A new track for government digital governance: How power grids generate user "health codes"

A more core part of data modeling is "finding feature points", which relies on human experience and insight.

It is understood that the characteristics of "scattered, disorderly and polluted" places are small scale, large number, strong concealment, and the distribution pattern is mostly concentrated in urban villages and industrial parks. In this regard, two typical settings based on feature points were constructed in the "scattered, disorderly and polluted" data governance case:

The first is to define the monthly electricity consumption threshold for “scattered, disorderly and polluting” places.

"If the monthly electricity consumption of ordinary residential users exceeds 600-800 kWh, it is already high. If it exceeds too much, it is a typical abnormal residential electricity consumption."

According to Yuan Chao, in 2018, based on the electricity consumption characteristics of typical "scattered, messy and polluted" places and the experience of government staff, the electricity consumption threshold was temporarily set at 3,000 kWh. In 2019, based on the electricity consumption characteristics of users in various regions of Guangzhou, differentiated electricity consumption thresholds were formulated, which can help to identify illegal operations disguised as residents.

In the information provided to the competent authorities, it is sufficient to confirm that the list of users with more than 3,000 kWh is displayed, without providing specific electricity consumption values. "This ensures the data security and privacy of users," said Yuan Chao.

The second is to find it based on the characteristics of the electricity load curve.

In addition to "quantity", "curve trend" is another dimension of judgment. The electricity load characteristics of "scattered, messy and polluted" users are significantly different from those of ordinary residential users. The electricity consumption curve of residential households has obvious peaks, valleys and flat periods; "scattered, messy and polluted" electricity consumption belongs to production-type electricity consumption, and the load is relatively stable throughout the day.

According to Yuan Chao, in the "scattered, disorderly and polluted" monitoring system, conducting investigations through electricity consumption data is only a means of big data analysis. In order to further improve the accuracy of the data model, the data screening mode sets a keyword index. By scanning the user's profile information, users who are obviously not "scattered, disorderly and polluted" places are removed from the suspected list. At the same time, based on past experience, more stringent standards have been formulated for screening in some river basins and some urban villages to reduce the number of fish that slip through the net.

//

Multi-party balance under algorithmic fairness

//

However, the inevitable human setting of algorithms and rules brings up another problem - algorithmic fairness.

Last July, a report by RAND Corporation titled “Algorithmic Fairness – Framework for Social Application” pointed out that data-driven algorithms mean that the algorithm’s commitment to objectivity is better described as a commitment to consistency. And this consistency does not mean absolute correctness or lack of bias.

In the treatment of scattered and disorderly pollution in Guangzhou, a monthly electricity consumption threshold of 3,000 kWh was designed, but the question that follows is, "Is 2,800 kWh not enough? Is 2,900 kWh a problem?"

In this case, the setting of the threshold involves the control of the tightness of the regulatory scale, as well as the summary of the rules and professional judgment of rationality. As the project executor, the power department needs to combine the assessment of the law enforcement degree of the competent department when giving opinions based on the characteristics of power load.

"In the process of building the data model, we are responsible for collecting their (district authorities') business requirements and organizing data engineers and development engineers to implement relevant functions," Yuan Chao said. "Based on the requirements, we will provide technical support for the model and details, but we will not decide the standards."

Yuan Chao said that in defining the standards, "when we first built the model, we fully accepted the feedback from staff from various towns and streets."

In the early stages of rule-setting, more extensive discussions, coordination, and finalization of details and consensus can avoid subsequent disputes to a certain extent.

Algorithms are based on human judgment and theoretically can be manipulated. Just like the problem of "price collusion" in antitrust enforcement, there is also the possibility of "algorithm collusion" in the field of digital supervision.

"The screening criteria proposed by town and street staff must also be approved by the relevant government departments." Yuan Chao also mentioned this point.

It is difficult for an algorithm to be 100% fair, but it can be adapted to local conditions as much as possible and take into account differentiated objective realities.

The setting of thresholds varies according to the regional economic development level and industrial structure. Taking Guangzhou as an example, the overall load level of Zengcheng and Conghua districts, which are located in the suburbs of the city, is relatively low, so the threshold for judging "scattered and disorderly pollution" has been lowered accordingly.

跨地域的用电特征差异更为显著。“拿广州的标准到东莞、沈阳，模型可能就跑不动了，需要结合当地的产业结构建模。”袁超表示。

According to reports, after screening out a list of suspected "scattered, disorderly and polluted" places, the list will be sent to the mobile phones of grassroots staff, and then the investigation tasks will be configured in the form of "scanning the code to grab the order". The already grabbed work orders will not appear again to avoid repeated investigations. After verifying the identity of the venue, production equipment and other information on site, the staff will input the evidence data through the mini program, and after approval by the town, district and city level, the "scattered, disorderly and polluted" venue will be determined.

This is equivalent to completing a bottom-up "capillary" environmental governance. In this model, supervision and accountability of grassroots governance are crucial, and digital traceability can play a role.

One of the reasons why "scattered, messy and dirty" recurs in some local areas is that some grassroots personnel are partial to each other and commit fraud when investigating and handling on the spot, and the higher authorities have no way to supervise. The solution to this case is that when grassroots personnel grab orders through their mobile phones, they have actually bound the responsibility of verification to personal information. Who is responsible for each point can be traced in the background.

Another point worth noting is that from "scanning the code to grab the order" to accurately locating the corresponding place according to "mobile phone navigation", grassroots personnel need to accurately match power data with geographic location data. This involves the issue of how to connect cross-border data standards and specifications.

In the case of "scattered, disorderly and polluting" governance, the local power department promoted the "address structuring" work in 2017, which kept the power user data consistent with the standard house number registration of the public security system, thereby realizing the function of fixed-point navigation based on electricity consumption data.

In addition, digital technologies and tools are updated at a high frequency. How to enable grassroots staff to master operational skills without making mistakes is also a realistic issue facing us.

//

The government’s “tandem function” is difficult to replace

//

When the "scattered, disorderly and polluted" data governance was initially designed, consideration was given to introducing utility data such as telecommunications data, electricity data and water data to implement multi-dimensional monitoring.

The idea proposed in the "Action Plan for the Remediation of "Scattered, Disorderly and Polluted" Sites" is to integrate data information such as electricity, water and communications, and build a big data supervision system for "scattered, disorderly and polluted" sites. Big data will be used to conduct statistics, analysis and monitoring of water and electricity consumption in various districts, towns, villages and industrial parks, and to issue early warnings for abnormal situations.

This will involve cross-border collaboration in at least three major industries, but in practice, "data breaking" across fields and enterprises has not yet been achieved.

Compared with the water system, which has a relatively loose organization and lags behind in intelligent data collection, the telecommunications and power sectors have a better foundation for data collection, aggregation and governance, and both have dedicated data analysis and processing teams, which can provide a foundation for rapid investment in government applications. However, this has led to a new problem: who should be selected to coordinate data modeling and analysis?

At this stage, the government department requires enterprises to provide data analysis results or desensitized information, and will not directly touch the specific data; while large infrastructure providers have little motivation to share their own collected data with another large-scale public utility company for free. This is not only due to the constraints of data security and industry regulations, but also the consideration of current corporate strategy.

Given that the “scattered, disorderly and polluted” screening work is aimed at “scenario-based users”, the energy consumption reflected by the power load data can better judge the real production and living conditions of “fixed-site” users compared to the mobile trajectory reflected by the signaling data provided by the telecommunications department. In the initial stage of the project, only power data was used to advance.