Garbage throwing optimization management method and system based on classification data mining

Abstract

The invention provides a garbage throwing optimization management method and system based on classification data mining, and relates to the technical field of garbage management, and the method comprises the steps: obtaining garbage throwing behavior data uploaded by garbage throwing collection equipment, and generating an original behavior sequence; performing time period division to generate a short-period classification sequence and a long-period classification sequence; respectively carrying out category difference processing of adjacent time periods and category aggregation calculation of each long-period stage to generate a short-period fluctuation index and a long-period stability index; performing cross-sequence correlation analysis to generate a correlation analysis result, and constructing a layered trigger factor comprising a short-period trigger factor and a long-period trigger factor; carrying out offset difference judgment, and when the offset difference between the short-period trigger factor and the long-period trigger factor exceeds a preset associated offset threshold value, generating a behavior mutation event; strategy migration processing is executed, and a migrated garbage throwing management strategy is generated; according to the invention, the autonomy and accuracy of garbage throwing optimization management are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of garbage management, in particular to a garbage throwing optimization management method and system based on classification data mining. BACKGROUND

[0002] The prior art generally uses a fixed garbage throwing management system, which collects the throwing behavior of residents in real time by installing a weighing sensor, a throwing port RFID identification device or a camera for each garbage can, and uploads the throwing amount, throwing time period, throwing type and other structured data to a background database. The background system generally uses a preset threshold rule or a simple clustering algorithm (such as K-means clustering based on time period) to statistically analyze the above data to generate garbage overflow prediction, garbage classification accuracy evaluation, and cleaning route planning management results. In actual deployment in communities, such systems rely on historical average or fixed rules to determine throwing behavior patterns, for example, using empirical statistics that "a certain type of garbage is thrown in large quantities at night" as the basis for subsequent scheduling, thereby achieving basic garbage classification management and cleaning scheduling automation.

[0003] However, the above prior art is prone to problems such as rigid data mining models and difficulty in adapting to real changes when facing scenarios with obvious throwing behavior heterogeneity. For example, in some communities in suburban areas, tenants frequently change, and the throwing habits of new residents (such as throwing large pieces of cardboard in the early morning for three consecutive days) can cause the actual overflow time of the garbage can to be suddenly advanced, but the rule model based on historical average still judges according to the original threshold of "not overflowing before 8 am", failing to dynamically identify this sudden change in behavior, resulting in a large amount of overflow of the garbage can at 4 am. Such technical defects are not just "low efficiency or high cost", but because existing models are difficult to automatically identify "abnormal throwing patterns" and "periodic changes", they cannot update behavior patterns in specific community customer structure change scenarios, causing scheduling output of the management system to lag and misjudge, resulting in garbage accumulation, odor spread, and an increase in subsequent manual intervention frequency and other practical problems. SUMMARY

[0004] The purpose of the present application is to provide a garbage throwing optimization management method and system based on classification data mining, which aims to solve the problems mentioned in the background.

[0005] To solve the above technical problems, the technical solution of the present application is as follows: In a first aspect, a garbage throwing optimization management method based on classification data mining, the method comprising: obtaining garbage throwing behavior data uploaded by a garbage throwing collection device, and generating an original behavior sequence representing continuous throwing behavior according to the data; According to the original behavior sequence, time period division is performed to generate short-period behavior data and long-period behavior data, and short-period classification sequences and long-period classification sequences are constructed respectively; According to the short-period classification sequences and the long-period classification sequences, category difference processing of adjacent time periods and category aggregation calculation of each long-period phase are performed respectively to generate a short-period fluctuation index and a long-period stability index; According to the short-period fluctuation index and the long-period stability index, cross-sequence correlation analysis is performed to generate a correlation analysis result, and a hierarchical trigger factor including a short-period trigger factor and a long-period trigger factor is constructed according to the correlation analysis result; According to the hierarchical trigger factor, offset difference determination is performed, and when the offset difference between the short-period trigger factor and the long-period trigger factor exceeds a preset correlation offset threshold, a behavior mutation event is generated; According to the behavior mutation event, strategy migration processing is performed to generate a migrated garbage disposal management strategy, which specifically includes an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage disposal classification correction prompt; According to the migrated garbage disposal management strategy, garbage disposal optimization management operations are performed, and the execution results are output to a garbage disposal management terminal.

[0006] Preferably, according to the short-period classification sequences and the long-period classification sequences, category difference processing of adjacent time periods and category aggregation calculation of each long-period phase are performed respectively to generate a short-period fluctuation index and a long-period stability index, including: The short-period classification sequences are segmented according to fixed-length short-period analysis windows, and in each analysis window, the category changes of adjacent time periods are compared in direction, judged in continuity, and accumulated in amplitude to generate short-period difference results within the short-period window; According to the short-period difference results, the category change patterns of consecutive multiple analysis windows are detected for consistency, and the consistency detection results are converted into a short-period fluctuation index; The long-period classification sequences are grouped according to long-period phases, and in each long-period group, long-period aggregation results are generated according to the category appearance frequency, the duration, and the stability degree of the dominant category within the phase, and the long-period stability index is calculated according to the long-period aggregation results.

[0007] Preferably, according to the short-period fluctuation index and the long-period stability index, cross-sequence correlation analysis is performed to generate a correlation analysis result, and a hierarchical trigger factor including a short-period trigger factor and a long-period trigger factor is constructed according to the correlation analysis result, including: The short-period fluctuation index and the long-period stability index are time-aligned, the trend difference value of the two is calculated in the aligned common time period, and a preliminary correlation result is generated according to the trend difference value; According to the trend difference value in the primary correlation result, the directionality, accumulativity and persistence in the alignment time period of the difference are segmented and analyzed, and a short-period trigger factor basis value for describing the short-period burst behavior and a long-period trigger factor basis value for describing the long-term deviation behavior are generated; According to the structural difference between the short-period trigger factor basis value and the long-period trigger factor basis value, the two are hierarchically reorganized, the short-period trigger factor is placed in the fast reaction layer, and the long-period trigger factor is placed in the trend recognition layer, forming a hierarchical trigger factor comprising two layers.

[0008] Preferably, according to the deviation difference between the short-period trigger factor and the long-period trigger factor, when the deviation difference exceeds a preset correlation deviation threshold, a behavior mutation event is generated, including: The short-period trigger factor and the long-period trigger factor of the hierarchical trigger factor are respectively subjected to deviation source filtering, the transient change caused by a single anomaly in the short-period trigger factor is removed, and the unstable deviation caused by short-term fluctuations in the long-period trigger factor is excluded, to generate filtered short-period trigger factors and long-period trigger factors; The continuous change amount of the deviation difference between the filtered short-period trigger factors and the long-period trigger factors is calculated, and the continuous change amount is accumulated in multiple consecutive time periods to form a deviation difference accumulation sequence; According to the accumulation amount and the duration in the deviation difference accumulation sequence, a double-condition judgment is performed, when the accumulation amount exceeds a preset correlation deviation threshold and the duration exceeds a preset first duration threshold, a behavior mutation event is generated, the behavior mutation event includes a mutation category for indicating a mutation type, a mutation amplitude for indicating a mutation intensity, and a mutation time period for indicating a mutation occurrence range.

[0009] Preferably, according to the behavior mutation event, a strategy migration process is performed to generate a migrated garbage disposal management strategy, specifically including an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage disposal classification correction prompt, including: According to the mutation amplitude and the mutation category corresponding to the behavior mutation event, the garbage can overflow prediction window is dynamically expanded or contracted, so that the time range of the prediction window changes proportionally with the mutation amplitude, and an adjusted prediction window is generated; According to the garbage categories involved in the behavior mutation event and their occurrence frequencies, the collection objects are prioritized, and by establishing a collection priority promotion factor, the high mutation categories obtain higher collection priority, and an adjusted collection sequence is generated; According to the garbage classification deviation corresponding to the behavior mutation event, the deviation category is taken as a key prompt category, the push frequency of the prompt of this category is improved, and the prompt content is adaptively updated, to generate an updated garbage disposal classification correction prompt.

[0010] Preferably, the short-period fluctuation indicator and the long-period stability indicator are time-aligned, the trend difference value is calculated in the aligned common time period, and the primary correlation result is generated according to the trend difference value, including: The time reference points are determined according to the short-period fluctuation indicator and the long-period stability indicator, and the time reference points are extended to form reference time series for describing the time distribution of the two types of indicators; The short-period fluctuation indicator and the long-period stability indicator are time-mapped according to the reference time series, so that the two types of indicators correspond on the same time axis to form an aligned indicator sequence; According to the aligned indicator sequence, the change direction, change amplitude and change speed of the short-period fluctuation indicator and the long-period stability indicator in the common time period are obtained, and interval difference extraction is performed to generate a trend difference value; According to the trend difference value, the deviation direction, deviation intensity and deviation duration in the interval are associated to obtain a primary correlation result for representing the cross-period behavior change relationship.

