Internet laundry quality evaluation method based on quality comprehensive evaluation index

Abstract

The application discloses an internet laundry quality evaluation method based on a quality comprehensive evaluation index, and particularly relates to the field of electrical appliance services, and constructs a full-automation quality control method from multi-source data collection, data preprocessing, index calculation, comprehensive evaluation, analysis display to dynamic early warning. Through offline and flow parallel collection, data comprehensive access is realized, data cleaning and standardization are completed by relying on a rule engine, multi-dimensional objective quantitative evaluation is realized by combining a combination weighting TOPSIS model, and visual analysis and EWMA dynamic early warning are matched, so that quality short boards and deterioration trends can be accurately identified and captured, data islands and human evaluation deviations can be effectively eliminated, quality problem early warning and hierarchical disposal can be realized, quality management efficiency and digitalization level can be greatly improved, and a data-driven closed-loop quality control capability is formed.

Description

Technical Field

[0001] This invention relates to the field of electrical appliance services, and more specifically to an internet-based laundry quality evaluation method based on a comprehensive quality evaluation index. Background Technology

[0002] With the development of online laundry platforms, processes such as order placement, garment pickup and delivery, washing and processing, rewashing, and after-sales complaints are gradually becoming online and data-driven. Platforms typically monitor quality through the following methods:

[0003] First, subjective quality assessment based on user ratings and star ratings; second, after-sales quality monitoring indicators based on complaint rate and negative review rate; third, internal quality assessment indicators based on quality inspection and spot check results; and fourth, timeliness indicators based on order completion rate and overdue rate.

[0004] However, existing methods for evaluating laundry quality still have the following problems:

[0005] 1. Ignoring the value of rewashed data:

[0006] When a platform or factory discovers that the washing quality of clothing is substandard, it will initiate a rewash operation. Rewashing is usually not directly reflected in user ratings; some rewashes are handled after user complaints, while others are proactively discovered during the quality inspection process. Rewashing records serve as internal evidence of quality issues. Many platforms do not use the rewash rate as a core indicator in their unified quality evaluation system, but rather conduct simple statistics or internal assessments.

[0007] 2. The complaint rate and rewash rate do not form a unified quality view:

[0008] The complaint rate represents user-perceived quality issues; the re-wash rate represents quality issues detected internally by the platform. When these two are not integrated, it can lead to:

[0009] Low complaint rate but high re-wash rate: This may be due to frequent internal rework, which increases costs but users have not yet noticed it.

[0010] High complaint rate but low rewash rate: This may be due to improper handling, failure to rewash, or lack of record-keeping, which affects user satisfaction and platform reputation.

[0011] Therefore, most existing assessment systems typically only look at the complaint rate or the internal rewash rate, lacking a unified and comparable comprehensive quality evaluation index.

[0012] 3. Lack of standardized algorithms and parameterization methods:

[0013] Currently, many quality rating systems use empirical weighting: for example, "complaint rate is 60%, positive review rate is 40%." There is a lack of interpretable weighting calculation algorithms (such as the analytic hierarchy process or entropy weighting). Therefore, there is an urgent need for adaptive weights that dynamically adjust with the data, enabling the standardization and comprehensive evaluation of "cost-based indicators" such as rewash rate and complaint rate.

[0014] 4. Quality evaluation results are difficult to use in a closed-loop online manner:

[0015] Even when quality scores are available, reports are often exported periodically for manual review, lacking real-time calculation and visualization; furthermore, there is a lack of linkage between quality scores and quality warnings and incentive mechanisms; and fine-grained (store, factory, employee, category) quality analysis is not achieved.

[0016] Therefore, there is an urgent need for a method for internet-based laundry platforms that can use rewash rate and complaint rate as core indicators, integrate other auxiliary indicators to form a unified comprehensive quality evaluation index, and provide a feasible system architecture to achieve automated calculation, visualization, and early warning triggering. Summary of the Invention

[0017] Therefore, this invention provides an internet laundry quality evaluation method based on a comprehensive quality evaluation index to solve the problems mentioned in the background art.

[0018] To achieve the above objectives, the present invention provides the following technical solution: an internet laundry quality evaluation method based on a comprehensive quality evaluation index, comprising the following steps:

[0019] Business data on orders, re-washing, complaints, stores, factories, delivery personnel, and clothing categories are extracted from the platform's order system, re-washing management system, and complaint work order system, and written into a business table and a dimension dictionary table, respectively. The order business table stores the corresponding dimension code as a foreign key, and forms an association mapping with the primary key of the dimension dictionary table through the foreign key to realize the association and matching between business data and basic dimension information. A dual-track architecture of parallel deployment of batch ETL offline collection and real-time streaming collection is used to jointly complete the construction of data collection units for multiple business systems.

[0020] The system uses the unique order number as the association key to connect multiple source business tables, removes missing data in key fields, fills non-key fields according to preset rules, removes duplicate orders, and generates a standardized order detail data table.

[0021] Select the start and end time interval of the assessment, group and aggregate according to at least one dimension of store, factory, deliveryman, and clothing category, and calculate at least two original cost indicators from the following: rewash rate, complaint rate, average complaint severity, and percentage of complaints after rewashing.

[0022] The range normalization method is used to eliminate the differences in the dimensions of the indicators, mapping all indicator values ​​to the interval [0,1], thus obtaining the normalized standardized matrix; based on the normalized standardized matrix... The TOPSIS comprehensive evaluation model is constructed; the objective weights of the indicators are calculated by the entropy weight method, and the subjective weights of the indicators are obtained by the expert scoring method; the combination coefficient is set to linearly weight and fuse the two types of weights, and the final combined weight is obtained after normalization. The combined weight is then applied to the TOPSIS comprehensive evaluation model.

[0023] The proximity and ranking of each evaluation object output by the TOPSIS comprehensive evaluation model are transformed into interpretable and actionable information to support quality improvement decisions. The time series of proximity of each object is used as input, and the exponentially weighted moving average (EWMA) algorithm is used to realize dynamic quality early warning.

[0024] Preferably, the business tables include the order master table of the platform order system, the rewash record table of the rewash management system, and the complaint work order table of the complaint work order system, which are used to record the real operational data of the entire laundry business process;

[0025] The business dimension codes of the order_main table are: order_id, user_id, store_id, factory_id, courier_id, category_code, order_status, create_time, and finish_time.

[0026] The business dimension codes for the rewash record table `rewash_record` are: `order_id`, `rewash_flag`, `rewash_reason`, and `rewash_time`.

