本文用gganimate包展示模拟疫情数据
本文篇幅较长,分为以下几个部分:
- 前言
- 效果展示
- 小结
- 附录:代码
前言
前文《R语言模拟疫情传播-RVirusBroadcast》已经介绍了一种用R语言模拟疫情传播的方法,不过当时为了实时模拟数据(没有事先准备好的数据集可用)的动态展示效果,笔者在RVirusBroadcast中使用的是R中的基础作图功能,在本文中我们将用gganimate包来生成动态图。
gganimate包是一个可以在ggplot作图基础上生成动态图的工具,像ggplot作图一样,使用gganimate包要先把数据集准备好。由于gganimate的运算量较大,所以只模拟了30天的数据。
效果展示
请注意下面的gif动图都是模拟生成的,纯属虚构,不具有现实意义,仅供电脑模拟实验所用。
当人们随意移动时
上面两张图是gganimate包生成的gif动图,利用magick包还可以将两张动图组合起来:
当人们控制自己的移动时
小结
- 与RVirusBroadcast程序用基础画图功能来动态展示数据不同的是,gganimate包是在ggplot作图的基础上生成动图的,它需要事先准备好数据集。
- 利用gganimate + magick可以组合动态图。
- gganimate还可以做很多有意思的事情,有兴趣的朋友可以到官网上去看一看。
参考
[1] “计算机仿真程序告诉你为什么现在还没到出门的时候” 视频地址:
https://www.bilibili.com/video/av86478875?spm_id_from=333.788.b_765f64657363.1
[2] VirusBroadcast (Java)程序源码:
https://github.com/KikiLetGo/VirusBroadcast
[3] gganimate包项目链接:
https://github.com/thomasp85/gganimate
附录:代码
###name:RVirusBroadcast-via-gganimate
###author:hxj7
###version:202002010
###note:本程序是"VirusBroadcast (in Java)"的R-动画版本,只能处理事先准备好的数据集
### VirusBroadcast (in Java) 项目链接:
### https://github.com/KikiLetGo/VirusBroadcast/tree/master/src
library(tibble)
library(dplyr)
library(ggplot2)
library(tidyr)
library(magick)
library(gganimate)
########## 模拟参数 ##########
ORIGINAL_COUNT <- 50 # 初始感染数量
BROAD_RATE <- 0.8 # 传播率
SHADOW_TIME <- 140 # 潜伏时间,14天为140
HOSPITAL_RECEIVE_TIME <- 10 # 医院收治响应时间
BED_COUNT <- 1000 # 医院床位
MOVE_WISH_MU <- -0.99 # 流动意向平均值,建议调整范围:[-0.99,0.99];
# -0.99 人群流动最慢速率,甚至完全控制疫情传播;
# 0.99为人群流动最快速率, 可导致全城感染
CITY_PERSON_SIZE <- 5000 # 城市总人口数量
FATALITY_RATE <- 0.02 # 病死率,根据2月6日数据估算(病死数/确诊数)为0.02
SHADOW_TIME_SIGMA <- 25 # 潜伏时间方差
CURED_TIME <- 50 # 治愈时间均值,从入院开始计时
CURED_SIGMA <- 10 # 治愈时间标准差
DIE_TIME <- 300 # 死亡时间均值,30天,从发病(确诊)时开始计时
DIE_SIGMA <- 50 # 死亡时间标准差
CITY_WIDTH <- 700 # 城市大小即窗口边界,限制不允许出城
CITY_HEIGHT <- 800
MAX_TRY <- 300 # 最大模拟次数,300代表30天
########## 生成人群点,用不同颜色代表不同健康状态。 ##########
# 用正态分布刻画人群点的分布
CITY_CENTERX <- 400 # x轴的mu值
CITY_CENTERY <- 400
PERSON_DIST_X_SIGMA <- 100 # x轴的sigma值
PERSON_DIST_Y_SIGMA <- 100
# 市民状态应该需要细分,虽然有的状态暂未纳入模拟,但是细分状态应该保留
STATE_NORMAL <- 0 # 正常人,未感染的健康人
STATE_SUSPECTED <- STATE_NORMAL + 1 # 有暴露感染风险
STATE_SHADOW <- STATE_SUSPECTED + 1 # 潜伏期
STATE_CONFIRMED <- STATE_SHADOW + 1 # 发病且已确诊为感染病人
STATE_FREEZE <- STATE_CONFIRMED + 1 # 隔离治疗,禁止位移
STATE_DEATH <- STATE_FREEZE + 1 # 病死者
STATE_CURED <- STATE_DEATH + 1 # 治愈数量用于计算治愈出院后归还床位数量,该状态是否存续待定
worldtime <- 0
NTRY_PER_DAY <- 10 # 一天模拟几次
getday <- function(t) (t - 1) %/% NTRY_PER_DAY + 1
# 生成人群数据
format_coord <- function(coord, boundary) {
if (coord < 0) return(runif(1, 0, 10))
else if (coord > boundary) return(runif(1, boundary - 10, boundary))
else return(coord)
}
set.seed(123)
people <- tibble(
id = 1:CITY_PERSON_SIZE,
x = sapply(rnorm(CITY_PERSON_SIZE, CITY_CENTERX, PERSON_DIST_X_SIGMA),