[0011] Preferably, the directionality, accumulation and persistence of the difference in the trend difference value are analyzed in segments according to the primary correlation result, to generate a short-period trigger factor basic value for describing short-period burst behavior and a long-period trigger factor basic value for describing long-term deviation behavior, including: The trend difference value is directionally divided according to the change direction of the trend difference value to form a directionality difference segment for representing the trend change direction; According to the directionality difference segment, the trend difference value inside the difference segment is accumulated, and an accumulation difference segment for representing the difference accumulation intensity is formed according to the accumulation change characteristics; According to the duration of the accumulation difference segment in the aligned time period, the difference segment is analyzed for persistence, and the difference segment whose duration exceeds a preset second duration threshold is identified, and a short-period trigger factor basic value is generated according to the difference segment; The difference segment that does not meet the persistence threshold is analyzed for long-term deviation, and a long-period trigger factor basic value is generated according to the deviation direction, deviation amplitude and deviation phase characteristics.

[0012] In a second aspect, a garbage disposal optimization management system based on classification data mining, the system comprises: A data acquisition module for acquiring garbage disposal behavior data uploaded by a garbage disposal collection device, and generating an original behavior sequence for representing continuous disposal behavior according to the garbage disposal behavior data, the garbage disposal behavior data including garbage disposal quantity, garbage disposal time and garbage disposal type; The sequence construction module is configured to divide time periods according to the original behavior sequence, generate short-period behavior data and long-period behavior data, and construct a short-period classification sequence and a long-period classification sequence, respectively; The index generation module is configured to perform category difference processing on adjacent time periods and category aggregation calculation on each long-period stage according to the short-period classification sequence and the long-period classification sequence, and generate a short-period fluctuation index and a long-period stability index; The correlation analysis module is configured to perform cross-sequence correlation analysis according to the short-period fluctuation index and the long-period stability index, generate a correlation analysis result, and construct a hierarchical trigger factor including a short-period trigger factor and a long-period trigger factor according to the correlation analysis result; The mutation recognition module is configured to perform offset difference determination according to the hierarchical trigger factor, and generate a behavior mutation event when an offset difference between the short-period trigger factor and the long-period trigger factor exceeds a preset correlation offset threshold; The strategy migration module is configured to perform strategy migration processing according to the behavior mutation event, and generate a migrated garbage disposal management strategy, which specifically includes an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage disposal classification correction prompt. The disposal management module is configured to perform garbage disposal optimization management operation according to the migrated garbage disposal management strategy, and output an execution result to a garbage disposal management terminal.

[0013] The above scheme of the present application at least has the following beneficial effects: Firstly, by continuously processing the multi-dimensional disposal behavior data uploaded by the garbage disposal collection equipment, the present application can reconstruct the scattered single disposal records into complete original behavior sequences, so that the behavior mode is no longer limited to the mean value statistics of fixed time periods. This sequential expression method can avoid the behavior mode missing problem caused by data breakpoints in the traditional rule model in structure, so that the subsequent analysis can directly reflect the real disposal rhythm changes of the residents, and provide a continuous and stable data basis for further identifying behavior trends and abnormalities.

[0014] On this basis, the present application divides the original behavior sequence into short-period and long-period time scales, and constructs a short-period classification sequence and a long-period classification sequence, respectively, so that the behavior characteristics can be expressed at different time granularities. The short-period sequence can sensitively capture the rapid fluctuations of disposal quantity or disposal category in a short period, and the long-period sequence is used to present the dominant change trend in different stages. Unlike the prior art which relies on a single time granularity to judge behavior, the period hierarchical structure of the present application effectively improves the ability of the model to distinguish between sudden behavior and long-term changes, and reduces the analysis distortion and misjudgment of the traditional method in the user group change scenario.

[0015] Further, the present application generates short-period fluctuation indicators and long-period stability indicators by performing category difference processing on short-period classification sequences and aggregated calculation on long-period classification sequences, so that behavior changes are quantitatively expressed in two dimensions of "rapid fluctuation" and "overall stability". Compared with the prior art which only uses simple threshold values or clustering results as the basis for judgment, the present application can extract more levels of information from the behavior structure itself, so that the system has higher sensitivity and analysis capability when facing changes in household delivery habits.

[0016] On this basis, the present application constructs hierarchical trigger factors through cross-sequence correlation analysis, which can convert the difference relationship between short-period fluctuation indicators and long-period stability indicators into trigger factors with hierarchical properties. Short-period trigger factors can express the impact of short-time sudden behavior, and long-period trigger factors can describe the degree of long-term deviation from the trend of behavior. Unlike the prior art which cannot distinguish between short-term abnormalities and long-term structural changes, the hierarchical trigger factors of the present application can enable the system to make more refined judgments according to the nature of behavior changes, significantly reducing scheduling errors due to mistaking short-term fluctuations as trend changes.

[0017] Further, by performing offset difference judgment on the hierarchical trigger factors, when the difference between the short-period trigger factors and the long-period trigger factors exceeds a preset correlation offset threshold, the present application can generate a behavior mutation event to explicitly identify that the behavior pattern has deviated from the original trend. This mechanism enables the system to identify structural changes in household delivery behavior in advance, avoiding the situation described in the background art where "midnight concentrated delivery leads to early overflow" is not recognized by the system, effectively reducing garbage overflow and environmental impact caused by scheduling lag.

[0018] Finally, by performing strategy migration processing according to the behavior mutation event, the present application can automatically update the garbage can overflow prediction window, garbage collection priority, and garbage delivery classification correction prompt, realizing dynamic adjustment of the management strategy. This strategy migration mechanism makes the system scheduling result no longer dependent on long-term fixed rules, but can evolve in real time with the change of delivery behavior, improving prediction accuracy, collection scheduling adaptability, and the relevance of classification guidance. In actual community operation, for example, when a certain type of paperboard garbage suddenly increases in a short period of time and significantly deviates from the long-term rule, the present application can generate a mutation event and automatically increase the collection priority of the related garbage can in time, avoiding the situation of midnight overflow, so that the management personnel can reduce the frequency of manual intervention and maintain the stability and cleanliness of the community environment. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of the garbage delivery optimization management method based on classification data mining provided by the embodiment of the present application. DETAILED DESCRIPTION

[0020] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be thoroughly understood, and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0021] As Figure 1 shown, the embodiments of the present application propose a garbage throwing optimization management method based on classification data mining, which comprises: obtaining garbage throwing behavior data uploaded by a garbage throwing collection device, and generating original behavior sequences for representing continuous throwing behaviors according to the garbage throwing behavior data, the garbage throwing behavior data including garbage throwing amount, garbage throwing time and garbage throwing type; dividing time periods according to the original behavior sequences, generating short-period behavior data and long-period behavior data, and respectively constructing short-period classification sequences and long-period classification sequences; According to the short-period classification sequence and the long-period classification sequence, the category difference processing of adjacent time periods and the category aggregation calculation of each long-period stage are performed respectively, and the short-period fluctuation index and the long-period stability index are generated; According to the short-period fluctuation index and the long-period stability index, cross-sequence correlation analysis is performed to generate correlation analysis results, and a hierarchical trigger factor including a short-period trigger factor and a long-period trigger factor is constructed according to the correlation analysis results; According to the hierarchical trigger factor, the offset difference is determined, and when the offset difference between the short-period trigger factor and the long-period trigger factor exceeds a preset correlation offset threshold, a behavior mutation event is generated; According to the behavior mutation event, a strategy migration process is performed to generate a migrated garbage throwing management strategy, which specifically includes an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage throwing classification correction prompt; According to the migrated garbage throwing management strategy, a garbage throwing optimization management operation is performed, and the execution result is output to a garbage throwing management terminal.

[0022] In the embodiments of the present application, by continuously processing the multi-dimensional garbage throwing behavior data uploaded by the garbage throwing collection device, a more actual behavior change rule-conforming original behavior sequence can be obtained, avoiding the problem of incomplete behavior mode caused by directly using scattered data, thereby providing a complete and continuous data basis for subsequent behavior analysis.

[0023] The original behavior sequence is divided into time periods, and short-period classification sequences and long-period classification sequences are constructed respectively, so that the throwing behavior can be characterized at different time scales. On the one hand, the short-period classification sequence can reflect the fluctuation of the throwing behavior in a small time range; on the other hand, the long-period classification sequence can reflect the trend change of the throwing behavior in a large time range. Such hierarchical sequence structure can improve the analysis ability of the throwing behavior characteristics, so that both the short-term fluctuations and the long-term changes of the throwing behavior can be effectively reflected.

