Agricultural process big data statistical analysis and alarm method

Abstract

The application discloses an agricultural process big data statistical analysis and alarm method, collects meteorological, soil, insect, irrigation state and equipment working condition data in agricultural production, carries out cleaning, time alignment and unified modeling on multi-source data, forms a monitoring sequence, carries out standardization and rank statistical processing on the monitoring sequence, identifies candidate abnormal points, further combines local significance test and interval distribution change test, confirms abnormal intervals, and then comprehensively scores mutation strength, significance strength and distribution change strength, generates early warning grades, early warning time and related index information, and outputs alarm results. The application realizes statistical analysis and graded early warning of abnormal changes in agricultural processes.

Description

Technical Field

[0001] This invention relates to the field of agricultural informatization and data processing technology, and in particular to a method for big data statistical analysis and alarm in agricultural processes. Background Technology

[0002] With the development of agricultural IoT and smart agriculture technologies, data on meteorological environment, soil conditions, insect infestation changes, irrigation control, and equipment operation during agricultural production are gradually being continuously collected and transmitted online. Agricultural process analysis and risk warning based on multi-source monitoring data have become important technical means for precision agricultural management. Currently, agricultural monitoring systems typically rely on threshold judgment or simple statistical methods to detect anomalies in single indicators or small amounts of data, and implement basic alarm functions through the platform.

[0003] Existing agricultural data analysis technologies still have significant shortcomings in multi-source data fusion and complex anomaly identification. On the one hand, traditional methods are mostly based on fixed thresholds or single statistical features, lacking the ability to characterize the overall distribution changes of time series data, making it difficult to effectively identify abrupt changes and implicit distribution transitions, and prone to false positives or false negatives. On the other hand, existing methods lack multi-stage verification mechanisms in the anomaly detection process, typically failing to combine mutation detection, significance testing, and distribution difference analysis for joint judgment, resulting in poor stability of anomaly identification results and difficulty in adapting to the multi-scale and multi-noise data characteristics of the agricultural environment. Furthermore, existing technologies often limit the expression of anomaly results to single indicators or simple alarms, lacking the ability to comprehensively quantify and grade the intensity of anomalies, making it difficult to support the needs of refined management and decision-making.

[0004] Therefore, how to provide a method for big data statistical analysis and alarm in agricultural processes is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention proposes a big data statistical analysis and alarm method for agricultural processes. It adopts multi-source data fusion, rank statistical modeling and multi-stage statistical testing methods to uniformly process and identify anomalies in monitoring data during agricultural production. It determines abnormal intervals through mutation point detection, significance verification and distribution change analysis, and completes anomaly classification and alarm output.

[0006] A method for big data statistical analysis and alarm of agricultural processes according to an embodiment of the present invention includes the following steps: S1. Collect data generated during agricultural production, perform cleaning, time alignment and unified data model mapping on the collected data, and generate a multi-source agricultural monitoring sequence set; S2. Receive a set of multi-source agricultural monitoring sequences, perform standardization processing according to preset indicator specifications, construct rank statistical sample sequences, and generate a pool of indicator sequences to be detected; S3. Receive the pool of index sequences to be detected, perform global mutation point localization processing based on Pettitt test on each index sequence, output candidate mutation points and corresponding mutation intensity results, and form a candidate mutation point dataset. S4. Receive the candidate mutation point dataset, construct the front and back local sample windows based on each candidate mutation point, perform the Mann-Whitney test to verify significance, output the significance strength results, and filter to obtain the set of effective candidate abnormal intervals. S5. Receive the set of valid candidate anomaly intervals, perform the improved Lepage test in each anomaly interval, generate the position offset result, scale change result and distribution transition intensity result, and form a distribution transition discrimination dataset. S6. Receive mutation intensity results, significance intensity results, and distribution transition discrimination dataset, construct a unified anomaly scoring vector and perform hierarchical threshold mapping, generate warning level, warning time, and related indicator information, and output alarm information.

[0007] Optionally, generating a multi-source agricultural monitoring sequence set includes: S11. Access the data streams from meteorological sensors, soil sensors, insect monitoring equipment, irrigation control terminals, and agricultural equipment controllers through the agricultural IoT gateway. Parse the original data messages according to the unified communication protocol, extract the monitoring indicator identifier, collection time, equipment number, and monitoring value, and construct the original multi-source data record set. S12. Perform data cleaning on the original multi-source data record set, remove data records with missing key fields according to the preset field integrity rules, and calculate the median and absolute deviation of the median of each indicator based on the historical sliding window. When the monitored value deviates from the median by more than a preset multiple threshold, outlier pruning or replacement is performed to obtain the cleaned multi-source data record set. S13. Perform time alignment processing on the cleaned multi-source data record set, set a unified sampling period, construct a standard time axis, and map the collection time of each data source to the standard time axis; when there is a time interval between adjacent valid sampling points, calculate the interpolated data of the target time point according to the linear proportional relationship of the changes in the values ​​of the two adjacent sampling points, and generate a time-aligned dataset. S14. Perform spatial identifier binding processing on the time-aligned dataset, determine the monitoring location based on the mapping relationship between the device number and the plot code, and encapsulate the indicator identifier, time index, spatial index and monitoring value into structured data units according to the unified agricultural monitoring data model to form a standardized multi-source agricultural monitoring data table. S15. Sort and serialize the standardized multi-source agricultural monitoring data table according to the indicator identifier and time index to generate a continuous time series with a single monitoring indicator as the unit, and summarize it to form a multi-source agricultural monitoring sequence set.

[0008] Optionally, generating the pool of sequences of indicators to be detected includes: S21. Receive a set of multi-source agricultural monitoring sequences, extract indicator identifiers, time indexes, spatial indexes and monitoring values ​​according to the preset agricultural monitoring indicator coding rules, and group the data according to the indicator identifiers to form a set of original indicator sequences divided by a single monitoring indicator. S22. Perform standardization processing on the original indicator sequence set. For each monitoring indicator, calculate its minimum and maximum values ​​within a preset time window, and map the monitoring values ​​at each time point to the interval between zero and one in a linear proportion to obtain the standardized indicator sequence set. S23. Perform time continuity processing on the standardized indicator sequence set, construct a standard time axis according to a unified sampling period, and sort each indicator sequence by time index; when there are missing time points, calculate the interpolated data of the missing time based on the ratio of the numerical changes of two adjacent valid sampling points, and generate a time-continuous indicator sequence set. S24. Construct a rank statistical sample sequence for the time-continuous indicator sequence set, sort the standardized monitoring values ​​in each indicator sequence in ascending order according to time sequence and assign a rank; when there are the same values, take the average of the corresponding ranks as the rank of the value, and generate the rank statistical sequence corresponding to each indicator. S25. The rank statistical sequence of each indicator is structured and encapsulated according to the indicator identifier, time index and spatial index to form an indicator data unit containing the rank value sequence. All indicator data units are summarized and indexed to generate a pool of indicator sequences to be detected.