[0027] The business dimension codes for the complaint ticket table (complaint_ticket) are: order_id, is_platform, complaint_type, complaint_level, refund_amount, and is_arbitration.

[0028] The dimension dictionary table comprises four tables: a store information table, a factory information table, a delivery person information table, and a clothing category lookup table. These tables provide a mapping between the codes and names of each business dimension. Each dimension dictionary uses its unique code as a primary key to uniquely identify a dimension object. The order master table stores the corresponding dimension code as a foreign key, which, along with the primary key of the dimension dictionary table, forms a mapping relationship, enabling the matching of business data with basic dimension information. The business data table and the dimension dictionary table are linked through primary and foreign keys.

[0029] The business tables are coded using four business dimensions: store_id, factory_id, courier_id, and category_code. These are linked to the store information table, factory information table, delivery person information table, and clothing category lookup table, respectively, and are used for display in the final reports and alert pages.

[0030] Preferably, the batch ETL offline data collection architecture extracts data from the platform's order system, rewash management system, and complaint work order system at fixed intervals, providing complete and traceable offline data for periodic quality evaluations. Specifically:

[0031] Input the business table and dimension dictionary table, perform incremental extraction every day at midnight, and perform a full data backtracking once a month at the end of the month; extract newly added or status-changed data from the previous collection period to the current time based on the timestamps create_time, finish_time, rewash_time, and complaint_time; use DataX or Sqoop tools, configure the database source connection information, and write the extraction script; the raw data is stored in the data warehouse ODS layered, stored by system and date partition, and the original fields are retained without modification;

[0032] The real-time streaming acquisition architecture continuously monitors changes in business data, providing low-latency data for real-time quality monitoring and dynamic early warning. Specifically:

[0033] The business table uses the unique order number (order_id) as the unique key for association, which is a unified primary key; the dimension dictionary table maps to the business table through foreign keys such as store_id, factory_id, courier_id, and category_code, and retains the original encoding without conversion during the data collection process.

[0034] Enable binlog functionality for business tables to record additions, modifications, and deletions of data in the order master table, reprocessing record table, and complaint work order table. Use the Canal component to monitor and capture changes in the logs in real time. The captured real-time data is uniformly pushed to the Kafka message middleware for temporary caching, and then the Flink computing engine reads and consumes the data stored internally in Kafka. Real-time data for orders, reprocessing, and complaints is written to the data warehouse DWD layer and persisted in real time by the unique order number (order_id). A standardized raw data table is output as input for subsequent data cleaning.

[0035] Preferably, based on a standardized original data table, the order number (order_id) is used as the globally unique association key. A left join is used to connect the order master table, the rewash record table, and the complaint work order table: using the order master table as the base table, the order number (order_id) is matched against the rewash record table and the complaint work order table. If a match is found, the corresponding rewash information and complaint information are written; if no match is found, the missing fields are filled with default values. Data with missing order numbers is directly removed, and for duplicate order numbers, the latest record is retained and redundant data is deleted. Finally, a standardized order detail data table containing full order information, rewash marking information, complaint classification information, and business dimension coding information is formed.

[0036] Preferably, let , Represents the set of all evaluation objects. Indicates the total number of evaluation objects; Indicates the first The evaluation targets include stores, factories, delivery services, and clothing categories; The evaluation period is ;

[0037] Rewashing rate: ;

[0038] in, Indicates the first time after cleaning The total number of rewash orders for each evaluation object, i.e. the number of orders with the rewash_flag set to 1; Indicates the first time after cleaning Total number of orders for each evaluated entity;

[0039] Complaint rate: ;

[0040] in, Indicates the first time after cleaning The number of complaint orders for each evaluation object, i.e. the number of orders with the complaint flag is_complaint set to 1;

[0041] Average complaint severity ;

[0042] in, This indicates the severity of the complaint corresponding to a single complaint order;

[0043] If the first There were no complaints about any of the orders within the evaluation category. The average severity of complaints is uniformly filled in as 0;

[0044] Regarding valid complaints that arose after rewashing, the percentage of complaints received after rewashing was [not specified]. ;

[0045] in, Indicates the first The number of complaints after each evaluation object is rewashed, that is, the number of orders with the rewash flag set to 1 and the complaint flag set to 1;

[0046] If the first The evaluated object had no repeat washing orders, that is... The percentage of complaints after rewashing Fill the value with 0% to avoid calculation errors caused by dividing by zero.

[0047] Preferably, the indicators include positive indicators and negative indicators; the positive indicator is the total order volume; the negative indicators include the repeat washing rate, complaint rate, average complaint severity, and the percentage of complaints after repeat washing.

[0048] For each indicator Iterate through all evaluation objects Calculate this indicator global minimum value and global maximum value ;

[0049] Inverse index normalization formula: ;

[0050] in, Indicates the first The evaluation object is in the first The original values ​​of each indicator; Represents the normalized i-th The evaluation object is in the first Standardized matrix for each indicator; Indicates that all evaluation objects are in the first... The maximum value of each indicator; This indicates that all evaluation objects are in the first... The minimum value of each indicator;

[0051] Positive index normalization formula: ;

[0052] If the maximum value and the minimum value of a certain indicator are equal ( To avoid division by zero errors, this indicator The normalization value of all objects is uniformly set to 1.0;

[0053] All after normalization All of them are positive indicators, and each row represents an evaluation object. Each column represents a normalized positive indicator. This provides standardized input data for the subsequent TOPSIS comprehensive evaluation model; the larger the value, the more important it is for the evaluation object. In terms of indicators The better the quality performance.

[0054] Preferably, based on the normalized standardized matrix Construct a comprehensive evaluation model for TOPSIS:

[0055] set up For the first Each indicator weight satisfies ;

[0056] Multiply each normalized index value by its corresponding weight to obtain a weighted matrix. :

[0057] ;

[0058] Each indicator The maximum value among all objects, i.e., the optimal performance, is the positive ideal solution. :

[0059] ;

[0060] ;

[0061] Each indicator The minimum value among all objects, i.e., the worst performance, is the positive ideal solution. :

[0062] ;

[0063] ;

[0064] The distance from each object to the positive and negative ideal solutions is calculated using Euclidean distance:

[0065] Distance to the ideal solution ;

[0066] Distance to negative ideal solution ;

[0067] Calculate proximity ;

[0068] in, Proximity The larger the value, the closer the object is to the positive ideal solution and the farther it is from the negative ideal solution, and the better its quality performance.