[0024] By performing category difference processing on the short-period classification sequence and aggregate calculation on the long-period classification sequence, short-period fluctuation indicators and long-period stability indicators can be obtained respectively. The short-period fluctuation indicator reflects the change amplitude in a short time period, and can reveal whether the throwing behavior changes rapidly; the long-period stability indicator reflects the concentration degree of the category in different stages, and can represent the stable trend of the throwing behavior in a long time period. The two types of indicators complement each other, which is conducive to building a more complete behavior feature system.

[0025] By using cross-sequence correlation analysis of the short-period fluctuation indicator and the long-period stability indicator, the behavior change relationship at two time scales can be identified, and hierarchical trigger factors can be constructed. Among them, the short-period trigger factor is used to express the influence of short-term fluctuations, and the long-period trigger factor is used to express the deviation effect of long-term trends. The construction of hierarchical trigger factors enables the system to identify potential sudden changes or trend deviations in the throwing behavior from multiple levels.

[0026] By performing deviation difference judgment on the hierarchical trigger factors, when the deviation difference between the short-period trigger factor and the long-period trigger factor exceeds the set deviation threshold, a behavior mutation event can be generated. The behavior mutation event can reflect the significant deviation between the throwing behavior and the expected pattern, providing a trigger basis for subsequent strategy adjustment.

[0027] Performing strategy migration processing according to the behavior mutation event can update the overflow prediction, the collection priority and the throwing classification prompt in the garbage throwing management strategy, so that the management strategy can be dynamically adjusted according to the behavior change. Through this dynamic migration method, the management strategy can be consistent with the actual throwing behavior, and the accuracy of throwing guidance, garbage collection and collection scheduling can be improved.

[0028] For example, in the daily operation scenario of a certain residential area, the garbage disposal behavior data is accessed to form an original behavior sequence, and the short-period and long-period classification sequences are generated after time period division. If a sudden increase in the amount of a certain type of garbage disposal occurs continuously in the short period, but the type of garbage does not remain stable in the long-period classification sequence, the short-period fluctuation index and the long-period stability index will form a significant difference in the association analysis stage. This difference is amplified by the hierarchical trigger factor, which may exceed the offset threshold, thereby generating a behavior mutation event, indicating that the disposal mode of the type of garbage has changed. The subsequent management strategy after migration adjusts the collection and disposal priority of the type of garbage and updates the corresponding disposal prompts, so that the collection and disposal arrangement can better meet the current residents' disposal behavior.

[0029] In a preferred embodiment of the present application, garbage disposal behavior data uploaded by garbage disposal collection equipment is obtained, and an original behavior sequence representing continuous disposal behavior is generated based thereon, specifically including: In the acquisition step, the data uploaded by the collection equipment deployed at the garbage disposal point is received in chronological order, and the garbage disposal amount, garbage disposal time and garbage disposal type in each data record are structured to enable the data to be stored in a unified format; The structured data is sorted by disposal time to ensure that there is no time overlap or interruption between the data, thereby ensuring the continuity of the behavior data in the time dimension; According to the sorted data, the garbage disposal amount, garbage disposal type and disposal time at different time points are integrated to form a continuously arranged data set. This data set is used to represent the time evolution of garbage disposal behavior and serves as an original behavior sequence for subsequent period construction and behavior change analysis; When the time interval between two adjacent data exceeds the maximum allowed interval, blank data or zero value data can be supplemented in the middle to keep the behavior sequence logically continuous, thereby avoiding the problem of time discontinuity caused by sparse disposal.

[0030] In a preferred embodiment of the present application, the original behavior sequence is divided into time periods to generate short-period behavior data and long-period behavior data, and short-period classification sequences and long-period classification sequences are constructed, specifically including: By reading the disposal time corresponding to each record in the original behavior sequence, the complete time range is divided into multiple time periods, where the length of the short-period time period can be selected as the hour level or other relatively shorter time units; the long-period time period is determined according to the time span of day, week or month; The data in the original behavior sequence is aggregated according to the short-period time period and the long-period time period, respectively, so that the data in each short-period time period and long-period time period is integrated into corresponding short-period behavior data and long-period behavior data; According to the number and proportion of garbage disposal types in each period in the short-period behavior data, the short-period classification sequence is formed by classifying the short-period behavior data into preset classification categories, which is used to describe the behavior change in a short time range; According to the main disposal types and their change trends in each stage in the long-period behavior data, the long-period classification sequence is formed by aggregating the long-period behavior data into long-period classification results, which is used to describe the overall trend and structural change; Among them, the short-period classification sequence can more sensitively reflect the short-time disposal fluctuation, and the long-period classification sequence can reflect the long-term behavior rule, and the combination of the two can provide a multi-dimensional data basis for subsequent index generation.

[0031] In a preferred embodiment of the present application, the preset correlation offset threshold is used to determine whether the offset difference between the short-period trigger factor and the long-period trigger factor is sufficient to indicate behavior mutation, specifically including: According to the historical garbage disposal behavior data, the difference range of the short-period trigger factor and the long-period trigger factor in the normal state is counted, and the offset difference interval in the normal behavior mode is obtained; The above offset difference interval is sorted, and according to the frequency of the offset difference, the duration and the stable range in the normal behavior, the upper limit value of the offset difference that can remain stable in the normal behavior is determined as the basis of the offset threshold setting; By analyzing the historical abnormal cases or simulating abnormal scenarios, the difference change of the trigger factor caused by abnormal behavior is observed, and the offset difference characteristics appearing in the abnormal behavior are compared with the normal range to determine the minimum offset difference amplitude that can represent abnormal behavior; The difference interval between the normal offset difference upper limit and the abnormal offset difference lower limit is taken as the adjustable space, and the appropriate value selected according to the application scene is taken as the preset correlation offset threshold; Among them, the threshold is not only used to detect the amplitude of the offset, but also can be combined with the setting rule of the offset duration to trigger the behavior mutation event together, so that the judgment of behavior mutation is more consistent with the actual disposal change.

[0032] In a preferred embodiment of the present application, according to the short-period classification sequence and the long-period classification sequence, the category difference processing of adjacent time periods and the category aggregation calculation of each long-period stage are performed respectively to generate short-period fluctuation indicators and long-period stability indicators, including: The short-period classification sequence is segmented according to a fixed-length short-period analysis window, and the category change of adjacent time periods in each analysis window is compared in direction, judged in continuity and accumulated in amplitude to generate the short-period difference result in the short-period window; According to the consistency detection result of the category change mode of the continuous multiple analysis windows, a short-period fluctuation index is obtained; The long-period classification sequence is grouped according to long-period stages, and in each long-period group, a long-period aggregation result is generated according to the category appearance frequency, duration and stability degree of the dominant category in the stage, and a long-period stability index is calculated according to the long-period aggregation result.

[0033] In the embodiment of the present application, the short-period classification sequence is segmented according to fixed-length analysis windows, so that the short-period behavior can be analyzed in a continuous and structured manner. By comparing the category change direction of adjacent time periods, judging the continuity and accumulating the amplitude, the change characteristics in a short time can be integrated into a short-period difference result, avoiding the fluctuation interference caused by single-point data, and helping to form a more reliable short-period feature expression.

[0034] The consistency detection of the continuous multiple analysis windows can identify whether there is a sustained change trend in the short-period change, and is more stable than the difference result of a single window. The short-period fluctuation index obtained thereby can reflect the overall degree of behavior change in the short period, thereby forming a more accurate description of the rapid fluctuation behavior.

[0035] The long-period classification sequence is grouped according to long-period stages, so that the long-period classification result can reflect a large time span. By aggregating the category appearance frequency, duration and stability degree of the dominant category, a concentrated expression of the long-term behavior trend can be formed. The long-period stability index obtained thereby can reflect the dominant law of the delivery behavior in different long-period stages, and provide a more explicit judgment basis for subsequent identification of trend changes.

[0036] This processing method can obtain feature indexes representing behavior fluctuation and behavior stability in the short-period and long-period time dimensions respectively, so that the revelation of the behavior change relationship in subsequent correlation analysis is more sufficient.

[0037] In a preferred embodiment of the present application, the short-period classification sequence is segmented according to fixed-length short-period analysis windows, and in each analysis window, the category change of adjacent time periods is compared in direction, judged for continuity and accumulated in amplitude to generate a short-period difference result in the short-period window, which specifically includes: The short-period classification sequence is segmented according to the set short-period analysis window length, so that each analysis window corresponds to a fixed time range; In each analysis window, the classification results of adjacent time periods are read in time sequence, and the category change direction between them is compared to determine whether the category change is increased, decreased or unchanged; According to the continuity of the category change between adjacent time periods, it is determined whether the behavior in a short time range presents a consistent change trend, such as continuous increase, continuous decrease, or unstable trend; According to the amplitude of the category change, the change amount between multiple adjacent time periods is accumulated, so that the total change amplitude in the analysis window can reflect the overall fluctuation; According to the combined features of directionality, continuity, and cumulative amplitude, the overall change in the analysis window is converted into a short-period difference result, which is used to represent the comprehensive characteristics of the behavior change in the short-period window.

