjacent node A s and the non-suspicious node set A nor , and the calculation expression is where represents the deviation of adjacent node A s from the mean value of the non-suspicious node set A nor on the sensor data type k, and calculate the first cross-validation consistency as in the formula, σ k is the historical data standard deviation on the sensor data type k, and the calculation expression is in the formula, Q is the size of the time window, is the data value of the sensor data type k at time point t, is the mean value of the sensor data type k within the time window Q, and the calculation expression is

[0032] Preferably, the logic for determining the fault path based on the network communication status and computing resource assessment is as follows:

[0033] The abnormal node conducts a Ping test to the adjacent edge node, attempts to establish a TCP connection with the adjacent edge node, and the adjacent edge node that conducts the Ping test and TCP connection checks the log to confirm whether there is received data. If the Ping test fails and the TCP connection fails, the abnormal fault is a network interruption;

[0034] If the Ping test is successful and the TCP connection fails, the abnormal fault is a port blockage or firewall restriction;

[0035] If the Ping test is successful and the TCP connection is successful, but the adjacent edge node does not receive the high-priority heartbeat signal, the abnormal fault is packet loss or congestion of the upstream device;

[0036] Collect the CPU context switch rate Csr, swap partition utilization rate SwapU, and task queue waiting time Qwt of the adjacent edge node that receives the high-priority heartbeat signal through system log collection, and use the logistic regression method to obtain the computing resource pressure index. The calculation expression is In the formula, δ1 and δ2 are respectively and The proportionality coefficients of, and both δ1 and δ2 are positive numbers. Ctxth is the context switch rate threshold, Queth is the task queue waiting time threshold. Set the computing resource pressure threshold as Crpth, compare the calculated computing resource pressure index Crp with the computing resource pressure threshold Crpth. If the computing resource pressure index Crp is greater than or equal to the computing resource pressure threshold Crpth, the computing resources of the edge node are insufficient, and it is determined that the abnormal fault is insufficient computing resources of the edge node;

[0037] If the computing resource pressure index Crp is less than the computing resource pressure threshold Crpth, the computing resources of the edge node are sufficient, and it is determined that the abnormal fault does not involve computing resource scheduling.

[0038] A greenhouse climate monitoring system includes an edge node management module, a node anomaly assessment module, a suspicious node review module, and a fault tracing module;

[0039] The edge node management module constructs a lightweight health self-check for the distributed edge nodes of greenhouse climate monitoring, is used to perform the first heartbeat detection, conduct the first cross-verification on adjacent edge nodes, and adjust the frequency of the first cross-verification in real time in combination with the memory occupancy rate, historical failure rate, and data stability of the edge node;

[0040] The node anomaly assessment module constructs a lightweight health self-check for the distributed edge nodes of greenhouse climate monitoring, is used to perform the first heartbeat detection, and classify the edge nodes, classifying the edge node types into suspicious nodes and non-suspicious nodes;

[0041] The suspicious node review module is used to perform a second node anomaly assessment on suspicious nodes using second cross-validation, review the suspicious nodes, and determine whether the suspicious nodes are detached from the distributed framework. If a suspicious node is detached from the distributed framework, the suspicious node that has detached from the distributed framework is marked as an abnormal node, and any adjacent edge node performs an intervention operation on the abnormal node;

[0042] The fault tracing module is used to, when the intervention fails, send a high-priority heartbeat signal from the abnormal node to the adjacent edge node. When the high-priority heartbeat signal does not reach the adjacent edge node, the network communication status is detected. When the high-priority heartbeat signal reaches the adjacent edge node but there is no response, the computing resources of the adjacent edge node that received the high-priority heartbeat signal are evaluated;

[0043] The fault path is determined based on the network communication status and the computing resource evaluation, and the staff is notified.

[0044] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0045] In the case of limited resources of edge computing devices, lightweight health self-checking with an adaptive period is adopted. When the monitoring is normal, the heartbeat frequency is reduced, the consumption of CPU, network bandwidth, and storage resources is reduced, the computing pressure is reduced, enabling edge nodes to focus on greenhouse environment monitoring, and improving real-time performance and stability. The heartbeat period is dynamically adjusted using the average network delay in the past N cycles to avoid misjudgment caused by network jitter.