[0069] For all evaluation objects, according to By sorting the evaluation objects from largest to smallest, a quality ranking is obtained, providing a quantitative basis for subsequent quality analysis and early warning.

[0070] Preferably, the combined weights adopt a dual-track design using the entropy weight method and the expert scoring method, and the weights of both methods satisfy the constraint that the sum is 1;

[0071] The entropy weight method is adopted, based on the standardized matrix. Calculate the information entropy and information utility value of each indicator:

[0072] Calculate the first The first indicator The proportion of each evaluation object :

[0073] ;

[0074] in, The total number of evaluation objects;

[0075] Calculate the first Entropy value under each indicator :

[0076] ;

[0077] ;

[0078] like , ;

[0079] Calculate information utility value :

[0080] ;

[0081] Calculate entropy weight ;

[0082] An expert scoring method was used, where business experts scored the importance of the indicators, and the weights were obtained after normalization.

[0083] Business, quality, and operations experts scored the importance of five indicators on a scale of 1 to 10; the average score for each indicator was calculated and then normalized to obtain its weight.

[0084] ;

[0085] Combination weights: setting combination coefficients ,according to Calculate the initial combined weights. If the sum of the combined weights is not 1, perform further normalization to make the sum of all indicator weights equal to 1; where the coefficients... Configure flexibly according to business needs; a larger value means more emphasis on expert experience, and a smaller value means more emphasis on objective data patterns.

[0086] The final combined weights are used as input parameters for the TOPSIS model to conduct a multi-dimensional comprehensive quality evaluation.

[0087] Preferably, the closeness of each evaluation object is based on the output of the TOPSIS comprehensive evaluation model. The system ranks and identifies the top N best-performing entities and the bottom N for overall quality across four categories: stores, factories, delivery personnel, and clothing categories. It also provides a breakdown analysis of low-performing entities to identify their weakest indicators and generate targeted improvement suggestions. Furthermore, it supports trend comparisons by time slices, enabling year-on-year and month-on-month analysis. Visualizing the quantitative evaluation results through bar charts, radar charts, line charts, and heatmaps transforms them into intuitive quality reports, supporting operational decisions and quality improvement.

[0088] Preferably, during the evaluation period Internally, based on the degree of relevance to each evaluation object. Using time series data as input, the exponentially weighted moving average (EWMA) algorithm is employed to implement dynamic quality early warning, including the following steps:

[0089] For each evaluation object The proximity sequence, through Real-time updates of EWMA estimates;

[0090] in, Indicates the object of evaluation In the EWMA estimate for the period; Represents the smoothing coefficient. ; Indicates the object of evaluation In the The closeness to the period; Indicates the initial value, taken from the evaluation object. Historical proximity mean;

[0091] Set upper and lower control limits based on the standard deviation of the EWMA estimate:

[0092] Upper control limit UCL: ;

[0093] Lower control limit LCL: ;

[0094] in, This represents the standard deviation of the EWMA value; The control limit coefficient is calculated as follows:

[0095] ;

[0096] in, The object of evaluation The mean of the historical proximity series;

[0097] When the evaluation object No. The closeness to the target date is lower than the lower control limit. When a yellow alert is triggered, indicating a mild abnormality, continued observation is required; when it is continuous... When the current level falls below the lower control limit (LCL), a red alert is triggered, prompting the operator to intervene in a timely manner to prevent quality issues from escalating.

[0098] The present invention has the following advantages:

[0099] II. Detailed Technical Effects (Applicable to Patent Specifications and Technical Solution Text)

[0100] 1. This invention adopts a parallel architecture of batch ETL offline acquisition and real-time streaming acquisition, which can simultaneously aggregate heterogeneous data from multiple sources such as orders, re-processing, complaints, and dimensional archives. It not only meets the needs of batch statistical analysis of historical data, but also captures dynamic changes in business in real time, taking into account both data integrity and real-time performance.

[0101] 2. This invention uses the order number as the unique association key to complete the concatenation of multiple tables. It automatically completes the deduplication of duplicate data, the removal of missing key fields, and the standardized filling of non-key fields through a preset rule engine. It also unifies the complaint severity mapping standard, effectively filters dirty data, unifies data standards, and ensures the accuracy of subsequent calculation and analysis results from the source.

[0102] 3. This invention uses a range normalization algorithm to uniformly map quality indicators of different dimensions and properties to the [0,1] interval, eliminating the differences between absolute values, percentages, and discrete levels, and providing standardized and unified input data for the comprehensive evaluation model.

[0103] 4. This invention constructs five core indicators including business volume, rewash level, complaint level, complaint severity, and rewash repair effect. It can conduct comprehensive evaluation of four types of objects—stores, laundry factories, delivery personnel, and clothing categories—from multiple perspectives, including business scale, basic service quality, and problem repair capabilities, with no blind spots in the evaluation dimensions.

[0104] 5. This invention integrates the objective weighting method of entropy weighting with the subjective weighting method of expert scoring. It obtains combined weights through linear weighting, which not only respects the distribution characteristics of the data itself, but also combines industry management experience and business control priorities, overcoming the limitations of a single weighting method and making the indicator weights more in line with actual management scenarios.

[0105] 6. This invention calculates the comprehensive proximity score and completes the ranking and classification based on the TOPSIS multi-attribute decision model, transforming the abstract service quality into a quantitative score that can be compared horizontally and tracked vertically, realizing fair benchmarking among different entities, replacing the traditional manual qualitative evaluation, and making the evaluation results objective and traceable.

[0106] 7. This invention supports the generation of a TOP-N list of high-quality objects and a list of the lowest-quality objects, and classifies the quality levels according to their proximity. Managers can quickly identify excellent benchmarks and entities that need rectification, thereby improving the targeting of management.

[0107] 8. This invention traces back the scores of individual indicators for low-quality entities, accurately identifies weak links, and automatically outputs rectification suggestions in combination with business rules, realizing a closed-loop analysis of overall evaluation, weakness identification, and solution guidance, which greatly shortens the problem investigation cycle.

[0108] 9. This invention supports year-on-year and month-on-month trend analysis on a weekly, monthly, and quarterly basis, and tracks quality fluctuations in real time; it displays data, rankings, and trends through visualization formats such as bar charts, radar charts, line charts, and heat maps, lowering the threshold for data interpretation, outputting standardized analysis reports, and directly supporting operational decisions.

[0109] 10. This invention introduces the EWMA (Exponentially Weighted Moving Average) algorithm to build a dynamic early warning model, which has a good ability to identify the slow decline and gradual deterioration of quality indicators, breaking through the traditional post-event statistical rectification mode and realizing early perception and early intervention of quality risks.

