Artificial intelligence application of computer-aided scheduling system

By introducing intelligent proxy technology into computer-aided scheduling systems, using artificial intelligence and machine learning to independently analyze data, the problem of the system being difficult to make scheduling decisions quickly in emergencies is solved, and more efficient and accurate scheduling assistance is achieved, improving System responsiveness.

CN119090190APending Publication Date: 2024-12-06INTERGRAPH CORP
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411098751.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2018-06-12
Filing Date
2019-06-10
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Existing computer-aided scheduling systems are difficult to make scheduling decisions quickly and accurately in emergencies, especially when dealing with complex and variable sequences of emergency events, which requires relying on users to manually filter large amounts of data, which is inefficient.

Method used

Using intelligent proxy technology, using artificial intelligence and machine learning to independently analyze data in computer-aided scheduling systems, generate useful notifications and recommendations, and help schedulers make decisions, including statistical agents, keyword agents, pattern agents, event matching agents, A combination of intelligent agents such as correlation agents, duplicate event agents and similarity agents.

Benefits of technology

It improves the efficiency and accuracy of scheduling decisions, reduces the burden of users to screen data, can handle complex emergency events in real time, provides automated scheduling assistance, and significantly improves the system's response capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090190A_ABST
    Figure CN119090190A_ABST
Patent Text Reader

Abstract

The invention discloses artificial intelligence application of a computer-aided scheduling system. Exemplary embodiments of the present invention provide a virtual dispatch assist system in which various types of intelligent agents (e.g., as part of a new CAD system architecture or as plug-ins of existing CAD systems) are deployed to analyze a large amount of historical operational data and provide various types of dispatch assist notifications and recommendations, for example, as part of a new CAD system architecture or as plug-ins of existing CAD systems. These scheduling assistance notifications and recommendations may be used by the scheduler or the CAD system itself (e.g., autonomously) to make scheduling decisions.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] This application is a divisional application of the invention patent application with the original application number 201980053820.8 (International application number: PCT / US2019 / 036368, application date: June 10, 2019, invention name: Artificial Intelligence Application of Computer-Aided Dispatch System). Technical Field

[0002] The present invention relates to a computer-aided dispatching system, and more particularly, to a computer-aided dispatching system that utilizes artificial intelligence to assist in making dispatching decisions. Background Art

[0003] In an emergency situation, time is of the essence when dispatching emergency responders. Emergency responders must not only be able to get to the incident location quickly, but they must also be able to dispatch the right type of emergency responder to handle the unique circumstances of the incident. For example, a hostage situation may require an emergency responder with hostage negotiation skills, while a potential drowning situation may require an emergency responder with water rescue skills.

[0004] A computer-aided dispatch (CAD) system is a computer system running specialized software that makes or assists dispatchers in making dispatch decisions for various types of assets: for example, personnel (e.g., police officers, firefighters, EMTs and other medical professionals, couriers, field service technicians, etc.), vehicles (e.g., police cars, fire trucks, ambulances, medical flight helicopters, boats, livery vehicles, etc.), equipment (e.g., special equipment required to respond to specific events or situations, such as "jaws of life" for car accidents), and facilities (e.g., hospitals or other medical facilities that can handle specific types of patients). For convenience, the term "unit" may be used herein to refer to one or more emergency responders and related equipment dispatched to an incident (e.g., a police unit may include a specific police car and one or more police officers who are using the police car).

[0005] CAD systems typically collect and store a large amount of information about various assets, such as status, availability, location, capabilities, and usage history. CAD systems typically also collect and store information about various emergency events, such as the location and type of emergency events, etc. Such information can be collected and stored in this way: automatically (e.g., through any of a variety of computer-based communication systems) and / or by operators who collect information and input the information into the CAD system. The CAD system can then help the dispatcher assign tasks to emergency responders, for example, by recommending which (which) emergency personnel and / or vehicles will be assigned to a specific emergency event based on criteria such as the following: the type of emergency, the distance of the emergency responder from the location of the emergency, the status of each available emergency responder (e.g., whether a specific emergency responder is currently responding to an emergency), the necessary equipment for the emergency (e.g., life-saving clamps), the necessary skills for the emergency (e.g., suicide negotiation skills, water rescue skills, etc.) or the minimum turn (for ladder fire trucks), etc.

[0006] After responding to an emergency, it is often necessary to transport a party to a facility, such as a hospital or detention center, for example. Therefore, in addition to tracking emergency personnel and vehicles, some CAD systems also collect and store information about each of the many available facilities, such as the type of facility, the location of the facility, and the services provided by the facility (e.g., general first aid versus specialized treatment), etc. The CAD system can then help the dispatcher assign emergency responders to the emergency and then to the facility, for example, making facility recommendations based on criteria such as the type and severity of treatment required by the patient or victim, the capabilities of the available facilities, and the proximity of the available facilities to the location of the emergency, etc. For example, the dispatcher usually prefers to assign an ambulance to the nearest hospital, but in some cases, the CAD system may recommend a second hospital that is farther away (e.g., the nearest hospital may only provide general first aid, while the patient in the ambulance may need specialized treatment such as cardiac care that can only be provided by the second hospital that is farther away).

[0007] Based on the large amount of data collected by the system (e.g., the current disposition and condition of various personnel, vehicles, equipment, and facilities that are candidates for handling the situation), the dispatch decision may not only be very complex, but also such dispatch decisions usually need to be made quickly. Generally speaking, the information from the call is input into the CAD system using a specified graphical user interface screen, and the CAD system can process the input information and other information stored in the system to make recommendations to the dispatcher. Such recommendations are usually based on rules specially programmed by a programmer, for example, the CAD system can recommend the nearest ambulance to respond to the incident and transport the patient to the nearest hospital. However, in some cases, the nearest ambulance may not be the best responder (e.g., the medical staff has not been trained to handle burn victims, or the ambulance is not properly equipped to handle burn victims), and the nearest hospital may not be the best facility to treat the patient (e.g., the nearest hospital may not be able to accept new patients, or may not have the right personnel available to treat burn victims). Moreover, the CAD system must be able to handle very complex sequences of related or unrelated events, for example, a series of thefts that occurred over a period of several days. Furthermore, the CAD system can use rule-based logic to provide recommendations to the dispatcher. Summary of the invention

[0008] According to one embodiment of the present invention, a computer-aided scheduling (CAD) system includes: a CAD database storing CAD data; and at least one server, wherein the at least one server includes a tangible non-transitory computer-readable medium storing an agent host subsystem and a notification subsystem. The agent host subsystem is configured to communicate with a plurality of intelligent agents, wherein each intelligent agent is configured to autonomously perform different scheduling-related analyses of the CAD data and generate scheduling-related notifications based on such analyses without user queries. The notification subsystem is configured to determine whether to present a given notification from a given intelligent agent to the user based at least in part on feedback received from the user.

[0009] In various alternative embodiments, the intelligent agent may include a statistical agent configured to detect outliers in the CAD data and generate a notification when an outlier is detected. The statistical agent may include a machine learning outlier detector trained to detect outliers in the CAD data.

[0010] Additionally or alternatively, the intelligent agent may include a keyword agent that generates a notification when at least one specified keyword is detected in an event.

[0011] Additionally or alternatively, the intelligent agent may include a pattern agent configured to detect patterns in the CAD data and generate notifications when patterns are detected in events.

[0012] Additionally or alternatively, the intelligent agent may include an event matching agent configured to detect similar events based on fuzzy logic analysis and generate notifications when similar events are detected based on the fuzzy logic analysis. The event matching agent may include a machine learning pattern detector trained to detect patterns in CAD data.

[0013] Additionally or alternatively, the intelligent agent may include a recurrent event agent configured to detect recurrent events and generate notifications when recurrent events are detected.

[0014] Additionally or alternatively, the intelligent agent may include a correlation agent that generates a notification when a combination of variables is determined to contribute to the occurrence of an event. The correlation agent may include a machine learning correlation detector that is trained to detect correlations in the CAD data.

[0015] Additionally or alternatively, the intelligent agent may include a similarity agent configured to detect similar events based on word frequency and generate a notification when similar events are detected based on word frequency.

[0016] Additionally or alternatively, the intelligent agent may include a rule agent that generates notifications when operational variables match preconfigured rules.

[0017] In any of the above embodiments, the notification presented to the user can provide a mechanism for the user to provide feedback about the notification. The system can allow notifications to be shared between users and / or with non-users. The system can attach notifications to related events. The intelligent agent can be implemented by at least one server (e.g., on-site) and / or can be implemented remotely from at least one server (e.g., via a cloud-based machine learning service).

[0018] In any of the above embodiments, the agent host subsystem, one or more intelligent agents, and / or notification subsystem may be embodied as a computer program in a computer program product including a tangible, non-transitory computer-readable medium.

[0019] The embodiments described and claimed herein have the effect of transforming a CAD system that may be passive or may require a user to sift through large amounts of data into a virtual dispatch assistance system that utilizes an intelligent agent that can utilize artificial intelligence and machine learning to proactively analyze CAD data and generate useful notifications to assist the dispatcher or automate dispatch operations. For at least these reasons, the activities described and claimed herein provide a technical solution (e.g., utilizing artificial intelligence and machine learning) to solve problems that arise directly in the technical field (e.g., computer-aided dispatch systems) in a manner that is not fully understood, routine, or conventional by those skilled in the art.

[0020] Additional implementations may be disclosed and claimed. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Advantages of various embodiments of the present invention will become more fully apparent to those skilled in the art from the following "Detailed Description" discussed with reference to the drawings summarized immediately below.

[0022] Figure 1 is a simplified schematic diagram of a computer-aided dispatch (CAD) system in accordance with certain exemplary embodiments.

[0023] Figure 2 is a schematic diagram illustrating a functional model of a virtual dispatch assistance system with an intelligent agent according to certain exemplary embodiments.

[0024] Figure 3 is a diagram showing a method according to some exemplary embodiments Figure 2 Schematic diagram of the data model of a virtual scheduling assistance system with intelligent agents.

[0025] Figure 4 An example notification format is shown.

[0026] Figure 5 Another example notification format with more detailed information and an OnCall map is shown.

[0027] Figure 6 Sharing notifications according to certain exemplary embodiments are shown.

[0028] Figure 7 A graphical user interface for attaching a notification to an event is shown, in accordance with certain exemplary embodiments.

[0029] Figure 8 A notification history list is shown in accordance with certain exemplary embodiments.

[0030] Fig. 9 is a schematic diagram illustrating a genetic algorithm workflow according to certain exemplary embodiments.

[0031] Fig.10 An example of a multiplication (crossover) operation is shown, in which the two fittest individuals (ie, parents) are split into two parts (the cut point may be random), and the resulting parts are combined so that each offspring includes a segment from each parent.

[0032] Fig.11 An example of a mutation operation is shown, where a portion of the elite population is subjected to mutations that may be random or rule-based.

[0033] Fig.12 A fuzzifier function that determines whether two events are close to each other is shown according to an exemplary embodiment.

[0034] Fig.13 A fuzzifier function for determining whether two events occur simultaneously is shown according to an exemplary embodiment.

[0035] Fig.14 is a graphical view of the results of a correlation analysis showing the averages of events for all codes correlated by day of the week (DOW) and a selected 20% influence range for TYCOD=4 (Assault / Sexual Violence) cases according to an exemplary embodiment.

[0036] Fig.15 is a graphical view of the results of a correlation analysis according to an exemplary embodiment, showing the average of events for all codes correlated by time of day (Hour of Day) and a selected 20% impact range for TYCOD=4 (Assault / Sexual Violence) conditions.

[0037] Fig.16 A fuzzy logic function for determining whether to show a notification to a user is shown according to one exemplary embodiment.

[0038] Fig.17 The main workflow of the StatisticAgent according to an exemplary embodiment is depicted.

[0039] Fig.18 Example baselines for the variables are shown.

[0040] Fig.19 An example of an outlier notification with statistical data is shown.

[0041] Fig. 20 An example updated baseline for the above example is shown.

[0042] Fig.21An example of related events shown in a map is shown, including the ability for a user to display event information such as identifying the original event and related events.