[0046] An adaptive adjustment strategy is adopted to adjust the cross-validation frequency in real time. Multiple adjacent nodes are compared with each other, and it is easier to detect abnormal nodes through data consistency verification. If a single node is abnormal but the data of adjacent nodes is consistent, the edge node needs further analysis instead of immediately being determined as an abnormal fault. The adaptive adjustment adjusts the cross-validation frequency in real time and makes reasonable decisions in different situations. When the computing load is high, the cross-validation frequency is reduced to give priority to ensuring the stable operation of the system. When the fault risk increases, the cross-validation frequency is automatically increased to detect abnormalities in a timely manner, further improving the flexibility of the monitoring operation, making full use of computing resources, and the adaptive first cross-validation collaboratively monitors through adjacent edge nodes. Even if an edge node has an abnormality, other edge nodes can still detect its data abnormality and quickly perceive it, effectively avoiding the impact of single-point failures on the monitoring status, ensuring the stability of greenhouse environment parameters, and providing a fault tracing method for troubleshooting network communication status and computing resource scheduling. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0048] Figure 1 It is the method flow chart of the present invention.

[0049] Figure 2 It is the system module diagram of the present invention. Specific embodiments

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0051] Embodiment 1: Please refer to Figure 1 As shown, the present invention is a method for monitoring the climate of a greenhouse. The specific steps include:

[0052] Build a lightweight health self-check for the distributed edge nodes of the greenhouse climate monitoring to perform the first heartbeat detection, conduct the first cross-verification on adjacent edge nodes, and adjust the frequency of the first cross-verification in real time in combination with the memory occupancy rate, historical failure rate, and data stability of the edge nodes;

[0053] Conduct the first node anomaly assessment on the edge nodes according to the results of the first heartbeat detection and the first cross-verification, and classify the edge nodes. The types of edge nodes include suspicious nodes and non-suspicious nodes;

[0054] Use the second cross-verification to conduct the second node anomaly assessment on the suspicious nodes, recheck the suspicious nodes, and determine whether the suspicious nodes are detached from the distributed framework. If the suspicious nodes are detached from the distributed framework, mark the suspicious nodes detached from the distributed framework as abnormal nodes, and any adjacent edge node will perform an intervention operation on the abnormal nodes;

[0055] If the intervention fails, the abnormal node sends a high-priority heartbeat signal to the adjacent edge node. When the high-priority heartbeat signal does not reach the adjacent edge node, the network communication status is detected. When the high-priority heartbeat signal reaches the adjacent edge node but there is no response, the computing resources of the adjacent edge node that receives the high-priority heartbeat signal are evaluated;

[0056] Determine the fault path according to the network communication status and the computing resource evaluation, and notify the staff.

[0057] The construction method for building a lightweight health self-check for the first heartbeat detection is as follows: Each edge node periodically sends a heartbeat signal to the cloud server to report the operating status of the edge node. The heartbeat signal includes the timestamp when the heartbeat signal is sent, the computing resource utilization rate of the edge node, the backlog status of the task queue, and the status of the monitoring sensor data. When it is detected that there is a delay in sending the heartbeat signal, insufficient computing resources of the edge node, backlog of the task queue, or invalid sensor data, it is determined that the heartbeat signal for the first heartbeat detection is abnormal; otherwise, it is determined that the heartbeat signal for the first heartbeat detection is not abnormal.

[0058] The sending period of the heartbeat signal for the first heartbeat detection is adaptively adjusted according to the computing resource status of the edge node and historical heartbeat anomaly data. The method of adaptive adjustment is as follows:

[0059] Define the initial cycle time R0 for the first heartbeat detection. Then the cycle time for the first heartbeat detection is Tlow = R0 * (1 + α1 * Rag + α2 * Ahr + α3 * And), where α1, α2, and α3 are the weight coefficients of the average CPU occupancy rate Rag, the abnormal heartbeat ratio Ahr in the past N cycles, and the average network delay And in the past N cycles, respectively, and α1, α2, and α3 are all positive numbers. Among them, the average CPU occupancy rate Rag is obtained through the greenhouse climate monitoring API, and the calculation expression for the abnormal heartbeat ratio Ahr in the past N cycles is The calculation expression for the average network delay And in the past N cycles is where ad i is the network delay value in the i-th cycle.