[0038] In a preferred embodiment of the present application, the category change patterns of consecutive multiple analysis windows are detected for consistency according to the short-period difference result, and the consistency detection result is converted into a short-period fluctuation indicator, specifically including: The short-period difference results generated in multiple adjacent short-period analysis windows are arranged in time sequence, and the change trend is checked for continuity; In the consecutive multiple analysis windows, it is determined whether the difference result presents a consistent change direction, such as whether it forms a change pattern of continuous increase, continuous decrease, or continuous instability; According to the consistency determination result, the behavior pattern features between consecutive windows are extracted, including whether there is short-term concentrated change, short-term periodic fluctuation, or short-term abnormal behavior; The above behavior pattern features are inductively processed according to the preset classification rules, so that the behavior pattern is expressed as quantifiable fluctuation features; According to the strength, frequency of occurrence, and duration of the fluctuation features, the consistency detection result is converted into a short-period fluctuation indicator, which is used to represent the overall fluctuation level of the behavior change in a short time range.

[0039] In a preferred embodiment of the present application, the long-period classification sequence is grouped according to long-period stages, the long-period aggregation result is generated in each long-period group according to the category occurrence frequency, duration, and stability degree of the dominant category in the stage, and the long-period stability indicator is calculated according to the long-period aggregation result, specifically including: The long-period classification sequence is divided according to the preset long-period stages, so that each stage can cover a long enough time range to reflect the long-term change law; In each long-period stage, the occurrence of different categories is counted, and the occurrence frequency and continuous occurrence time span of each category in the stage are recorded; On the basis of the statistical frequency and duration, the dominant category is determined according to the dominant degree of different categories in the stage, and the stability of the dominant category is further determined, i.e., whether the dominant category continuously maintains the main position in the stage; According to the comprehensive performance of frequency, duration and dominant category stability, the classification characteristics of this stage are generated into long-period aggregation results to represent the main behavior characteristics of the long-period stage; According to the aggregation results, the stability of each long-period stage is analyzed to form a long-period stability index, which can reflect whether the long-term behavior remains stable or changes.

[0040] In a preferred embodiment of the present application, cross-sequence correlation analysis is performed according to the short-period fluctuation index and the long-period stability index to generate correlation analysis results, and a hierarchical trigger factor including a short-period trigger factor and a long-period trigger factor is constructed according to the correlation analysis results, including: The short-period fluctuation index and the long-period stability index are time-aligned, the trend difference value of the two indexes is calculated in the aligned time period, and a primary correlation result is generated according to the trend difference value; According to the trend difference value in the primary correlation result, the directionality, accumulativity and persistence of the difference are analyzed in segments to generate a short-period trigger factor basic value for describing short-period burst behavior and a long-period trigger factor basic value for describing long-term deviation behavior; According to the structural difference between the short-period trigger factor basic value and the long-period trigger factor basic value, the two are hierarchically reorganized, the short-period trigger factor is placed in the fast reaction layer, and the long-period trigger factor is placed in the trend recognition layer to form a hierarchical trigger factor including two layers.

[0041] In the embodiment of the present application, the short-period fluctuation index and the long-period stability index are time-aligned, so that the two types of indexes can be compared in the common time period, avoiding analysis deviation caused by time scale difference. The indexes after time alignment can accurately reflect the relationship between short-term changes and long-term trends in the same time range.

[0042] By calculating the trend difference value in the aligned time period, the difference degree between the short-period fluctuation and the long-period stability can be obtained, including the difference in change direction, the difference in change amplitude and the difference in change speed. This difference analysis can reveal whether the delivery behavior deviates in short-term fluctuation and long-term trend, laying a foundation for further constructing trigger factors.

[0043] The trend difference value is analyzed in segments, so that the directionality, accumulativity and persistence of the difference can be distinguished and expressed respectively. Directionality analysis can identify the change direction of the difference, accumulativity analysis can identify whether the difference is cumulative, and persistence analysis can identify whether the difference continues for a certain length of time. Through further analysis of these difference characteristics, short-period trigger factor basic values and long-period trigger factor basic values can be generated, so that the characteristics of behavior changes in different time dimensions are independently expressed.

[0044] By hierarchical reorganization of the short-period trigger factor base value and the long-period trigger factor base value, a hierarchical trigger factor structure with different reaction speeds and different behavioral characteristics can be formed, thereby providing a more comprehensive basis for subsequent identification of behavioral mutations.

[0045] In a preferred embodiment of the present application, according to the structural differences between the short-period trigger factor base value and the long-period trigger factor base value, the two are hierarchically reorganized, the short-period trigger factor is placed in the fast reaction layer, and the long-period trigger factor is placed in the trend identification layer, forming a hierarchical trigger factor containing two layers, specifically including: The short-period trigger factor base value and the long-period trigger factor base value are classified according to their characteristic sources, wherein the short-period trigger factor base value comes from short-period burst behavior analysis, and the long-period trigger factor base value comes from long-term deviation behavior analysis; According to the characteristics of the short-period trigger factor base value, such as strong burst of behavior change and fast response speed, it is classified into the fast reaction layer, so that the layer can timely reflect the behavior change occurring within a short time range; According to the characteristics of the long-period trigger factor base value, such as strong cumulative change and long-term deviation trend of behavior, it is classified into the trend identification layer, so that the layer is used to represent the deviation degree of behavior change within a longer period; According to the different effects of the two levels, the structural relationship between the two types of trigger factors is combined, so that the fast reaction layer and the trend identification layer form a hierarchical arrangement relationship from top to bottom, thereby generating a hierarchical trigger factor with a double-layer structure; The hierarchical trigger factor can simultaneously reflect the burst change of short-period behavior and the deviation characteristics of long-period behavior, and provide a more comprehensive information basis for subsequent deviation difference determination.

[0046] In a preferred embodiment of the present application, a second duration threshold is preset for determining whether the deviation difference maintains a stable change trend within a plurality of consecutive time periods, specifically including: According to the duration characteristics of short-period changes and long-period changes in historical garbage disposal behavior data, the duration distribution range of the deviation difference under normal behavior is counted, and the common duration interval under normal state is identified; After counting the historical data, it is determined that the deviation difference appearing within a shorter duration in normal behavior is mostly sporadic change, and the deviation difference appearing within a longer duration is more likely to reflect the trend change of behavior; According to the above analysis, the maximum duration in normal behavior is compared with the minimum duration in abnormal behavior, forming a duration demarcation range that can distinguish between normal temporary fluctuations and long-term trend deviations; In combination with the management scenarios, such as the garbage disposal frequency in residential areas, the distribution of daily disposal peak, and other factors, a time value that can reliably reflect the behavior change persistence within the above range is selected as a preset second persistence threshold; When the offset difference accumulation sequence remains more than the preset second persistence threshold in multiple consecutive time periods, it is considered that the behavior change has a persistence characteristic, and thus it is determined that the behavior mutation event is established.

[0047] In a preferred embodiment of the present application, the offset difference is determined according to the hierarchical trigger factors, and when the offset difference between the short-period trigger factor and the long-period trigger factor exceeds the preset associated offset threshold, a behavior mutation event is generated, including: The short-period trigger factor and the long-period trigger factor of the hierarchical trigger factors are respectively subjected to offset source filtering, so as to eliminate the transient change caused by single abnormality in the short-period trigger factor and exclude the unstable offset caused by short-term fluctuation in the long-period trigger factor, and filtered short-period trigger factors and long-period trigger factors are generated; The continuous change amount of the offset difference between the filtered short-period trigger factors and the long-period trigger factors is calculated, and the continuous change amount is accumulated in multiple consecutive time periods to form an offset difference accumulation sequence; According to the accumulation amount and the duration in the offset difference accumulation sequence, a double-condition judgment is performed, and when the accumulation amount exceeds the preset associated offset threshold and the duration exceeds the preset first persistence threshold, a behavior mutation event is generated, and the behavior mutation event includes a mutation category indicating the mutation type, a mutation amplitude indicating the mutation intensity, and a mutation time period indicating the mutation occurrence range.

[0048] In the embodiment of the present application, the short-period trigger factor and the long-period trigger factor are respectively subjected to offset source filtering, so as to eliminate the transient change caused by single abnormality in the short-period trigger factor and eliminate the unstable offset caused by short-term fluctuation in the long-period trigger factor, so that the filtered trigger factors can better represent the real change trend of the behavior and reduce the risk of misjudgment.

[0049] By calculating the continuous change amount of the offset difference between the filtered short-period trigger factors and the long-period trigger factors, and forming an offset difference accumulation sequence, the change of the offset difference in multiple consecutive time periods can be reflected, and the persistence and accumulation of the behavior change can be reflected at the same time. Compared with a single point offset value, the accumulation sequence can provide more comprehensive behavior offset information.