format_coord, boundary = CITY_WIDTH), # (x, y) 为人群点坐标
y = sapply(rnorm(CITY_PERSON_SIZE, CITY_CENTERY, PERSON_DIST_Y_SIGMA),
format_coord, boundary = CITY_HEIGHT),
state = STATE_NORMAL, # 健康状态
infected_time = 0, # 感染时刻
confirmed_time = 0, # 确诊时刻
freeze_time = 0, # 隔离时刻
cured_moment = 0, # 痊愈时刻,为0代表不确定
die_moment = 0 # 死亡时刻,为0代表未确定,-1代表不会病死
) %>%
mutate(tx = rnorm(CITY_PERSON_SIZE, x, PERSON_DIST_X_SIGMA), # target x
ty = rnorm(CITY_PERSON_SIZE, y, PERSON_DIST_Y_SIGMA),
has_target = T, is_arrived = F)
# 随机选择初始感染者
peop_id <- sample(people$id, ORIGINAL_COUNT)
people$state[peop_id] <- STATE_SHADOW
people$infected_time[peop_id] <- worldtime
people$confirmed_time[peop_id] <- worldtime +
max(rnorm(length(peop_id), SHADOW_TIME / 2, SHADOW_TIME_SIGMA), 0)
########## 生成床位点 ##########
HOSPITAL_X <- 720 # 第一张床位的x坐标
HOSPITAL_Y <- 80 # 第一张床位的y坐标
NBED_PER_COLUMN <- 100 # 医院每一列有多少张床位
BED_ROW_SPACE <- 6 # 一行中床位的间距
BED_COLUMN_SPACE <- 6 # 一列中床位的间距
bed_ncolumn <- ceiling(BED_COUNT / NBED_PER_COLUMN)
hosp_beds <- tibble(id = 1, x = 0, y = 0, is_empty = T, state = STATE_NORMAL) %>%
slice(-1)
if (BED_COUNT > 0) {
hosp_beds <- tibble(
id = 1:BED_COUNT,
x = HOSPITAL_X + rep(((1:bed_ncolumn) - 1) * BED_ROW_SPACE,
each = NBED_PER_COLUMN)[1:BED_COUNT],
y = HOSPITAL_Y + 10 - BED_COLUMN_SPACE +
rep((1:NBED_PER_COLUMN) * BED_COLUMN_SPACE, bed_ncolumn)[1:BED_COUNT],
is_empty = T,
person_id = 0 # 占用床位的患者的序号,床位为空时为0
)
}
########## 准备数据 ##########
npeople_total <- CITY_PERSON_SIZE
npeople_shadow <- ORIGINAL_COUNT
npeople_confirmed <- npeople_freeze <- npeople_cured <- npeople_death <- 0
nbed_need <- 0
peop_point <- people %>% select(x, y, state) %>% mutate(worldtime = worldtime)
bed_point <- hosp_beds %>% select(x, y, is_empty) %>% mutate(worldtime = worldtime)
peop_stat <- tibble(npeople_death = npeople_death, npeople_cured = npeople_cured,
nbed_need = nbed_need, npeople_freeze = npeople_freeze,
npeople_confirmed = npeople_confirmed,
npeople_shadow = npeople_shadow, worldtime = worldtime)
########## 更新人群数据 ##########
# 市民流动意愿以及移动位置参数
MOVE_WISH_SIGMA <- 1
MOVE_DIST_SIGMA <- 50
SAFE_DIST <- 2 # 安全距离
worldtime <- worldtime + 1
get_min_dist <- function(person, peop) { # 一个人和一群人之间的最小距离
min(sqrt((person["x"] - peop$x) ^ 2 + (person["y"] - peop$y) ^ 2))
}
for (i in 1:MAX_TRY) {
# 如果已经隔离或者死亡了,就不需要处理了
#
# 处理已经确诊的感染者(即患者)
peop_id <- people$id[people$state == STATE_CONFIRMED &
people$die_moment == 0]
if ((npeop <- length(peop_id)) > 0) {
people$die_moment[peop_id] <- ifelse(
runif(npeop, 0, 1) < FATALITY_RATE, # 用均匀分布模拟确诊患者是否会死亡
people$confirmed_time + max(rnorm(npeop, DIE_TIME, DIE_SIGMA), 0), # 发病后确定死亡时刻
-1 # 逃过了死神的魔爪
)
}