[0009] Optionally, the candidate mutation point dataset includes: S31. Receive the pool of indicator sequences to be detected, extract the time-continuous rank statistical sequence corresponding to a single monitoring indicator according to the indicator identifier, and obtain a set of sample pairs sorted by time index. ,in, For the first Time index of each sampling moment The standardized monitoring value corresponding to the stated time. The total number of samples for the monitored indicator sequence; S32, to Perform full sequence sorting to generate an ordered column. ,in, For the first The rank of each sample; when there are identical values, the "same-value rank expansion" rule is applied: the rank interval occupied by any group of identical values ​​after sorting the entire sequence. Integer ranks are assigned sequentially to samples within the same numerical group according to their chronological order in the time index. This yields a discrete integer sequence; S33. Construct the Pettitt test statistic sequence, first calculating the rank prefix sum. ,in, Indicates the preceding The sum of the rank values ​​of each sample; then for each candidate segmentation position Calculate the Pettitt statistic: ; in, For position The Pettitt statistic at the given location characterizes the intensity of the abrupt change at that location; all according to Write the statistics curve in ascending order And record each symbols As an indicator of the direction of mutation; S34. Perform maximum absolute value location on the statistic curve and calculate... ,in, To determine the global maximum mutation strength; construct the maximum platform index set. Determine the left boundary of the platform With the right boundary of the platform The platform median index was used as the candidate mutation point index. The output candidate mutation point time index is The output mutation strength result is The direction of the output mutation is ; S35. Perform bidirectional consistency verification: Reverse the order column in chronological order. By repeatedly performing the Pettitt test statistic sequence construction on the reverse order column and the maximum absolute value localization step on the statistic curve, the reverse candidate mutation point index is obtained. and its largest platform set Mapping the inverted index back to the forward time axis yields... And convert the mapping platform set into a positive index range. ;when Falling into the range At that time, keep Unchanged; when Not falling into the range At that time, perform the following update: If the interval and interval If there is an intersection, then the median index of the intersection is taken as the updated value. ; If the two intervals have no intersection, then in the set Selecting from and The index with the smallest distance is used as the updated index. ; And simultaneously update the candidate mutation point time index as follows The mutation direction is identified as The result of maintaining mutation strength is ; S36. Calculate the Pettitt significance probability for the updated candidate mutation points and write it into the candidate mutation point record. The significance probability is approximated by the Pettitt test. ; in, The significance probability of candidate mutation points is calculated; and a single-index candidate mutation point record is generated, which includes an index identifier and a candidate mutation point time index. Platform Boundaries Mutation intensity results Mutation direction identifier and significance probability ; S37. Summarize all single-index candidate mutation point records to form a candidate mutation point dataset.

[0010] Optionally, the output significance intensity results can be used to filter and obtain a set of valid candidate outlier intervals, including: S41. Receive the candidate mutation point dataset, and for each candidate mutation point record, read the indicator identifier and candidate mutation point time index. Mutation direction identifier and platform boundary index and And extract the time-continuous standardized sample sequences of the corresponding indicators from the pool of indicator sequences to be detected. ,in, For time index Standardized monitoring values ​​at the location The total number of samples in the index sequence; S42. Construct a local sample window for platform exclusion based on the platform boundary for candidate mutation points, and define the platform exclusion interval as the index interval. And construct a front window and a rear window outside the aforementioned interval: Front window from index Select samples sequentially to the left, then the back window starts from the index. Select samples sequentially to the right; Set two sets of fixed window capacities and ,in For short window capacity, For long window capacity and ; Generate short-window and front-window sample sets in chronological order. With short window and back window sample set , Then, generate a long-window front window sample set using the same rules. With long window and back window sample set ,make ; When the platform excludes a range of samples that are too close to the sequence boundary, resulting in insufficient available samples, it continues to expand the index sampling outward on the same side according to the "far-end filling" rule until the corresponding window capacity is met, and records the actual index range used. S43, respectively, for short-window sample pairs Compared with long window samples Rank statistic construction for performing the Mann-Whitney test: and Merge into a sample set ,right Sort the samples in ascending order of their values ​​and assign them ranks; when there are duplicate values, take the average of the ranks corresponding to the duplicate values ​​as the rank of the values; calculate the sum of the ranks of the front window sample sets respectively. Rank sum of the back window sample set ,in, For set The sum of the rank values ​​of all samples within the sample. For set The sum of the rank values ​​of all samples within the sample; S44. Calculate the Mann-Whitney statistic based on the rank sum of the short window and the long window, respectively. And generate the corresponding significance intensity results: ; in, The Mann-Whitney statistic. For the front window sample size, For the sample size of the back window, The sum of the rank of the samples in the front window; the sum of the rank of the samples in the short window is calculated. Calculations for long windows yielded ; S45. Perform significance determination for both short and long windows, and implement dual-scale consistency constraints: and Convert them respectively to the corresponding two-sided test significance probabilities and And obtain the significance intensity respectively. and The significance level is taken as the absolute measure of the deviation of the corresponding statistic from the neutral position; if and only if both conditions are met. and At that time, the candidate mutation points were confirmed to pass the Mann-Whitney significance test, among which Preset significance level; S46. Under the condition that the candidate mutation point passes the significance verification, perform directional consistency confirmation: calculate the Mann-Whitney directional identifier based on the rank sum difference between the front window and the back window. ,when Time setting ,when Time setting and will Mutation direction identifier Perform consistency matching; when The candidate mutation point is retained if it is valid, otherwise it is discarded. S47. Generate valid candidate anomaly intervals for the retained candidate mutation points: Using the earliest time index of the front window and the latest time index of the back window actually used in the long window construction as the boundaries of the anomaly intervals, generate anomaly intervals. And the significance intensity result is defined as Write the abnormal range record; S48. Summarize all retained abnormal interval records to form a set of valid candidate abnormal intervals.