[0110] 11. This invention divides the warning into yellow and red levels based on statistical control limits, distinguishes between mild and severe anomalies, and matches differentiated handling strategies to facilitate managers to respond and handle matters in an orderly manner according to risk level. Detailed Implementation

[0111] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0112] This invention proposes a method for evaluating the quality of internet-based laundry services based on a comprehensive quality evaluation index, comprising the following steps:

[0113] Business data on orders, re-washing, complaints, stores, factories, delivery personnel, and clothing categories are extracted from the platform's order system, re-washing management system, and complaint work order system, and written into a business table and a dimension dictionary table, respectively. The order business table stores the corresponding dimension code as a foreign key, and forms an association mapping with the primary key of the dimension dictionary table through the foreign key to realize the association and matching between business data and basic dimension information. A dual-track architecture of parallel deployment of batch ETL offline collection and real-time streaming collection is used to jointly complete the construction of data collection units for multiple business systems.

[0114] The system uses the unique order number as the association key to connect multiple source business tables, removes missing data in key fields, fills non-key fields according to preset rules, removes duplicate orders, and generates a standardized order detail data table.

[0115] Select the start and end time interval of the assessment, group and aggregate according to at least one dimension of store, factory, deliveryman, and clothing category, and calculate at least two original cost indicators from the following: rewash rate, complaint rate, average complaint severity, and percentage of complaints after rewashing.

[0116] The range normalization method is used to eliminate the differences in the dimensions of the indicators, mapping all indicator values ​​to the interval [0,1], thus obtaining the normalized standardized matrix; based on the normalized standardized matrix... The TOPSIS comprehensive evaluation model is constructed; the objective weights of the indicators are calculated by the entropy weight method, and the subjective weights of the indicators are obtained by the expert scoring method; the combination coefficient is set to linearly weight and fuse the two types of weights, and the final combined weight is obtained after normalization. The combined weight is then applied to the TOPSIS comprehensive evaluation model.

[0117] The proximity and ranking of each evaluation object output by the TOPSIS comprehensive evaluation model are transformed into interpretable and actionable information to support quality improvement decisions. The time series of proximity of each object is used as input, and the exponentially weighted moving average (EWMA) algorithm is used to realize dynamic quality early warning.

[0118] The business tables include the order master table of the platform order system, the rewash record table of the rewash management system, and the complaint work order table of the complaint work order system, which are used to record the real operational data of the entire laundry business process;

[0119] The business dimension codes of the order_main table are: order_id, user_id, store_id, factory_id, courier_id, category_code, order_status, create_time, and finish_time.

[0120] The business dimension codes for the rewash record table `rewash_record` are: `order_id`, `rewash_flag`, `rewash_reason`, and `rewash_time`.

[0121] The business dimension codes for the complaint ticket table (complaint_ticket) are: order_id, is_platform, complaint_type, complaint_level, refund_amount, and is_arbitration.

[0122] The dimension dictionary table comprises four tables: a store information table, a factory information table, a delivery person information table, and a clothing category lookup table. These tables provide a mapping between the codes and names of each business dimension. Each dimension dictionary uses its unique code as a primary key to uniquely identify a dimension object. The order master table stores the corresponding dimension code as a foreign key, which, along with the primary key of the dimension dictionary table, forms a mapping relationship, enabling the matching of business data with basic dimension information. The business data table and the dimension dictionary table are linked through primary and foreign keys.

[0123] The business tables are coded using four business dimensions: store_id, factory_id, courier_id, and category_code. These are linked to the store information table, factory information table, delivery person information table, and clothing category lookup table, respectively, and are used for display in the final reports and alert pages, as detailed below:

[0124] Table 1. Order Main Table (order_main)

[0125] O001 U001 S01 F01 C01 CAT01 Completed 2025-01-01 2025-01-02 O002 U002 S02 F01 C02 CAT02 Completed 2025-01-01 2025-01-02

[0126] Table 2 Rewash Record Table (rewash_record)

[0127] O001 1 Not clean 2025-01-02

[0128] Table 3 Complaint Ticket Form (complaint_ticket)

[0129] O002 Damaged clothing 2 50 0

[0130] Table 4 Store Information Table (store_info)

[0131] S01 Store A S02 Store B

[0132] Table 5 Factory Information Table (factory_info)

[0133] F01 Central laundry plant

[0134] Table 6 Delivery Person Information Table (courier_info)

[0135] C01 Zhang San C02 Li Si

[0136] Table 7 Clothing Category Comparison Table (category_info)

[0137] CAT01 down jacket CAT02 shirt

[0138] Table 8 Standardized Order Details Data Table

[0139] O001 S01 F01 C01 CAT01 2025-01-01 2025-01-02 1 Not clean 0 0 0 O002 S02 F01 C02 CAT02 2025-01-01 2025-01-02 0 none 1 2 50

[0140] Table 9 Explanation of the relationships and meanings between Tables 1-8

[0141] order_id Order Number Order Master Table Globally unique primary key user_id User ID Order Master Table - store_id Store code Order Master Table Foreign key, related store dictionary table factory_id Factory Code Order Master Table Foreign key, related factory dictionary table courier_id Deliveryman Code Order Master Table Foreign key, associated with the delivery person dictionary table category_code Clothing category codes Order Master Table Foreign keys, related category dictionary tables order_status Order status Order Master Table Completed / Cancelled create_time Order time Order Master Table - finish_time Completion Time Order Master Table - rewash_flag Should it be rewashed? Rewashing Record Form 0 = not washed, 1 = washed rewash_reason Reasons for rewashing Rewashing Record Form Nothingness is emptiness rewash_time Rewashing time Rewashing Record Form Nothingness is emptiness is_complaint Should I file a complaint? Complaint form 0 = No complaint, 1 = Complaint complaint_type Complaint types Complaint form Nothingness is emptiness complaint level Complaint severity level Complaint form Level 1 / 2 / 3 refund_amount Refund amount Complaint form If there is none, then it is 0. is_arbitration Platform arbitration Complaint form 0 = No, 1 = Yes

[0142] The batch ETL offline data collection architecture extracts data from the platform's order system, rewash management system, and complaint work order system at fixed intervals, providing complete and traceable offline data for periodic quality evaluations. Specifically:

[0143] Input the business table and dimension dictionary table, perform incremental extraction every day at midnight, and perform a full data backtracking once a month at the end of the month; extract newly added or status-changed data from the previous collection period to the current time based on the timestamps create_time, finish_time, rewash_time, and complaint_time; use DataX or Sqoop tools, configure the database source connection information, and write the extraction script; the raw data is stored in the data warehouse ODS layered, stored by system and date partition, and the original fields are retained without modification;

[0144] The real-time streaming acquisition architecture continuously monitors changes in business data, providing low-latency data for real-time quality monitoring and dynamic early warning. Specifically:

[0145] The business table uses the unique order number (order_id) as the unique key for association, which is a unified primary key; the dimension dictionary table maps to the business table through foreign keys such as store_id, factory_id, courier_id, and category_code, and retains the original encoding without conversion during the data collection process.

