Method, internet of things system and storage medium for determining street cleaning route in smart city

1 . A method for determining a street cleaning route in a smart city implemented based on an Internet of Things system for determining the street cleaning route in the smart city, wherein the Internet of Things system for determining a street cleaning route in a smart city includes a management platform, a sensor network platform, an object platform, a user platform, and a service platform, the management platform includes at least one management sub-platform, and the sensor network platform includes at least one sensor network sub-platform; one of the at least one sensor network sub-platform corresponds to a target area; one of the at least one management sub-platform corresponds to one of the sensor network sub-platforms; street monitoring information of the target area is obtained based on the object platform and transmitted to the management sub-platform corresponding to the sensor network sub-platform based on the sensor network sub-platform corresponding to the target area; the sensor network platform is configured as a communication network and gateway; the object platform is configured as a monitoring device, a cleaning vehicle, and a relevant device of each target area; the user platform is configured as a terminal device, which feedbacks a fallen leaf cleaning route and related information of the target area to a user, the method is executed by the management platform, and the method comprises:

obtaining, based on the object platform, the street monitoring information of the target area through the sensor network platform;

determining, according to the street monitoring information, distribution of fallen leaves on the street, the distribution of fallen leaves including a total amount of fallen leaves and a count of fallen leaf piles, wherein the determining the distribution of fallen leaves on the street includes

determining the distribution of fallen leaves by processing the street monitoring information through an object identification model, the object identification model being a machine learning model or a related algorithm;

wherein the object identification model is a trained convolutional neural network or an object detection algorithm with set parameters; the object identification model is trained by training data labelled with the fallen leaf piles; the training data of the object identification model includes a training sample and a sample label; the training sample is a historical monitoring image containing fallen leaves on ground, the sample label is a fallen leaf pile that is labelled in the historical monitoring image; and the determining the distribution of fallen leaves by processing the street monitoring information through an object identification model includes:

determining the fallen leaves in a monitoring image of the street monitoring information through the object identification model;

determining the fallen leaf piles based on a clustering algorithm; and

determining the distribution of fallen leaves based on the fallen leaves and the fallen leaf piles;

determining, based on the total amount of fallen leaves and the count of fallen leaf piles, cleaning difficulty of each street in the target area; wherein the count of fallen leaf piles is positively correlated with the cleaning difficulty, and the total amount of fallen leaves is positively correlated with the cleaning difficulty;

determining, based on the total amount of fallen leaves and the cleaning difficulty, at least one street to be cleaned from the target area;

determining, based on the at least one street to be cleaned and a position of each cleaning vehicle, at least one continuous action sequence by iterative processing, the continuous action sequence including cleaning actions for each of streets to be cleaned;

wherein one round of the iterative processing includes:

randomly determining a street to be cleaned from at least one street to be cleaned that has not yet been cleaned to determine the corresponding cleaning actions based on a state of the cleaning vehicle after a previous round of the iterative processing;

selecting cleaning actions with a relatively large reward value from the randomly determined cleaning actions for state transition;

determining a state of the cleaning vehicle after a current round of the iterative processing based on a result of the state transition; and

repeating the iterative processing until cleaning tasks of all the streets to be cleaned are completed;

for any continuous action sequence of the at least one continuous action sequence, determining a reward value of each cleaning action of the continuous action sequence by processing the continuous action sequence based on a preset evaluation function, and determining the return value of each continuous action sequence of the at least one continuous action sequence by recording a total reward value of each cleaning action of the continuous action sequence as a return value of the continuous actions, the preset evaluation function being a trained evaluation function during a reinforcement learning;

adopting a continuous action sequence with a largest return value of the at least one continuous action sequence, and determining, according to an order where the cleaning vehicle arrives at each street in the continuous action sequence, the return value of each continuous action sequence, the fallen leaf cleaning route of the target area;

sending the fallen leaf cleaning route to each cleaning vehicle in the object platform through the sensor network sub-platform corresponding to the management sub-platform;

receiving a fallen leaf cleaning route query instruction issued by the user platform through the service platform; and

sending the fallen leaf cleaning route to the user platform through the service platform, including:

storing the fallen leaf cleaning route of each target area on the service platform; and

in response to determining that the fallen leaf cleaning route query instruction is received, sending the fallen leaf cleaning route to the user platform by calling the fallen leaf cleaning route according to a target area where the user is located through the service platform.