# 如果患者已经确诊,且(世界时刻-确诊时刻)大于医院响应时间,
# 即医院准备好病床了,可以抬走了
peop_id <- people$id[people$state == STATE_CONFIRMED &
worldtime - people$confirmed_time >= HOSPITAL_RECEIVE_TIME]
if ((npeop <- length(peop_id)) > 0) {
if ((nbed_empty <- sum(hosp_beds$is_empty)) > 0) { # 有空余床位
nbed_use <- min(npeop, nbed_empty)
bed_id <- hosp_beds$id[hosp_beds$is_empty][1:nbed_use]
# 更新患者信息
peop_id2 <- sample(peop_id, nbed_use) # 这里是随机选择,理论上应该按症状轻重
people$x[peop_id2] <- hosp_beds$x[bed_id]
people$y[peop_id2] <- hosp_beds$y[bed_id]
people$state[peop_id2] <- STATE_FREEZE
people$freeze_time[peop_id2] <- worldtime
# 更新床位信息
hosp_beds$is_empty[bed_id] <- F
hosp_beds$person_id[bed_id] <- peop_id2
}
}
# TODO 需要确定一个变量用于治愈时长。
# 为了说明问题,暂时用一个正态分布模拟治愈时长并且假定治愈的人不会再被感染
peop_id <- people$id[people$state == STATE_FREEZE &
people$cured_moment == 0]
if ((npeop <- length(peop_id)) > 0) { # 正态分布模拟治愈时间
people$cured_moment[peop_id] <- people$freeze_time[peop_id] +
max(rnorm(npeop, CURED_TIME, CURED_SIGMA), 0)
}
peop_id <- people$id[people$state == STATE_FREEZE & people$cured_moment > 0 &
worldtime >= people$cured_moment]
if ((npeop <- length(peop_id)) > 0) { # 归还床位
people$state[peop_id] <- STATE_CURED
hosp_beds$is_empty[! hosp_beds$is_empty & hosp_beds$person_id %in% peop_id] <- T
people$x[peop_id] <- sapply(rnorm(npeop, CITY_CENTERX, PERSON_DIST_X_SIGMA),
format_coord, boundary = CITY_WIDTH) # (x, y) 为人群点坐标
people$y[peop_id] <- sapply(rnorm(npeop, CITY_CENTERY, PERSON_DIST_Y_SIGMA),
format_coord, boundary = CITY_HEIGHT)
people$tx[peop_id] <- rnorm(npeop, people$x[peop_id], PERSON_DIST_X_SIGMA)
people$ty[peop_id] <- rnorm(npeop, people$y[peop_id], PERSON_DIST_Y_SIGMA)
people$has_target[peop_id] <- T
people$is_arrived[peop_id] <- F
}
# 处理病死者
peop_id <- people$id[people$state %in% c(STATE_CONFIRMED, STATE_FREEZE) &
worldtime >= people$die_moment & people$die_moment > 0]
if (length(peop_id) > 0) { # 归还床位
people$state[peop_id] <- STATE_DEATH
hosp_beds$is_empty[! hosp_beds$is_empty & hosp_beds$person_id %in% peop_id] <- T
}
# 处理发病的潜伏期感染者
peop_id <- people$id[people$state == STATE_SHADOW &
worldtime >= people$confirmed_time]
if ((npeop <- length(peop_id)) > 0) {
people$state[peop_id] <- STATE_CONFIRMED # 潜伏者发病
}
# 处理未隔离者的移动问题
peop_id <- people$id[
! people$state %in% c(STATE_FREEZE, STATE_DEATH) &
rnorm(CITY_PERSON_SIZE, MOVE_WISH_MU, MOVE_WISH_SIGMA) > 0] # 流动意愿
if ((npeop <- length(peop_id)) > 0) { # 正态分布模拟要移动到的目标点
pp_id <- peop_id[! people$has_target[peop_id] | people$is_arrived[peop_id]]
if ((npp <- length(pp_id)) > 0) {
people$tx[pp_id] <- rnorm(npp, people$tx[pp_id], PERSON_DIST_X_SIGMA)
people$ty[pp_id] <- rnorm(npp, people$ty[pp_id], PERSON_DIST_Y_SIGMA)
people$has_target[pp_id] <- T
people$is_arrived[pp_id] <- F
}
# 计算运动位移
dx <- people$tx[peop_id] - people$x[peop_id]