[0043] Fig. 22 A more detailed example of similarity analysis according to an exemplary embodiment is shown.

[0044] Fig.23 The architecture of EventClassificationAgent according to an exemplary embodiment is shown.

[0045] Fig.24 A detailed example of event classification is shown.

[0046] Fig.25 Another example of event classification is shown.

[0047] Fig.26 An architecture for speech-to-text integration according to an exemplary embodiment is shown.

[0048] Fig. 27 An example is shown of how a ClusteringAnalysisAgent may help identify an optimal location to place a trailer based on a history of towing events, according to an exemplary embodiment.

[0049] Fig.28 An example graphical user interface screen showing an optimized route is shown according to an exemplary embodiment.

[0050] Fig.29 A dedicated regression graph for a time estimation agent according to an exemplary embodiment is shown.

[0051] Fig.30 An example graphical user interface screen showing staged locations of groups is shown according to an example implementation.

[0052] Fig.31 An example coverage criteria definition is shown according to one exemplary embodiment.

[0053] Fig.32 An example Location Redundancy Details screen is shown according to an example implementation.

[0054] Fig.33 An example of area coverage relocation according to an exemplary embodiment is shown.

[0055] Fig.34 An example CAD system screen is shown showing various types of notifications overlaid on a map.

[0056] Fig.35Some example notifications are shown in more detail.

[0057] Fig.36 Example similarity notification details are shown.

[0058] Fig.37 Example mode notification details are shown.

[0059] Fig.38 Example keyword notification details are shown.

[0060] Fig.39 is a schematic diagram showing the relationship between a CAD scheduling server (referred to herein as OnCall 10), a SmartAdvisor AgentHoster, and a CAD scheduling client workstation according to an exemplary embodiment.

[0061] Fig.40 A SmartAdvisor AgentHoster graphical user interface screen is shown according to an exemplary embodiment.

[0062] Figures 41 to 51 A class model of SmartAdvisor and various intelligent agents according to an exemplary embodiment is shown.

[0063] It should be noted that the aforementioned figures and the elements depicted therein are not necessarily drawn to consistent scale or in any scale. Unless the context dictates otherwise, the same elements are represented by the same figure numerals. DETAILED DESCRIPTION

[0064] definition

[0065] As used in this specification and the appended claims, the following terms shall have the indicated meanings unless the context requires otherwise.

[0066] Generally speaking, rule-based logic is a computer program in which decision logic is specifically programmed by a programmer to produce specific outputs based on specific types of inputs. For a simple example of rule-based logic, a CAD system might include a rule that the closest available responder should respond to an event.

[0067] Generally speaking, machine learning (ML) is the application of artificial intelligence (AI) to provide software systems with the ability to learn and improve themselves based on experience without being explicitly programmed. Instead, programmers model the problem to be solved rather than programming the solution itself, and the ML model is trained to perform a given task based on sample data (called "training data") in order to learn how to produce a specific type of output based on a specific type of input. Therefore, machine learning is useful for difficult problems that may have many feasible solutions. ML solutions may exhibit unexpected behavior due to artificial intelligence and may be improved over time based on additional training data accumulated over time. Computer programs that use machine learning may be referred to as intelligent agents in this article.

[0068] introduce

[0069] Artificial Intelligence (AI) and Machine Learning (ML) are being used more frequently today across a wide range of technologies and applications. TM A list of exemplary applications of machine learning is provided, including agriculture, anatomy, adaptive websites, affective computing, banking, bioinformatics, brain-computer interfaces, cheminformatics, computer networks, computer vision, credit card fraud detection, data quality, DNA sequence classification, economics, financial market analysis, general gaming, handwriting recognition, information retrieval, insurance, Internet fraud detection, linguistics, machine learning control, machine perception, machine translation, marketing, medical diagnostics, natural language processing, natural language understanding, online advertising, optimization, recommender systems, robotic locomotion, search engines, sentiment analysis, sequence mining, software engineering, speech recognition, structural health monitoring, grammatical pattern recognition, telecommunications, theorem proving, time series forecasting, and user behavior analysis. It should be recognized that ML programs often fail to provide expected results, for example, due to lack of suitable data, data bias, and improper choice of tasks and algorithms, among others.

[0070] The inventors of the present invention contemplate using AI and ML in a computer-aided dispatch (CAD) system to provide real-time, embedded analytics that can be used by the dispatcher or by the CAD system itself (e.g., autonomously) to make dispatch decisions. An obvious difficulty in applying AI and ML to CAD systems is to translate the theoretical advantages of AI / ML into a concrete implementation, i.e., to provide insightful and practical outputs for making dispatch decisions, and to use dispatcher feedback on an ongoing basis to improve learning and future outputs. Moreover, since dispatch decisions typically involve a wide range of potential event types (involving complex interrelationships between the types of data collected (e.g., data about events, event outcomes, personnel, vehicles, equipment, facilities, etc.)), AI / ML implementations need to be flexible to allow the incorporation of new types of data sources and new types of outputs that can be used for dispatch decisions. Therefore, the present invention is not merely equivalent to "applying AI / ML to a CAD system", but rather provides specific methods that can use AI / ML in a CAD system to improve dispatch decisions.

[0071] Figure 1 1 is a simplified schematic diagram of a computer-aided dispatch (CAD) system 100 according to certain exemplary embodiments. Among other aspects, the components of the CAD system 100 can be logically divided into multiple subsystems, including a CAD server subsystem 110, a dispatcher subsystem 120, and an optional personnel subsystem 130. Implementations of the exemplary CAD system 100 can (and typically do) have multiple instances of each subsystem. For example, the exemplary CAD system 100 may include: a personnel subsystem 130 for each emergency responder, multiple dispatcher subsystems 120 that enable multiple emergency dispatchers to process emergency calls and alarms within a single dispatch center and / or across multiple dispatch centers, and multiple CAD server subsystems 110 at one or more data centers that provide matters such as high availability services and load balancing.

[0072] The CAD server subsystem 110 includes: a CAD scheduling server 111 connected to one or more dispatcher subsystems 120 via a first communication network, a mobile responder server 112 connected to one or more personnel subsystems 130 via a second communication network, a database system 113 (in which the CAD scheduling server 111 and / or the mobile responder service 112 maintain various types of data), and optionally a machine learning service 114 that utilizes machine learning to track data over time (the machine learning service can be part of the CAD scheduling server 111, or can be separate from the CAD scheduling server 111). As discussed in U.S. Patent Application No. 15 / 384,874, incorporated by reference above, (e.g., the CAD scheduling server 111) can query the optional machine learning service 114 for analytical data. The machine learning service 114 can be implemented on-site with the CAD scheduling server subsystem 110, or can be implemented remotely, for example, based on the cloud.

[0073] The dispatcher subsystem 120 includes a CAD dispatch client workstation 121 used by the emergency dispatcher to manage dispatch-related information. For example, the dispatcher can use the CAD dispatch client workstation 121 to enter information about incidents and emergency responders, view the status of incidents and responders, dispatch emergency responders, and interact with emergency responders, such as by providing updated incident information, among other things.

[0074] As discussed in U.S. Patent Application No. 15 / 384,874, incorporated by reference above, the optional personnel subsystem 130 includes a monitoring device 131 (e.g., a wristband-type device) that can be worn by an emergency responder and a mobile responder client device 132 (e.g., a smartphone-type device running a specially configured mobile responder client application). The wearable monitoring device 131 communicates (e.g., "pairs") with the mobile responder client device 132 via a wireless communication system (e.g., Bluetooth or WiFi). The mobile responder client device 132 communicates with the CAD server subsystem 110 via a second communication network, which may include a wireless communication system such as a cellular phone / data network.

[0075] The mobile responder client device 132 is specifically configured (e.g., using dedicated hardware and / or a purpose-built client application) to communicate various types of information between the wearable monitoring device 131 and the CAD server subsystem 110, although certain types of information may be exchanged between the mobile responder client device 132 and the CAD server subsystem 110 to the exclusion of the wearable monitoring device 131. Among other things, as discussed in more detail below, the mobile responder client device 132 may communicate information from the wearable monitoring device 131 and / or the mobile responder client device 132 itself to the CAD server subsystem 110, e.g., to enable monitoring of emergency responders, and may also receive information from the CAD server subsystem 110 for presentation to the emergency responders on the mobile responder client device 132 and / or on the wearable monitoring device 131 via communication from the mobile responder client device 132. The mobile responder client device 132 may provide a dedicated graphical user interface through which the emergency responders may send and / or receive various types of information.

[0076] According to certain exemplary embodiments, the wearable monitoring device 131 includes one or more interface devices to collect status information about the emergency responder (referred to herein as "biometric" information for convenience). For example, the wearable monitoring device 131 may include one or more of a heart rate sensor, a skin temperature sensor, a galvanic skin response sensor, a blood oxygen level sensor, and / or other sensors for collecting biometric information. The wearable monitoring device 131 sends the information obtained from the interface device to the mobile responder client device 132, which can be configured to process the information and / or send the information to the CAD server subsystem 110 for processing.

[0077] Wearable monitoring device 131 and / or mobile responder client device 132 may include other types of interface devices, such as, for example, a microphone (e.g., a microphone that enables an emergency responder to speak to an emergency dispatcher or is used to detect gunfire or other sounds), a speaker (e.g., a speaker that enables an emergency dispatcher to speak to an emergency responder or an audible alarm is issued to the emergency responder by CAD server subsystem 110 or emergency dispatcher), a camera (e.g., a camera that enables an emergency responder to record pictures or videos for evidentiary purposes and / or to send to CAD server subsystem 110 or emergency dispatcher), a tactile output device such as a vibrator device (e.g., a tactile output device such as a vibrator device (e.g., a tactile output device such as a vibrator device) The server subsystem 110 or emergency dispatcher issues a tactile alert to the emergency responder), a "team emergency alert" input (e.g., a button that enables the emergency responder to issue an alert to the CAD server subsystem 110 or emergency dispatcher), a motion sensor such as an accelerometer or gyroscope (e.g., the motion sensor monitors whether the emergency responder is moving or stationary), a posture sensor (e.g., the posture sensor monitors whether the emergency responder is standing upright or lying down), a temperature sensor (e.g., the temperature sensor monitors the temperature of the environment in which the emergency responder is working), and / or a location sensor such as a GPS sensor (e.g., the location sensor provides location information to the emergency dispatcher), etc. Information obtained from such interface devices can be processed by the mobile responder client device 132 and / or sent to the CAD server subsystem 110 for processing.

[0078] Thus, for example, the first responder may be monitored by information obtained solely from wearable monitoring device 131 , or may be monitored by a combination of information obtained from wearable monitoring device 131 and information obtained from mobile responder client device 132 .

[0079] The mobile responder client device 132 can send information to the CAD server subsystem 110 periodically (e.g., via a Web service API) or upon request from a responder or scheduler. Each paired wearable monitoring device 131 and mobile responder client device 132 is uniquely identifiable to the CAD server subsystem 110, and security is preferably implemented, for example, using HTTPS-based communications between the mobile responder client device 132 and the CAD server subsystem 110 to prevent data leakage.

[0080] The CAD server subsystem 110 maintains various types of information in the database system 113. For example, the CAD server subsystem 110 maintains information about various types of events that may occur, information about each event that does occur, and information about each emergency responder, among other things.

[0081] Exemplary embodiments of the present invention provide a virtual dispatch assistance system (sometimes referred to herein as "SmartAdvisor") in which various types of intelligent agents (sometimes referred to herein as "SmartAdvisor agents") are deployed, for example as part of a new CAD system architecture or as plug-ins to an existing CAD system, to analyze large amounts of historical operational data in a database system 113 and provide various types of dispatch assistance notifications and recommendations that can be used by the dispatcher or the CAD system itself (e.g., autonomously) to make dispatch decisions. The inventors foresee that the virtual dispatch assistance system and related intelligent agents that employ machine learning, artificial intelligence, and real-time analysis will provide significant improvements and key market differentiators over traditional CAD systems and will be a springboard for more advanced features, such as cloud-enabled features and real-time (even proactive / predictive) crime centers and "safe city" applications. In an exemplary embodiment, the intelligent agents run autonomously in the CAD server subsystem 110 and / or the optional machine learning service 114, where each intelligent agent is configured to autonomously and proactively perform different dispatch-related analyses of data in the CAD database and generate dispatch-related notifications based on such analyses without user queries. In an exemplary embodiment, notifications are processed on a per-user basis by an optional machine learning filter so that only notifications relevant to a particular user are presented to that user, although alternative embodiments may provide some or all of the notifications from the intelligent agent to the user in an unfiltered manner. Thus, the intelligent agent and optional notification filters provide input to the scheduling decision process in a manner that the inventors believe has not heretofore been available in CAD systems.