[0146] Enable binlog functionality for business tables to record additions, modifications, and deletions of data in the order master table, reprocessing record table, and complaint work order table. Use the Canal component to monitor and capture changes in the logs in real time. The captured real-time data is uniformly pushed to the Kafka message middleware for temporary caching, and then the Flink computing engine reads and consumes the data stored internally in Kafka. Real-time data for orders, reprocessing, and complaints is written to the data warehouse DWD layer and persisted in real time by the unique order number (order_id). A standardized raw data table is output as input for subsequent data cleaning.

[0147] Based on standardized raw data tables, using the unique order number (order_id) as the globally unique join key, a left join is used to concatenate the order master table, rewash record table, and complaint ticket table: Using the order master table as the base table, the order number (order_id) is matched against the rewash record table and the complaint ticket table. If a match is found, the corresponding rewash information and complaint information are written; otherwise, the missing fields are filled with default values. Data with missing order numbers is directly removed, and for duplicate order numbers, the latest record is retained while redundant data is deleted. This results in a standardized order detail data table containing full order information, rewash marking information, complaint classification information, and business dimension coding information. Data cleaning is performed using the following rule engine and simple statistical algorithms:

[0148] The deduplication rule uses the order number (order_id) as the grouping basis and selects the latest record corresponding to the maximum value of the order status (order_status) in each group to remove duplicate orders.

[0149] The missing value handling rules are as follows: for the key fields of order number (order_id), store code (store_id), and factory code (factory_id), the corresponding record will be directly removed when they are missing; when the rewash flag (rewash_flag) is missing, it will be filled with 0 by default; when the complaint flag is missing, it will be filled with 0 by default; and when the delivery person code is missing, it will be filled with unknown.

[0150] The severity of complaints is calculated using a discrete numerical mapping algorithm, which maps general complaints to 1, refund complaints to 2, and platform arbitration complaints to 3, thus forming a standardized complaint severity field.

[0151] Table 10 Data before cleaning

[0152] O001 S01 F01 C01 1 NULL Washing defects 2 O001 S01 F01 C01 NULL NULL Washing defects 1 O002 NULL F02 C02 NULL 1 Refund 3 O003 S02 F01 NULL NULL NULL Platform Arbitration 3

[0153] Table 11 Data after cleaning

[0154] O001 S01 F01 C01 1 0 1 Deduplication and retain status=2 O003 S02 F01 unknown 0 0 3 Fill missing fields

[0155] Table 12 Multidimensional Indicator Calculation

[0156] Stores S01 100 5 5.00% 3 3.00% 1.33 20.00% factory F01 500 18 3.60% 12 2.40% 1.75 11.11% deliveryman C01 50 2 4.00% 1 2.00% 1.00 0.00% Clothing CAT01 200 12 6.00% 6 3.00% 2.00 25.00%

[0157] Within the evaluation period [Tstart, Tend], statistics are collected for each evaluation object. The original cost-type indicators (i.e., the dimension types in Table 12) include rewash rate, complaint rate, average complaint severity, and the percentage of complaints after rewashing.

[0158] In Table 12; Total Orders: The total number of valid order records within the statistical group;

[0159] Rewash rate: The rewash rate is calculated as a percentage by dividing the number of orders with the rewash_flag set to 1 by the total number of orders.

[0160] Complaint rate: The complaint rate is calculated as a percentage, calculated by dividing the number of orders with the complaint flag is_complaint set to 1 by the total number of orders.

[0161] Average Complaint Severity: Based on the complaint severity mapping rules (general complaints = 1, refund complaints = 2, platform arbitration complaints = 3), the average severity of complaint orders is calculated:

[0162] The number of orders with the complaint flag `is_complaint` set to 1, i.e., the total number of complaint orders within the current group, is denoted as... The severity value corresponding to a single complaint order. , These correspond to general complaints, refund complaints, and platform arbitration complaints, respectively; average complaint severity. ;

[0163] If there are no complaint orders in the group, that is The average severity of complaints is uniformly filled in as 0;

[0164] The total number of rewash orders within this dimension group, i.e., the number of orders with the rewash_flag set to 1, is denoted as... The number of rewashed and complained orders within this dimension group, i.e., the number of orders with a rewash flag of 1 and a complaint flag of 1, is denoted as... The percentage of complaints after rewashing ;

[0165] If there are no rewash orders within this dimension group ( This indicator is uniformly filled with 0% to avoid calculation errors when dividing by 0;

[0166] Only valid complaints generated after the rewashing process are counted; order cancellations and invalid complaints are not included in the numerator.

[0167] The metrics include positive and negative metrics; the positive metric is the total order volume; the negative metrics include the repeat washing rate, complaint rate, average complaint severity, and the percentage of repeat complaints after the repeat washing.

[0168] For each indicator Iterate through all evaluation objects Calculate this indicator global minimum value and global maximum value ;

[0169] Inverse index normalization formula: ;

[0170] in, Indicates the first The evaluation object is in the first The original values ​​of each indicator; Represents the normalized i-th The evaluation object is in the first Standardized matrix for each indicator; This indicates that all evaluation objects are in the first... The maximum value of each indicator; This indicates that all evaluation objects are in the first... The minimum value of each indicator;

[0171] Positive index normalization formula: ;

[0172] If the maximum value and the minimum value of a certain indicator are equal ( To avoid division by zero errors, this indicator The normalization value of all objects is uniformly set to 1.0;

[0173] All after normalization All of them are positive indicators, and each row represents an evaluation object. Each column represents a normalized positive indicator. This provides standardized input data for the subsequent TOPSIS comprehensive evaluation model; the larger the value, the more important it is for the evaluation object. In terms of indicators The better the quality performance.

[0174] Table 13 Indicator Explanation

[0175] Total orders Positive indicators The higher the value, the better. Rewashing rate contrarian indicators The smaller the value, the better. Complaint rate contrarian indicators The smaller the value, the better. Average complaint severity contrarian indicators The smaller the value, the better. The percentage of complaints after rewashing contrarian indicators The smaller the value, the better.