[0050] According to the double-condition judgment of the accumulation amount and the duration of the offset difference accumulation sequence, whether the behavior truly mutates can be effectively identified. Only when the accumulation degree of the offset difference reaches the set threshold and the duration exceeds the set range, a behavior mutation event is generated, so that the judgment condition of the behavior mutation is more reliable.

[0051] The generated behavior mutation event includes a mutation category, a mutation amplitude and a mutation time period, can clearly describe the nature, intensity and occurrence range of the behavior change, and provides sufficient information basis for subsequent strategy migration processing.

[0052] In a preferred embodiment of the present application, the short-period trigger factor and the long-period trigger factor of the layered trigger factor are respectively subjected to offset source filtering, the instantaneous change caused by single abnormality in the short-period trigger factor is removed, and the unstable offset caused by short-term fluctuation in the long-period trigger factor is excluded, to generate filtered short-period trigger factor and long-period trigger factor, specifically including: The change of the short-period trigger factor in the continuous multiple time periods is obtained, and the sudden increase or sudden decrease of the trigger factor in a single time period is identified; According to the change persistence of the short-period trigger factor in adjacent time periods, it is judged whether the sudden increase or sudden decrease occurs only in a single time period. When the change only appears in an isolated time period and does not continue in the time periods before and after it, the change is identified as an instantaneous change; The above instantaneous change is removed from the short-period trigger factor, so that the short-period trigger factor can reflect the real and continuous behavior fluctuation in a short time range; The change trend of the long-period trigger factor in a long time range is obtained, and the small offset caused by short-term disturbance is identified; According to the phase change characteristics of the long-period trigger factor, it is judged whether the offset only appears in a local time period and does not form a trend change, and when the offset is only a short-term disturbance, it is excluded from the long-period trigger factor; Through the above processing, the filtered short-period trigger factor and long-period trigger factor are generated, so that both of them maintain their own corresponding behavior characteristics, and the subsequent offset difference calculation is not affected by noise.

[0053] In a preferred embodiment of the present application, the offset difference continuous change quantity between the filtered short-period trigger factor and the long-period trigger factor is calculated, the continuous change quantity is accumulated in multiple continuous time periods, and an offset difference accumulation sequence is formed, specifically including: The values of the filtered short-period trigger factor and the long-period trigger factor in each time period are obtained, and the difference between the two is calculated in the same time period to obtain the offset difference of the time period; According to the change of the offset difference in adjacent time periods, it is judged whether the offset difference presents a trend of continuous increase, continuous decrease or stable change; The change of the offset difference in multiple continuous time periods is accumulated, so that the accumulation result can reflect the overall change degree of the offset difference in time, and the deviation caused by a single time period to the behavior judgment is avoided; When the time period of the consistent and continuous change of the offset difference is long, the continuous change amount is accumulated to form an offset difference accumulation sequence, so that the sequence can reflect the accumulation trend of the offset difference in a long time range; The offset difference accumulation sequence is used for subsequent continuous judgment, so that the identification of the behavior mutation event can consider the characteristics of the offset amplitude and the offset duration.

[0054] In a preferred embodiment of the present application, a double condition judgment is performed according to the accumulation amount and the duration in the offset difference accumulation sequence, and when the accumulation amount exceeds a preset correlation offset threshold and the duration exceeds a preset first duration threshold, a behavior mutation event is generated, which includes a mutation category for indicating a mutation type, a mutation amplitude for indicating a mutation intensity, and a mutation time period for indicating a mutation occurrence range, and specifically includes: The offset difference accumulation sequence is analyzed to identify a time period in which the accumulation amount reaches a peak value, and the size of the accumulation amount is recorded; According to the accumulation trend of the offset difference accumulation sequence, it is judged whether the accumulation amount exceeds a preset correlation offset threshold, so that the judgment result can reflect whether the offset difference reaches a change degree that needs attention; The time period of the continuous offset in the offset difference accumulation sequence is counted to identify whether the duration exceeds a preset first duration threshold; When the accumulation amount exceeds the preset correlation offset threshold and the duration exceeds the preset first duration threshold, the case that both conditions are met is identified as a behavior mutation event; According to the direction and category of the offset change in the offset difference accumulation sequence, the mutation category involved in the behavior mutation event is marked; According to the size of the accumulation amount in the offset difference accumulation sequence, the mutation amplitude is converted for indicating the intensity of the mutation event; According to the start time and the end time of the continuous offset, the mutation time period is determined, so that the occurrence range of the mutation event can be clearly expressed; The finally generated behavior mutation event can be used as the basis for strategy migration processing, so that the subsequent garbage disposal management strategy adjustment has a clear data source.

[0055] In a preferred embodiment of the present application, strategy migration processing is performed according to the behavior mutation event to generate a migrated garbage disposal management strategy, specifically including an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage disposal classification correction prompt, including: According to the mutation amplitude and the mutation category corresponding to the behavior mutation event, the garbage can overflow prediction window is dynamically expanded or contracted, so that the time range of the prediction window changes proportionally with the mutation amplitude, and an adjusted prediction window is generated; According to the garbage category involved in the behavior mutation event and its occurrence frequency, the cleaning object is prioritized, a cleaning priority promotion factor is established, the high mutation category obtains a higher cleaning priority, and an adjusted cleaning sequence is generated; According to the garbage classification deviation corresponding to the behavior mutation event, the deviation category is taken as a key prompt category, the push frequency of the category prompt is improved, the prompt content is adaptively updated, and an updated garbage disposal classification correction prompt is generated.

[0056] In the embodiment of the application, the garbage can overflow prediction window is dynamically expanded or contracted according to the behavior mutation event, so that the prediction window can be adjusted according to the behavior change. When the mutation event reflects that a certain type of garbage has a significant increase in a certain time period, expanding the prediction window can reflect the possible overflow risk in advance. When the behavior trend changes in the direction of decrease, the prediction window is contracted to reduce unnecessary prediction redundancy. Thus, the prediction result can be closer to the actual disposal rhythm, and the timeliness of prediction adjustment can be improved.

[0057] The cleaning object is prioritized according to the garbage category involved in the behavior mutation event and its occurrence frequency, so that the cleaning sequence can be automatically rearranged according to the behavior change. When the disposal behavior of a certain type of garbage is high frequency or has a significant increase in the mutation event, increasing the cleaning priority of this type of garbage can help reduce the accumulation risk and make the cleaning arrangement more consistent with the current disposal state. The priority relationship between categories can be further clarified by a priority promotion factor, so that the cleaning task arrangement is more clear.

[0058] The prompt content is updated according to the garbage classification deviation in the behavior mutation event, so that the prompt information can be adjusted according to the actual change of the disposal behavior. When a certain type of garbage shows a classification deviation trend in the mutation event, increasing the push frequency of the category prompt and updating the prompt content can help guide the disposal behavior to return to the correct classification method and reduce the continuous occurrence of disposal errors. Overall, this method can make the disposal guidance, cleaning scheduling and overflow prediction linked, and improve the coordination of the garbage management process.

[0059] In a preferred embodiment of the application, the garbage can overflow prediction window is dynamically expanded or contracted according to the mutation amplitude and mutation category corresponding to the behavior mutation event, so that the time range of the prediction window changes proportionally with the mutation amplitude, and an adjusted prediction window is generated, specifically including: Identify the mutation category contained in the behavior mutation event, and determine whether the mutation category belongs to a garbage type with a higher risk of garbage can overflow, such as wet garbage or garbage with large volume and easy to occupy capacity; According to the mutation amplitude recorded in the behavior mutation event, the degree of change in the disposal amount is determined. When the mutation amplitude is large, it indicates that a higher disposal load may be generated in a short time. The mutation amplitude is matched with a preset window adjustment rule, for example, when the mutation amplitude is high, the prediction window is appropriately expanded forward, so that the system can recognize the overflow trend earlier; when the mutation amplitude is low, the prediction window is appropriately contracted according to the characteristics of the mutation category, to avoid that the prediction range is too large and the prediction result is deviated; In the prediction window adjustment process, the adjustment ratio is determined according to the delivery behavior characteristics of the mutation category, for example, for a garbage category with high delivery concentration, the window expansion ratio can be appropriately increased; Based on the above adjustment result, a new prediction window time range is formed, which can more accurately reflect the influence of the sudden behavior on the overflow trend; Finally, the adjusted prediction window is output as the basis for subsequent overflow prediction.