[0082] Figure 2 is a schematic diagram illustrating a functional model of a virtual scheduling assistance system with a smart agent and optional notification filters according to certain exemplary embodiments. The virtual scheduling assistance system includes, among other aspects, four main subsystems, namely, a SmartAdvisor input subsystem, a SmartAdvisor commons subsystem, a SmartAdvisor agent subsystem, and a SmartAdvisor notification subsystem.

[0083] The SmartAdvisor input subsystem accepts input from one or more OnCall systems (e.g., 911 and / or various inbound and online input systems), desktop systems (e.g., from dispatchers), administrator systems, and test data systems (e.g., data used to train smart agents).

[0084] The SmartAdvisor Common Resource Subsystem includes software and hardware components that allow the use of smart agents within the system, including a receiver (REST) ​​API that allows smart agents to run within the system and access CAD data and other tools.

[0085] The SmartAdvisor agent subsystem includes an agent host module ("AgentHoster") that communicates with various intelligent agents, wherein each intelligent agent is configured or trained to generate notifications based on a specific type of analysis of historical operational data (e.g., one intelligent agent can detect patterns across events, another intelligent agent can detect similarities between events, and so on). The virtual scheduling assistance system can support almost any type of intelligent agent, and add or remove intelligent agents to the system in almost any desired combination of intelligent agents. Intelligent agents can be implemented on-site with the CAD scheduling server subsystem 110 and / or can be implemented remotely, for example, through a cloud-based machine learning service. Therefore, AgentHoster can be configured to host an on-site agent and connect to a remote agent or a cloud-based agent, for example, to pass events and other CAD data to the agent and receive notifications and other data returned from these agents. Figure 2 Some possible types of intelligent agents that may be included in the virtual scheduling assistance system are shown, including: Statistic Agent, Rule Agent, Pattern Agent, Keyword Agent, Event Match Agent, Correlation Agent, Recurrent Agent, Location Agent, Template Agent, and Similarity Agent; these intelligent agents and some other possible types of intelligent agents are described in more detail below.

[0086] The SmartAdvisor notification subsystem includes modules that determine which types of events to present to the scheduler and that process feedback from the scheduler in response to notifications, for example, to adjust the types of future events presented to the scheduler or to provide feedback for retraining one or more intelligent agents to improve the functionality of the intelligent agents. All of these subsystems can access a database system 113 (referred to herein as the "SmartAdvisor database" and depicted as accessible to all four SmartAdvisor subsystems) to access historical operational data and to store new data. Figure 1The components shown can be Figure 2 The components of the virtual dispatch assistance system shown are implemented as part of the CAD dispatch server 111, or distributed between the CAD dispatch server 111 and the machine learning service 114 (for example, a portion of the intelligent agent can be implemented in the machine learning service 114). Therefore, the intelligent agent can be cloud-based and / or integrated / on-premises.

[0087] In general, each intelligent agent has a specific "skill" and actively works to continuously analyze operational data and other data to provide notifications about specific environmental events. In an exemplary embodiment, the CAD data is passed to the AgentHoster via a REST API, which stores the data for historical purposes and provides access to the CAD data for the intelligent agent. The AgentHoster provides notifications from the intelligent agent to the SmartAdvisor notification subsystem so that it can be displayed on the dispatcher's client device. The dispatcher can then provide feedback (e.g., the utility or value of a particular notification), and the virtual dispatch assistance system can learn the dispatcher's notification preferences from the feedback and improve its own performance to generate future notifications. Fig.39 is a schematic diagram showing the relationship between a CAD scheduling server (referred to herein as OnCall 10), a SmartAdvisor AgentHoster, and a CAD scheduling client workstation according to an exemplary embodiment. Fig.40 A SmartAdvisor AgentHoster graphical user interface screen according to an exemplary embodiment is shown. Here, a user can view the available agents, input history, notifications, and other details of the operation of the Intelligent Agent.

[0088] Figure 3 is a diagram showing a method according to some exemplary embodiments Figure 2 Schematic diagram of the data model of a virtual dispatch assistance system with an intelligent agent. Here, the intelligent agent uses local operational data and other data (such as predefined customer preferences, behaviors, and goals) to generate notifications and recommendations, which can be filtered through a set of rules (which themselves can be machine learning) to determine which notifications and recommendations are ultimately presented to the dispatcher.

[0089] The following is a sample event data structure used in certain exemplary embodiments of the invention:

[0090]

[0091]

[0092] SmartAdvisor will provide notice (by Figure 2 In this context, notifications are the product or output of intelligent analysis, which convey relevant information to users. The following is a suggested notification data structure format according to a specific exemplary embodiment:

[0093]

[0094]

[0095] Figure 4 An example notification format is shown.

[0096] Figure 5 Another example notification format with more detailed information and an OnCall map is shown.

[0097] In an exemplary embodiment, the notification may include a user interface through which the user can provide feedback on the value of the notification, such as a LIKE / DISLIKE button or other type of rating input (e.g., high / medium / low or a rating of 1 to 10). SmartAdvisor is based on user feedback (by Figure 2 The notifications are classified (represented by the Feedback box in ) to predefine whether the user is willing to receive specific types of notifications. In an exemplary embodiment, the technology used is a Bayesian classifier, which uses the characteristics of previously liked and disliked notifications as criteria for future notifications. The system learns by tracking user feedback provided in previously presented notifications. The filtering can be user-specific, for example, learning the preferences of each user separately. In order to ensure some randomness in the filter, a small portion of these notifications (for example, 5% by default, but optionally configurable by the administrator) are not classified, so that the user may receive a small portion of unwanted / disliked notifications. Related notifications can be filtered as groups, for example, if the original event notification is not shown to the user, then notifications related to the original event may not be shown to the user. Notification filtering is performed by Figure 2 This is done by using the DelivererManager box in the .

[0098] In an exemplary embodiment, a user can create a shared notification, which in an exemplary embodiment is a portable link (URL) that can be exchanged between users to convey relevant SmartAdvisor findings and results in the presentation of a web page that replicates the original notification information and optionally also replicates other links and information. The system can restrict shared notifications to users of the system (even specific users), or can allow these shared notifications to be shared with other users outside the system. Figure 6 Sharing notifications according to certain exemplary embodiments are shown.

[0099] In an exemplary embodiment, a user can attach a notification to an event (e.g., as a special note or comment), optionally along with additional information, such as a list of users who liked or disliked the notification. Attaching a notification to an event can, among other things, enable the user to understand how the notification was created and why the information is relevant. The attached notification can be used as part of an audit or standard procedure review. Figure 7 A graphical user interface for attaching a notification to an event is shown, in accordance with certain exemplary embodiments.

[0100] In an exemplary embodiment, notifications are stored in the SmartAdvisor database along with user feedback, allowing the history of notifications to be reviewed. Figure 8 A notification history list is shown in accordance with certain exemplary embodiments.

[0101] Machine Learning and CAD Applications

[0102] Machine learning is a broad term that can be used in many different ways. Real-world applications often integrate more than one type of learning. There are many libraries of "ready-to-use" algorithms that utilize Bayesian, neural networks, and many clustering functions. All of these techniques usually involve some awkward modeling and data formatting. Although much emphasis is placed on deep learning today, it is not the only way to create machine learning systems. Genetic programming and rule-based programming can also be used for optimization and triggering alerts from other analytical / learning machines.

[0103] The following is a brief discussion of some flavors of machine learning and some possible applications of these flavors of machine learning in CAD systems.

[0104] The Symbolic learning style is based on rules and deductive reasoning. Knowledge is represented by a set of rules that can be represented with decision trees and random forests, and the output is based on reasoning from these trees / rules. Rules can be updated using fuzzy logic, and these trees can be trained. In terms of possible CAD applications, the Symbolic learning style involves storing a set of rules in the system, so it is relatively easy to implement, can take into account existing data or more exhaustive datasets, and can produce strong reasoning results on large datasets. Here are some examples of possible real-time CAD alerts based on the Symbolic learning style:

[0105] The number of events has unexpectedly increased in the last two hours.

[0106] Number of severe incidents above normal levels - alert supervisor.

[0107] Dangerous incident within 200 metres of a public school - evacuation may be required.

[0108] Fire reported close to sensitive location - Land Safety Alert required.

[0109] The Bayesian learning style is based on statistics and probability. Knowledge is collected in terms of probabilities, and the occurrence of events changes the probability of other events. Learning is based on these statistics. Applying statistics can reveal rules that lead to conclusions. Finding correlations in the data can lead to existing causal relationships, which can be explored by the CAD system. In terms of possible CAD applications, the Bayesian learning style can support rule-based intelligence using probability (for example, using Bayes' theorem and Markov chains). Here are some examples of possible CAD applications based on the Bayesian learning style:

[0110] There is a 95% chance of a new incident in Area X within the next hour - send out a patrol?

[0111] There has been a relevant increase in crashes in Zone Z over the past 3 days - dispatch a routine check?

[0112] There was a rare increase in domestic violence incidents on Sundays last month - most of the incidents occurred in the A, D and G zones.

[0113] Analogizer learning style is based on group similarities, clusters, and patterns. Learning comes from finding similarities in data. The algorithm looks for events that are closer to each other and classifies the events by assigning similar features to the group. It uses algorithms like nearest neighbor, support vector, and clustering. Here are some examples of possible CAD applications based on Analogizer learning models:

[0114] Try to categorize events by grouping them according to similarities.

[0115] Grouping closer events can lead to better geographic decisions.

[0116] This analysis can be used to reveal endemic disease events (dependent on location).

[0117] Once clusters of similar events are identified, they can be treated in the same way, or previously successful responses can be applied to them, and the CAD system can learn from past successes.

[0118] The evolutionary learning style is based on natural evolution and genetic algorithms. This is essentially a natural learning algorithm, i.e., survival of the fittest. A population of possible solutions (DNA) is evaluated, and the fittest survive to produce the next generation. It can also use "crossover" and "mutation" techniques to create new alternatives. In general, there are only two basic requirements: 1) the ability to generate the "DNA" of a solution, and 2) the ability to evaluate the solution. Using a lot of computer power, genetic and evolutionary algorithms can create unique and innovative solutions. Genetic algorithms can help improve other learning styles (e.g., neural networks, analogies, etc.).

[0119] Genetic algorithms use natural evolution techniques (Darwinian evolution) to evolve solutions. Solutions are represented by DNA-like structures. A fitness function is associated with the solution (DNA) and is used to evaluate fitness. The population of possible solutions (DNA) is randomly initialized and evaluated, and the fittest individuals survive to produce the next generation. Optimization can be understood as evolving towards the optimal direction. Genetic algorithms can create unique and innovative solutions. Since the control over the solution is low, two separate runs may give two completely different solutions. Restrictions on the solution can be easily modeled as penalties on the fitness function.

[0120] Fig. 9 Schematic diagram showing the genetic algorithm workflow according to certain exemplary embodiments. According to predetermined criteria, each of a number of successive population generations (starting from an initial population generation) is analyzed to select and sort the most suitable (elite) members in these populations. These elite members represent the most adapted members of the current generation (if a predetermined stopping criterion is reached, then they may be the most adapted in general). For each successive population generation, a new population is generated by multiplication (crossover) and / or mutation of the elite members of the previous generation.