2 . The method of claim 1 , wherein the determining a reward value of each cleaning action of the continuous action sequence by processing the continuous action sequence based on a preset evaluation function includes:

for each cleaning action, obtaining a first street feature of the street to be cleaned corresponding to the cleaning action, the first street feature including a current total amount of fallen leaves on the street to be cleaned;

determining a positive reward value of the cleaning action by processing the first street feature based on a reward function; and

determining, according to the positive reward value, the reward value of the cleaning action.

3 . The method of claim 2 , wherein the first street feature further includes ground cleanliness and/or ground dryness.

4 . The method of claim 2 , wherein the determining, according to the positive reward value, the reward value of the cleaning action includes:

obtaining a second street feature of the street to be cleaned, the second street feature including a distance between the street to be cleaned and a street to be cleaned corresponding to a previous cleaning action;

determining a reverse penalty value of the cleaning action by processing the second street feature based on a penalty function; and

determining the reward value of the cleaning action according to the positive reward value and the reverse penalty value.

5 . The method of claim 4 , wherein the second street feature further includes a dispersion degree of fallen leaves and a generation rate of fallen leaves, and the dispersion degree of fallen leaves and the generation rate of fallen leaves are positively correlated with the reverse penalty value.

6 . The method of claim 5 , wherein the generation rate of fallen leaves is determined based on a manner including:

obtaining auxiliary evaluation information, the auxiliary evaluation information including a tree condition of the street to be cleaned and/or a wind condition during a preset period of time, and the tree condition including a tree planting density and a tree age; and

determining, according to the auxiliary evaluation information, the generation rate of fallen leaves.

7 . The method of claim 1 , further comprising:

determining the cleaning difficulty of each street in the target area according to wind strength; wherein an input of a wind speed prediction model is a wind condition before a current moment, an output is a wind condition during a period of time in a future after the current moment, and the wind condition includes the wind strength; the wind speed prediction model is a machine learning model, and the wind speed prediction model is obtained through training.

8 . A non-transitory computer-readable storage medium storing computer instructions, wherein when the computer instructions are executed by a processor, the method for determining the street cleaning route in the smart city of claim 1 is implemented.

9 . An Internet of Things system for determining a street cleaning route in a smart city including a management platform, a sensor network platform, an object platform, a user platform, and a service platform, wherein the management platform includes at least one management sub-platform, and the sensor network platform includes at least one sensor network sub-platform; one of the at least one sensor network sub-platform corresponds to a target area; one of the at least one management sub-platform corresponds to one of the sensor network sub-platforms; street monitoring information of the target area is obtained based on the object platform and transmitted to the management sub-platform corresponding to the sensor network sub-platform based on the sensor network sub-platform corresponding to the target area; the sensor network platform is configured as a communication network and gateway; the object platform is configured as a monitoring device, a cleaning vehicle, and a relevant device of each target area; the user platform is configured as a terminal device, which feedbacks a fallen leaf cleaning route and related information of the target area to a user and the management platform is configured to:

obtain, based on the object platform, the street monitoring information of the target area through the sensor network platform;

determine, according to the street monitoring information, distribution of fallen leaves on the street, the distribution of fallen leaves including a total amount of fallen leaves and a count of fallen leaf piles, wherein to determine the distribution of fallen leaves on the street, the management platform is further configured to:

determine the distribution of fallen leaves by processing the street monitoring information through an object identification model, the object identification model being a machine learning model or a related algorithm;

wherein the object identification model is a trained convolutional neural network or an object detection algorithm with set parameters; the object identification model is trained by training data labelled with the fallen leaf piles; the training data of the object identification model includes a training sample and a sample label; the training sample is a historical monitoring image containing fallen leaves on a ground, the sample label is a fallen leaf pile that is labelled in the historical monitoring image; and the determining the distribution of fallen leaves by processing the street monitoring information through an object identification model includes:

determine the fallen leaves in a monitoring image of the street monitoring information through the object identification model;

determine the fallen leaf piles based on a clustering algorithm; and

determine the distribution of fallen leaves based on the fallen leaves and the fallen leaf piles;