dy <- people$ty[peop_id] - people$y[peop_id]
move_dist <- sqrt(dx ^ 2 + dy ^ 2)
people$is_arrived[peop_id][move_dist < 1] <- T # 判断是否到达目标点
pp_id <- peop_id[move_dist >= 1]
if ((npp <- length(pp_id)) > 0) {
udx <- sign(dx[move_dist >= 1]) # x轴运动方向
udy <- sign(dy[move_dist >= 1])
# 是否到了边界
pid_x <- (1:npp)[people$x[pp_id] + udx < 0 | people$x[pp_id] + udx > CITY_WIDTH]
pid_y <- (1:npp)[people$y[pp_id] + udy < 0 | people$y[pp_id] + udy > CITY_HEIGHT]
# 更新到了边界的点的信息
people$x[pp_id[pid_x]] <- people$x[pp_id[pid_x]] - udx[pid_x]
people$y[pp_id[pid_y]] <- people$y[pp_id[pid_y]] - udy[pid_y]
people$has_target[unique(c(pp_id[pid_x], pp_id[pid_y]))] <- F
# 更新没有到边界的点的信息
people$x[pp_id[! pp_id %in% pid_x]] <- people$x[pp_id[! pp_id %in% pid_x]] +
udx[! pp_id %in% pid_x]
people$y[pp_id[! pp_id %in% pid_y]] <- people$y[pp_id[! pp_id %in% pid_y]] +
udy[! pp_id %in% pid_y]
}
}
# 处理健康人被感染的问题
# 通过一个随机幸运值和安全距离决定感染其他人
normal_peop_id <- people$id[people$state == STATE_NORMAL]
other_peop_id <- people$id[! people$state %in% c(STATE_NORMAL, STATE_CURED)]
if (length(normal_peop_id) > 0) {
normal_other_dist <- apply(people[normal_peop_id, ], 1, get_min_dist,
peop = people[other_peop_id, ])
normal2other_id <- normal_peop_id[normal_other_dist < SAFE_DIST &
runif(length(normal_peop_id), 0, 1) < BROAD_RATE]
if ((n2other <- length(normal2other_id)) > 0) {
people$state[normal2other_id] <- STATE_SHADOW
people$infected_time[normal2other_id] <- worldtime
people$confirmed_time[normal2other_id] <- worldtime +
max(rnorm(n2other, SHADOW_TIME / 2, SHADOW_TIME_SIGMA), 0)
}
}
# 更新后的数据
npeople_confirmed <- sum(people$state >= STATE_CONFIRMED)
npeople_death <- sum(people$state == STATE_DEATH)
npeople_freeze <- sum(people$state == STATE_FREEZE)
npeople_shadow <- sum(people$state == STATE_SHADOW)
npeople_cured <- sum(people$state == STATE_CURED)
nbed_need <- npeople_confirmed - npeople_cured - npeople_death - BED_COUNT
nbed_need <- ifelse(nbed_need > 0, nbed_need, 0) # 不足病床数
# 更新结果
peop_point <- rbind(peop_point, people %>% select(x, y, state) %>%
mutate(worldtime = worldtime))
bed_point <- rbind(bed_point, hosp_beds %>% select(x, y, is_empty) %>%
mutate(worldtime = worldtime))
peop_stat <- rbind(peop_stat, c(npeople_death, npeople_cured, nbed_need,
npeople_freeze, npeople_confirmed,
npeople_shadow, worldtime))
# 更新世界时间
worldtime <- worldtime + 1
print(i)
}
########## 设置画图参数 ##########
person_color <- data.frame( # 不同健康状态的颜色不同
label = c("健康", "潜伏", "确诊", "隔离", "治愈", "死亡"),
state = c(STATE_NORMAL, STATE_SHADOW, STATE_CONFIRMED, STATE_FREEZE,
STATE_CURED, STATE_DEATH),
color = c(
"lightgreen", # 健康
"#EEEE00", # 潜伏期
"red", # 确诊
"#FFC0CB", # 隔离
"green", # 治愈
"black" # 死亡
), stringsAsFactors = F
)
