Shulou(Shulou.com)11/24 Report--

In 2014, the application of electronic face order in the field of express delivery started the digitization of the industry, which has entered the 10th year so far. Like online car-hailing and takeout, express delivery has high data complexity and technology dependence, which requires real-time and precision, but the fulfillment link of express delivery is longer, and the service ecology is more complex. a high degree of automation also makes its "combination of software and hardware" and "combination of man and machine" more obvious.

Take Shentong as an example, 50 million parcels per day, 1 billion operations, and millions of customer service; the time limit for receiving and receiving the signature is less than 44 hours, but the average pre-sale and post-sale period is 14 days, and the amount of data has expanded to the order of 10 billion. The complex service ecology, the gathering of "people, cars, freight yards and machines" elements, and hundreds of thousands of practitioners in the network-facing such a scenario, to achieve the implementation of the contract with the price of a bottle of water and to control costs with "per cent" is inseparable from the breakthrough in digital intelligence. Every technical person in it will find it "interesting" and challenging.

The original intention of this article is to explore the future of express delivery technology through the review of the process of digital intelligence of express delivery, the practice and thinking of Shentong, to arouse more attention inside and outside the industry.

I. the starting point of digitization in the express delivery industry

The digitization of express delivery begins with the electronic face sheet. In 2014, the rookie typed out an electronic noodle order, and the following year the electronic noodle order gradually became popular in the express industry, replacing the traditional five-couplet order, making the industry leap from the "marker pen" to the digital age. The first prize of the first National Postal Industry Science and Technology Progress Award was awarded to the rookie electronic noodle list, which shows its contribution to the express delivery industry.

1.1 Electronic sheet-digitization

The advantages of electronic noodles in environmental protection, standardization of information entry and production cost are obvious, while the core innovation is the "three-segment code".

Before the three-paragraph code, the package roughly divides the destination by zip code and address. The postcode is a coarse-grained static geographical region description, because China has a vast territory, rapid urbanization and complex terminal area. Express sorting mainly depends on the address for manual sorting. Sorters need to memorize the corresponding area of each address and mark it with a marker, which is not only inefficient, but also very error-prone. The postcode has been unable to meet the fast-growing needs of express sorting and delivery.

The three-segment code with real-time self-learning and adaptive ability redefines the three-level delivery area of express delivery. according to the recipient address, the destination assignment code, destination network code and courier code are calculated dynamically and intelligently according to the recipient address, which greatly improves the information granularity. The three-segment code can quickly and accurately identify the destination of the package through big data's learning, thus sorting the package to the correct transport path, which greatly improves the sorting efficiency and reduces the missorting rate. At the same time, through the analysis of the behavior characteristics of couriers, the three-paragraph code can accurately identify its conventional distribution range, accurately match the corresponding parcels and complete efficient delivery.

1.2 Smart Operations-Automation

With the popularity of electronic noodles, the matching automatic sorting equipment has gradually become the mainstream of the industry. Distribution center cross-belt, DWS (three-in-one) and so on have become standard, and large outlets have also moved from manual to automation, solving the efficiency of package operation and greatly improving the efficiency and production capacity of the whole network.

Express operation includes transportation in addition to sorting. At present, express companies are also actively exploring the two major links of trunk lines and terminals, such as L3-level intelligent trunk driving carried out by Shentong and Debang. Cainiao, Zhongtong and other "the last 100 meters" unmanned vehicle development and application, Shunfeng, Yuantong and other exploration in the field of drones.

Second, the train of thought and practice of Shentong from digitalization to digital intelligence.

2.1Digitized 1.0 (2015-2018): small-scale trial run automatic sorting equipment

2.2 Digital 2.0 (2019-2022): from lean management to incentive integration

In 2019, Shentong and Cainiao reached a cooperation to strengthen digital information cooperation and promote the digital upgrading of Shentong. Since 2019, Shentong's technical team has expanded rapidly, and the "butler" product matrix has been launched one after another. Shentong completed the cloud business of the whole station in 2020, becoming the first company in the express delivery industry to use public cloud. As a result, Shentong has entered the digital 2.0 stage.