[0121] The algorithm can be illustrated by an example using a DNA analogy. Each population generation consists of a set of individual DNA. The initial population can be completely random. The population evolves for many generations. The fitness function associates the chromosomes (individuals) with values ​​that are proportional to the quality of the solution. This function will define the fitness of the solution to select the most tuned ones. Population evaluation is usually the most complex and time consuming part of the genetic algorithm. It can utilize heuristic evaluation and can accommodate local improvements (e.g., cultural gene algorithms - genetic local search). The elite population is defined as the percentage of the most fit ones (e.g., 20%). The elite population is then reproduced / evolved to ensure that the best solution moves towards a better solution. This reproduction / evolution involves multiplication (crossover) and mutation. Fig.10 An example of a multiplication (crossover) operation is shown, in which the two fittest individuals (ie, parents) are split into two parts (the cut point can be random), and then the resulting parts are combined so that each offspring includes a segment from each parent. Fig.11 An example of a mutation operation is shown, where a portion of the elite population is subjected to mutations that may be random or rule-based.

[0122] Evolutionary learning models are particularly useful for optimization and multi-objective solutions. Here are some examples of possible CAD applications based on evolutionary learning models:

[0123] Vehicle routing with special restrictions.

[0124] Facility location and capacity for best response.

[0125] Relocation / mobilization optimization.

[0126] Automatic scheduling planning.

[0127] The connectionist learning style is based on artificial neural networks and is often considered the "state of the art" in machine learning, with deep learning, multi-level neural networks as an example. Learning is stored in a manner similar to the "synapses" of the brain. The basic structure is an artificial neural network. The network is trained by applying inputs and known outputs. This creates "synapses" and over time the network is able to give the correct output to other inputs. The inputs / outputs can be almost any type of data. The connectionist learning style can be very useful for classification (supervised classification) and usually requires a lot of training. Here are some examples of possible CAD applications based on connectionist learning models:

[0128] Image recognition.

[0129] Event classification (e.g., type, urgency, etc.).

[0130] Event detection.

[0131] Alert classification.

[0132] Time and distance measurement (e.g., better ETA).

[0133] Technical Overview

[0134] The following is an overview of some AI / ML techniques that may be employed by various intelligent agents, according to exemplary embodiments.

[0135] Rule-based logic

[0136] In the context of SmartAdvisor, a rule is a logical expression that evaluates to either a true statement or a false statement. We can say that each rule has an expression like this:

[0137] if (logical expression) then (issue a notification)

[0138] This technology usually involves symbolic processing and rule evaluation engines, but in some cases it is possible to reduce logical expressions to lists of variables and limits in a structured way, for example:

[0139] if(variable.operator.limit)then(issue notification)

[0140] Or in a more general form:

[0141] if((variable1.operator.limit1)AND

[0142] (variable2.operator.limit2)AND...)

[0143] then(send notification)

[0144] Here, we only consider adding operator.AND. to the expression list, since operator.OR. is the same as having two rules.

[0145] In an exemplary embodiment, the logical operators of the logical expression are:

[0146]

[0147] The variables are system variables that are converted to real numbers, and the limits are real numbers or other system variables. To evaluate the expression, the system takes the values ​​of the variables and the limits and executes the operator, repeating the operation for all variable-limit pairs.

[0148] Fuzzy Logic

[0149] Boolean logic deals with true and false, which are usually associated with 1 and 0, while fuzzy logic is a variation of logic in which values ​​can be any number between 0 and 1. Fuzzy logic includes 0 and 1 as extreme cases of truth or facts, but also includes various truth states in between, so it takes into account partial truths and is often used to facilitate the expression of rules and facts such as "near", "far away", "close to", and "like".

[0150] The goal of fuzzy logic is to relate a collection of partial truths to a final logical decision of true or false. This is an approach similar to human reasoning and is therefore considered part of the application of artificial intelligence (AI): representing generalized human cognitive abilities in software so that AI systems can find solutions to unfamiliar tasks.

[0151] Lotfi Zadeh, the inventor of fuzzy logic, observed that unlike computers, humans make decisions considering possibilities like "definitely", "maybe", "I don't know", "probably", "definitely" as variations of yes (true) or no (false), but in the end, humans sometimes have to make a "yes" or "no" decision. A typical algorithm for a fuzzy logic problem can be:

[0152] 1. Define input variables

[0153] 2. Use Fuzzyfier to transform these variables into fuzzy sets

[0154] 3. Construct a set of rules that map inputs to outputs

[0155] 4. Implement these rules in FuzzIntelligence

[0156] 5. Evaluate input and generate output

[0157] Fuzzy logic can be implemented in software or even hardware in systems with various sizes and capabilities. In an exemplary embodiment, the fuzzy logic system is simplified into two modules, referred to herein as Fuzzyfier and FuzzyIntelligence. The Fuzzyfier module transforms a generic input (which can appear in various formats) into a set of fuzzy values ​​(i.e., numbers from 0 to 1) that are input to the FuzzyIntelligence module. The FuzzyIntelligence module simulates the human reasoning process by performing fuzzy reasoning on fuzzy values ​​based on IF-THEN rules to produce logical outputs. The third component can be used for "defuzzification" and transforms the logical output into an output action or value.

[0158] Another useful application is membership functions. Membership functions allow quantification of linguistic terms and graphical representation of fuzzy sets. The membership function of a fuzzy set A on a universe of discourse X can be defined as μA:X→[0,1].

[0159] Among some of the advantages of fuzzy logic are its simplicity and the possibility to transform complex problems in many fields into simple solutions in a manner similar to human decision making. One disadvantage of using fuzzy logic is that there is no general approach or automatic learning; instead, each problem must be adapted to each situation.

[0160] As an example of using a fuzzy logic system to match events, let's consider creating a fuzzy logic engine that solves the problem:

[0161] Two events are likely to be the same if they occur close together and are reported at about the same time. They can have the same type, but sometimes users describe events differently, resulting in different types. Given two events (e1, e2), are they the same?

[0162] So, given two events e1(x1,y1,t1,s1) and e2(x2,y2,t2,s2), where:

[0163] x1,x2,y1,y2 – the reported UTM location of the event

[0164] t1, t2 – date-time of the event

[0165] s1,s2 – the types of events 1 and 2

[0166] The Fuzzyfier module must transform this data into fuzzy input. In order to make the final decision, the system needs the following fuzzy information:

[0167] Are the two events close to each other?

[0168] Do the events occur simultaneously?

[0169] Are both events of the same type?

[0170] In one example of determining whether two events are close to each other, let d be the Euclidean distance from one event to the other. <d max , then we can say that the two events are close to each other, where, for example, d max = 100m. But if d = 101m, are the two events still close to each other? What if they are 105m apart? In order to have a consistent solution, the exemplary embodiment may use a Fig.12 The fuzzer function for the function shown, where:

[0171]

[0172] Here, we use As d max With this function, an event is considered “close” if Fuzzy(d)<0.5 (or any other number less than 1.0).

[0173] To determine whether events occur simultaneously, a similar generalized fuzzer function (e.g., Fig.13 The same method can be applied to time, where:

[0174]

[0175] To determine whether two events are of the same type, we can consider the exact type or a family of types, or even the scale of types. One exemplary embodiment considers the exact type.

[0176] Using these three types of fuzzy information (i.e., events are close together, events occur at the same time, and events are of the same type), we can now formulate a Fuzzy Intelligence function, for example, as represented by the following decision tree (which can be transformed into these logical expressions):

[0177] IF(near and same_time)THEN return True(Same)

[0178] IF(not_near and not_same_event)THEN return False(not Same)

[0179] IF(not_near and same_time)THEN

[0180] IF(same_type)THEN return True(Same Event)

[0181] ELSE return False (not Same)

[0182] Other functions of the same type may be implemented with different results.

[0183] Basic Statistics

[0184] For basic statistics of events in CAD operations, the exemplary embodiment will be based on the following assumptions: events are distributed with a normal (Gaussian) distribution. Therefore, the concepts of mean and standard deviation can help represent the data. Maximum and minimum values ​​are also monitored, as well as four different time frames: hour (short time frame), day, day of the week (DOW - a number between 1 and 7), and month (long time frame). In the exemplary embodiment, the following information is tracked for each event type:

[0185]

[0186] The following is an overview of the statistical algorithm according to an exemplary embodiment:

[0187] Given a new event ei of type Ti

[0188] Handling new events

[0189] Processing statistics of all events - EVENT

[0190] Calculate count, mean, standard deviation, and maximum

[0191] Check event rules - send notification

[0192] Process statistics of events of type T - TYPE[t]

[0193] Calculate count, average value (AVR), standard deviation (STD), maximum value

[0194] Check Type Rules - Issue Notice

[0195] Save event status

[0196] in:

[0197]

[0198] Now, let limit be the percentage at which notifications are allowed to be issued, e.g. limit = 1.25 (25% higher than the desired value). So, using the above example, a notification will appear when the following conditions are met:

[0199] DT_N_HOUR>DT_AVR_HODBefore*limit

[0200] DT_N_HOUR>DT_MAX_HODBefore (set new maximum value for the hour)

[0201] DT_N_DAY>DT_AVR>DOWCurrent*limitDT_N_DAY>DT_MAX_DOWCurrent (set a new maximum value for the DOW)

[0202] Correlation Statistics

[0203] This is a simple algorithm that detects possible correlations between variables by identifying outliers in a simple statistical analysis of a set of unrelated events. The study applied this analysis to a sample of data and can be processed whenever a relatively large amount of data is available.

[0204] Given a set of unrelated events E = {e1, e2, ...e n}, where each event has a separate set of characteristics c ij , where: e i =(c i1 ,c i2 ,…,c im ), we can assume some known finite set of values.

[0205] Each value of each characteristic is assumed to be independent of the value of another characteristic, although each characteristic may have its own probability distribution.

[0206] For example, let c1 be the day of the week and c2 be the type of event. It can be assumed that the types of events are evenly distributed across the days of the week, regardless of the type of event. (That is, the probability of p(c1) is equal to p(c1 / c2)

[0207] p(c1)~p(c1|c2=C)

[0208] If a particular type of event is more common on a certain day or week, then there is a correlation between that type of event and that day of the week. If we compare the probability of that event occurring on that day to the probability of all events occurring on that day, we can track this trend. This indicates a possible correlation of feature c1 with value c2=C.

[0209] p(c1|c2=C)≠p(c1)

[0210] This reasoning is a direct application of Bayesian inference, where, as a prior probability of execution time, we directly compute the probability that P(c1) is equal to the day (as if there were no relationship). And P(c1|c2=C), the posterior probability of execution time given a day. When they deviate, the data indicates a relationship between the two variables.

[0211] In the example, suppose that it is likely that the time to complete an event is shorter on weekends than on other days; we can say that there is a possible correlation between the time to complete the event and weekends.

[0212] All of the above relationships require a level of significance in the data sample. To identify these phenomena, it is possible to restrict the data set to a small size.

[0213] Algorithmically, given a set of events and associated variables, we identify pairs of variables that may be correlated. If the frequency of occurrence of the data for that pair of variables is higher than expected, a correlation is observed.

[0214] The probability of an event variable appearing in an event is the ratio of the number of events in which the variable appears to the total number of possible events. Therefore, if a variable affects the occurrence of an event, the probability conditional on the variable will be greater than the probability in the absence of the variable. By comparing these two probabilities, we can identify the possible effects:

[0215] In the event data set E = {e1, e2, ... e n}, given two features c1 and c2,

[0216] Among them, e i =(c i1 ,c i2 ,…,c im ),

[0217] Repeat for each value C of c2

[0218] Assume p(c2=C) and p(c1)>α (where α=significance level)

[0219] If p(c1|c2=C)>β*p(c1) (where β=influence level)

[0220] Then c1 is affected by c2=C

[0221] The main evidence of an influence between variables in an event is if there is unexpected behavior in the presence of a given factor. The occurrence of events is assumed to follow a random pattern. When the assumption of randomness is not confirmed, influence between variables is identified. The software seeks to identify such repeated behaviors that deviate from those that might be considered expected. We want to identify sets of events that consistently behave outside the mean of the behavior of other events, so-called "outliers."