[0060] For the heartbeat detection with a fixed cycle, even when the system is stable, it will still occupy a certain amount of computing resources and network bandwidth. In the case of limited resources of edge computing devices, unnecessary high-frequency heartbeats may cause the system burden to increase and affect the actual business processing. By adopting the lightweight health self-check with an adaptive cycle, when the monitoring is normal, the heartbeat frequency is reduced, the consumption of CPU, network bandwidth, and storage resources is reduced, the computing pressure is reduced, enabling the edge node to focus on greenhouse environment monitoring, and improving real-time performance and stability. The fixed heartbeat cycle does not consider network fluctuations and is prone to misjudgment. A short network delay may cause the heartbeat signal to time out, making the system think that the device has failed and triggering unnecessary alarms. However, in fact, the edge node may still be operating normally, just with momentary network congestion. By dynamically adjusting the heartbeat cycle using the average network delay in the past N cycles, misjudgment caused by network jitter can be avoided.

[0061] The method for cross-verifying adjacent edge nodes is as follows:

[0062] Each edge node periodically sends its own sensor data to adjacent edge nodes, and calibrates the sensor data as Sep k , where k is the number of sensor data types, and k = {1, 2, 3…, p}, p is the total number of sensor data types, and p is a positive integer. Calculate the data deviation of adjacent edge nodes at the same time point, and the calculation expression is In the formula, u and v are the numbers of adjacent edge nodes. Set the deviation threshold of different sensor data types as Sth k , if is greater than or equal to Sth k , then the sensor data of this edge node is abnormal relative to its adjacent edge node, and mark this edge node as a trembling node. If an edge node is a trembling node relative to its M adjacent nodes, then update and mark this edge node as a disabled node;

[0063] It should be noted that M is a positive integer, and the specific value of M is set by professionals in the field according to the scale and structure of the distributed framework.

[0064] The frequency at which each edge node periodically sends its own sensor data to adjacent edge nodes is:[[]]

[0065] Adopt an adaptive adjustment strategy to adjust the cross-validation frequency. The calculation expression of the cross-validation frequency is Crv = β1*Mel + β2*Hfr + β3*Dst. In the formula, β1, β2, and β3 are the weight coefficients of the memory occupancy rate Mel, the historical failure rate Hfr, and the data stability Dst respectively, and β1, β2, and β3 are all positive numbers. Among them, the memory occupancy rate Mel is obtained through the system resource monitoring of the edge node. The calculation expression of the historical failure rate Hfr is Tw is the historical time window. The sensor data is collected H times periodically within the historical time window Tw. The calculation expression of the data stability Dst is In the formula, X g is the sensor data of the gth one, is the average value of H times of sensor data, and the calculation expression is

[0066] Compared with the traditional method, adjacent edge nodes perform cross-validation and adopt an adaptive adjustment strategy to adjust the cross-validation frequency in real time. Compare multiple adjacent nodes with each other. It is easier to detect abnormal nodes through data consistency verification. Relying only on the self-check of a single node may increase the risk of misjudgment due to single-point failure or sensor drift, affecting the accuracy of greenhouse climate monitoring. In data consistency verification, if a single node is abnormal but the data of adjacent nodes is consistent, then this edge node needs further analysis instead of immediately being determined as an abnormal fault;

[0067] Traditional methods use a fixed frequency for cross - validation, which increases the burden when computing resources are scarce, resulting in an inability to respond quickly at critical moments. The adaptive approach adjusts the cross - validation frequency in real - time, making reasonable decisions in different situations. When the computing load is high, it reduces the cross - validation frequency to prioritize the stable operation of the system. When the risk of failure increases, it automatically raises the cross - validation frequency to detect anomalies in a timely manner, further enhancing the flexibility of the monitoring operation and making full use of computing resources.

[0068] When a failure occurs in a certain edge node using traditional methods, other edge nodes may have difficulty perceiving it in a timely manner, resulting in the persistence of the failure. However, the adaptive first cross - validation monitors through the cooperation of adjacent edge nodes. Even if an anomaly occurs in a certain edge node, other edge nodes can still detect data anomalies and quickly perceive them, effectively avoiding the impact of single - point failures on the monitoring status and ensuring the stability of the environmental parameters in the greenhouse.

[0069] The logic for performing the first node anomaly assessment and classifying edge nodes is as follows:

[0070] When it is determined that the heartbeat signal of the lightweight health self - check is abnormal and the edge node is marked as a disabled node, record the disabled node and transmit a disabled signal to the cloud server and other edge nodes, and record the disabled node as a suspicious node.

[0071] When it is determined that the heartbeat signal of the lightweight health self - check is not abnormal and the edge node is marked as a disabled node, record the disabled node as a non - suspicious node.