[0011] Optionally, the distributed transition discrimination dataset includes: S51. Receive the set of valid candidate abnormal intervals, and for each abnormal interval record, read the indicator identifier and the abnormal interval boundary. Time index of candidate mutation points And extract time-continuous standardized sample sequences of the same indicator from the pool of indicator sequences to be detected. ,in For time indexing, For time index Standardized monitoring values ​​at the location The total number of samples in the index sequence; S52. Construct a two-sample input set: indexed by time interval Extract the sample set within the interval and with A reference set is formed by sampling from the nearest to the farthest point in time on both sides outside the interval, with the center as the reference set. ,in Reference subset from the left With reference subset on the right Composition, satisfying and When the reference sample on either side is insufficient, the sampling order on the other side is fixed, and the sample is continuously supplemented on the available side according to the time interval until the sample is obtained. and record and Each covers a specific time index range; S53, For sets Perform a single rank assignment to construct the positional verification input: Sort the sample values ​​in ascending order and assign them ranks to obtain a merged order column. When encountering the same sample value, take the average rank of the rank interval occupied by the same sample value as the rank of the sample value; calculate the set. In-sample rank sum ,in For set Sum the rank values ​​of all samples and calculate the standardized statistic of positional offset. ; in, For the standardized statistics of position offset, The sample size within the interval is 1. For reference sample size; S54. Perform a second rank assignment on the same merged order column to construct a metric test. Input: using the center rank... Based on this, calculate the absolute distance from the rank of each sample to the center rank, and then sort all absolute distance values ​​in ascending order before assigning them a new rank. When encountering the same absolute distance value, use the average distance rank. Calculate the set... The sum of the rank of the corresponding distance rank ,in For set The sum of the rank of the distances of all samples within the range, and with The standardized statistic for scale change is obtained by subtracting the expected rank sum of distances under the null hypothesis and dividing by the standard deviation of the rank sum of distances under the null hypothesis. ; S55. Generate joint statistics and produce results on the intensity of distribution transitions: ,in For the improved Lepage joint statistic, Standardized statistics for scale change; The symbol determines the position offset direction identifier, in order to The symbol determines the direction of scale change and... Record the abnormal interval as the result of the distribution transition intensity; S56. For each anomalous interval, output the location offset result, scale change result, and distribution transition intensity result, and include the index identifier and anomalous interval boundary. Candidate mutation point time index , , Location offset direction indicator, scale change direction indicator, and joint statistics Encapsulated as a distributed transition discrimination record; S57. Summarize the distribution transition discrimination records to form a distribution transition discrimination dataset.

[0012] Optionally, the generated alarm information includes: S61, Receive mutation intensity results Significance results and the joint statistics in the distribution transition discriminant dataset Based on the indicator identifier, three types of intensity indicators for the same candidate anomaly interval are correlated and matched to construct a three-dimensional anomaly feature vector. ,in The global mutation strength is obtained from the Pettitt test. The significance level is obtained from the Mann-Whitney test. The intensity of the distributed transitions obtained from the improved Lepage test; S62. Perform uniform scale mapping on the three-dimensional anomaly feature vectors, and obtain them within the preset historical sample intervals. , , The maximum and minimum values ​​are obtained, and each component is mapped to a standard interval of zero to one in a linear proportion to obtain the normalized intensity components. , , ; S63. Construct a unified anomaly scoring function, according to preset weight coefficients. , , The standardized intensity components are weighted and summed to generate a uniform anomaly score. ; S64. Standardized anomaly scoring values Perform hierarchical threshold mapping processing and set incremental threshold intervals. ,when It was determined to be a Level 1 warning at that time. It was determined to be a Level II warning at that time. It was determined to be a Level III warning at that time. The system is then classified as a Level 4 warning and a corresponding warning level identifier is generated. S65. Determine the warning time and related indicator information, wherein the warning time is taken from the time index of the candidate mutation point. And record the boundaries of the abnormal intervals. The duration of the anomaly; related indicator information includes indicator identifiers, monitoring location identifiers, and corresponding... , , With uniform anomaly score ; S66. Encapsulate the warning level, warning time, abnormal duration range, related indicator information, and statistical intensity results into a structured alarm information record, and write it into the alarm database through the alarm release interface, while sending the alarm information to the preset alarm terminals.

[0013] The beneficial effects of this invention are: This invention introduces a global mutation point localization method based on the Pettt test to identify structural changes in time series, achieving accurate characterization of mutation location and intensity, improving the global perception capability of anomaly detection, and overcoming the problem that existing technologies are unable to effectively identify sudden changes.

[0014] This invention introduces the Mann-Whitney test for significance verification based on mutation points, and combines a dual-scale window and directional consistency constraint mechanism to achieve secondary screening of candidate anomalies, thereby improving the reliability and stability of anomaly judgment and solving the problem of unstable results caused by the lack of verification links in anomaly identification in the prior art.

[0015] This invention performs an improved Lepage test within anomaly intervals to jointly analyze the positional shift and scale changes of data, thereby achieving a comprehensive characterization of distribution transition characteristics, enhancing the ability to identify complex anomaly patterns, and compensating for the shortcomings of existing technologies in simultaneously reflecting mean changes and fluctuation changes, thus further improving the overall anomaly detection accuracy. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a big data statistical analysis and alarm method for agricultural processes proposed in this invention; Figure 2 This is a schematic diagram of the mutation point localization process based on the Pettitt test proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A method for big data statistical analysis and alarm in agricultural processes, comprising the following steps: S1. Collect meteorological monitoring data, soil monitoring data, insect monitoring data, irrigation status data and equipment operating condition data generated during agricultural production. Perform cleaning, time alignment and unified data model mapping on the collected data to generate a multi-source agricultural monitoring sequence set. S2. Receive a set of multi-source agricultural monitoring sequences, perform standardization processing according to preset indicator specifications, construct rank statistical sample sequences, and generate a pool of indicator sequences to be detected; S3. Receive the pool of index sequences to be detected, perform global mutation point localization processing based on Pettitt test on each index sequence, output candidate mutation points and corresponding mutation intensity results, and form a candidate mutation point dataset. S4. Receive the candidate mutation point dataset, construct the front and back local sample windows based on each candidate mutation point, perform the Mann-Whitney test to verify significance, output the significance strength results, and filter to obtain the set of effective candidate abnormal intervals. S5. Receive the set of valid candidate anomaly intervals, perform the improved Lepage test in each anomaly interval, generate the position offset result, scale change result and distribution transition intensity result, and form a distribution transition discrimination dataset. S6. Receive mutation intensity results, significance intensity results, and distribution transition discrimination dataset, construct a unified anomaly scoring vector and perform hierarchical threshold mapping, generate warning level, warning time, and related indicator information, and output alarm information.