[0222] For example, we can test the hypothesis that traffic accidents occur more often on weekends by comparing the likelihood of a traffic accident occurring on a certain day of the week to the frequency of all types of accidents occurring on that day. If there is a day of the week effect, the probability will be significantly higher on that day. Note that this method is tied to the absolute number of events, i.e. there must be a large number of events for the ratio to have some statistical significance. This technique is similar to the technique known as "Market Basket Analysis", which is a common technique for finding product purchase associations in supermarket purchase analysis. Identify unusual characteristics and consumption patterns.

[0223] This type of research allows analyzing the correlation of any two variables recorded for an event. But this must not be confused with a causal relationship between the variables. This means that the fact that two variables are linked is not evidence that one is the cause of the other. Common causes affect both the variables and the causal correlation.

[0224] To illustrate the application, we will analyze historical data of a typical emergency medical system. The system is responsible for answering emergency calls within the city. Each call is classified according to a predefined list of complaint types (TYCOD). Let's see how the day of the week, time of day and location affect these types of events. In addition to the complaint type and date, the following are recorded: geographically precise service location, arrival time, sub-classification of the incident type, service team, etc.

[0225] An exemplary study correlating event type (TYCOD) with day of week (DOW) with an impact factor of 20% identified the following impacts:

[0226] At DOW=01, the impact of TYCOD=04 [01,3%] is higher than the average [057,2%]

[0227] At DOW=07, the impact of TYCOD=04 [01,3%] is higher than the average [026,2%]

[0228] At DOW=01, the impact of TYCOD=23 [01,4%] is higher than the average [035,6%]

[0229] At DOW=07, the impact of TYCOD=29 [05.2%] is higher than the average [023.3%] At DOW=01, the impact of TYCOD=30 [02.2%] is higher than the average [025.9%]

[0230] in:

[0231]

[0232] The results reported that the number of incidents of aggression, drug overdose, traffic accidents, and trauma increased above average on weekends (DOW) (where 1 = Sunday and 7 = Saturday). For example, a closer look at these incidents might show that they are associated with increased alcohol consumption on weekends.

[0233] This may be a common reason for more accidents these days. In an exemplary embodiment, the correlation analysis will indicate that there is a cause to be investigated, but not what the cause is. The analysis also provides a graphical view of each of these codes. For example, Fig.14Shown are the averages of events for all codes correlated by day of week (DOW) and the selected 20% impact range for the case of TYCOD=4 (assault / sexual violence), Fig.15 Shown are the averages of events for all codes correlated by time of day and a selected 20% impact range for the TYCOD=4 (Assault / Sexual Violence) case.

[0234] It can be observed that when DOW=1 and 7, the frequency of events occurring increases greatly, indicating a correlation.

[0235] Natural Language Processing (NLP)

[0236] Natural language processing refers to communicating with computers using natural language, as opposed to structured command language. This is the field of artificial intelligence and involves reading and writing. For this technical note, we will limit the application to reading text and making decisions based on it. In this study, we will not consider the use of speech, although it is a part of NLP.

[0237] NLP is usually divided into NLU (Natural Language Understanding), which is understanding textual input, and NLG (Natural Language Generation), which is producing meaningful phrases and sentences. We will only look at NLU.

[0238] Language is very rich in form and structure, so understanding natural language is very difficult. There are many levels of understanding:

[0239] Lexical – is very primal, word-level, and considers words as belonging to a language (“bread” is an English word)

[0240] Semantic – Also associated with language and considers the meaning of words. (“Bread” is a food made from flour and usually baked)

[0241] Syntax – considers the use of words in a phrase. (Bread is delicious means that “bread” has the property of “delicious”)

[0242] Referential – Consider using a referential pronoun (Bread is delicious. It is made of flour – it refers to bread)

[0243] The same phrase / word can have multiple interpretations and can lead to multiple meanings and analyses.

[0244] NLP usually has five steps: lexical, syntax / parsing, semantics, discourse, and pragmatics.

[0245] Lexical Analysis - It involves identifying and analyzing the structure of words. The lexicon of a language refers to the collection of words and phrases in a language. Lexical analysis divides the entire block of txt into paragraphs, sentences, and words.

[0246] Syntactic Analysis (Parsing)- It involves grammatical analysis of the words in a sentence and arranging the words in a way that shows the relationship between the words. An English syntactic analyzer rejects sentences like "The school goes to boy".

[0247] Semantic Analysis- It derives the exact meaning or dictionary meaning from the text. It checks whether the text makes sense or not. This is done by mapping syntactic structures and objects in the task domain. A semantic analyzer ignores sentences like “hot ice cream”.

[0248] Discourse Integration - The meaning of any sentence depends on the meaning of the sentence immediately preceding it. In addition, it also brings with it the meaning of the immediately preceding sentence.

[0249] Pragmatic Analysis- During this, what is said is reinterpreted in terms of its meaning. It involves deriving those aspects of language that require real-world knowledge.

[0250] Some AI applications can be implemented at the lexical level, while others require syntactic and semantic understanding.

[0251]

[0252] Text pattern recognition

[0253] One of the possibilities to find patterns in a text is to define a regular expression pattern to be highlighted whenever it appears on the text. A regular expression in a pattern consists of one or more literal characters, operators, and other text structures. Most programming languages ​​have tools for extracting text that matches a regular expression. Regular expressions are described by a set of constructs, and when the input text matches these expressions, we have the expected pattern.

[0254] Here are some example REGEX expressions for text pattern recognition:

[0255]

[0256] Keyword Identification

[0257] One of the possibilities to find patterns in the text is to define a set of predefined keywords and when some of these keywords are found in the event description, the system generates appropriate notifications, for example to specific users who need this information.

[0258] Notification likelihood analysis

[0259] User feedback (e.g., thumbs up or thumbs down) will teach the virtual assistant whether similar notifications should be sent to the user. In other words, the system can estimate the likelihood probability of a notification, e.g., whether the user will like the notification based on previous information. An exemplary way of doing this is through a Bayesian classifier, as described below.

[0260] The result of the classifier is an estimate of the probability of the likelihood (of a notification) given the previous feedback information from the user for the notifications that the user has received so far. We can also estimate the probability of not liking the information.

[0261] Let PL(Ni) be the probability of likelihood, where close to 1.0 means the user is likely to like the notification. Let PD(Ni) be the probability that the user will dislike the notification based on previous dislikes. Using PL(Ni) and PD(Ni), we can use fuzzy logic and create fuzzy intelligent logic to decide whether to show a notification to the user, for example, Fig.16 shown.

[0262] The exemplary embodiment also includes some serendipity, for example, by providing the user with some random small amount of notifications, regardless of category. This gives the user a chance to check his or her preferences.

[0263] Bayesian Classifier

[0264] A Bayesian classifier is a technique used to automatically classify notifications. It is based on the Bayesian statistical theorem, which estimates the conditional probability of an event based on the prior knowledge that another related event has occurred. In this case, we can estimate the probability that the user will like the notification, that is, the notification is of class C, given that it is described by a set of parameters T (notification description). In other words, we will estimate the likelihood that the user will like the notification based on the previous likes given by the user in other annotations.

[0265] The technology is similar to how spam is sorted in email or important versus promotional emails are sorted in an inbox. It looks for the frequency of certain words and makes a prediction, which the user can correct with new feedback.

[0266] Assume that Bayes' theorem is expressed as:

[0267]

[0268] in:

[0269] P(C|T) is the conditional probability of a class given a set of parameters T, i.e. the desired outcome;

[0270] P(T|C) is the conditional probability of a given parameter in class C (estimated from the history of previous occurrences);

[0271] P(C) is the probability of the class, also estimated based on previous events;

[0272] P(T) is the probability of occurrence of a given parameter, also estimated from previous events, since the parameter set consists of a single term:

[0273] T={W1,W2...W n}.

[0274] For notifications we can for example have: notification type, notification location, notification text (set of words), etc.

[0275] P(T|C) can be expressed as:

[0276] P(T|C)=P(W1|C)*P(W2|C)*...*P(W n |C)

[0277] as well as

[0278] P(C|T)=P(W1|C)*P(W2|C)*...*P(W n |C))*P(C)

[0279] The implementation can be structured into two phases: training and testing. In the training phase, parsing, eliminating rejected word phrases, eliminating rejected words, and counting the remainder of each class. In the testing phase, receiving text input, parsing the input, and calculating the probability that the input belongs to each class, thus selecting the most likely class.

[0280] The probability is estimated by counting the words, such as:

[0281]

[0282] as well as

[0283]

[0284] Event pattern analysis

[0285] The characteristic of the event pattern analysis goal is to highlight events that share similar patterns. There are many ways to define patterns in this context, but one exemplary embodiment will consider two different approaches: repeated event identification and event similarity.

[0286] Repeated event identification

[0287] This feature looks for repeated incidents of the same location and type in a historical database. The goal is to identify repeat offenders, while fairly simple crimes may have aggravated consequences, such as domestic violence.

[0288] For this notification, we can use the same event matching technique with fuzzy logic comparison of events, where the tolerance for distance is very small and the type is the same, and the date is not considered in the analysis.

[0289] Event Similarity

[0290] This is the technology involved in evaluating short texts and assessing their distance (or in other words, their similarity). Two texts can be considered similar if they refer to the same event. The definition can simply consider the syntax of the text (the words and terms used in the text) or the semantics (meaning) of the words and text. This will be explored by NLP (Natural Language Processing) techniques to find patterns when comparing events and social posts. An exemplary embodiment defines cosine text similarity as a value between 0 (no similarity) and 1 (very similar or comparable).

[0291] Distance can also be considered when dealing with fuzzy logic comparisons, such as close, farther, closer, earlier, later, etc. For example, given two feature sets:

[0292] x i =(x i1 , x i2 …x id ) and x j =(x j1 , x j2 , …x jd )

[0293] Where d is the dimension of the feature set, the relationship between the two sets can be evaluated as a distance. Distance can be used to measure continuous features, while the concept of similarity can be used for qualitative variables.

[0294] The general distance formula is called the Minkowski distance and is expressed as:

[0295]

[0296] If n = 1, then the formula represents the Manhattan distance or "city block" distance, while if n = 2, then the formula represents the Euclidean distance, as shown in the following table (from

[0297] http: / / www.molmine.com / magma / analysis / distance.htm):

[0298]

[0299] Cosine text similarity

[0300] To define the distance between one short text description (d1) and another short description (d2), we can use some analogies of vector distance. A popular metric is the Euclidean distance, which is defined by the root of the square of the difference between two vectors:

[0301]

[0302] Here, A·B considers the inner product of two vectors, giving a unique real value for this measure. In order to compare two measures, we need to define the scale. Another very common measure between vectors that has its own scale is the cosine of the angle between the two vectors, which can be expressed as:

[0303]

[0304] If cos is close to 1, the two vectors are close to each other, and if cos is zero, the two vectors are separated from each other; this provides a good way to calculate the percentage of similarity. Consider each short text description as a sequence of words:

[0305] d = {w1, w2, ..., w N}

[0306] So

[0307] d1={w 11 , w 12 ,…,w 1N}

[0308] d2={w 21 , w 22 ,…,w 2N}

[0309] as well as

[0310] d1·d2=d1·d2=+w 12 w 22 +…+w 1N w 2N

[0311] or

[0312]

[0313] As the formula shows, the more words they share, the closer they are. However, in this description, words cannot be considered equally important. For example, in a police description, the characteristics of a suspect are more important than an armed robbery or a list of stolen items. To account for this difference between words, an exemplary embodiment uses the TF-IDF algorithm as the weight (fi ),as follows:

[0314]

[0315] TF-IDF algorithm

[0316] TF-IDF stands for "Term Frequency - Inverse Document Frequency". This index tends to filter out uncommon words from short texts:

[0317] tfidf(w,d,D)=tf(w,d)·idf(w,D)

[0318] in:

[0319] tf(w, d) - represents the frequency of word w in the document; and

[0320] idf(w,D) - measures the inverse frequency of a term (w) in the entire document collection (D).