[0072] When it is determined that the heartbeat signal of the lightweight health self - check is abnormal and the edge node is not marked as a disabled node, record the edge node as a non - suspicious node.

[0073] When it is determined that the heartbeat signal of the lightweight health self - check is abnormal and the edge node is marked as a disabled node, record the edge node as a non - suspicious node.

[0074] Use the second cross - validation to perform a second node anomaly assessment on suspicious nodes. The method for re - checking suspicious nodes is as follows:

[0075] Adopt a method based on dynamic - weight random sampling to re - check suspicious nodes, mark the suspicious node as S e , calculate the suspicious node S e The weights of all adjacent edge nodes {A1, A2, …, A d 0, and the calculation expression is W s = γ1*Dst + γ2*Mel + γ3*Fvc, where W s is the suspicious node S e All adjacent edge nodes {A1, A2, …, Ad The weight of}, Dst is the adjacent node A s The data stability in the previous historical time window, Mel is the adjacent node A s The memory occupancy rate of, Fvc is the adjacent node A s The consistency of the first cross-validation last time. Among them, d is the total number of all adjacent edge nodes, s is the serial number of the adjacent edge nodes, and s = {1, 2, 3…, d}, d is a positive number. Using the weighted random sampling method, according to the weight W s The distribution of the suspicious node S e Randomly select ran adjacent edge nodes for cross-validation. The value of ran is a positive integer and ran is less than or equal to d. Mark the selected ran adjacent edge nodes as control nodes. Each control node independently measures the data of the suspicious node S e And compare it with its own data. Set the deviation threshold as Devth. Denote the suspicious node S e The number of control nodes whose deviation from the suspicious node exceeds the deviation threshold Devth is Bec. Denote the suspicious node S e The number of control nodes whose deviation from the suspicious node does not exceed the deviation threshold Devth is Nec. Then calculate the review index as Set the review threshold as Reth. If the review index Rei is greater than or equal to the review threshold Reth, it is determined that the suspicious node breaks away from the distributed framework. If the review index Rei is less than the review threshold Reth, it is determined that the suspicious node does not break away from the distributed framework;

[0076] If the suspicious node breaks away from the distributed framework, any one of the adjacent edge nodes performs an intervention operation on the suspicious node that breaks away from the distributed framework. If the suspicious node does not break away from the distributed framework, no intervention operation is performed.

[0077] The acquisition method of Fvc is: Collect the sensor data of the adjacent node A s In the first cross-validation and calibrate it as It represents the data of the kth sensor data type, and there are p sensor data types in total;

[0078] Collect the node set A nor Recorded as non-suspicious nodes in the abnormal assessment of the first node and calibrate it as Among them In the formula, A f Is an edge node in the node set A of non-suspicious nodes nor In Represents the node set A of non-suspicious nodes norThe mean value of the data of all edge nodes in the sensor data type k;

[0079] Calculate the adjacent node A s The node set A of non-suspicious nodes nor The deviation of is calculated by the formula where represents the adjacent node A s The deviation of the mean value of the node set A of non-suspicious nodes on the sensor data type k nor Calculate the first cross-validation consistency as In the formula, σ k Is the standard deviation of historical data on the sensor data type k, and the calculation formula is In the formula, Q is the size of the time window, Is the data value of the sensor data type k at the time point t, Is the mean value of the sensor data type k within the time window Q, and the calculation formula is

[0080] Traditional random sampling methods usually adopt a fixed sampling probability, which may cause some abnormal nodes to be misjudged as normal, or normal nodes to be misjudged as abnormal. The sampling strategy of the first cross-validation is relatively single and cannot dynamically adapt to different fault scenarios. Assign a higher sampling weight to the edge nodes with abnormal suspicion, increase the probability of their cross-validation, improve the accuracy of the review, and adaptively adjust the sampling strategy by combining factors such as historical abnormal records and consistency scores to reduce the interference to normal nodes;

[0081] Traditional methods use global consistency detection or full-coverage cross-validation, which leads to an increased communication burden on edge nodes, occupies computing resources, and frequent verification increases the interaction between edge nodes and the cloud, affecting the real-time performance of data transmission. The second cross-validation only focuses on reviewing the suspicious nodes screened by the first cross-validation, rather than global traversal, reducing unnecessary calculations. At the same time, it dynamically adjusts the cross-validation frequency to reduce unnecessary communication at high frequencies, making the monitoring operation more efficient, adapting to different fault modes, and improving the adaptability and fault tolerance to complex environments.