bed_color <- data.frame(
is_empty = c(T, F),
color = c("#F8F8FF", "#FFC0CB"),
stringsAsFactors = F
)
stat_color <- data.frame(
state = c("npeople_death", "npeople_cured", "nbed_need",
"npeople_freeze", "npeople_confirmed",
"npeople_shadow"),
label = c("死亡", "治愈", "不足\n床位", "隔离", "累计\n确诊", "潜伏"),
color = c("black", "green", "#FFE4E1", "#FFC0CB", "red", "#EEEE00"),
stringsAsFactors = F
)
########## 画出动态图 ##########
# 散点图
peop_point <- peop_point %>%
left_join(person_color, by = "state")
bed_point <- bed_point %>%
left_join(bed_color, by = "is_empty")
p_scatter <- ggplot() +
geom_point(aes(x = x, y = y), data = peop_point, color = peop_point$color) +
geom_point(aes(x = x, y = y), data = bed_point, color = bed_point$color) +
labs(title = "疫情传播模拟第{current_frame}次", x = "", y = "") +
theme(plot.title = element_text(hjust = .5),
axis.text = element_blank(),
axis.ticks = element_blank(),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
panel.background = element_blank()) +
guides(color = "none") +
transition_manual(worldtime)
#p_scatter
#Sys.sleep(10)
# 条形图
peop_stat <- peop_stat %>%
gather(state, value, -worldtime) %>%
left_join(stat_color, by = "state") %>%
arrange(worldtime)
abline_data <- data.frame(
label = c("床位总数", "城市总人口"),
value = c(BED_COUNT, CITY_PERSON_SIZE),
type = c(3, 1),
stringsAsFactors = F
)
abline_type <- abline_data$type
names(abline_type) <- abline_data$label
p_bar <- ggplot() +
geom_bar(aes(x = state, y = value), data = peop_stat, fill = peop_stat$color,
stat = "identity", position = "dodge") +
scale_x_discrete(limits = stat_color$state, labels = stat_color$label,
expand = c(0, 0)) +
scale_y_continuous(expand = c(0, 0), limits = c(0, CITY_PERSON_SIZE * 1.05),
breaks = seq(0, CITY_PERSON_SIZE, length.out = 6)) +
geom_text(aes(x = state, y = value + CITY_PERSON_SIZE / 25, label = value),
data = peop_stat) +
geom_hline(aes(yintercept = value, linetype = label), data = abline_data,
color = "gray") +
scale_linetype_manual(values = abline_type, limits = abline_data$label) +
labs(title = "人群变化模拟第{current_frame}次", x = "", y = "", linetype = "") +
theme_classic() +
theme(plot.title = element_text(hjust = .5),
axis.line.y = element_blank(),
axis.ticks.y = element_blank(),
legend.position = "bottom") +
coord_flip() +
transition_manual(worldtime)
#p_bar
#Sys.sleep(10)
# 组合散点图和条形图
p_scatter_gif <- animate(p_scatter, width = 300, height = 240)
p_bar_gif <- animate(p_bar, width = 240, height = 240)
scatter_mgif <- image_read(p_scatter_gif)
bar_mgif <- image_read(p_bar_gif)
new_gif <- image_append(c(bar_mgif[1], scatter_mgif[1]))
for(i in 2:length(bar_mgif)){
combined <- image_append(c(bar_mgif[i], scatter_mgif[i]))
new_gif <- c(new_gif, combined)
}
new_gif
Sys.sleep(10)