[0060] In a preferred embodiment of the present application, according to the garbage category involved in the behavior mutation event and its occurrence frequency, the cleaning object is prioritized, a cleaning priority promotion factor is established, the high mutation category obtains a higher cleaning priority, and an adjusted cleaning sequence is generated, which specifically includes: The garbage category involved in the behavior mutation event is analyzed, and it is identified that the mutation behavior of which garbage type is more obvious, and the frequency of its occurrence in a preset time period is recorded; According to the combination characteristics of the occurrence frequency and the mutation amplitude, the emergency degree of the mutation behavior is determined, for example, the category with high mutation amplitude and high frequency has more obvious influence on the cleaning demand; According to the importance of the mutation behavior, a cleaning priority promotion factor is constructed, so that the mutation category can be promoted in the priority sorting; the promotion factor can be set according to the capacity occupation characteristics, use scenario and management rules of the garbage category; The cleaning objects are sorted according to the priority adjusted by the promotion factor, so that the garbage bin of the mutation category can be preferentially arranged for cleaning; According to the sorting result, a final cleaning sequence is generated, and the sequence is used for generating subsequent scheduling instructions to improve the rationality of the cleaning arrangement.

[0061] In a preferred embodiment of the present application, according to the garbage classification deviation corresponding to the behavior mutation event, the deviation category is taken as a key prompt category, the push frequency of the prompt of the category is increased, and the prompt content is adaptively updated, an updated garbage delivery classification correction prompt is generated, which specifically includes: The mutation category recorded in the behavior mutation event is analyzed to determine whether the category is related to the garbage classification deviation, for example, a large amount of a certain garbage is incorrectly put in; The mutation category is taken as a key prompt category, the push frequency of the prompt of the category is increased according to the frequency and severity of the deviation, so that the user can receive the correction information earlier; According to the deviation type recorded in the behavior mutation event, the prompt content is adaptively updated, for example, the prompt text is modified to emphasize the correct classification method, or the targeted suggestion is added, so that the prompt content can better fit the actual classification deviation; In the prompt content updating process, the prompt logic can be optimized according to the historical deviation data, so that the updated prompt content can more effectively guide the user to correct the classification behavior; Generate the final garbage disposal classification correction prompt and continuously push it in subsequent disposal behaviors to prompt the user to adjust the disposal habits in time and reduce the frequency of classification errors.

[0062] In a preferred embodiment of the present application, the short-period fluctuation index and the long-period stability index are time-aligned, the trend difference value of the two is calculated in the common time period after alignment, and the preliminary correlation result is generated according to the trend difference value, including: According to the short-period fluctuation index and the long-period stability index, time reference points are determined, and the time reference points are expanded to form reference time series for describing the time distribution of the two types of indexes; According to the reference time series, the short-period fluctuation index and the long-period stability index are time-mapped so that the two types of indexes correspond on the same time axis to form an aligned index sequence; According to the aligned index sequence, the change direction, change amplitude and change speed of the short-period fluctuation index and the long-period stability index in the common time period are obtained, and interval difference extraction is performed to generate a trend difference value; According to the trend difference value, the deviation direction, deviation intensity and deviation duration in the interval are associated to obtain a preliminary correlation result for representing the cross-period behavior change relationship.

[0063] In the embodiments of the present application, by determining the time reference points of the short-period fluctuation index and the long-period stability index and performing expansion processing, a unified time distribution structure covering the two types of indexes can be formed, solving the problem that data of different time scales cannot be directly compared. The formation of the reference time series enables the indexes to be analyzed in consistent time sequence in subsequent steps.

[0064] By time-mapping based on the reference time series, the short-period fluctuation index and the long-period stability index can correspond on the same time axis, avoiding alignment deviation caused by time scale difference, so that the aligned indexes more accurately reflect the actual behavior change relationship. The aligned index sequence obtained by mapping can provide stable data correspondence for subsequent difference extraction.

[0065] By extracting the difference characteristics of change direction, change amplitude and change speed in the same time period, the difference structure of the two types of indexes in the behavior change can be comprehensively reflected, so that the trend difference value can reflect the deviation between the short-term change and the long-term trend at the same time.

[0066] By associating the direction, intensity and persistence of the trend difference value, a primary association result can be formed to distinguish whether the behavior difference has an internal association. The primary association result generated thereby can provide a more definite data direction in the trigger factor construction stage, making the generation of the trigger factor more reliable.

[0067] In a preferred embodiment of the present application, the time reference points are determined according to the short-period fluctuation index and the long-period stability index, and the time reference points are expanded to form reference time series for describing the time distribution of the two types of indexes, specifically including: The start time and end time of each short period are extracted from the time record corresponding to the short-period fluctuation index, which are used as the initial time reference points of the short period; The start time and end time of each long period are extracted from the time record corresponding to the long-period stability index, which are used as the initial time reference points of the long period; Since the time span of the short period and the long period is different, in order to enable the two types of indexes to be aligned and analyzed under the same time structure, the initial reference points are expanded by a preset time range to form reference time points containing complete time intervals; The expanded multiple time reference points are arranged in time sequence so that there is no overlap or gap between the reference time points, so as to constitute a reference time series with complete time coverage; The reference time series is used in the subsequent time mapping step to enable the different period indexes to be compared and analyzed on a unified time axis.

[0068] In a preferred embodiment of the present application, the short-period fluctuation index and the long-period stability index are time-mapped according to the reference time series, so that the two types of indexes correspond on the same time axis to form an aligned index sequence, specifically including: Each time point in the reference time series is read as the target alignment time of the short-period fluctuation index and the long-period stability index; In the time mapping process, the short-period index value corresponding to the time point is determined by searching for the short period in the short-period fluctuation index that is adjacent to or contains the time point; Similarly, by searching for a long period adjacent to or containing the reference time point in the long period stability index, the long period index value corresponding to the time point is determined; The short period index value and the long period index value are arranged in the time sequence of the reference time sequence, so that the two types of indexes corresponding to the time can form a one-to-one correspondence; Finally, the alignment index sequence is formed, so that the short period fluctuation index and the long period stability index can compare the change trend in the same time period, laying a foundation for subsequent trend difference extraction.

[0069] In a preferred embodiment of the application, according to the alignment index sequence, the change direction, change amplitude and change speed of the short period fluctuation index and the long period stability index in the common time period are obtained, and interval difference extraction is performed to generate a trend difference value, specifically including: The short period fluctuation index values corresponding to the continuous time points in the alignment index sequence are read, and the size relationship between adjacent time points is compared to determine the change direction of the short period fluctuation index in the interval; According to the difference value of the short period fluctuation index, the amplitude of the change is judged, and the interval with too large change amplitude is marked as a rapid fluctuation interval; The long period stability index is analyzed by the same method as the short period, and the trend direction, amplitude and change speed are determined from the change amount of adjacent time points; The direction, amplitude and speed obtained by the short period index and the long period index in the common time period are analyzed, including judging whether the trend directions are consistent, whether the change amplitudes deviate from each other, and whether the change speeds have a gap; The above difference content is converted into a trend difference value by a textual description method, so that the trend difference value can represent whether the behaviors of the two time scales deviate in the time interval and the deviation degree; The trend difference value is used as basic data for subsequent association judgment, and is used to identify the relationship between the short period change and the long period trend.

[0070] In a preferred embodiment of the application, according to the trend difference value, the deviation direction, deviation intensity and deviation duration in the interval are associated to obtain a primary association result for representing the cross-period behavior change relationship, specifically including: The directionality feature recorded in the trend difference value is analyzed, and by comparing whether the short period change direction and the long period change direction are consistent, the deviation direction is identified, and if the deviation direction is opposite, it indicates that the behavior change has a reverse feature; According to the deviation amplitude information in the trend difference value, the deviation intensity is identified, and the deviation intensity is divided into mild, moderate or severe, so that the intensity level can more intuitively represent the difference between the two types of behaviors; According to the length of the trend difference value in the continuous multiple alignment time periods, it is judged whether the deviation duration is long enough to reflect whether the cross-period deviation is persistent; The three characteristics of deviation direction, deviation intensity and deviation duration are comprehensively combined to make the correlation determination of the trend difference, and a primary correlation result capable of describing the relationship between short-period behavior change and long-period trend is formed; The primary correlation result is used in the subsequent trigger factor basis value generation step, so that the behavior characteristic expression can be further hierarchically processed.

[0071] In a preferred embodiment of the present application, according to the trend difference value in the primary correlation result, the directionality, accumulation and persistence of the difference in the alignment time period are segmented and analyzed, and short-period trigger factor basis values for describing short-period burst behavior and long-period trigger factor basis values for describing long-term deviation behavior are generated, including: According to the change direction of the trend difference value, the trend difference value is directionally divided to form a directionality difference segment for indicating the trend change direction; According to the directionality difference segment, the trend difference value inside the difference segment is accumulated, and an accumulation difference segment for indicating the difference accumulation intensity is formed according to the accumulation change characteristics; According to the duration of the accumulation difference segment in the alignment time period, the persistence of the difference segment is analyzed, the difference segment whose duration exceeds a preset second duration threshold is identified, and a short-period trigger factor basis value is generated according to the difference segment; The difference segment that does not meet the duration threshold is analyzed for long-term deviation, the trend is aggregated according to the deviation direction, deviation amplitude and deviation phase characteristics, and a long-period trigger factor basis value is generated.