[0082] When the adjacent edge node fails to intervene in the abnormal node, trace the reason for the intervention failure, and the abnormal node sends a high-priority heartbeat signal to the adjacent edge node;

[0083] When the high-priority heartbeat signal does not reach the adjacent edge node, detect the network communication status. The method for detecting the network communication status is:

[0084] The abnormal node performs a Ping test on the adjacent edge node, attempts to establish a TCP connection with the adjacent edge node, and the adjacent edge node that performs the Ping test and the TCP connection checks the log to confirm whether there is received data. If the Ping test fails and the TCP connection fails, the abnormal fault is a network interruption;

[0085] If the Ping test is successful and the TCP connection fails, the abnormal fault is a port blockage or firewall restriction;

[0086] If the Ping test is successful and the TCP connection is successful, but the adjacent edge node does not receive the high-priority heartbeat signal, the abnormal fault is packet loss or congestion in the upstream device;

[0087] When the high-priority heartbeat signal arrives at the adjacent edge node but there is no response, the computing resources of the adjacent edge node that receives the high-priority heartbeat signal are evaluated;

[0088] Determine the fault path and notify the staff.

[0089] The method for evaluating computing resources is as follows:

[0090] Collect the CPU context switching rate Csr, swap partition usage SwapU, and task queue waiting time Qwt of the adjacent edge node that receives the high-priority heartbeat signal through the system log, and use the logistic regression method to obtain the computing resource pressure index. The calculation expression is In the formula, δ1 and δ2 are respectively and The proportionality coefficients of, and both δ1 and δ2 are positive numbers. Ctxth is the context switching rate threshold, Queth is the task queue waiting time threshold. Set the computing resource pressure threshold as Crpth, compare the calculated computing resource pressure index Crp with the computing resource pressure threshold Crpth. If the computing resource pressure index Crp is greater than or equal to the computing resource pressure threshold Crpth, the computing resources of the edge node are insufficient, and the abnormal fault is determined to be insufficient computing resources of the edge node;

[0091] If the computing resource pressure index Crp is less than the computing resource pressure threshold Crpth, the computing resources of the edge node are sufficient, and it is determined that the abnormal fault does not involve computing resource scheduling;

[0092] After obtaining the fault path, notify the staff of the cause of the abnormal fault.

[0093] It should be noted that the context switching rate threshold Ctxth and the task queue waiting time threshold Queth are set by professionals in the field according to the system task load.

[0094] Example 2: Please refer toFigure 2 As shown in the figure, the present invention is a greenhouse climate monitoring system, including an edge node management module, a node anomaly assessment module, a suspicious node review module, and a fault tracing module;

[0095] The edge node management module constructs a lightweight health self-check for the distributed edge nodes of greenhouse climate monitoring, used to perform the first heartbeat detection, conduct the first cross-verification on adjacent edge nodes, and adjust the frequency of the first cross-verification in real time in combination with the memory occupancy rate, historical failure rate, and data stability of the edge nodes;

[0096] The node anomaly assessment module is used to perform the first node anomaly assessment on the edge nodes according to the results of the first heartbeat detection and the first cross-verification, classify the edge nodes, and divide the edge node types into suspicious nodes and non-suspicious nodes;

[0097] The suspicious node review module is used to perform the second node anomaly assessment on the suspicious nodes by using the second cross-verification, review the suspicious nodes, judge whether the suspicious nodes are detached from the distributed framework. If the suspicious nodes are detached from the distributed framework, mark the suspicious nodes detached from the distributed framework as abnormal nodes, and any adjacent edge node performs an intervention operation on the abnormal nodes;

[0098] The fault tracing module is used to, when the intervention fails, send a high-priority heartbeat signal from the abnormal node to the adjacent edge nodes. When the high-priority heartbeat signal does not reach the adjacent edge nodes, detect the network communication status. When the high-priority heartbeat signal reaches the adjacent edge nodes but there is no response, perform a computing resource assessment on the adjacent edge nodes that receive the high-priority heartbeat signal;

[0099] Determine the fault path according to the network communication status and the computing resource assessment, and notify the staff.