[0176] Table 14 Indicator Normalization

[0177] F01 1.00 0.79 0.72 0.50 0.79 F02 0.33 0.38 0.45 0.20 0.47 F03 0.83 0.66 0.57 0.68 0.68 F04 0.00 0.00 0.00 0.00 0.00 F05 0.50 1.00 1.00 0.80 1.00

[0178] The combined weights are designed using a dual-track approach: entropy weighting and expert scoring. The weights from both methods satisfy the constraint that the sum of the weights is 1.

[0179] The entropy weight method is adopted, based on the standardized matrix. Calculate the information entropy and information utility value of each indicator:

[0180] Calculate the first The first indicator The proportion of each evaluation object :

[0181] ;

[0182] in, The total number of evaluation objects;

[0183] Calculate the first Entropy value under each indicator :

[0184] ;

[0185] ;

[0186] like , ;

[0187] Calculate information utility value :

[0188] ;

[0189] Calculate entropy weight ;

[0190] An expert scoring method was used, where business experts scored the importance of the indicators, and the weights were obtained after normalization.

[0191] Business, quality, and operations experts scored the importance of five indicators on a scale of 1 to 10; the average score for each indicator was calculated and then normalized to obtain its weight.

[0192] ;

[0193] Combination weights: setting combination coefficients ,according to Calculate the initial combined weights. If the sum of the combined weights is not 1, perform further normalization to make the sum of all indicator weights equal to 1; where the coefficients... Configure flexibly according to business needs; a larger value means more emphasis on expert experience, and a smaller value means more emphasis on objective data patterns.

[0194] The final combined weights are used as input parameters for the TOPSIS model to conduct a multi-dimensional comprehensive quality evaluation.

[0195] Table 15 Original Data

[0196] F01 500 3.6 2.4 1.75 11.11 F02 300 5.2 3.1 2.00 18.00 F03 450 4.1 2.8 1.60 13.50 F04 200 6.5 4.0 2.20 25.00 F05 350 3.0 1.9 1.50 8.00

[0197] Table 16 After combining weights ( )

[0198] Total orders 0.15 0.10 0.125 Rewashing rate 0.25 0.30 0.275 Complaint rate 0.25 0.30 0.275 Average complaint severity 0.20 0.20 0.200 The percentage of complaints after rewashing 0.15 0.10 0.125 total 1.00 1.00 1.00

[0199] Based on the normalized standardized matrix Construct a comprehensive evaluation model for TOPSIS:

[0200] set up For the first Each indicator weight satisfies ;

[0201] Multiply each normalized index value by its corresponding weight to obtain a weighted matrix. :

[0202] ;

[0203] Each indicator The maximum value among all objects, i.e., the optimal performance, is the positive ideal solution. :

[0204] ;

[0205] ;

[0206] Each indicator The minimum value among all objects, i.e., the worst performance, is the positive ideal solution. :

[0207] ;

[0208] ;

[0209] The distance from each object to the positive and negative ideal solutions is calculated using Euclidean distance:

[0210] Distance to the ideal solution ;

[0211] Distance to negative ideal solution ;

[0212] Calculate proximity ;

[0213] in, Proximity The larger the value, the closer the object is to the positive ideal solution and the farther it is from the negative ideal solution, and the better its quality performance.

[0214] For all evaluation objects, according to By sorting the evaluation objects from largest to smallest, a quality ranking is obtained, providing a quantitative basis for subsequent quality analysis and early warning.

[0215] Table 17. Comprehensive Evaluation Results of the TOPSIS Comprehensive Evaluation Model

[0216] F05 0.88 1 Grade A F01 0.75 2 Grade A F03 0.62 3 Grade B F02 0.35 4 Class C F04 0.12 5 Class D

[0217] The closeness of each evaluation object output based on the TOPSIS comprehensive evaluation model The system ranks and identifies the top N best-performing entities and the bottom N for overall quality across four categories: stores, factories, delivery personnel, and clothing categories. It also provides a breakdown analysis of low-performing entities to identify their weakest indicators and generate targeted improvement suggestions. Furthermore, it supports trend comparisons by time slices, enabling year-on-year and month-on-month analysis. Visualizing the quantitative evaluation results through bar charts, radar charts, line charts, and heatmaps transforms them into intuitive quality reports, supporting operational decisions and quality improvement.

[0218] Table 18 lists the top and bottom performers in the TOPSIS comprehensive evaluation model.

[0219] TOP-2 F05 0.88 The rewash rate, complaint rate, and percentage of complaints after rewashing were all optimal. TOP-2 F01 0.75 Largest order volume, overall balanced performance Last digit - 2 F02 0.35 Low order volume, high re-wash rate, and high complaint rate. Last digit - 2 F04 0.12 Lowest order volume, worst quality indicators.

[0220] Table 19 Examples of Indicator Breakdown for Low-Performance Entities

[0221] Total orders 200 0.00 serious shortcomings Enhance order fulfillment capacity and optimize production capacity Rewashing rate 6.5% 0.00 serious shortcomings Optimize washing processes and strengthen factory quality inspection. Complaint rate 4.0% 0.00 serious shortcomings Investigate for damage and stains, and establish a complaint review mechanism. Average complaint severity 2.20 0.00 serious shortcomings Reduce the proportion of complaints related to refunds and platform arbitration. The percentage of complaints after rewashing 25.00% 0.00 serious shortcomings Improve the quality of the rewashing process to avoid secondary complaints.

[0222] Table 20: Example of Time Trend Comparison (Year-on-Year / Month-on-Month)

[0223] F01 0.75 0.72 +4.17% 0.70 +7.14% rise F05 0.88 0.89 -1.12% 0.85 +3.53% smooth F04 0.12 0.18 -33.33% 0.25 -52.00% deterioration

[0224] During the evaluation period Internally, based on the degree of relevance to each evaluation object. Using time series data as input, the exponentially weighted moving average (EWMA) algorithm is employed to implement dynamic quality early warning, including the following steps:

[0225] For each evaluation object The proximity sequence, through Real-time updates of EWMA estimates;

[0226] in, Indicates the object of evaluation In the EWMA estimate for the period; Represents the smoothing coefficient. ; Indicates the object of evaluation In the The closeness to the period; Indicates the initial value, taken from the evaluation object. Historical proximity mean;

[0227] Set upper and lower control limits based on the standard deviation of the EWMA estimate:

[0228] Upper control limit UCL: ;

[0229] Lower control limit LCL: ;

[0230] in, This represents the standard deviation of the EWMA value; The control limit coefficient is calculated as follows:

[0231] ;

[0232] in, The object of evaluation The mean of the historical proximity series;

[0233] When the evaluation object No. The closeness to the target date is lower than the lower control limit. When a yellow alert is triggered, indicating a mild abnormality, continued observation is required; when it is continuous... When the current level falls below the lower control limit (LCL), a red alert is triggered, prompting the operator to intervene in a timely manner to prevent quality issues from escalating.