[0019] In this embodiment, generating a multi-source agricultural monitoring sequence set includes: S11. Access the data streams from meteorological sensors, soil sensors, insect monitoring equipment, irrigation control terminals, and agricultural equipment controllers through the agricultural IoT gateway. Parse the original data messages according to the unified communication protocol, extract the monitoring indicator identifier, collection time, equipment number, and monitoring value, and construct the original multi-source data record set. S12. Perform data cleaning on the original multi-source data record set, remove data records with missing key fields according to the preset field integrity rules, and calculate the median and absolute deviation of the median of each indicator based on the historical sliding window. When the monitored value deviates from the median by more than a preset multiple threshold, outlier pruning or replacement is performed to obtain the cleaned multi-source data record set. S13. Perform time alignment processing on the cleaned multi-source data record set, set a unified sampling period, construct a standard time axis, and map the collection time of each data source to the standard time axis; when there is a time interval between adjacent valid sampling points, calculate the interpolated data of the target time point according to the linear proportional relationship of the changes in the values ​​of the two adjacent sampling points, and generate a time-aligned dataset. S14. Perform spatial identifier binding processing on the time-aligned dataset, determine the monitoring location based on the mapping relationship between the device number and the plot code, and encapsulate the indicator identifier, time index, spatial index and monitoring value into structured data units according to the unified agricultural monitoring data model to form a standardized multi-source agricultural monitoring data table. S15. Sort and serialize the standardized multi-source agricultural monitoring data table according to the indicator identifier and time index to generate a continuous time series with a single monitoring indicator as the unit, and summarize it to form a multi-source agricultural monitoring sequence set for subsequent statistical analysis and alarm processing.

[0020] In this embodiment, generating the pool of index sequences to be detected includes: S21. Receive a set of multi-source agricultural monitoring sequences, extract indicator identifiers, time indexes, spatial indexes and monitoring values ​​according to the preset agricultural monitoring indicator coding rules, and group the data according to the indicator identifiers to form a set of original indicator sequences divided by a single monitoring indicator. S22. Perform standardization processing on the original indicator sequence set. For each monitoring indicator, calculate its minimum and maximum values ​​within a preset time window, and map the monitoring values ​​at each time point to the interval between zero and one in a linear proportion to obtain a standardized indicator sequence set with unified dimensions. S23. Perform time continuity processing on the standardized indicator sequence set, construct a standard time axis according to a unified sampling period, and sort each indicator sequence by time index; when there are missing time points, calculate the interpolated data of the missing time based on the ratio of the numerical changes of two adjacent valid sampling points, and generate a time-continuous indicator sequence set. S24. Construct a rank statistical sample sequence for the time-continuous indicator sequence set, sort the standardized monitoring values ​​in each indicator sequence in ascending order according to time sequence and assign a rank; when there are the same values, take the average of the corresponding ranks as the rank of the value, and generate the rank statistical sequence corresponding to each indicator. S25. The rank statistical sequence of each indicator is structured and encapsulated according to the indicator identifier, time index and spatial index to form an indicator data unit containing the rank value sequence. All indicator data units are summarized and indexed to generate a pool of indicator sequences to be detected for subsequent mutation point detection and anomaly discrimination processing.

[0021] In this embodiment, forming the candidate mutation point dataset includes: S31. Receive the pool of indicator sequences to be detected, extract the time-continuous rank statistical sequence corresponding to a single monitoring indicator according to the indicator identifier, and obtain a set of sample pairs sorted by time index. ,in, For the first Time index of each sampling moment This is the standardized monitoring value corresponding to that moment. This represents the total number of samples for the monitored indicator sequence. S32, to Perform full sequence sorting to generate an ordered column. ,in, For the first The rank of each sample; when there are identical values, the "same-value rank expansion" rule is applied: the rank interval occupied by any group of identical values ​​after sorting the entire sequence. Integer ranks are assigned sequentially to samples within the same numerical group according to their chronological order in the time index. This yields a discrete integer order sequence used for calculating the Pettt statistic; S33. Construct the Pettitt test statistic sequence, first calculating the rank prefix sum. ,in, Indicates the preceding The sum of the rank values ​​of each sample; then for each candidate segmentation position Calculate the Pettitt statistic: ; in, For position The Pettitt statistic at a given location is used to characterize the intensity of the mutation at that position; all according to Write the statistics curve in ascending order And record each symbols As an indicator of the direction of mutation; S34. Perform maximum absolute value location on the statistic curve and calculate... ,in, To determine the global maximum mutation strength; construct the maximum platform index set. Determine the left boundary of the platform With the right boundary of the platform The platform median index was used as the candidate mutation point index. The output candidate mutation point time index is The output mutation strength result is The direction of the output mutation is ; S35. Perform bidirectional consistency verification to complete the second improvement / innovation: reverse the order column in time. By repeatedly performing the Pettitt test statistic sequence construction on the reverse order column and the maximum absolute value localization step on the statistic curve, the reverse candidate mutation point index is obtained. and its largest platform set Mapping the inverted index back to the forward time axis yields... And convert the mapping platform set into a positive index range. ;when Falling into the range At that time, keep Unchanged; when Not falling into the range At that time, perform the following update: If the interval and interval If there is an intersection, then the median index of the intersection is taken as the updated value. ; If the two intervals have no intersection, then in the set Selecting from and The index with the smallest distance is used as the updated index. ; And simultaneously update the candidate mutation point time index as follows The mutation direction is identified as The result of maintaining mutation strength is ; S36. Calculate the Pettitt significance probability for the updated candidate mutation points and write it into the candidate mutation point record. The significance probability is approximated by the Pettitt test. ; in, The significance probability of candidate mutation points is calculated; and a single-index candidate mutation point record is generated, which includes at least an index identifier and a candidate mutation point time index. Platform Boundaries Mutation intensity results Mutation direction identifier and significance probability ; S37. Repeat steps S31 to S36 for each index sequence in the index sequence pool to be detected, summarize all single index candidate mutation point records, form a candidate mutation point dataset, and use it for the significance verification processing in step S4 of claim 1.