[0072] In the embodiment of the present application, by directionally dividing according to the change direction of the trend difference value, the difference value can be distinguished according to different change trends, so that the directionality of short-period change and long-term change is clearly expressed. The formation of the directionality difference segment can help the subsequent steps to more accurately identify the category characteristics of the difference.

[0073] By accumulating the directionality difference segment, the difference degree in the continuous time period can be integrated into an accumulation difference segment, so that the accumulation change intensity of the difference is reflected. The accumulation difference segment can be used to identify whether the change shows a persistent accumulation trend, thereby providing a basis for subsequent differentiation of short-term burst change and long-term deviation change.

[0074] By analyzing the duration of the accumulation difference segment in the alignment time period, the part whose duration exceeds the set threshold can be identified, thereby generating a short-period trigger factor basis value. The basis value can reflect whether the short-period behavior change has a burst, and provides a data basis for subsequent burst behavior identification.

[0075] The trend aggregation processing is performed on the difference segments that do not reach the persistence threshold, and the offset direction, offset amplitude and phase change characteristics of the difference segments are integrated to form a long-period trigger factor basic value. The basic value can reflect the offset of behavior change in the long-term dimension, and provide support for subsequent identification of trend anomalies.

[0076] In a preferred embodiment of the present application, the trend difference value is directionally divided according to the change direction of the trend difference value to form a directional difference segment for representing the trend change direction, specifically including: The records of the trend difference value in consecutive alignment time periods are read, and the size change of the trend difference value between adjacent time periods is compared to determine whether the trend difference value is rising, falling or remaining unchanged; According to the change direction obtained by the judgment, the trend difference value is divided into a plurality of directional difference segments in time sequence, for example, the trend difference value continuously in the rising direction is classified into one directional difference segment, and the trend difference value continuously in the falling direction is classified into another directional difference segment; In the division process, when the trend difference value changes from rising to falling or from falling to rising, it is regarded as a direction change point, and the difference segment is divided at the point to ensure that the direction inside each directional difference segment is consistent; The directional difference segment after division is recorded, so that each difference segment can independently represent the trend difference direction of a certain stage, and provide a structured basis for subsequent cumulative analysis.

[0077] In a preferred embodiment of the present application, according to the directional difference segment, the trend difference value inside the difference segment is cumulatively processed, and a cumulative difference segment for representing the difference accumulation intensity is formed according to the cumulative change characteristics, specifically including: In each directional difference segment, the trend difference values in the segment are read one by one, and the difference values are accumulated according to the increase and decrease amplitude, so that the accumulation result can reflect the overall difference intensity in the directional difference segment; When the cumulative processing is performed, the continuous change of the trend difference value is judged to identify the growth trend or reduction trend of the difference value in the segment, and the changes in the same direction are added up, so that the cumulative amount can express the total influence of the difference change in the segment; When the cumulative result of the trend difference value in a directional difference segment reaches a significant level, the segment is marked as a cumulative difference segment to represent that the difference change intensity in the segment is high; The cumulative characteristics obtained by the cumulative processing of each directional difference segment are recorded and arranged into cumulative difference segments, which are used for subsequent persistence analysis.

[0078] In a preferred embodiment of the present application, the persistence analysis is performed on the duration of the cumulative difference segment within the alignment time period, and the difference segment with a duration exceeding a preset second duration threshold is identified, and a short-period trigger factor base value is generated based on the difference segment, specifically including: determining the start time and end time corresponding to each cumulative difference segment, and calculating the duration of the difference segment; comparing the duration with a preset duration threshold, and determining that the difference segment has sufficient persistence to reflect short-period burst behavior when the duration of the cumulative difference segment exceeds the preset second duration threshold; for the difference segment satisfying the duration threshold condition, generating a short-period trigger factor base value based on the direction, cumulative amount and change characteristics of the trend difference value in the segment, so that the base value can reflect the characteristics of short-term burst behavior; recording the generated short-period trigger factor base value and using it for subsequent hierarchical trigger factor construction.

[0079] In a preferred embodiment of the present application, long-term offset analysis is performed on the difference segment that does not satisfy the duration threshold, and a long-period trigger factor base value is generated based on the offset direction, offset amplitude and offset phase characteristics, specifically including: identifying the difference segment that does not reach the duration threshold as an offset characteristic that may affect long-term behavior, rather than short-term burst behavior; reading the trend difference value in the difference segment, and identifying the direction characteristics of long-term offset based on the change direction of the difference value, such as a persistent overall upward or overall downward trend; judging the strength of long-term offset based on the change amplitude of the trend difference value in the segment, so that the offset strength can reflect the scale of long-term behavior change; analyzing the change phase of the difference segment in different periods, such as initial offset, stable offset and late offset, and identifying the structural characteristics of long-term offset through phase change; comprehensively processing the offset direction, offset amplitude and offset phase characteristics to form a trend aggregation result, and generating a long-period trigger factor base value based thereon to represent the long-term offset behavior characteristics; using the long-period trigger factor base value for subsequent hierarchical trigger factor construction to represent the long-term behavior change trend.

[0080] In a preferred embodiment of the present application, the preset second duration threshold is used to determine whether the change of the trend difference value has sufficient persistence within the short period, so as to ensure that the short-period trigger factor base value reflects the true burst behavior, specifically including: According to the historical changes of the trend difference values in multiple short-period analysis windows, the duration of the difference values under normal behavior is counted, and the time difference between the possible temporary fluctuation and the persistent fluctuation in the short period is identified; According to the statistical results, it is determined that the burst behavior in the short period usually has obvious continuity, and the accidental difference usually occurs in a single or a small number of windows, and by distinguishing the continuity and non-continuity changes, the accuracy of the burst behavior judgment can be effectively improved; The time period in which the trend difference continuously appears is subjected to cluster analysis, the distribution of different duration types is determined, such as short duration, medium duration and long duration, and the shortest duration that the burst behavior may have is judged; According to the actual rules of garbage throwing behavior, such as the length of the concentrated segment of resident throwing time, the typical duration of burst-type throwing behavior and other factors, one of the duration types is selected as a preset second duration threshold value which can accurately reflect the shortest duration of the burst behavior; When the trend difference continuously exceeds the preset second duration threshold value in the aligned time period, it is considered that the difference has the persistence of the burst behavior, so as to generate a short-period trigger factor basic value.

[0081] Embodiments of the present application also provide a garbage throwing optimization management system based on classified data mining, which comprises: A data acquisition module is configured to acquire garbage throwing behavior data uploaded by a garbage throwing collection device, and generate original behavior sequences representing continuous throwing behavior according to the garbage throwing behavior data, wherein the garbage throwing behavior data comprises garbage throwing amount, garbage throwing time and garbage throwing type; A sequence construction module is configured to divide time periods according to the original behavior sequences, generate short-period behavior data and long-period behavior data, and construct short-period classification sequences and long-period classification sequences respectively; An index generation module is configured to perform category difference processing of adjacent time periods and category aggregation calculation of each long-period stage according to the short-period classification sequences and the long-period classification sequences, and generate short-period fluctuation indexes and long-period stability indexes; An association analysis module is configured to perform cross-sequence association analysis according to the short-period fluctuation indexes and the long-period stability indexes, generate association analysis results, and construct hierarchical trigger factors including short-period trigger factors and long-period trigger factors according to the association analysis results; A mutation identification module is configured to perform offset difference determination according to the hierarchical trigger factors, and generate a behavior mutation event when the offset difference between the short-period trigger factors and the long-period trigger factors exceeds a preset association offset threshold value; A policy migration module is configured to perform a policy migration process according to the behavior mutation event, and generate a migrated garbage throwing management policy, which specifically includes an updated garbage can overflow prediction window, an adjusted garbage collection priority, and a reset garbage throwing classification correction prompt. A throwing management module is configured to perform a garbage throwing optimization management operation according to the migrated garbage throwing management policy, and output an execution result to a garbage throwing management terminal.

[0082] It should be noted that the system is a system corresponding to the above method, and all the implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0083] Embodiments of the present application also provide a computing device, comprising a processor and a memory storing a computer program, wherein the computer program is executed by the processor to perform the method described above. All the implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0084] Embodiments of the present application also provide a computer readable storage medium storing instructions, which, when executed on a computer, cause the computer to perform the method described above. All the implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0085] The above is the preferred embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered within the scope of protection of the present application.