(1) lean management

When talking about digitization, we must talk about "lean" or "fine". One of the key points of Shentong at this stage is to do lean management.

At this stage, Shentong manages the cost, quality and timeliness to every link, everyone. The data communicate with each other layer by layer, the goal is clear, and the responsibility lies with people. Every link of the operation can be controlled through the four stages of planning, control, monitoring and assessment. This phase mainly focuses on planning, top-down control. At present, the construction of Shentong Digital 2.0 has been completed, and the operating cost of single ticket has dropped by 9.7% in 2021.

(2) incentive blending

Lean management does not solve all problems in a complex delivery ecology. The management of franchise express involves headquarters, provinces, centers, outlets, as well as a lot of social resources, in which the roles involved are cross-regional and cross-organizational. Every role involved is an economic rational person. In the stage of extensive management, Lean is a very good way of management, but it can not stimulate the inherent potential of rational people. The top-down game makes the information transaction cost become very large, at the same time, digital lean management is very easy to copy, can not become the core of differentiation.

With the integration of incentives, the core focuses on the reform of business management mechanism, changes the way of pure control, designs incentive compatible mechanisms, and increases incentives for individuals and small teams on the basis of the plan, so as to make the whole ecology more dynamic, such as team counting, loading rate incentive, gridding service, network direct operation, and so on.

Note: in the mechanism design theory founded by Hurwiez, "incentive compatibility" refers to: in the market economy, every rational economic man will have a self-interested side, and his individual behavior will act according to the rules of self-interest; if there can be an institutional arrangement to make the actor pursue individual interests, it coincides with the goal of enterprises to maximize collective value, this institutional arrangement is "incentive compatibility".

2.3 Digital 3.0 means Digital Intelligence: on-site decision-making (2022 to present)

In the digital stage, from lean management in the early stage to incentive integration in the later stage, we have completed a series of product construction, including operating platform Kunlun, operating platform, network product matrix, key customer butler, merchant product super merchant platform.

In particular, the reform of the management mechanism of "incentive integration" not only activates the vitality of the organization, but also magnifies the authority of the front line and on the spot, and puts forward very high requirements for the ability of individuals and small teams. At the same time, in the process of digitization, more and more data are online, and the scale and dimensions of data explode. Shentong has more than 300,000 employees and more than 4000 outlets, with transit direct operation and network joining as the main organizational structure, and its organizational complexity leads to a straight rise in data complexity. If we continue to use the management mode of the data 2.0 era, it is difficult to achieve accurate on-site decision-making, and it is easy to produce a large number of blind areas of decision-making. Therefore, our numerical intelligence focuses on solving the problem of "on-site decision-making", allowing people who hear the sound of gunfire to make decisions.

Third, towards the numerical intelligence of on-site decision-making, the core key competence and practice of Shentong.

The process of Shentong digital construction is jointly driven by "business + technology". If the digital products are not combined with business process reengineering and management mechanism reform, it will be a useless castle in the air.

Numerical intelligence is different, it is not strongly dependent on business process refactoring, but it has high technical requirements for engineering and algorithms. The system should not only find the problem, but also know the cause and solution of the problem.

The technical characteristics of the digital intelligence stage are real-time, intelligence, automation +, and the key capabilities are real-time computing and prediction, intelligent algorithms, deep integration of software and hardware, comprehensive AI empowerment, and data asset accumulation.

3.1 Real-time

Today, diagnosis based on historical data can help the business to improve iteratively, but it is difficult to support accurate on-site real-time decision-making. Sometimes real-time is not enough, the field resources, deployment and operations of the distribution center are difficult to change in real time, so it needs to have the ability of algorithm prediction.

-Prophet engine: real-time computing and prediction

Digital intelligence focuses on real-time on-site management and intelligent decision-making. Shentong builds a "prophet engine" to meet the above needs, which supports real-time data calculation of spatio-temporal index, has the ability to predict time, space and state, and truly achieves macro and micro "one account".