[0022] In this embodiment, the output significance intensity results and the selection of a set of valid candidate anomaly intervals include: S41. Receive the candidate mutation point dataset, and for each candidate mutation point record, read the indicator identifier and candidate mutation point time index. Mutation direction identifier and platform boundary index and And extract the time-continuous standardized sample sequences of the corresponding indicators from the pool of indicator sequences to be detected. ,in, For time index Standardized monitoring values ​​at the location This represents the total number of samples in the index sequence. S42. Construct a local sample window for "platform exclusion" based on the platform boundary for candidate mutation points, and define the platform exclusion interval as the index interval. And construct the front and rear windows outside this interval: Front window from index Select samples sequentially to the left, then the back window starts from the index. Select samples sequentially to the right; Set two sets of fixed window capacities and ,in For short window capacity, For long window capacity and ; Generate short-window and front-window sample sets in chronological order. With short window and back window sample set ,make Then, generate a long-window front window sample set using the same rules. With long window and back window sample set ,make ; When the platform excludes a range of samples that are too close to the sequence boundary, resulting in insufficient available samples, it continues to expand the index sampling outward on the same side according to the "far-end filling" rule until the corresponding window capacity is met, and records the actual index range used. S43, respectively, for short-window sample pairs Compared with long window samples Rank statistic construction for performing the Mann-Whitney test: and Merge into a sample set ,right Sort the samples in ascending order and assign them ranks; when there are duplicate values, take the average of the ranks corresponding to those duplicate values ​​as the rank of that value; calculate the sum of the ranks of the front window sample sets respectively. Rank sum of the back window sample set ,in, For set The sum of the rank values ​​of all samples within the sample. For set The sum of the rank values ​​of all samples within the sample; S44. Calculate the Mann-Whitney statistic based on the rank sum of the short window and the long window, respectively. And generate the corresponding significance intensity results: ; in, The Mann-Whitney statistic. For the front window sample size, For the sample size of the back window, The sum of the rank of the samples in the front window; the sum of the rank of the samples in the short window is calculated. Calculations for long windows yielded ; S45. Perform significance determination for both short and long windows, and implement dual-scale consistency constraints: and Convert them respectively to the corresponding two-sided test significance probabilities and And obtain the significance intensity respectively. and The significance level is taken as the absolute measure of the deviation of the corresponding statistic from the neutral position; if and only if both conditions are met. and At that time, the candidate mutation points were confirmed to pass the Mann-Whitney significance test, among which Preset significance level; S46. Under the condition that the candidate mutation point passes the significance verification, perform directional consistency confirmation: calculate the Mann-Whitney directional identifier based on the rank sum difference between the front window and the back window. ,when Time setting ,when Time setting and will Mutation direction identifier Perform consistency matching; when If the candidate mutation point is valid, retain it; otherwise, discard it. S47. Generate valid candidate anomaly intervals for the retained candidate mutation points: Using the earliest time index of the front window and the latest time index of the back window actually used in the long window construction as the boundaries of the anomaly intervals, generate anomaly intervals. And the significance intensity result is defined as Write an anomaly interval record; the anomaly interval record must at least contain the indicator identifier and the time index of the candidate mutation point. Platform Boundaries Abnormal interval boundaries Short window significance probability Long window significance probability and significance intensity results ; S48. Summarize all retained abnormal interval records to form a set of valid candidate abnormal intervals.

[0023] In this embodiment, forming the distributed transition discrimination dataset includes: S51. Receive the set of valid candidate abnormal intervals, and for each abnormal interval record, read the indicator identifier and the abnormal interval boundary. Time index of candidate mutation points And extract time-continuous standardized sample sequences of the same indicator from the pool of indicator sequences to be detected. ,in For time indexing, For time index Standardized monitoring values ​​at the location This represents the total number of samples in the index sequence. S52. Construct a two-sample input set: indexed by time interval Extract the sample set within the interval and with A reference set is formed by sampling from the nearest to the farthest point in time on both sides outside the interval, with the center as the reference set. ,in Reference subset from the left With reference subset on the right Composition, satisfying and When the reference sample on either side is insufficient, the sampling order on the other side is fixed, and the sample is continuously supplemented on the available side according to the time interval until the sample is obtained. and record and Each covers a specific time index range; S53, For sets Perform a single rank assignment to construct the positional verification input: Sort the sample values ​​in ascending order and assign them ranks to obtain a merged order column. When encountering the same sample value, take the average rank of the rank interval occupied by that sample value as the rank of that sample value; calculate the set. In-sample rank sum ,in For set Sum the rank values ​​of all samples and calculate the standardized statistic of positional offset. ; in, For the standardized statistics of position offset, The sample size within the interval is 1. For reference sample size; S54. Perform a second rank assignment on the same merged order column to construct a metric test. Input: using the center rank... Based on this, calculate the absolute distance from the rank of each sample to the center rank, and then sort all absolute distance values ​​in ascending order before assigning them a new rank. When encountering the same absolute distance value, use the average distance rank. Calculate the set... The sum of the rank of the corresponding distance rank ,in For set The sum of the rank of the distances of all samples within the range, and with The standardized statistic for scale change is obtained by subtracting the expected rank sum of distances under the null hypothesis and dividing by the standard deviation of the rank sum of distances under the null hypothesis. ; S55. Generate joint statistics and produce distribution transition strength results. ,in For the improved Lepage joint statistic, Standardized statistics for scale change; The symbol determines the position offset direction identifier, in order to The symbol determines the direction of scale change and... Record the abnormal interval as the result of the distribution transition intensity; S56. For each anomalous interval, output the location offset result, scale change result, and distribution transition intensity result, and include the index identifier and anomalous interval boundary. Candidate mutation point time index , , Location offset direction indicator, scale change direction indicator, and joint statistics Encapsulated as a distributed transition discrimination record; S57. Summarize the distribution transition discrimination records to form a distribution transition discrimination dataset.

[0024] In this embodiment, the generated alarm information includes: S61, Receive mutation intensity results Significance results and the joint statistics in the distribution transition discriminant dataset Based on the indicator identifier, three types of intensity indicators for the same candidate anomaly interval are correlated and matched to construct a three-dimensional anomaly feature vector. ,in, The global mutation strength is obtained from the Pettitt test. The significance level is obtained from the Mann-Whitney test. The intensity of the distributed transitions obtained from the improved Lepage test; S62. Perform uniform scale mapping on the three-dimensional anomaly feature vectors, and obtain them within the preset historical sample intervals. , , The maximum and minimum values ​​are obtained, and each component is mapped to a standard interval of zero to one in a linear proportion to obtain the normalized intensity components. , , ; S63. Construct a unified anomaly scoring function, according to preset weight coefficients. , , The standardized intensity components are weighted and summed to generate a uniform anomaly score. : ; in, To standardize anomaly scoring values, Each weighting coefficient is determined based on a preset scoring strategy; S64. Standardized anomaly scoring values Perform hierarchical threshold mapping processing, setting at least three incremental threshold intervals. ,when It was determined to be a Level 1 warning at that time. It was determined to be a Level II warning at that time. It was determined to be a Level III warning at that time. The system is then classified as a Level 4 warning and a corresponding warning level identifier is generated. S65. Determine the warning time and related indicator information, wherein the warning time is taken from the time index of the candidate mutation point. And record the boundaries of the abnormal intervals. The duration of the anomaly; related indicator information includes indicator identifiers, monitoring location identifiers, and corresponding... , , With uniform anomaly score ; S66. Encapsulate the warning level, warning time, abnormal duration range, related indicator information, and statistical intensity results into a structured alarm information record, and write it into the alarm database through the alarm publishing interface. At the same time, send alarm information to the preset alarm terminals to realize alarm output.

