موتور قاعده کارآمد برای سیستم ساختمان هوشمند Efficient Rule Engine for Smart Building System

۱ INTRODUCTION WITH the development of the Wireless Sensor and Actuator Networks (WSANs), smart building systems have been extensively studied in recent years [1], [2]. The primary objective of such system is to control electric appliances intelligently according to the environmental information collected by sensors for energy conservation in buildings. The smart control process is usually performed according to certain rules. The rules triggered by events can be expressed as the form of condition-action. For example, a rule can be described as “when someone works in the office with dim light, the corresponding lamp is turned on automatically”. In a smart building system, rule engine is an important component that can provide flexible control. The essence of a rule engine subsystem is to separate logics and data, so as to make logics as independent and maintainable parts. In a smart building system, detected environment data may be sound, image, temperature, smoke/gas concentration, humidity, etc. Sensors around a certain monitoring region collect environmental data and report them to the server within a regular sampling period. The server analyzes and processes the data for identifying events, matching the rules, and executing the corresponding actions. The events are often sudden environmental changes such as sound, light, fire (temperature, smoke concentration) and surface vibration. Generally, the frequency of reporting data is far greater than that of generating events. In order to filter a great deal of redundant data and improve the efficiency and accuracy of event generation, we design an effective event preprocessing mechanism according to static properties of the data itself (e.g., a geographical position, type of node, etc.). Take the rule – if Temperature > 60◦C, then an alarm sound – for example, we can filter the data from two aspects: 1) we filter the data reported by all the sensors except temperature sensors according to the type of node; 2) we filter the data that is collected beyond related monitoring region by the geographical position. In this way, the real-time performance of event generation is improved. With the development of smart building systems, the rapid expansion of events and rules cause the rule engine encounter two main problems: how to filter plenty of meaningless events and how to improve rule matching efficiency. In this paper, we consider dynamic factors (e.g., time and combinational conditions) to further promote the operation efficiency of rules. Considering that many rules are triggered by conditions which are composed of several events instead of a single one, it is crucial to design an efficient rule matching mechanism to promote the real-time performance of rule engine [3]. Many current business rule engines (CLIPS [4], JESS [5], DROOLS [6], BizTalk [7], etc.) are employed to provide better flexibility and reduce the cost of designing, developing and delivering software. The traditional algorithms, including the RETE for rule engine [8], [9], [10], mainly focus on the complex processing mechanism of rule engine with a large rule set and limited event throughput. However, in WSANs based smart building systems, thousands of deployed sensors and actuators produce abundant data. As shown in Fig. 1, in a WASNs based smart building system, many kinds of sensors are deployed for collecting environment information, each electric device is equipped with an actuator for receiving control commands, and each user subscribes multiple rules to customize required services. As a result, there are lots of events contributing to a large scale of rule set. In addition, many urgent events generated in smart building systems often have real-time response requirements. Existing rule engines mainly focus on traditional business scope and omit the problem of data load. Moreover, these engines are generally too heavy and complex to handle plenty of events, thus cannot be applied to a smart building system directly. On the other hand, traditional algorithms including the RETE cannot guarantee the quick matching between plenty of events and rules, and thus are not suitable for the system with lots of subscribed rules and produced events. In this paper, aiming at large-scale smart building systems, we propose an efficient rule engine with high data load and large rule set, which can match events and execute rules in real time. First, by analyzing the features of data in a smart building system, we find that although the reported data is abundant, the execution frequency of triggered rules is relative low. By filtering the unnecessarily processed data in time, we realize an efficient rule engine. In addition, with the increase of the scale of rule set, the rule conditions become more complex and rule executions are more frequently triggered. Hence, the performance of the rule engine can be further promoted by adjusting rule execution schemes dynamically according to the current system states. Our main contributions can be summarized as follows. • For the large scale smart building system containing abundant events and rules, we design a highefficient rule engine for quick matching between events and rules and rule execution. • We construct a minimal perfect hash table based on MPHF, in which the key set is composed of all the atomic trigger events. As an effective filtering table, the minimal perfect hash table discards the majority of unnecessarily processed data with only O(1) time overhead. Our proposed engine adaption scheme, based on the rule matching feedback, can significantly reduce the rule matching overhead adaptively. • We implement the proposed rule engine, and further verify it by a real smart building system. The experiment results show that our solution improve the performance of rule execution even with overwhelming data and large rule set. The remainder of this paper is organized as follows. We discuss the related work in Section 2. Section 3 describes the preliminaries. The proposed efficient rule engine is detailed in Section 4. Section 5 analyzes the operational complexity of our solution. In Section 6, we give our experimental study and simulations. We conclude the paper in Section 7.