3.2 Intelligence

Intelligent algorithms: training and reasoning of large-scale and low-cost neural network models

If the neural network model is used on a large scale in the express delivery industry, is it anti-aircraft artillery to hit mosquitoes? The answer is no.

First of all, the neural network model is different from the previous machine learning algorithm, the biggest advantage is that it does not rely on feature engineering. It is an extremely complex process for a package to be collected and dispatched, in which a large number of features will be generated, and it is very difficult to select features manually. Therefore, in many application scenarios, neural network is the best choice.

Secondly, the rapid progress of pre-training large model technology has greatly reduced the threshold for the use of neural networks. Express companies do not need to build large-scale GPU clusters themselves, as long as they reuse the open source pre-training model and adapt or finetune in business. Therefore, the neural network model applied by express companies has fewer layers and smaller data magnitude than the top model, which can meet the demand with lower cost and high performance-to-price ratio.

We have carried out the application of neural network technology in the following scenarios, such as visual AI, intelligent customer service, complaint probability prediction, package value prediction and so on.

-- data assets: packages, locations, merchants, consumers, vision

The advantages and disadvantages of intelligent algorithm model depend on the algorithm, parameters, data scale and quality. In the industry, it is rare to optimize the model results through algorithm optimization or innovation, coupled with the mature development of neural network algorithms, the optimization space for feature selection and parameter tuning becomes smaller, so the quality of the model mainly depends on the scale and quality of the data.

Data as the most important factor affecting the quality of the algorithm model, data assets will become one of the core competitiveness of our future technology. In the field of express delivery, in addition to parcels, there are also regional, merchant, consumer and visual data.

Case 1 limitation control tower: saving a delayed bus

The aging control tower is a tool for on-site management and command of the transfer center. The data time granularity is refined to seconds, and the real-time computing complexity is very high. The limitation control tower is divided into two scenarios: outbound and inbound. Taking outbound as an example, the transit center can perceive in real time the delivery of the network, the central operation, the departure of the center, the status of the trunk line on the way, and the entry of the destination center. At the same time, the system will intelligently predict anomalies and recommend practical actions.

A real live case on April 3, 2023:

3:30 ●, Zhongshan Distribution Center: received a delay warning from the departure control tower. Through the GPS track and arrival forecast, the bus from Zhongshan to Hangzhou is expected to arrive at the center of Hangzhou by three and a half hours. The estimated arrival time is 6:02, and the deadline for delivery will be missed at 5:15. The dispatcher of Zhongshan Control Tower advised the driver to arrive as early as possible by phone through the system, so that it would be possible to recover.

● 5:30 Hangzhou Distribution Center: after the driver increased speed, the shuttle bus arrived in Hangzhou 30 minutes ahead of schedule. Although it also missed the deadline for delivery, it only missed 15 minutes, and there was still one hour before the deadline for delivery. The Zhongshan dispatcher saw this information in the system and communicated with the Hangzhou Center through the phone of the system, hoping to give priority to the timeliness of this train.

● 6:05 Hangzhou Distribution Center: the shuttle bus begins to unload.

● 6:30 Hangzhou Distribution Center: the deadline for Hangzhou first delivery is up. The train, which was delayed for three hours, grabbed the unloading capacity of half an hour through the information of the departure control tower, caught up with a delivery and saved the limitation of half a truckload of parcels.

Case 2 on-demand delivery: deliver the package to the consumer in the most comfortable way

On-demand delivery is a "mixed distribution" mode to realize the integration of differentiated delivery, personalized delivery and standardized delivery. Several forecasting models are used in on-demand delivery, such as package complaint probability prediction, consumer preference prediction and so on, in which data assets also play a key role.