Claims

1. A waste disposal optimization management method based on classification data mining, characterized in that, The method includes: Acquire waste disposal behavior data uploaded by waste disposal collection devices, and generate an original behavior sequence to represent continuous disposal behavior based on the data; The original behavior sequence is divided into time periods to generate short-cycle behavior data and long-cycle behavior data, and short-cycle classification sequences and long-cycle classification sequences are constructed respectively. Based on the short-cycle classification sequence and the long-cycle classification sequence, the category difference processing of adjacent time periods and the category aggregation calculation of each long-cycle stage are performed to generate short-cycle fluctuation index and long-cycle stability index. Cross-sequence correlation analysis is performed based on short-cycle volatility indicators and long-cycle stability indicators to generate correlation analysis results, and a hierarchical triggering factor including short-cycle triggering factors and long-cycle triggering factors is constructed based on these results. The offset difference is determined based on the hierarchical triggering factor. When the offset difference between the short-cycle triggering factor and the long-cycle triggering factor exceeds the preset associated offset threshold, a behavioral mutation event is generated. Based on the behavioral mutation event, the strategy migration process is executed to generate the migrated waste disposal management strategy, which includes an updated waste bin overflow prediction window, an adjusted waste collection priority, and a reset waste disposal classification correction prompt. Execute waste disposal optimization management operations according to the migrated waste disposal management strategy, and output the execution results to the waste disposal management terminal.

2. The waste disposal optimization management method based on classification data mining according to claim 1, characterized in that, Based on short-cycle and long-cycle classification sequences, class difference processing is performed between adjacent time periods, and class aggregation calculation is performed for each long-cycle stage to generate short-cycle volatility indicators and long-cycle stability indicators, including: The short-period classification sequence is segmented according to a fixed-length short-period analysis window. Within each analysis window, the directional comparison, continuity judgment and amplitude accumulation of the category changes in adjacent time periods are performed to generate short-period difference results within the short-period window. Consistency detection is performed on the category change patterns of multiple consecutive analysis windows based on the short-cycle difference results, and the consistency detection results are transformed into short-cycle fluctuation indicators. Long-period classification sequences are grouped according to long-period stages. Within each long-period group, long-period clustering results are generated based on the frequency of category occurrence, duration, and stability of the dominant category within the stage. Long-period stability indices are then calculated based on the long-period clustering results.

3. The waste disposal optimization management method based on classification data mining according to claim 1, characterized in that, Cross-series correlation analysis is performed based on short-cycle volatility indicators and long-cycle stability indicators to generate correlation analysis results. Based on these results, a hierarchical triggering factor system is constructed, including both short-cycle and long-cycle triggering factors, comprising: The short-cycle volatility indicator and the long-cycle stability indicator are time-aligned, and the trend difference value between the two is calculated within the common time period after alignment. The initial correlation result is generated based on the trend difference value. Based on the trend difference values ​​in the primary correlation results, the directionality, cumulativeity, and persistence of the difference in the alignment time period are segmented for analysis, generating basic values ​​of short-term triggering factors to describe short-term burst behavior and basic values ​​of long-term triggering factors to describe long-term offset behavior. Based on the structural differences between the base values ​​of short-cycle trigger factors and long-cycle trigger factors, the two are hierarchically reorganized, with the short-cycle trigger factor placed in the rapid response layer and the long-cycle trigger factor placed in the trend identification layer, forming a layered trigger factor with two layers.

4. The waste disposal optimization management method based on classification data mining according to claim 1, characterized in that, Based on the hierarchical triggering factors, offset difference is determined. When the offset difference between the short-cycle triggering factor and the long-cycle triggering factor exceeds a preset correlation offset threshold, a behavioral mutation event is generated, including: Offset source filtering is performed on the short-cycle trigger factors and long-cycle trigger factors of the layered trigger factors respectively. The instantaneous changes caused by a single anomaly in the short-cycle trigger factors are removed, and the unstable offsets caused by short-term fluctuations in the long-cycle trigger factors are excluded, thus generating filtered short-cycle trigger factors and long-cycle trigger factors. Calculate the continuous change in the offset difference between the filtered short-period triggering factor and the long-period triggering factor, and accumulate the continuous change over multiple consecutive time periods to form an offset difference accumulation sequence. A dual-condition determination is made based on the cumulative amount and duration in the cumulative offset sequence. When the cumulative amount exceeds a preset associated offset threshold and the duration exceeds a preset first duration threshold, a behavioral mutation event is generated. The behavioral mutation event includes a mutation category to indicate the mutation type, a mutation amplitude to indicate the mutation intensity, and a mutation time period to indicate the mutation occurrence range.

5. The waste disposal optimization management method based on classification data mining according to claim 1, characterized in that, Based on the behavioral mutation event, a strategy migration process is executed to generate a migrated waste disposal management strategy. This includes an updated waste bin overflow prediction window, an adjusted waste collection priority, and reset waste sorting correction prompts, including: The prediction window for overflowing trash cans is dynamically expanded or contracted based on the magnitude and type of the behavioral mutation event, so that the time range of the prediction window changes proportionally with the magnitude of the mutation, and an adjusted prediction window is generated. Based on the waste categories involved in behavioral mutation events and their frequency of occurrence, the collection objects are prioritized. By establishing a collection priority enhancement factor, high mutation categories are given higher collection priority, and an adjusted collection order is generated. Based on the waste sorting deviations corresponding to behavioral mutation events, the deviation category is used as a key prompt category. The frequency of push notifications for this category is increased, and the prompt content is adaptively updated to generate updated waste disposal and sorting correction prompts.

6. The waste disposal optimization management method based on classification data mining according to claim 3, characterized in that, The short-cycle volatility indicator and the long-cycle stability indicator are time-aligned. The trend difference between the two is calculated within their common aligned time period, and a preliminary correlation result is generated based on this trend difference, including: Time reference points are determined based on short-cycle fluctuation indicators and long-cycle stability indicators, and these time reference points are expanded to form a reference time series that describes the time distribution of the two types of indicators. Based on the reference time series, short-cycle volatility indicators and long-cycle stability indicators are time-mapped so that the two types of indicators correspond on the same time axis, forming an aligned indicator sequence. Based on the alignment indicator sequence, the direction, magnitude, and speed of change of short-cycle volatility indicators and long-cycle stability indicators within the same time period are obtained, and interval differences are extracted to generate trend difference values. Based on the trend difference value, the deviation direction, deviation intensity and deviation duration within the interval are correlated to obtain the primary correlation results used to characterize the relationship of cross-cycle behavior change.

7. The waste disposal optimization management method based on classification data mining according to claim 3, characterized in that, Based on the trend difference values ​​in the primary correlation results, a segmented analysis is performed on the directionality, cumulativeity, and persistence of the differences within the alignment time period. This generates basic values ​​for short-term triggering factors to describe short-term burst behavior and basic values ​​for long-term offset behavior, including: The trend difference values ​​are divided directionally according to the direction of change of the trend difference values, forming directional difference segments to represent the direction of trend change; Based on the directional difference segment, the trend difference values ​​within the difference segment are accumulated, and a cumulative difference segment is formed to represent the intensity of the cumulative difference based on the characteristics of the cumulative change. Based on the duration of the cumulative difference segment within the alignment time period, a persistence analysis is performed on the difference segment to identify the difference segment whose duration exceeds a preset second persistence threshold, and a short-cycle trigger factor base value is generated based on the difference segment. Long-term offset analysis is performed on the difference segments that do not meet the continuous threshold. Trend aggregation is performed based on the offset direction, offset magnitude and offset stage characteristics to generate the base value of long-term trigger factor.

8. A waste disposal optimization management system based on classification data mining, characterized in that, The system, used in any one of claims 1 to 7, comprises: The data acquisition module is used to acquire waste disposal behavior data uploaded by the waste disposal collection device, and generate an original behavior sequence to represent continuous disposal behavior based on the data. The waste disposal behavior data includes waste disposal amount, waste disposal time and waste disposal type. The sequence construction module is used to divide the original behavior sequence into time periods, generate short-cycle behavior data and long-cycle behavior data, and construct short-cycle classification sequences and long-cycle classification sequences respectively. The indicator generation module is used to perform category difference processing between adjacent time periods and category aggregation calculation for each long period based on short-period classification sequences and long-period classification sequences, to generate short-period volatility indicators and long-period stability indicators. The correlation analysis module is used to perform cross-series correlation analysis based on short-cycle volatility indicators and long-cycle stability indicators, generate correlation analysis results, and construct hierarchical triggering factors including short-cycle triggering factors and long-cycle triggering factors based on them. The mutation identification module is used to determine the offset difference based on the hierarchical triggering factor. When the offset difference between the short-cycle triggering factor and the long-cycle triggering factor exceeds the preset associated offset threshold, a behavioral mutation event is generated. The strategy migration module is used to perform strategy migration processing based on behavioral mutation events, and generate the migrated waste disposal management strategy, which includes an updated waste bin overflow prediction window, an adjusted waste collection priority, and a reset waste disposal classification correction prompt. The waste disposal management module is used to perform waste disposal optimization management operations according to the migrated waste disposal management strategy, and output the execution results to the waste disposal management terminal.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images