[0234] Table 21. Proximity time series of factory F01 as an example.

[0235] 1 0.72 0.72 normal 2 0.73 0.722 normal 3 0.70 0.718 normal 4 0.68 0.710 normal 5 0.65 0.698 normal 6 0.60 0.678 Yellow Alert 7 0.58 0.658 Yellow Alert 8 0.55 0.636 Red Alert

[0236] #imgpt219#; #imgpt220#; Lower control limit #imgpt221#;

[0237] Table 22 Summary of Early Warning Results

[0238] F01 0.55 0.636 0.65 Red Alert Immediately investigate the process and quality control links. F05 0.88 0.870 0.70 normal Maintain current operating strategy F03 0.62 0.620 0.55 normal Continuous monitoring is sufficient. F02 0.35 0.350 0.40 Yellow Alert Conduct quality reviews and optimize rewashing processes. F04 0.12 0.120 0.30 Red Alert Initiate a special rectification campaign, and suspend production for rectification if necessary.

[0239] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for evaluating the quality of internet-based laundry services based on a comprehensive quality evaluation index, characterized by: Includes the following steps: Business data on orders, re-washing, complaints, stores, factories, delivery personnel, and clothing categories are extracted from the platform's order system, re-washing management system, and complaint work order system, and written into a business table and a dimension dictionary table, respectively. The order business table stores the corresponding dimension code as a foreign key, and forms an association mapping with the primary key of the dimension dictionary table through the foreign key to realize the association and matching between business data and basic dimension information. A dual-track architecture of parallel deployment of batch ETL offline collection and real-time streaming collection is used to jointly complete the construction of data collection units for multiple business systems. The system uses the unique order number as the association key to connect multiple source business tables, removes missing data in key fields, fills non-key fields according to preset rules, removes duplicate orders, and generates a standardized order detail data table. Select the start and end time interval of the assessment, group and aggregate according to at least one dimension of store, factory, deliveryman, and clothing category, and calculate at least two original cost indicators from the following: rewash rate, complaint rate, average complaint severity, and percentage of complaints after rewashing. The range normalization method is used to eliminate the differences in the dimensions of the indicators, mapping all indicator values ​​to the interval [0,1], thus obtaining the normalized standardized matrix; based on the normalized standardized matrix... The TOPSIS comprehensive evaluation model is constructed; the objective weights of the indicators are calculated by the entropy weight method, and the subjective weights of the indicators are obtained by the expert scoring method; the combination coefficient is set to linearly weight and fuse the two types of weights, and the final combined weight is obtained after normalization. The combined weight is then applied to the TOPSIS comprehensive evaluation model. The proximity and ranking of each evaluation object output by the TOPSIS comprehensive evaluation model are transformed into interpretable and actionable information to support quality improvement decisions. The time series of proximity of each object is used as input, and the exponentially weighted moving average (EWMA) algorithm is used to realize dynamic quality early warning.

2. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 1, characterized in that: The business tables include the order master table of the platform order system, the rewash record table of the rewash management system, and the complaint work order table of the complaint work order system, which are used to record the real operational data of the entire laundry business process; The business dimension codes of the order_main table are: order_id, user_id, store_id, factory_id, courier_id, category_code, order_status, create_time, and finish_time. The business dimension codes for the rewash record table `rewash_record` are: `order_id`, `rewash_flag`, `rewash_reason`, and `rewash_time`. The business dimension codes for the complaint ticket table (complaint_ticket) are: order_id, is_platform, complaint_type, complaint_level, refund_amount, and is_arbitration. The dimension dictionary table comprises four tables: a store information table, a factory information table, a delivery person information table, and a clothing category lookup table. These tables provide a mapping between the codes and names of each business dimension. Each dimension dictionary uses its unique code as a primary key to uniquely identify a dimension object. The order master table stores the corresponding dimension code as a foreign key, which, along with the primary key of the dimension dictionary table, forms a mapping relationship, enabling the matching of business data with basic dimension information. The business data table and the dimension dictionary table are linked through primary and foreign keys. The business tables are coded using four business dimensions: store_id, factory_id, courier_id, and category_code. These are linked to the store information table, factory information table, delivery person information table, and clothing category lookup table, respectively, and are used for display in the final reports and alert pages.

3. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 2, characterized in that: The batch ETL offline data collection architecture extracts data from the platform's order system, rewash management system, and complaint work order system at fixed intervals, providing complete and traceable offline data for periodic quality evaluations. Specifically: Input the business table and dimension dictionary table, perform incremental extraction every day at midnight, and perform a full data backtracking once a month at the end of the month; extract newly added or status-changed data from the previous collection period to the current time based on the timestamps create_time, finish_time, rewash_time, and complaint_time; use DataX or Sqoop tools, configure the database source connection information, and write the extraction script; the raw data is stored in the data warehouse ODS layered, stored by system and date partition, and the original fields are retained without modification; The real-time streaming acquisition architecture continuously monitors changes in business data, providing low-latency data for real-time quality monitoring and dynamic early warning. Specifically: The business table uses the unique order number (order_id) as the unique key for association, which is a unified primary key; the dimension dictionary table maps to the business table through foreign keys such as store_id, factory_id, courier_id, and category_code, and retains the original encoding without conversion during the data collection process. Enable binlog functionality for business tables to record additions, modifications, and deletions of data in the order master table, reprocessing record table, and complaint work order table. Use the Canal component to monitor and capture changes in the logs in real time. The captured real-time data is uniformly pushed to the Kafka message middleware for temporary caching, and then the Flink computing engine reads and consumes the data stored internally in Kafka. Real-time data for orders, reprocessing, and complaints is written to the data warehouse DWD layer and persisted in real time by the unique order number (order_id). A standardized raw data table is output as input for subsequent data cleaning.

4. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 3, characterized in that: Based on standardized raw data tables, and using the unique order number (order_id) as the globally unique join key, a left join is used to connect the main order table, the rewash record table, and the complaint work order table. Using the main order table as the base table, the order number (order_id) is matched against the rewash record table and the complaint work order table. If a match is found, the corresponding rewash information and complaint information are written; otherwise, the missing fields are filled with default values. Data with missing order numbers is directly removed, and for duplicate order numbers, the latest record is retained while redundant data is deleted. This results in a standardized order detail data table containing full order information, rewash marking information, complaint classification information, and business dimension coding information.

5. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 4, characterized in that: set up , Represents the set of all evaluation objects. Indicates the total number of evaluation objects; Indicates the first The evaluation targets include stores, factories, delivery services, and clothing categories; The evaluation period is ; Rewashing rate: ; in, Indicates the first time after cleaning The total number of rewash orders for each evaluation object, i.e. the number of orders with the rewash_flag set to 1; Indicates the first time after cleaning Total number of orders for each evaluated entity; Complaint rate: ; in, Indicates the first time after cleaning The number of complaint orders for each evaluation object, i.e. the number of orders with the complaint flag is_complaint set to 1; Average complaint severity ; in, This indicates the severity of the complaint associated with a single complaint order; If the first There were no complaints about any of the orders within the evaluation category. The average severity of complaints is uniformly filled in as 0; Regarding valid complaints that arose after rewashing, the percentage of complaints received after rewashing was [not specified]. ; in, Indicates the first The number of complaints after each evaluation object is rewashed, that is, the number of orders with the rewash flag set to 1 and the complaint flag set to 1; If the first The evaluated object had no repeat washing orders, that is... The percentage of complaints after rewashing Fill the value with 0% to avoid calculation errors caused by dividing by zero.

6. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 5, characterized in that: The metrics include positive and negative metrics; the positive metric is the total order volume; the negative metrics include the repeat washing rate, complaint rate, average complaint severity, and the percentage of repeat complaints after the repeat washing. For each indicator Iterate through all evaluation objects Calculate this indicator global minimum value and global maximum value ; Inverse index normalization formula: ; in, Indicates the first The evaluation object is in the first The original values ​​of each indicator; Represents the normalized i-th The evaluation object is in the first Standardized matrix for each indicator; This indicates that all evaluation objects are in the first... The maximum value of each indicator; Indicates that all evaluation objects are in the first... The minimum value of each indicator; Positive index normalization formula: ; If the maximum value and the minimum value of a certain indicator are equal ( To avoid division by zero errors, this indicator The normalization value of all objects is uniformly set to 1.0; All after normalization All of them are positive indicators, and each row represents an evaluation object. Each column represents a normalized positive indicator. This provides standardized input data for the subsequent TOPSIS comprehensive evaluation model; the larger the value, the more important it is for the evaluation object. In indicators The better the quality performance.

7. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 6, characterized in that: Based on the normalized standardized matrix Construct a comprehensive evaluation model for TOPSIS: set up For the first Each indicator weight satisfies ; Multiply each normalized index value by its corresponding weight to obtain a weighted matrix. : ； Each indicator The maximum value among all objects, i.e., the optimal performance, is the positive ideal solution. : ； ； Each indicator The minimum value among all objects, i.e., the worst performance, is the positive ideal solution. : ； ； The distance from each object to the positive and negative ideal solutions is calculated using Euclidean distance: Distance to the ideal solution ; Distance to negative ideal solution ; Calculate proximity ; in, Proximity The larger the value, the closer the object is to the positive ideal solution and the farther it is from the negative ideal solution, and the better its quality performance. For all evaluation objects, according to By sorting the evaluation objects from largest to smallest, a quality ranking is obtained, providing a quantitative basis for subsequent quality analysis and early warning.

8. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 7, characterized in that: The combined weights are designed using a dual-track approach: entropy weighting and expert scoring. The weights from both methods satisfy the constraint that the sum of the weights is 1. The entropy weight method is adopted, based on the standardized matrix. Calculate the information entropy and information utility value of each indicator: Calculate the first The first indicator The proportion of each evaluation object : ； in, The total number of evaluation objects; Calculate the first Entropy value under each indicator : ； ； like , ; Calculate information utility value : ； Calculate entropy weight ; An expert scoring method was used, where business experts scored the importance of the indicators, and the weights were obtained after normalization. Business, quality, and operations experts scored the importance of five indicators on a scale of 1 to 10; the average score for each indicator was calculated and then normalized to obtain its weight. ； Combination weights: setting combination coefficients ,according to Calculate the initial combined weights. If the sum of the combined weights is not 1, perform further normalization to make the sum of all indicator weights equal to 1; where the coefficients... Configure flexibly according to business needs; a larger value means more emphasis on expert experience, and a smaller value means more emphasis on objective data patterns. The resulting combined weights are used as input parameters for the TOPSIS model to conduct a multi-dimensional comprehensive quality evaluation.

9. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 8, characterized in that: The closeness of each evaluation object output based on the TOPSIS comprehensive evaluation model The rankings are the top N and bottom-ranked evaluation objects with the best overall quality across four categories: stores, factories, delivery personnel, and clothing categories. The system breaks down and analyzes low-performing targets to identify their weakest indicators and provides targeted improvement suggestions. It supports trend comparison by time slices, enabling year-on-year / month-on-month analysis. Through visualization formats such as bar charts, radar charts, line charts, and heat maps, the quantitative evaluation results are transformed into intuitive quality reports to support operational decisions and quality improvement.

10. The internet laundry quality evaluation method based on a comprehensive quality evaluation index according to claim 9, characterized in that: During the evaluation period Internally, based on the degree of relevance to each evaluation object. Using time series data as input, the exponentially weighted moving average (EWMA) algorithm is employed to implement dynamic quality early warning, including the following steps: For each evaluation object The proximity sequence, through Real-time updates of EWMA estimates; in, Indicates the object of evaluation In the EWMA estimate for the period; Represents the smoothing coefficient. ; Indicates the object of evaluation In the The closeness to the period; Indicates the initial value, taken from the evaluation object. Historical proximity mean; Set upper and lower control limits based on the standard deviation of the EWMA estimate: Upper control limit UCL: ; Lower control limit LCL: ; in, This represents the standard deviation of the EWMA value; The control limit coefficient is calculated as follows: ； in, The object of evaluation The mean of the historical proximity series; When the evaluation object No. The closeness to the target date is lower than the lower control limit. When a yellow alert is triggered, indicating a mild abnormality, continued observation is required; when it is continuous... When the current level falls below the lower control limit (LCL), a red alert is triggered, prompting the operator to intervene in a timely manner to prevent quality issues from escalating.

Images