[0100] In the case of limited resources of edge computing devices, this application adopts an adaptive-cycle lightweight health self-check. When the monitoring is normal, it reduces the heartbeat frequency, reduces the consumption of CPU, network bandwidth, and storage resources, reduces the computing pressure, enables the edge nodes to focus on greenhouse environment monitoring, and improves real-time performance and stability. It dynamically adjusts the heartbeat cycle by using the average network delay in the past N cycles to avoid misjudgment caused by network jitter.

[0101] Adopt an adaptive adjustment strategy to adjust the cross-validation frequency in real time. Compare multiple adjacent nodes with each other. It is easier to detect abnormal nodes through data consistency verification. If a single node is abnormal but the data of adjacent nodes is consistent, then this edge node needs further analysis instead of being immediately determined as an abnormal fault. The adaptive adjustment adjusts the cross-validation frequency in real time and makes reasonable decisions under different circumstances. When the computing load is high, reduce the cross-validation frequency to give priority to ensuring the stable operation of the system. When the fault risk increases, automatically increase the cross-validation frequency to detect abnormalities in a timely manner, further improving the flexibility of monitoring operations, making full use of computing resources. The first cross-validation of adaptiveness is through collaborative monitoring of adjacent edge nodes. Even if an edge node has an abnormality, other edge nodes can still detect its data abnormality and quickly perceive it, effectively avoiding the impact of single-point failures on the monitoring status, ensuring the stability of the environmental parameters of the greenhouse, and providing a fault tracing method for troubleshooting network communication status and computing resource scheduling.

[0102] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0103] Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system can refer to the corresponding process in the foregoing method embodiment and will not be elaborated here.

[0104] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claimed rights.

Claims

1. A greenhouse climate monitoring method, characterized in that: The specific steps include: Build a lightweight health self-check for distributed edge nodes for greenhouse climate monitoring, which is used to perform the first heartbeat detection, perform the first cross-validation on adjacent edge nodes, and adjust the frequency of the first cross-validation in real time based on the edge node's memory usage, historical failure rate, and data stability; Performing a first node abnormality assessment on the edge nodes according to the results of the first heartbeat detection and the first cross-validation, and classifying the edge nodes, where the edge node types include suspicious nodes and non-suspicious nodes; Use the second cross-validation to perform the second node abnormality assessment on the suspicious node, review the suspicious node, and determine whether the suspicious node is out of the distributed framework. If the suspicious node is out of the distributed framework, mark the suspicious node out of the distributed framework as an abnormal node, and any adjacent edge node will intervene in the abnormal node. If the intervention fails, the abnormal node sends a high-priority heartbeat signal to the adjacent edge node. When the high-priority heartbeat signal does not reach the adjacent edge node, the network communication status is detected. When the high-priority heartbeat signal reaches the adjacent edge node but there is no response, the computing resource evaluation is performed on the adjacent edge node that received the high-priority heartbeat signal. Determine the fault path based on network communication status and computing resource evaluation, and notify staff.

2. A greenhouse climate monitoring method according to claim 1, characterized in that: The method for constructing a lightweight health self-check for the first heartbeat detection is: Each edge node periodically sends a heartbeat signal to the cloud server to report the running status of the edge node. The heartbeat signal includes the timestamp of the heartbeat signal, the utilization rate of the computing resources of the edge node, the backlog status of the task queue and the status of the monitoring sensor data. When a delay in the sending of the heartbeat signal, insufficient computing resources of the edge node, a backlog of the task queue or invalid sensor data is detected, it is determined that the heartbeat signal of the first heartbeat detection is abnormal; Otherwise, it is determined that the heartbeat signal of the first heartbeat detection is not abnormal.

3. A greenhouse climate monitoring method according to claim 2, characterized in that: The heartbeat signal sending period of the first heartbeat detection is adaptively adjusted according to the computing resource status of the edge node and the historical heartbeat abnormality data. The adaptive adjustment method is: Define the initial cycle time R0 of the first heartbeat detection, then the cycle time of the first heartbeat detection is Tlow = R0*(1+α1*Rag+α2*Ahr+α3*And), where α1, α2 and α3 are weight coefficients of the average CPU occupancy rate Rag, the abnormal heartbeat ratio Ahr in the past N cycles and the average network delay And in the past N cycles, and α1, α2 and α3 are all positive numbers. The average CPU occupancy rate Rag is obtained through the greenhouse climate monitoring API, and the calculation expression of the abnormal heartbeat ratio Ahr in the past N cycles is The calculation expression of the average network delay And in the past N cycles is: In the formula, ad i is the network delay value of the i-th cycle.