[0025] Example: To verify the feasibility of the proposed agricultural process big data statistical analysis and alarm method, this invention was applied to a production monitoring and anomaly alarm scenario at a large-scale agricultural planting base. This base comprises multiple contiguous planting areas, covering a mixed production area of ​​food crops and cash crops. Meteorological monitoring terminals, soil monitoring terminals, insect monitoring terminals, irrigation control terminals, and agricultural machinery status acquisition terminals have been deployed on-site. These terminals continuously output monitoring data such as air temperature, air humidity, rainfall, wind speed, light intensity, soil moisture content, soil temperature, insect population, irrigation valve status, water pump operating status, equipment current, and equipment vibration. The base's original system primarily used a fixed threshold method for alarms, triggering a notification when a certain indicator exceeded a preset upper limit or fell below a preset lower limit. In actual operation, base managers found that this method is difficult to detect slow-accumulating anomalies and multi-factor coupled anomalies in a timely manner. For example, in the early stage of local irrigation blockage, the valve status does not show a fault, but the soil moisture content has already shown a continuous deviation; in the early stage of insect infestation, although the number of insects has not reached the extreme value, the fluctuation range has increased significantly; in the early stage of equipment failure, both current and vibration show slight changes, and the single threshold method is difficult to form an effective alarm, which often leads to on-site treatment only after the crops are affected or the equipment is shut down.

[0026] In this scenario, the present invention is deployed on an agricultural data analysis platform. The platform accesses the data streams from various monitoring terminals through an agricultural IoT gateway, parses the raw messages, and extracts indicator identifiers, collection times, device numbers, spatial locations, and monitoring values ​​to form a raw multi-source data record set. The system first performs data cleaning, removing records with missing key fields, and trims or replaces outliers that significantly deviate from the normal distribution based on historical statistical results within a sliding window. After cleaning, the system constructs a standard time axis according to a unified sampling period, mapping data from different frequencies and sources to the same time reference. For gaps between adjacent valid sampling points, interpolation is used to fill in the missing time data. Subsequently, the system completes spatial binding based on the mapping relationship between device numbers and plot codes, encapsulating indicator identifiers, time indices, spatial indices, and monitoring values ​​into unified structured data units, and generating continuous time series according to a single monitoring indicator, forming a multi-source agricultural monitoring sequence set.

[0027] Due to significant differences in the dimensions of various monitoring indicators, the system continues to standardize each indicator sequence, mapping the monitoring values ​​at each time point to a unified numerical range and constructing a rank statistical sample sequence in chronological order. For samples with the same numerical value, the average rank method is used to ensure consistency in subsequent statistical tests. After completing the rank statistical modeling, the system performs global mutation point localization processing based on the Pettitt test for each indicator sequence. By constructing a full-sequence statistical curve, it searches for candidate positions with the largest absolute value of the statistical value and outputs the candidate mutation point and its corresponding mutation intensity. For time-reverse sequences, the system performs the same calculation simultaneously, confirming the consistency between the forward and reverse results to improve the stability of candidate mutation point localization. After obtaining the candidate mutation point, the system constructs local sample windows before and after the candidate position, excluding samples within the plateau interval, forming two sets of before and after samples: a short window and a long window. Then, the Mann-Whitney test is performed to verify the significance of the difference before and after the candidate mutation point. Only candidate points that simultaneously meet the dual-scale significance constraint and are in the same direction are retained as valid candidate anomaly intervals. For the abnormal intervals that pass the screening, the system further extracts samples within the interval and constructs a reference sample set outside the interval, performs an improved Lepage test, and calculates the positional shift results, scale change results, and joint statistics to characterize the distribution transition characteristics of the abnormal interval in terms of numerical level and fluctuation structure.

[0028] After completing the above analysis, the system correlates and matches mutation intensity, significance intensity, and distribution transition intensity to construct a three-dimensional anomaly feature vector. It then performs uniform scale mapping on each component and performs weighted calculation using preset weights to generate a unified anomaly score. The system outputs the corresponding warning level according to the score interval and simultaneously records the time index of candidate mutation points, the duration of the anomaly interval, related plots, and related indicator information, forming a structured alarm record, which is written to the alarm database and sent to the management terminal. On-site management personnel use this information to quickly investigate irrigation pipelines, areas of insect infestation change, and equipment operating status. In actual operation, the soil moisture content of a certain plot did not fall below the original lower threshold for a period of time, but the system detected a significant mutation in the sequence during a local period. After significance verification and distribution transition discrimination, it was confirmed as a valid anomaly interval. Upon inspection, management personnel found a blockage in the drip irrigation branch pipe and promptly cleared it, preventing continuous water loss in the local crop. In another equipment operating area, although the current and vibration data did not trigger the traditional threshold alarm, the system identified distributed transition characteristics. After on-site verification, it was confirmed that this was a precursor to bearing wear. Subsequently, maintenance was arranged to avoid equipment downtime.

[0029] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for big data statistical analysis and alarm in agricultural processes, characterized in that, include: S1. Collect data generated during agricultural production, perform cleaning, time alignment and unified data model mapping on the collected data, and generate a multi-source agricultural monitoring sequence set; S2. Receive a multi-source agricultural monitoring sequence set, perform standardization processing according to the preset indicator specifications, construct a rank statistical sample sequence, and generate a pool of indicator sequences to be detected; S3. Receive the pool of index sequences to be detected, perform global mutation point localization processing based on Pettitt test on each index sequence, output candidate mutation points and corresponding mutation intensity results, and form a candidate mutation point dataset. S4. Receive the candidate mutation point dataset, construct the front and back local sample windows based on each candidate mutation point, perform the Mann-Whitney test to verify significance, output the significance strength results, and filter to obtain the set of effective candidate abnormal intervals. S5. Receive the set of valid candidate anomaly intervals, perform the improved Lepage test in each anomaly interval, generate the position offset result, scale change result and distribution transition intensity result, and form a distribution transition discrimination dataset. S6. Receive mutation intensity results, significance intensity results, and distribution transition discrimination dataset, construct a unified anomaly scoring vector and perform hierarchical threshold mapping, generate warning level, warning time, and related indicator information, and output alarm information.

2. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The generation of a multi-source agricultural monitoring sequence set includes: S11. Access the data streams from meteorological sensors, soil sensors, insect monitoring equipment, irrigation control terminals, and agricultural equipment controllers through the agricultural IoT gateway. Parse the original data messages according to the unified communication protocol, extract the monitoring indicator identifier, collection time, equipment number, and monitoring value, and construct the original multi-source data record set. S12. Perform data cleaning on the original multi-source data record set, remove data records with missing key fields according to the preset field integrity rules, and calculate the median and absolute deviation of the median of each indicator based on the historical sliding window. When the monitored value deviates from the median by more than a preset multiple threshold, outlier pruning or replacement is performed to obtain the cleaned multi-source data record set. S13. Perform time alignment processing on the cleaned multi-source data record set, set a unified sampling period, construct a standard time axis, and map the collection time of each data source to the standard time axis; when there is a time interval between adjacent valid sampling points, calculate the interpolated data of the target time point according to the linear proportional relationship of the changes in the values ​​of the two adjacent sampling points, and generate a time-aligned dataset. S14. Perform spatial identifier binding processing on the time-aligned dataset, determine the monitoring location based on the mapping relationship between the device number and the plot code, and encapsulate the indicator identifier, time index, spatial index and monitoring value into structured data units according to the unified agricultural monitoring data model to form a standardized multi-source agricultural monitoring data table. S15. Sort and serialize the standardized multi-source agricultural monitoring data table according to the indicator identifier and time index to generate a continuous time series with a single monitoring indicator as the unit, and summarize it to form a multi-source agricultural monitoring sequence set.

3. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The pool of sequences to be detected includes: S21. Receive a set of multi-source agricultural monitoring sequences, extract indicator identifiers, time indexes, spatial indexes and monitoring values ​​according to the preset agricultural monitoring indicator coding rules, and group the data according to the indicator identifiers to form a set of original indicator sequences divided by a single monitoring indicator. S22. Perform standardization processing on the original indicator sequence set, calculate the minimum and maximum values ​​of each monitoring indicator within a preset time window, and map the monitoring values ​​at each time point to the interval between zero and one in a linear proportion to obtain the standardized indicator sequence set. S23. Perform time continuity processing on the standardized indicator sequence set, construct a standard time axis according to a unified sampling period, and sort each indicator sequence by time index; when there are missing time points, calculate the interpolated data of the missing time based on the ratio of the numerical changes of two adjacent valid sampling points, and generate a time-continuous indicator sequence set. S24. Construct a rank statistical sample sequence for the time-continuous indicator sequence set, sort the standardized monitoring values ​​in each indicator sequence in ascending order according to time sequence and assign a rank; when there are the same values, take the average of the corresponding ranks as the rank of the value, and generate the rank statistical sequence corresponding to each indicator. S25. The rank statistical sequence of each indicator is structured and encapsulated according to the indicator identifier, time index and spatial index to form an indicator data unit containing the rank value sequence. All indicator data units are summarized and indexed to generate a pool of indicator sequences to be detected.

4. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The candidate mutation point dataset includes: S31. Receive the pool of indicator sequences to be detected, extract the time-continuous rank statistical sequence corresponding to a single monitoring indicator according to the indicator identifier, and obtain a set of sample pairs sorted by time index. ,in, For the first Time index of each sampling moment The standardized monitoring value corresponding to the stated time. The total number of samples for the monitored indicator sequence; S32, to Perform full sequence sorting to generate an ordered column. ,in, For the first The rank of each sample; when there are identical values, the "same-value rank expansion" rule is applied: the rank interval occupied by any group of identical values ​​after sorting the entire sequence. Integer ranks are assigned sequentially to samples within the same numerical group according to their chronological order in the time index. This yields a discrete integer sequence; S33. Construct the Pettitt test statistic sequence, first calculating the rank prefix sum. ,in, Indicates the preceding The sum of the rank values ​​of each sample; then for each candidate segmentation position Calculate the Pettitt statistic: ； in, For position The Pettitt statistic at the given location characterizes the intensity of the abrupt change at that location; all according to Write the statistics curve in ascending order And record each symbols As an indicator of the direction of mutation; S34. Perform maximum absolute value location on the statistic curve and calculate... ,in, To determine the global maximum mutation strength; construct the maximum platform index set. Determine the left boundary of the platform With the right boundary of the platform The platform median index was used as the candidate mutation point index. The output candidate mutation point time index is The output mutation strength result is The direction of the output mutation is ; S35. Perform bidirectional consistency verification: Reverse the order column in chronological order. By repeatedly performing the Pettitt test statistic sequence construction on the reverse order column and the maximum absolute value localization step on the statistic curve, the reverse candidate mutation point index is obtained. and its largest platform set Mapping the inverted index back to the forward time axis yields... And convert the mapping platform set into a positive index range. ;when Falling into the range At that time, keep Unchanged; when Not falling into the range At that time, perform the following update: If the interval and interval If there is an intersection, then the median index of the intersection is taken as the updated value. ; If the two intervals have no intersection, then in the set Selecting from and The index with the smallest distance is used as the updated index. ; And simultaneously update the candidate mutation point time index as follows The mutation direction is identified as The result of maintaining mutation strength is ; S36. Calculate the Pettitt significance probability for the updated candidate mutation points and write it into the candidate mutation point record. The significance probability is approximated by the Pettitt test. ； in, The significance probability of candidate mutation points is calculated; and a single-index candidate mutation point record is generated, which includes an index identifier and a candidate mutation point time index. Platform Boundaries Mutation intensity results Mutation direction identifier and significance probability ; S37. Summarize all single-index candidate mutation point records to form a candidate mutation point dataset.

5. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The output shows the significance level results, and the set of valid candidate anomaly intervals is obtained by filtering: S41. Receive the candidate mutation point dataset, and for each candidate mutation point record, read the indicator identifier and candidate mutation point time index. Mutation direction identifier and platform boundary index and And extract the time-continuous standardized sample sequences of the corresponding indicators from the pool of indicator sequences to be detected. ,in, For time index Standardized monitoring values ​​at the location The total number of samples in the index sequence; S42. Construct a local sample window for platform exclusion based on the platform boundary for candidate mutation points, and define the platform exclusion interval as the index interval. And construct a front window and a rear window outside the aforementioned interval: Front window from index Select samples sequentially to the left, then the back window starts from the index. Select samples sequentially to the right; Set two sets of fixed window capacities and ,in For short window capacity, For long window capacity and ; Generate short-window and front-window sample sets in chronological order. With short window and back window sample set , Then, generate a long-window front window sample set using the same rules. With long window and back window sample set ,make ; When the platform excludes a range of samples that are too close to the sequence boundary, resulting in insufficient available samples, it continues to expand the index sampling outward on the same side according to the "far-end filling" rule until the corresponding window capacity is met, and records the actual index range used. S43, respectively, for short-window sample pairs Compared with long window samples Rank statistic construction for performing the Mann-Whitney test: and Merge into a sample set ,right Sort the samples in ascending order of their values ​​and assign them ranks; when there are duplicate values, take the average of the ranks corresponding to the duplicate values ​​as the rank of the values; calculate the sum of the ranks of the front window sample sets respectively. Rank sum of the back window sample set ,in, For set The sum of the rank values ​​of all samples within the sample. For set The sum of the rank values ​​of all samples within the sample; S44. Calculate the Mann-Whitney statistic based on the rank sum of the short window and the long window, respectively. And generate the corresponding significance intensity results: ； in, The Mann-Whitney statistic. For the front window sample size, For the sample size of the back window, The rank sum of the samples in the front window; the short window is calculated as follows: Calculations for long windows yielded ; S45. Perform significance determination for both short and long windows, and implement dual-scale consistency constraints: and Convert them respectively to the corresponding two-sided test significance probabilities and And obtain the significance intensity respectively. and The significance level is taken as the absolute measure of the deviation of the corresponding statistic from the neutral position; if and only if both conditions are met. and At that time, the candidate mutation points were confirmed to pass the Mann-Whitney significance test, among which Preset significance level; S46. Under the condition that the candidate mutation point passes the significance verification, perform directional consistency confirmation: calculate the Mann-Whitney directional identifier based on the rank sum difference between the front window and the back window. ,when Time setting ,when Time setting and will Mutation direction identifier Perform consistency matching; when The candidate mutation point is retained if it is valid, otherwise it is discarded. S47. Generate valid candidate anomaly intervals for the retained candidate mutation points: Using the earliest time index of the front window and the latest time index of the back window actually used in the long window construction as the boundaries of the anomaly intervals, generate anomaly intervals. And the significance intensity result is defined as Write the abnormal range record; S48. Summarize all retained abnormal interval records to form a set of valid candidate abnormal intervals.

6. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The distributed transition discrimination dataset includes: S51. Receive the set of valid candidate abnormal intervals, and for each abnormal interval record, read the indicator identifier and the abnormal interval boundary. Time index of candidate mutation points And extract time-continuous standardized sample sequences of the same indicator from the pool of indicator sequences to be detected. ,in For time indexing, For time index Standardized monitoring values ​​at the location The total number of samples in the index sequence; S52. Construct a two-sample input set: indexed by time from interval Extract the sample set within the interval and with A reference set is formed by sampling on both sides of the interval from near to far in time, centered on the interval. ,in Reference subset from the left With reference subset on the right Composition, satisfying and When the reference sample on either side is insufficient, the sampling order on the other side is fixed, and the sample is continuously supplemented on the available side according to the time interval until the sample is obtained. and record and Each covers a specific time index range; S53, on sets Perform a single rank assignment to construct the positional verification input: Sort the sample values ​​in ascending order and assign them ranks to obtain a merged order column. When encountering the same sample value, take the average rank of the rank interval occupied by the same sample value as the rank of the sample value; calculate the set. Inner Sample Rank Sum ,in For set Sum the rank values ​​of all samples and calculate the standardized statistic of positional offset. ； in, For the standardized statistics of position offset, The sample size within the interval is 1. For reference sample size; S54. Perform a second rank assignment on the same merged order column to construct a metric test. Input: using the center rank... Based on this, calculate the absolute distance from the rank of each sample to the center rank, and then sort all absolute distance values ​​in ascending order before assigning them a new rank. When encountering the same absolute distance value, use the average distance rank. Calculate the set... The sum of the rank of the corresponding distance rank ,in For set The sum of the rank of the distances of all samples within the range, and with The standardized statistic for scale change is obtained by subtracting the expected rank sum of distances under the null hypothesis and dividing by the standard deviation of the rank sum of distances under the null hypothesis. ; S55. Generate joint statistics and produce results on the intensity of distribution transitions: ,in For the improved Lepage joint statistic, Standardized statistics for scale change; The symbol determines the position offset direction identifier, in order to The symbol determines the direction of scale change and... Record the abnormal interval as the result of the distribution transition intensity; S56. For each anomalous interval, output the location offset result, scale change result, and distribution transition intensity result, and include the index identifier and anomalous interval boundary. Candidate mutation point time index , , Location offset direction indicator, scale change direction indicator, and joint statistics Encapsulated as a distributed transition discrimination record; S57. Summarize the distribution transition discrimination records to form a distribution transition discrimination dataset.

7. The method for big data statistical analysis and alarm of agricultural processes according to claim 1, characterized in that, The generated alarm information includes: S61, Receive mutation intensity results Significance results and the joint statistics in the distribution transition discriminant dataset Based on the indicator identifier, three types of intensity indicators for the same candidate anomaly interval are correlated and matched to construct a three-dimensional anomaly feature vector. ,in The global mutation strength is obtained from the Pettitt test. The significance level is obtained from the Mann-Whitney test. The intensity of the distributed transitions obtained from the improved Lepage test; S62. Perform uniform scale mapping on the three-dimensional anomaly feature vectors, and obtain them within the preset historical sample intervals. , , The maximum and minimum values ​​are obtained, and each component is mapped to a standard interval of zero to one in a linear proportion to obtain the normalized intensity components. , , ; S63. Construct a unified anomaly scoring function, according to preset weight coefficients. , , The standardized intensity components are weighted and summed to generate a uniform anomaly score. ; S64. Standardized anomaly scoring values Perform hierarchical threshold mapping processing and set incremental threshold intervals. ,when It was determined to be a Level 1 warning at that time. It was determined to be a Level II warning at that time. It was determined to be a Level III warning at that time. The system is then classified as a Level 4 warning and a corresponding warning level identifier is generated. S65. Determine the warning time and related indicator information, wherein the warning time is taken from the time index of the candidate mutation point. And record the boundaries of the abnormal intervals. The duration of the anomaly; related indicator information includes indicator identifiers, monitoring location identifiers, and corresponding... , , With uniform anomaly score ; S66. Encapsulate the warning level, warning time, abnormal duration range, related indicator information, and statistical intensity results into a structured alarm information record, and write it into the alarm database through the alarm release interface, while sending the alarm information to the preset alarm terminals.

Images