How to dispatch a package depends on the consumer characteristics, package characteristics and delivery address. Previously, we have been focused on the consumer and the package itself, but in fact, the AOI characteristics of the delivery address are the key, such as the walking distance from the post station, whether there is an elevator, whether the doorman is easy to lose, and so on. After the implementation of on-demand delivery, the application of algorithms and data enables front-line couriers to get "stupid" delivery instructions from "applicant APP", which greatly improves the efficiency and quality of delivery.

3.3 Automation +

Automation has been widely used in the express delivery industry since 2015, and the intelligence of trunk lines and terminals has also developed rapidly. Different from general automation, automatic sorting and intelligent transportation in express delivery industry should fully consider efficiency, cost and self-adaptability. the ultimate goal of "Automation +" is to achieve the "Pareto optimization" of man-machine cooperation efficiency through deep integration of software and hardware and comprehensive empowerment of AI. If expressed by mathematical formula, (software × hardware) automation = automation +.

-- automatic sorting

Shentong has invested 10 billion yuan in infrastructure construction in the past three years, introduced a large number of advanced automation equipment, and increased the proportion of self-developed equipment.

 self-developed high-speed crossing belt, the sorting efficiency is increased by 30%, and the energy consumption is reduced by 50%.

The ultra-high speed cross-belt system is driven by permanent magnet synchronous motor, the maximum speed can reach 3.5 m / s, the efficiency is increased by 30%, and the energy consumption is reduced by 50%. Self-developed pneumatic grid, through the baffle automatic switching, can meet the needs of ultra-high-speed cross-belt grid, and does not increase the occupied area.

Software and hardware transformation of  transmission line to realize adaptive energy saving of equipment

Through the customization of Shentong standard modbus protocol and the application of DTU (Data Transfer Unit), the real-time status and load of the transmission line are analyzed in real time, and the working frequency of the transmission line is adjusted adaptively to achieve energy saving without affecting the operation efficiency and intelligently reduce equipment energy consumption.

High-precision UHF RFID application to accurately track the location and status of goods

RFID tag can mark a unique identification code for each piece of goods. By reading the RFID tag, it can accurately track the location, status, quantity and other information of goods, achieve accurate tracking of goods, and effectively reduce the loss of goods and the circulation cycle of goods. For example, through the RFID environmental protection bag instead of the traditional collection bag, the return package line can be realized instead of manual pulling bag, and the error rate of manual pulling bag can be reduced. At the same time, the environmental protection bag can be counted automatically and the life cycle of environmental protection bag can be managed visually.

-- autopilot

In recent years, with the continuous upgrading of autopilot technology at home and abroad, the technical application of autopilot in express trunk lines and terminal distribution is gradually moving towards commercialization.

● autopilot: since the second half of 2021, Shentong team has tested and investigated the application of self-driving technology on several trunk lines by working with a number of third-party companies to test and investigate the application of self-driving technology on trunk lines. It has run more than 300000 kilometers safely and can achieve 95% autopilot on highway sections. Autopilot can effectively reduce manpower and improve the transport efficiency of trunk vehicles. at the same time, in theory, it can greatly reduce the risk of traffic accidents and provide more protection for the safety of drivers.

Unmanned vehicle at the end of ●: the unmanned vehicle at the end has great restrictions on the distribution area. At present, we are conducting small-scale pilot projects in open roads and closed parks in cities such as universities, as well as from outlets to post stations. The intelligent unmanned express runs at a speed of less than 30 kilometers and can carry about 800 pieces of express delivery. The vehicle has a range of up to 150 kilometers and can run 24 hours a day.

In the process of the centralization and automation of the express delivery industry, Shentong started early but lagged behind after several twists and turns, which led to the company's difficult situation. In the last three years, we have seized the industry opportunity of the transition from digitization to digital intelligence and focused on infrastructure construction with the determination to change planes for engines. Through the combination of lean management and incentives, the business can pick up quickly and start running.

Now, Shentong is continuing to explore the application of real-time and intelligence in the field of express delivery, looking forward to becoming Shentong's differentiated competitiveness and bringing more inspiration to the industry in terms of digital intelligence.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.