4. A greenhouse climate monitoring method according to claim 1, characterized in that: The method for performing the first cross validation on adjacent edge nodes is: Each edge node periodically sends its own sensor data to the adjacent edge nodes, and the sensor data is calibrated as Sep k , where k is the number of the sensor data type, and k = {1, 2, 3…, p}, p is the total number of sensor data types, and p is a positive integer. The data deviation of adjacent edge nodes at the same time point is calculated. The calculation expression is: In the formula, u and v are the numbers of adjacent edge nodes, and the deviation threshold of different sensor data types is set as Sth k ,like Greater than or equal to Sth k , then the sensor data of the edge node is abnormal relative to its adjacent edge nodes, and the edge node is marked as a trembling node. If the edge node is a trembling node relative to its M adjacent nodes, the edge node is updated and marked as a disabled node.

5. A greenhouse climate monitoring method according to claim 4, characterized in that: The frequency of the first cross-validation is adjusted in real time based on the edge node's memory usage, historical failure rate, and data stability: Adaptive adjustment strategy is adopted to adjust the cross-validation frequency. The calculation expression of cross-validation frequency is Crv=β1*Mel+β2*Hfr+β3*Dst, where β1, β2 and β3 are the weight coefficients of memory occupancy Mel, historical failure rate Hfr and data stability Dst, respectively, and β1, β2 and β3 are all positive numbers. Among them, memory occupancy Mel is obtained through system resource monitoring of edge nodes, and the calculation expression of historical failure rate Hfr is Tw is the historical time window. Sensor data is collected H times in the historical time window Tw. The calculation expression of data stability Dst is: Where, X g is the sensor data of the gth sensor, is the mean value of H sensor data, and the calculation expression is 6. A greenhouse climate monitoring method according to claim 4, characterized in that: The logic for evaluating the first node abnormality and classifying edge nodes is as follows: When it is determined that the heartbeat signal of the lightweight health self-check is abnormal and the edge node is marked as a disabled node, the disabled node is recorded and the disabled signal is transmitted to the cloud server and other edge nodes, and the disabled node is recorded as a suspicious node; When it is determined that the heartbeat signal of the lightweight health self-check is not abnormal and the edge node is marked as a disabled node, the disabled node is recorded as a non-suspicious node; When the heartbeat signal of the lightweight health self-check is determined to be abnormal and the edge node is not marked as a disabled node, the edge node is recorded as a non-suspicious node; When the heartbeat signal of the lightweight health self-check is determined to be abnormal and the edge node is marked as a disabled node, the edge node is recorded as a non-suspicious node.

7. A greenhouse climate monitoring method according to claim 6, characterized in that: The second cross validation is used to evaluate the abnormality of the second node on the suspicious node. The method for reviewing the suspicious node is as follows: The suspicious nodes are reviewed by using a method based on dynamic weight random sampling, and the suspicious nodes are marked as S e , calculate the suspicious node S e All adjacent edge nodes {A1,A2,…,A d The weight of} is calculated as W s =γ1*Dst+γ2*Mel+γ3*Fvc, where W s is a suspicious node S e All adjacent edge nodes {A1,A2,…,A d }, Dst is the weight of the adjacent node A s Data stability in the previous historical time window, Mel is the adjacent node A s The memory usage of Fvc is the memory usage of the adjacent node A. s The first cross-validation consistency of the last time, where d is the total number of all adjacent edge nodes, s is the sequence number of the adjacent edge node, and s = {1, 2, 3..., d}, d is a positive number, and a weighted random sampling method is used, according to the weight W s The distribution of suspicious nodes S e Randomly select ran adjacent edge nodes for cross-validation, where the value of ran is a positive integer and ran is less than or equal to d, and mark the selected ran adjacent edge nodes as control nodes. Each control node independently measures the suspicious node S e The data is compared with its own data, and the deviation threshold is set as Devth. The suspicious node S is recorded e The number of control nodes whose deviation from the control node exceeds the deviation threshold Devth is Bec, and the suspicious node S is recorded as e The number of control nodes whose deviation from the control node does not exceed the deviation threshold Devth is Nec, and the calculation recheck index is The recheck threshold is set to Reth. If the recheck index Rei is greater than or equal to the recheck threshold Reth, the suspicious node is determined to have left the distributed framework. If the recheck index Rei is less than the recheck threshold Reth, the suspicious node is determined to have not left the distributed framework. If the suspicious node leaves the distributed framework, any adjacent edge node will intervene in the suspicious node that has left the distributed framework; if the suspicious node does not leave the distributed framework, no intervention will be performed.