[0321] The value of tf(w, d) is either zero (if the word does not appear in the document) or greater than 1, and we can assume 1 and consider each word individually. The use of square root and logarithmic functions helps to improve the ratio between the distributions and boost less frequent words:

[0322]

[0323] Exemplary Intelligent Agent

[0324] The inventors have contemplated various types of intelligent agents that can be incorporated into the present system, and possibly, different intelligent agents and different intelligent agent types are deployed for different clients. The agent host typically allows the addition, removal, and replacement of intelligent agents as needed or desired. Without limitation, the following are some exemplary intelligent agents contemplated by the inventors.

[0325] A StatisticAgent may be included to monitor the operating variables of the system and to notify of outliers. Outliers are variables whose values ​​exceed expected limits. In an exemplary embodiment, when the allowed limits are predefined, the expected values ​​are determined by measuring the variables during a training period. The main motivation for monitoring operating variables is that normal operation should be expected to remain substantially stable, i.e., the expected values ​​of the monitored variables should remain within acceptable limits, and therefore, when the variables cross the limits, something abnormal may be occurring in the operation, and an alarm must be sounded. The more stable the operation, the more important the deviation alarms are, because they may reflect that something unexpected has occurred in the operation. If the operation is very unstable, i.e., the number of times an event recurs is small, then the relevance of the deviation alarm is small, because the operation naturally changes the number of occurrences without external reasons.

[0326] For the purposes of the following discussion, the following terms are defined:

[0327] Categorical variables are operational variables expressed in terms of categories. For example, event type is a categorical variable. Categorical variables are monitored by counting the variable over a period of time and observing the frequency of that category among all categories.

[0328] Continuous variables are operational variables expressed in real values. For example, the time to an event is expressed in minutes. Monitoring of continuous variables can also be done by estimating the average value over a period of time.

[0329] An outlier is a measure of a monitored variable that is outside of some expected limit. The expected value is determined by measuring the variable during a training period, and the allowed limits are predefined.

[0330] Training is the period during which the variables are monitored to define the expected value for each class (if categorical) or the mean value (if continuous).

[0331] Check - a period after the training period to see if there are outliers on the validation variable. This period can be repeated indefinitely.

[0332] Thresholds are limits expressed as percentage deviations at which a variable is considered an outlier, for example a threshold of 50% means that if the frequency during the examination exceeds 50% of the expected value, there is an outlier.

[0333] Support is the minimum frequency of a variable that is considered statistically relevant. If the variable frequency is below the support value, no outliers are released. The support value is used to avoid estimating statistics from very rare events.

[0334] In an exemplary embodiment, variables are classified into two types: categorical variables (which have values ​​defined by category, such as event type), and continuous variables (whose values ​​range continuously from minimum to maximum, such as time to event). Each variable will be managed differently. Categorical statistics will be measured by the number of events per unit of time, such as measured in hours. It should be noted that continuous variables can be monitored in categories, for example, by organizing them into multiple categories.

[0335] In an exemplary embodiment, each variable is defined by the following data structure:

[0336]

[0337] Time periods can be normalized to hours, shifts, days, weeks, months, and any other time expression as follows:

[0338] Time Hours Hour 1 Flights 8 sky 24 week 168 moon 720 (approx.)

[0339] The following are the monitored variables for the CAD operation in the exemplary embodiment:

[0340] Variable Code Variable Types Variable Description Typical inspection period type category Event Type sky STY category Event Subtype HOD category Time of day sky DWO category Day of the week week DRG category Scheduling Group sky ETA continuous Arrival time of the event Less than a day

[0341] In the exemplary embodiment, StatisticAgent has three main classes, as follows:

[0342] StatisticAgent – ​​Main class that stores a list of CategoricalVariables to monitor and continuously receives events, parses the monitored variables and passes them to the monitored variables. When variables identify outliers, they issue notifications back to the agent.

[0343] CategoricalVariable - Its main focus is monitoring variables. It automatically identifies various categories and monitors their occurrence. During the initial learning phase, the input is used to train the model defining a baseline, which is the expected value that is checked against the monitored variable that is checked regularly.

[0344] CategoricalVariableStats – This class holds a list of categories and frequencies.

[0345] In an exemplary embodiment, the agent is set up by adding a list of monitored categorical variables (a table of monitored categorical variables for CAD operations is recommended). After this setup, if events are received, the agent continues to follow the main workflow for each event. Fig.17Describe the main workflow of the statistical agent (StatisticAgent) according to an exemplary embodiment. First, the external caller enters a series of events in order and gets a list of notifications as a return (method-Trigger). Then, the agent extracts the features of each category variable from the event, and sends the feature data to the corresponding CategorialVariable that monitors the category variable, requesting a list of possible outliers from them (method-getOutliersNote). During the training phase, the monitor stores features and calculates the frequency of each variable (method-Monitor.Add). At the end of the training phase, the monitor is cloned to the baseline (method-Monitor.Close). The baseline will be sorted and the frequency will be calculated (Baseline.SorteByCategory, Baseline.setFrequency). During each inspection cycle, the features are stored in the monitor (method-Monitor.Add). At the end of the inspection cycle, the monitor is compared with the baseline via the Check method, and a set of outliers can be issued (method-Check). At the end, the outliers are transformed into notifications, and these notifications are returned to the caller as output (method-GetNotes).

[0346] Fig.18 Example baselines for the variables are shown.

[0347] Fig.19 An example of an outlier notification with statistical data is shown.

[0348] In this example, MISCELLANEOUS events increased by 68% (24 events expected, 40 events registered), and GARBAGE MISSES increased by 60.8% (7 expected, 12 measured) during that period (24 hours). SIGNAGE REPAIR also decreased by 71.3% during the same period. An updated baseline incorporating that week's data was generated because the algorithm is constantly learning. Fig. 20 An example updated baseline for the above example is shown.

[0349] Here are some examples of key statistics notifications:

[0350] Example Additional Information 10 outliers were found in this hour (16h00) Individual outlier notifications 4 outliers were found in the last week of March 20, 2019 Individual outlier notifications Three outliers were found in that month (March 2019). Individual outlier notifications

[0351] Here are some examples of outlier notifications:

[0352]

[0353] A statistical graph may be presented to the user. The statistical graph displays the outliers of the statistical analysis in graphical form. In an exemplary embodiment, the graph shows the expected and actual values ​​and identifies why this is an outlier. Once the learning phase is complete, the baseline graph may be used to show the expected values ​​in a separate notification. If the number of items is very large, a limited number of items may be displayed, for example, the top 20% or by limiting the number of bars.

[0354] A RuleAgent may be included to monitor some user defined rules involving operational variables and send notifications when the rules are verified. In an exemplary embodiment, the list of available operational variables is predefined and may include statistical monitored variables (see StatisticAgent above). Limits and thresholds may be fixed values ​​or may be combinations with system variables. Rules and notifications associated with rules are user defined. In an exemplary embodiment, rules are generally defined as follows:

[0355] If (logical expression) then send a notification

[0356] The following are example rule and notification configurations:

[0357] If ((number of robberies in the last hour) > 5)

[0358] (“Alert the Chief of Police that there have been more than five robberies in the last hour”) Here are some examples of rule alert notifications:

[0359] Example Additional Information 5 rules checked in recent events Rules separate notice 2 rules checked in the last hour Rules separate notice

[0360] Here are some examples of rule-alone notifications:

[0361] Example Three armed robberies occurred in the last four hours and were close to each other. The incident is 50 feet from a public school. Consider evacuation procedures.

[0362] A PatternAgent may be included to recognize predefined pattern expressions (e.g., type, description, notes, comments, location) in event descriptions and issue a notification to the user along with an associated action with the notification. A pattern expression describes a specified configuration of numbers, letters, and / or symbols, e.g., a regular expression. The action may incorporate the identified pattern therein, e.g., a URL / link or API call related to an external action. PatternAgent operates on a list of named patterns that have associated regular expressions and actions associated with the regular expressions. In an exemplary embodiment, a set of predefined expressions is provided, but users are free to create their own expressions using regular expression syntax. There are regular expressions that represent phone numbers, emails, hashtags, license plate numbers, social security numbers, and the like.

[0363] Here are some examples of pattern alert notifications:

[0364] Example Additional Information 5 patterns found in Event C2344-AS Single mode notification 2 patterns found in Event 2340-ABC Single mode notification

[0365] Here are some examples of individual modal notifications:

[0366]

[0367] A KeywordAgent may be included to detect keywords in the event description and alert the user to actions related to these keywords. In an exemplary embodiment, keywords must be detected word by word. Every word on the keyword must exist, and typos are not allowed. These words may appear in different orders, and they are still considered. Words are not case sensitive. Partial words are also considered, so, for example, the keyword "Manifest" will match the word "Manifestation" in the event description, while the keyword "Manifestation" will not match the word "Manifest" in the event description. Typos and synonyms are not considered at this stage, but alternative embodiments may consider typos and synonyms, for example, using a simple word proximity metric to capture typos. Synonyms can be used with existing natural language processing (NLP) tools.

[0368] Here are some examples of matched and unmatched keywords:

[0369]

[0370] Here are some examples of keyword alert notifications:

[0371]

[0372] Here is an example of a single keyword notification:

[0373] The following keywords [keyword list] were recently reported in incident [EventID]. Action Required: Please notify Detective Joe (435-09029)

[0374] An EventMatchAgent may be included to identify similar events among currently open events, even if the description (e.g., event type, location, date) is slightly different. In an exemplary embodiment, the comparison uses a fuzzy logic approach that can accommodate some variation and thereby identify some minor variations. Fuzzy logic takes into account words like "close" and "near" in the comparison. It turns these "fuzzy" words into comparison logic. In an exemplary embodiment, a fuzzy logic tolerance parameter [Fuzzy] is defined, for example, by default 25%.

[0375] The following is an example of a matching event alert notification:

[0376] Example Additional Information 3 events share the same characteristics as event C234509 Related events on the map

[0377] Fig.21 An example of related events shown in a map is shown, including the ability for a user to display event information such as identifying the original event and related events.

[0378] A CorrelationAgent may be included to analyze a pair of variables and their possible correlation with the occurrence of an event. A notification is issued when there is a certain probability that the combination of these variables may be contributing to the occurrence of an event (it should be noted that this is a correlation, not necessarily a causal relationship). In an exemplary embodiment, the unit in which the statistical data is estimated is the number of events per unit time (in hours), and the variables are measured by counting the events, so each new event is counted and measured to define the expected value and the outlier. In an exemplary embodiment, the technology used is called "a priori" analysis, and the following is defined for the CorrelationAgent: variable 1, variable 2, inspection period, threshold, and support list. Variable 1 and variable 2 are the operating variables to be analyzed. The inspection is the period after the training period to verify whether there are outliers in the variables. The threshold is the limit probability of being considered as an outlier, for example, 50% means that if the frequency during the inspection exceeds 50% of the expected value, there is an outlier. Support is the minimum frequency that is considered relevant. If the variable frequency is lower than the support value, the outlier is not measured. Lift is the percentage increase in the expected value. The lift must be higher than the threshold value of the correlation to be considered.

[0379] Here are some examples of relevance alert notifications:

[0380] Example Additional Information 10 correlations were found in the last week. Individual dependency notifications Found 4 correlations in the last shift Individual dependency notifications

[0381] Here are some examples of individual dependency notifications:

[0382]

[0383] A RepeatedEventAgent (sometimes called a RecurrentAgent) may be included to identify past (perhaps repeating) events that occurred within a specified time span [TSPAN, e.g., TSPAN=180 (6 months) or TSPAN=365 (1 year)] and alert the dispatcher that this may be a recurring event. Sometimes a simple event (e.g., a domestic disturbance) may expand into a larger event due to its recurrence, and in such cases is sometimes important enough to alert on-site personnel. Sometimes the dispatcher prefers to assign a team that may already be aware of the incident. Being reminded of past events may be useful in assigning a response team. The agent needs access to past events that may be stored in a CAD system. In an exemplary embodiment, the search is done based on location and event type. In an exemplary embodiment, the RepeatedEventAgent considers exact locations (e.g., the same home address) rather than approximate locations, however alternative embodiments may expand the search (e.g., the same neighborhood).