determine, based on the total amount of fallen leaves and the count of fallen leaf piles, cleaning difficulty of each street in the target area; wherein the count of fallen leaf piles is positively correlated with the cleaning difficulty, and the total amount of fallen leaves is positively correlated with the cleaning difficulty;

determine, based on the total amount of fallen leaves and the cleaning difficulty, at least one street to be cleaned from the target area;

determine, based on the at least one street to be cleaned and a position of each cleaning vehicle, at least one continuous action sequence by iterative processing, the continuous action sequence including cleaning actions for each of streets to be cleaned;

wherein one round of the iterative processing includes:

randomly determining a street to be cleaned from at least one street to be cleaned that has not yet been cleaned to determine the corresponding cleaning actions based on a state of the cleaning vehicle after a previous round of the iterative processing;

selecting cleaning actions with a relatively large reward value from the randomly determined cleaning actions for state transition;

determining a state of the cleaning vehicle after a current round of the iterative processing based on a result of the state transition; and

repeating the iterative processing until cleaning tasks of all the streets to be cleaned are completed;

for any continuous action sequence of the at least one continuous action sequence, determine a reward value of each cleaning action of the continuous action sequence by processing the continuous action sequence based on a preset evaluation function, and determine the return value of each continuous action sequence of the at least one continuous action sequence by recording a total reward value of each cleaning action of the continuous action sequence as a return value of the continuous actions, the preset evaluation function being a trained evaluation function during a reinforcement learning; and

adopt a continuous action sequence with a largest return value of the at least one continuous action sequence, and determine, according to an order where the cleaning vehicle arrives at each street in the continuous action sequence the return value of each continuous action sequence, the fallen leaf cleaning route of the target area;

send the fallen leaf cleaning route to each cleaning vehicle in the object platform through the sensor network sub-platform corresponding to the management sub-platform;

receive a fallen leaf cleaning route query instruction issued by the user platform through the service platform; and

send the fallen leaf cleaning route to the user platform through the service platform, wherein to send the fallen leaf cleaning route to the user platform through the service platform, the management platform is further configured to:

store the fallen leaf cleaning route of each target area on the service platform; and

in response to determining that the fallen leaf cleaning route query instruction is received, send the fallen leaf cleaning route to the user platform by calling the fallen leaf cleaning route according to a target area where the user is located through the service platform.

10 . The Internet of Things system of claim 9 , wherein the management platform is further configured to:

for each cleaning action, obtain a first street feature of the street to be cleaned corresponding to the cleaning action, the first street feature including a current total amount of fallen leaves on the street to be cleaned;

determine a positive reward value of the cleaning action by processing the first street feature based on a reward function; and

determine, according to the positive reward value, the reward value of the cleaning action.

11 . The Internet of Things system of claim 10 , wherein the first street feature further includes ground cleanliness and/or ground dryness.

12 . The Internet of Things system of claim 10 , wherein the management platform is further configured to:

obtain a second street feature of the street to be cleaned, the second street feature including a distance between the street to be cleaned and a street to be cleaned corresponding to a previous cleaning action;

determine a reverse penalty value of the cleaning action by processing the second street feature based on a penalty function; and

determine the reward value of the cleaning action according to the positive reward value and the reverse penalty value.

13 . The Internet of Things system of claim 12 , wherein the second street feature further includes a dispersion degree of fallen leaves and a generation rate of fallen leaves, and the dispersion degree of fallen leaves and the generation rate of fallen leaves are positively correlated with the reverse penalty value.

14 . The Internet of Things system of claim 13 , wherein the management platform is further configured to:

obtain auxiliary evaluation information, the auxiliary evaluation information including a tree condition of the street to be cleaned and/or a wind condition during a preset period of time, and the tree condition including a tree planting density and a tree age; and

determine, according to the auxiliary evaluation information, the generation rate of fallen leaves.

15 . The Internet of Things system of claim 9 , wherein the management platform is further configured to:

determine the cleaning difficulty of each street in the target area according to wind strength; wherein an input of a wind speed prediction model is a wind condition before a current moment, an output is a wind condition during a period of time in a future after the current moment, and the wind condition includes the wind strength; the wind speed prediction model is a machine learning model, and the wind speed prediction model is obtained through training.