8. A greenhouse climate monitoring method according to claim 7, characterized in that: Neighboring node A s The method for obtaining the consistency of the first cross-validation last time is: Collect adjacent nodes A s The sensor data in the first cross validation is calibrated as Data representing the k-th sensor data type, there are a total of p sensor data types; Collect the node set A recorded as non-suspicious nodes in the first node anomaly assessment nor The average data is calibrated as in In the formula, A f A is a set of non-suspicious nodes nor An edge node in A represents the node set A of non-suspicious nodes nor The data mean of all edge nodes in sensor data type k; Calculate the adjacent nodes A s The node set A with non-suspicious nodes nor The calculation expression of the deviation is in, Represents the adjacent node A s The node set A with non-suspicious nodes on sensor data type k nor The deviation from the mean, the first cross validation consistency is calculated as In the formula, σ k is the standard deviation of historical data on sensor data type k, and the calculation expression is Where Q is the size of the time window, is the data value of sensor data type k at time point t, is the data mean of sensor data type k in time window Q, and the calculation expression is 9. A greenhouse climate monitoring method according to claim 1, characterized in that: The logic for determining the fault path based on network communication status and computing resource evaluation is as follows: The abnormal node performs a Ping test on the adjacent edge node and tries to establish a TCP connection with the adjacent edge node. The adjacent edge node that performs the Ping test and TCP connection checks the log to confirm whether there is any received data. If the Ping test fails and the TCP connection fails, the abnormal fault is a network interruption. If the Ping test succeeds and the TCP connection fails, the abnormal failure is port blocking or firewall restriction; If the Ping test is successful and the TCP connection is successful, but the adjacent edge node does not receive the high-priority heartbeat signal, the abnormal fault is packet loss or congestion in the upstream device; The CPU context switching rate Csr, swap partition usage SwapU, and task queue waiting time Qwt of adjacent edge nodes that receive high-priority heartbeat signals are collected through system logs, and the computing resource pressure index is obtained using the logistic regression method. The calculation expression is: In the formula, δ1 and δ2 are and The proportional coefficient is , and δ1 and δ2 are both positive numbers, Ctxth is the context switching rate threshold, Queth is the task queue waiting time threshold, and the computing resource pressure threshold is set to Crpth. The calculated computing resource pressure index Crp is compared with the computing resource pressure threshold Crpth. If the computing resource pressure index Crp is greater than or equal to the computing resource pressure threshold Crpth, the computing resources of the edge node are insufficient, and the abnormal fault is determined to be insufficient computing resources of the edge node; If the computing resource pressure index Crp is less than the computing resource pressure threshold Crpth, the computing resources of the edge node are sufficient, and it is determined that the abnormal fault does not involve computing resource scheduling.

10. A greenhouse climate monitoring system, used to implement a greenhouse climate monitoring method according to any one of claims 1 to 9, comprising an edge node management module, a node anomaly assessment module, a suspicious node review module and a fault tracing module; The edge node management module builds a lightweight health self-check for the distributed edge nodes of greenhouse climate monitoring, which is used to perform the first heartbeat detection, perform the first cross-validation on adjacent edge nodes, and adjust the frequency of the first cross-validation in real time based on the memory usage, historical failure rate and data stability of the edge nodes; The node anomaly assessment module is used to perform a first node anomaly assessment on the edge node according to the results of the first heartbeat detection and the first cross-validation, and classify the edge nodes into suspicious nodes and non-suspicious nodes; The suspicious node review module is used to perform a second node abnormality assessment on the suspicious node using the second cross-validation, review the suspicious node, and determine whether the suspicious node is out of the distributed framework. If the suspicious node is out of the distributed framework, the suspicious node out of the distributed framework is marked as an abnormal node, and any adjacent edge node intervenes in the abnormal node. The fault tracing module is used to send a high-priority heartbeat signal from the abnormal node to the adjacent edge node when the intervention fails. When the high-priority heartbeat signal does not reach the adjacent edge node, the network communication status is detected. When the high-priority heartbeat signal reaches the adjacent edge node but there is no response, the computing resource evaluation is performed on the adjacent edge node that received the high-priority heartbeat signal. Determine the fault path based on network communication status and computing resource evaluation, and notify staff.

Images