[0384] The following is an example of a recurring event alert notification:

[0385] Example Additional Information Event 23A-ESR has occurred 4 times in the past year List of events and event data

[0386] A SimilarityAgent may be included to find events similar to the current event (e.g., a new event entered into the system) based on the event's description (e.g., type, description, notes, comments, location, etc.). Two similar events share relevant information, although they may not have the same type, date, or location. In an exemplary embodiment, the similarity index takes into account the words used in the event description and notes. In an exemplary embodiment, the algorithm evaluates less common words, so when they appear in different events, the similarity index grows. The agent learns by analyzing the frequency of words used in the event corpus, and it is constantly learning and updating the word list and creating a dictionary model. In an exemplary embodiment, at this stage, typo correction or synonyms are not considered, although alternative embodiments may consider typos and synonyms. Clustering can help find patterns in geographic information and other forms of data. Embodiments can find patterns in geographic distributions and include other information in the analysis. Clustering techniques can help find "clusters" of similar events. Text / keyword analysis is used to infer potential non-obvious relationships between events. Through a dashboard / monitor (or OnCall notification), the system can display potential linked events and other information that may be relevant to the user. When a pattern becomes frequent, the main event prediction module / workflow can be extended for SimilarityAgent.

[0387] In an exemplary embodiment, the following are defined for SimilarityAgent: similarity threshold, rejection word list, tokenizer list, and rejection number. The similarity threshold [similarity] is an indicator from 0 to 1, above which two descriptions are considered similar, for example, the default [similarity=0.8]=80%. The rejection word list (file) is a list of words that are not considered in the analysis, usually very common words with no relevant information. The tokenizer list is a list of symbols used as word separators. Rejection numbers (yes / no) defines whether to reject all numbers in the description (default=yes).

[0388] In an exemplary embodiment, SimilarityAgent is based on a combination of tf-idf (e.g., https: / / en.wikipedia.org / wiki / Tf%E2%80%93idf) and cosine similarity of two events (e.g., https: / / en.wikipedia.org / wiki / Cosine_similarity), where the frequency of words in the event description is analyzed, less common words are considered more important, and then cosine similarity is used to determine the final similarity score. In an exemplary embodiment, users can mark specific keywords, such as locations, people, places, etc. of interest. If these keywords appear, a notification is issued. The embodiment can also find similarities in tweets and other social media text messages. Search conditions (e.g., topic tags or locations) can be used to select such social media text messages for analysis.

[0389] Here are some examples of similarity alert notifications:

[0390] Example Additional Information 5 events similar to this event [EventID] Individual Similarity Notification 1 event similar to this event [EventID] Individual Similarity Notification

[0391] The following is a sample individual similarity notification:

[0392] Example Additional Information There is an 80% similarity between these events, suggesting a possible correlation. Similar events and data reports

[0393] Fig. 22 A more detailed example of similarity analysis according to an exemplary embodiment is shown. Here, SimilarityAgent includes a trained classifier that compares the details of a new event with the details of a past event (shown in the figure as "List Test Events" obtained from https: / / data.louisvilleky.gov / ) and identifies related events that are considered by the trained classifier to have at least 80% similarity.

[0394] An EventClassificationAgent may be included to define event agents, types, priorities, or urgency based on event descriptions and other available data. Here, an automatic classifier is trained using historical data (e.g., a pre-classified set of events). The EventClassificationAgent may use feedback to learn from mistakes and improve classification over time. Without limitation, some exemplary applications supported by the EventClassificationAgent include: speeding up the call answering process, selecting agents in multi-agent sites, defining the urgency of incoming messages, detecting the sentiment of incoming messages, and automatic event creation (pre-creation).

[0395] Fig.23 The architecture of EventClassificationAgent according to an exemplary embodiment is shown.

[0396] This exemplary embodiment uses the following Bayesian classifier (conditional probability):

[0397] P(A|B)=P(B|A)P(A) / P(B) – Bayes’ Theorem

[0398] More specifically, event classification is based on the following:

[0399] Given a text (T), the probability of a class (C)

[0400] Given a word (Wi) in a text (T), the probability of a class (C)

[0401] P(C|T)=P(T|C)P(T) / P(C)

[0402] P(C|T)=P(W1|C)*P(W2|C)*...P(T) / P(C)

[0403] Training = Estimate P(Wi|C) = Create bag of words for each class.

[0404] Testing = Compute P(C|T) for each class and then sort the results by probability.

[0405] Continuous learning = reviewing bag of words and statistics.

[0406] Fig.24 A detailed example of event classification is shown.

[0407] Fig.25 Another example of event classification is shown.

[0408] EventClassificationAgent may include speech-to-text integration for automatically classifying events received via speech-based systems. Fig.26 An architecture for speech-to-text integration according to an exemplary embodiment is shown.

[0409] A ClusteringAnalysisAgent may be included to identify clusters of events. Without limitation, some exemplary applications include: optimal positioning of service centers (e.g., helping place service centers to improve response time; can be applied to patrol cars, rescue and emergency vehicles; minimizing the distance between events and service centers), finding patterns among events (e.g., finding areas in the data where event density has given characteristics; identifying and assessing the spread of events on a map, locating areas where a class of events is more prevalent), and monitoring endemic phenomena (e.g., diseases like dengue and Zika are often associated with the presence of stagnant water, and continued monitoring of the formation of event clusters can be an early indicator of such endemic diseases).

[0410] Fig. 27 An example is shown of how the ClusteringAnalysisAgent can help identify the best location to place a trailer based on the history of towing events.

[0411] Using similar methods, similar techniques can be applied to plan EMS, police teams, utilities, fire departments, maintenance teams, and other ICAD-like resources.

[0412] A RouteOptimizationAgent may be included to analyze a given set of events and propose a route that visits the maximum number of events. The agent may continuously optimize the route and suggest the best allocation for a particular group. The dispatcher may accept the proposal and manually assign the event to the group. In an unrestricted situation, route optimization may be used to minimize the route by distance or time (e.g., to save travel time), maximize the number of events that fit in a shift (e.g., 6 hours), start and end at a given location (e.g., a base station), and schedule events according to event type to take an average execution time. Route optimization may take into account the time to move from one event to another (such as from a traffic perspective, if available).

[0413] To gain insight into the route optimization problem, assume that there are 30 tasks to do, but only 10 can be completed. The question then is how to select the best 10 tasks to complete in a selected order. Number of possibilities = 30 x 29 x 28 x 27 x 26 x 25 x 24 x 23 x 22 x 21 = 109 billion alternatives. Even if it took 0.0001 seconds to evaluate each possibility, selecting the best set of 10 of them would take 345 years. An alternative approach might be to always select the closest tasks (e.g., you could just select the first 10), but this might not result in completing the best or most important 10. Therefore, the exemplary embodiment applies a genetic algorithm such as discussed herein to select the best route. In the unrestricted case, some algorithmic considerations include: maximizing the importance of an event or its associated benefit; maximizing the number of executed events if all events are of equal importance; minimizing route transportation costs; minimizing resource costs by using the cheapest resources; scheduling the execution of events to end at or near a given location; limiting the amount of time to execute events to a given shift; skipping events for which no resources are available; considering traffic and other information that affects routing; considering fees and service level agreements (SLAs); and modeling group and event type restrictions.

[0414] Fig.28 An example graphical user interface screen showing an optimized route.

[0415] A TimeEstimationAgent may be included to provide time estimates such as estimated time of arrival (ETA) and time to complete an event. Time estimates may be based on specialized regression analysis (e.g., DNN) based on factors such as, but not limited to, time of day, event type, group type, day of the week, location information (e.g., departure point and / or destination), and other available / relevant information (e.g., weather, traffic, etc.), such as Fig.29 shown.

[0416] Historical events can be used to train a regression service, which can be continuously updated to improve time estimates. The scheduler is informed of better time estimates.

[0417] AreaCoverageAgent can be included to provide location and relocation suggestions for locating field teams, such as responding to future events quickly based on such matters as the number of available teams, the maximum time / radius to the event, and the expected coverage redundancy. Future events can be predicted based on trends, historical data, and external related information. Then, the location where the team is most likely to be needed can be predicted. Here, machine learning is used to help find the best and near-optimal solutions to maximize coverage and redundancy and reduce response time. AreaCoverageAgent can use cluster analysis, route optimization analysis, and time estimation analysis as part of the location / relocation determination. The output can include station locations and the distribution of teams between station locations.

[0418] Fig.30 An example graphical user interface screen showing the staging locations of the teams is shown. The problem can be modeled as a maximum coverage location problem, e.g., given a facility associated with multiple teams and a radius R covered by the facility, the number of teams covering each event / demand is maximized and the number of teams on the facility is minimized.

[0419] Fig.31 An example coverage criteria definition is shown. In this example, facility F1 covers requirements d3, d4, d6 (3 requirements) with one group, while facility F2 covers requirements d1, d2, d3, d4, d5, d7 with two groups. In this example, the quality of service is different, for example, events d3 and d4 are covered by 3 groups; events d1, d2, d5 and d7 are covered by 2 groups; event d6 is covered by 1 group.

[0420] Demands can be represented by discrete points in space. Facilities are located at selected demand points. Each facility can have one or more groups. Each demand is covered by at least one group (unless there are not enough resources). These groups are shared by all demands within the arrival radius of the facility. Agents can consider linear distance (Euler) or Manhattan distance. Reaching a demand means reaching the center of the demand (not its radius which is used to simplify the model). This approach tends to place more groups where there are more events and limits the number of redundant groups for a demand as a quality of service indicator. The number and location of facilities are the output of the algorithm. During positioning, the algorithm can include / exclude new facilities. During relocation, the number and location of facilities are usually fixed. Each time relocation is performed, some improvements to the distribution of groups can be proposed.

[0421] In an exemplary embodiment, the inputs to the algorithm include a vector (X, Y, event) of demand = N locations, a radius R, a desired group for each demand (e.g., maximum U = 3), and the number of groups P. The location output of the algorithm may include a vector of facilities that maximizes demand coverage, the number of groups per facility, and a coverage list for the demand. The relocation output of the algorithm may include a relocation of the groups that maintains maximum demand coverage, and a coverage list for the demand.

[0422] Fig.32 An example Location Redundancy Details screen is shown.

[0423] Fig.33 An example of area coverage relocation is shown.

[0424] Other contemplated intelligent agents include: a safety agent that analyzes data to identify safety issues or improvements (e.g., when or under what conditions police officers are most likely to be injured); a well-being agent that analyzes personnel data (e.g., work hours, rest time, physiological parameters, etc.) to identify fatigue or stress; a workload agent that analyzes data to identify workload issues or improvements (including plans); a team recommendation agent that analyzes data to recommend the most appropriate team for a given incident; a natural language and social stream analysis agent that generates context-based notifications or provides input to other agents; a predictive policing agent that analyzes data to identify potential criminal activity and provides recommendations for proactive and reactive responses; and various automated agents, such as agents for making automated assignment and dispatch decisions.

[0425] Figure 34 to Figure 38 Some exemplary graphical user interface screens are shown according to an exemplary CAD system implementation. Fig.34 An example CAD system screen is shown showing various types of notifications overlaid on a map. Fig.35 Some example notifications are shown in more detail. Here, the example notifications include notifications from SimilarityAgent, PatternAgent, and KeywordAgent. The user can click on a notification to view the details of the notification. Fig.36 Example similarity notification details are shown indicating a similarity threshold determination between events (ie, 81.6% similarity) and the first few words that matched the similarity (ie, PRIVATE, PROPERTY). Fig.37 Example pattern notification details are shown indicating that a pattern was found based on the phone number and providing event details and suggested actions. Fig.38Example keyword notification details are shown indicating keywords identified in an event and providing event details and suggested actions.

[0426] The following table shows the types of algorithms used in various intelligent agents according to one exemplary embodiment:

[0427]

[0428] Figures 41 to 51 A class model of SmartAdvisor and various intelligent agents according to an exemplary embodiment is shown.

[0429] Miscellaneous

[0430] It should be noted that the headings above are used for convenience and should not be construed as limiting the invention in any way.

[0431] Various embodiments of the present invention may be implemented, at least in part, in any conventional computer programming language. For example, some embodiments may be implemented in a procedural programming language (e.g., "C") or in an object-oriented programming language (e.g., "C++"). Other embodiments of the present invention may be implemented as pre-configured independent hardware elements and / or pre-programmed hardware elements (e.g., application specific integrated circuits, FPGAs, and digital signal processors), or other related components.

[0432] In alternative embodiments, the disclosed apparatus and methods (e.g., see the various flowcharts above) can be implemented as a computer program product for use with a computer system. Such an implementation may include a series of computer instructions fixed on a tangible, non-transitory medium such as a computer-readable medium (e.g., a floppy disk, CD-ROM, ROM, or fixed disk). The series of computer instructions may specifically implement all or part of the functionality previously described herein with reference to the system.

[0433] It will be clear to those skilled in the art that such computer instructions may be written in many programming languages ​​for use with many computer architectures or operating systems. Moreover, such instructions may be stored in any memory device (such as semiconductor, magnetic, optical or other memory device) and may be transmitted using any communication technology such as optical, infrared, microwave or other transmission technology.

[0434] In other ways, such a computer program product can be distributed as a removable medium with an accompanying printed or electronic document (e.g., shrink-wrapped software), preloaded with a computer system (e.g., preloaded on a system ROM or fixed disk), or distributed from a server or electronic bulletin board over a network (e.g., the Internet or the World Wide Web). In fact, some embodiments can be implemented in a software as a service model ("SAAS") or a cloud computing model. Of course, some embodiments of the present invention can be implemented as a combination of software (e.g., a computer program product) and hardware. Other embodiments of the present invention are implemented entirely in hardware or entirely in software.

[0435] The computer program logic that implements all or part of the functionality previously described herein may be executed on a single processor at different times (e.g., concurrently), or may be executed on multiple processors at the same or different times, and may run under a single operating system process / thread or under different operating system processes / threads. Thus, the term "computer process" generally refers to the execution of a set of computer program instructions, regardless of whether different computer processes are executed on the same or different processors, and regardless of whether different computer processes are running under the same operating system process / thread or different operating system processes / threads.

[0436] It is important to note that embodiments of the present invention may employ conventional components, such as conventional computers (e.g., off-the-shelf PCs, mainframes, microprocessors), conventional programmable logic devices (e.g., off-the-shelf FPGAs or PLDs), conventional hardware components (e.g., off-the-shelf ASICs or discrete hardware components), and conventional AI / ML algorithms, which, when programmed or configured to perform the non-conventional methods described herein, produce non-conventional devices or systems. Thus, the inventions described herein are non-conventional because, even if embodiments are implemented using conventional components, the resulting devices and systems (e.g., a CAD system with intelligent agents and intelligent notification subsystems) are necessarily non-conventional because, absent special programming or configuration, the conventional components are unable to inherently perform the described non-conventional functions.

[0437] The activities described and claimed herein provide technical solutions to problems that arise directly in the art. These solutions are generally not well understood, routine, or conventional, and nevertheless provide practical applications that can transform and improve computers and computer routing systems.

[0438] Although various creative embodiments have been described and illustrated herein, it will be readily foreseeable by those of ordinary skill in the art that a variety of other means and / or structures for performing functions and / or obtaining results and / or one or more of the advantages described herein, and each of such changes and / or modifications is considered to be within the scope of the creative embodiments described herein. More generally, it will be readily appreciated by those skilled in the art that all parameters, dimensions, materials, and configurations described herein are meant to be exemplary, and that actual parameters, dimensions, materials, and / or configurations will depend on one or more specific applications in which the teachings of the present invention are used. Those skilled in the art will recognize or be able to determine many equivalents of the specific creative embodiments described herein using only routine experiments. Therefore, it should be understood that the aforementioned embodiments are presented only as examples, and within the scope of the appended claims and their equivalents, the creative embodiments may be practiced differently from as specifically described and claimed. The creative embodiments of the present disclosure are directed to each individual feature, system, article, material, kit, and / or method described herein. In addition, any combination of two or more such features, systems, articles, materials, kits and / or methods (to the extent such features, systems, articles, materials, kits and / or methods are not mutually inconsistent) is included within the inventive scope of the present disclosure.

[0439] Various inventive concepts may be embodied as one or more methods for which examples have been provided. The actions performed as part of the method may be arranged in any suitable manner. Therefore, even if shown as sequential actions in an illustrative embodiment, embodiments may be constructed in which the actions are performed in a different order than illustrated, which may include performing some actions simultaneously.

[0440] As defined and used herein, all definitions should be understood to control over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.

[0441] As used herein in the specification and claims, the indefinite article should be understood to mean "at least one", unless explicitly indicated to the contrary.

[0442] As used herein in the specification and claims, the phrase "and / or" should be understood to mean "either or both" of the elements so combined, i.e., elements that are present connectedly in some cases and separately in other cases. Multiple elements listed with "and / or" should be interpreted in the same manner, i.e., "one or more" of the elements so combined. In addition to the elements explicitly identified by the "and / or" clause, other elements may optionally be present, whether related or unrelated to those specifically identified. Thus, as a non-limiting example, a reference to "A and / or B", when used in conjunction with open language such as "comprising": in one embodiment, may refer to only A (optionally including elements other than B); in another embodiment, to only B (optionally including elements other than A); in yet another embodiment, to both A and B (optionally including other elements); etc.

[0443] As used herein in the specification and claims, "or" should be understood to have the same meaning as "and / or" defined above. For example, when separating items in a list, "or" or "and / or" should be interpreted as inclusive, that is, including at least one, but also including more than one element in a plurality of elements or element lists, and optionally including other unlisted items. Only explicitly indicating the opposite terms (such as "only one of them" or "just one of them", or when used in the claims, "consisting of...") refers to including exactly one element in a plurality of elements or element lists. In general, the term "or" used herein, when before exclusive terms (such as "any one", "one of them", "only one of them" or "just one of them"), should only be interpreted as indicating exclusive alternatives (that is, "one or the other, but not both"). "Substantially consisting of...", when used in the claims, should have its ordinary meaning used in the field of patent law.

[0444] As used herein in the specification and claims, with respect to a list of one or more elements, the phrase "at least one" should be understood to refer to at least one element selected from any one or more elements in the list of elements, but does not necessarily include at least one of each element explicitly listed in the list of elements, and does not exclude any combination of elements in the list of elements. This definition also allows that in addition to the elements explicitly identified in the list of elements referred to by the phrase "at least one", other elements may optionally be present, whether related or unrelated to those elements specifically identified. Thus, as a non-limiting example, "at least one of A and B" (or, equivalently, "at least one of A or B", or equivalently, "at least one of A and / or B") may: in one embodiment, refer to at least one (optionally including more than one) A, and there is no B (and optionally including elements other than B); in another embodiment, refer to at least one (optionally including more than one) B, and there is no A (and optionally including elements other than A); in yet another embodiment, refer to at least one (optionally including more than one) A and at least one (optionally including more than one) B (and optionally including other elements); and so on.

[0445] In the claims and in the description above, all transitional phrases such as "comprising," "including," "carrying," "having," "containing," "involving," "holding," "consisting of," and the like shall be construed as open-ended, i.e., meaning including, but not limited to, including. Only the transitional phrases "consisting of" and "consisting essentially of" shall be closed or semi-closed transitional phrases, respectively, as set forth in Section 2111.03 of the U.S. Patent Office Manual of Patent Examining Procedures.

[0446] Although the above discussion discloses various exemplary embodiments of the present invention, it should be understood that various modifications that will achieve some advantages of the present invention may be made by those skilled in the art without departing from the true scope of the present invention. Unless the context otherwise requires, any reference to the "invention" is intended to refer to exemplary embodiments of the present invention and should not be interpreted as referring to all embodiments of the present invention. The described embodiments are to be considered in all respects merely as illustrative and not restrictive.

Claims

1. A computer-aided scheduling CAD system, the CAD system comprising: A CAD database storing CAD data; as well as at least one server including a tangible, non-transitory computer-readable medium storing a proxy host subsystem and a notification subsystem, wherein: The agent host subsystem is configured to communicate with a plurality of intelligent agents, each intelligent agent being configured to autonomously perform different scheduling-related analyses of the CAD data and generate scheduling-related notifications based on such analyses without being queried; and The notification subsystem is configured to implement a machine learning filter that uses characteristics of previous notifications provided to a given user and user feedback associated with such previous notifications to classify notifications and determine which types of notifications the given user wants to receive and which types of notifications the given user does not want to receive, and selectively presents future notifications to the given user based at least in part on the determination of which types of notifications the given user wants to receive and which types of notifications the given user does not want to receive, wherein the notification system is configured to present the desired notifications along with multiple unwanted notifications to the given user via a notification interface, enabling the given user to provide additional feedback on such unwanted notifications to indicate whether the given user wants to receive such unwanted notifications in the future, and wherein the machine learning filter is updated based on such additional feedback to selectively present future notifications.

2. The system according to claim 1, wherein: The notifications presented to the given user provide a mechanism for the given user to provide feedback regarding the value of the notifications to the given user.

3. The system according to claim 2, wherein: The feedback includes like / dislike indications.

4. The system according to claim 2, wherein: The feedback includes a value rating.

5. The system according to claim 1, wherein: The machine learning filter implements a Bayesian classifier to classify notifications.

6. The system according to claim 1, wherein: The notification system also uses predefined user preferences to determine which types of notifications the given user wants to receive.

7. The system according to claim 1, wherein: The notification system is configured to learn notification preferences for a plurality of users respectively.

8. The system according to claim 1, wherein: The number of explicitly unwanted notifications presented to the user can be configured by an administrator.

9. The system according to claim 1, wherein: The notification system is configured to correlate a plurality of related notifications and selectively present all or none of the related notifications.

10. The system according to claim 1, wherein: At least one of the intelligent agents is a machine learning intelligent agent that is retrained based on the user feedback.

11. A computer program product, the computer program product comprising a tangible non-transitory computer readable medium, the tangible non-transitory computer readable medium embodying a computer program, the computer program when executed on a computer implementing a computer process comprising: an agent host subsystem configured to communicate with a plurality of intelligent agents, each intelligent agent configured to autonomously perform different dispatch-related analyses of computer-aided dispatch (CAD) data and generate dispatch-related notifications based on such analyses without being queried; and a notification subsystem configured to implement a machine learning filter that uses characteristics of previous notifications provided to a given user and user feedback associated with such previous notifications to classify notifications and determine which types of notifications the given user wants to receive and which types of notifications the given user does not want to receive, and selectively present future notifications to the given user based at least in part on the determination of which types of notifications the given user wants to receive and which types of notifications the given user does not want to receive, wherein The notification system is configured to present desired notifications along with multiple unwanted notifications to the given user via a notification interface, enabling the given user to provide additional feedback on such unwanted notifications to indicate whether the given user wants to receive such unwanted notifications in the future, and wherein the machine learning filter is updated based on such additional feedback to selectively present future notifications.

12. The computer program product according to claim 11, wherein: The notifications presented to the given user provide a mechanism for the given user to provide feedback regarding the value of the notifications to the given user.

13. The computer program product of claim 12, wherein: The feedback includes like / dislike indications.

14. The computer program product of claim 12, wherein: The feedback includes a value rating.

15. The computer program product of claim 11, wherein: The machine learning filter implements a Bayesian classifier to classify notifications.

16. The computer program product of claim 11, wherein: The notification system also uses predefined user preferences to determine which types of notifications the given user wants to receive.

17. The computer program product of claim 11, wherein: The notification system is configured to learn notification preferences for a plurality of users respectively.

18. The computer program product of claim 11, wherein: The number of explicitly unwanted notifications presented to the user can be configured by an administrator.

19. The computer program product of claim 11, wherein: The notification system is configured to correlate a plurality of related notifications and selectively present all or none of the related notifications.

20. The computer program product of claim 11, wherein: At least one of the intelligent agents is a machine learning intelligent agent that is retrained based on the user feedback.

Citation Information

Patent Citations

  • Computer-Aided Dispatch Systems and Methods Utilizing Biometrics to Assess Responder Condition and Suitability

    US20180174430A1

  • Apparatus and system for harvesting magnetic energy

    US20190036368A1